Jonathan Dancy - Practical Thought - Essays On Reason, Intuition, and Action-Oxford University Press (2021)

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 9/6/2021, SPi
Practical Thought
Practical Thought
Essays on Reason, Intuition, and Action
JONATHAN DANCY
1
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Jonathan Dancy 2021
The moral rights of the author have been asserted
First Edition published in 2021
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2020952993
ISBN 978–0–19–886560–5
DOI: 10.1093/oso/9780198865605.001.0001
Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
Contents
Preface vii
Introduction: No More Answers? 1

1. The Logical Conscience 15
SECTION 1: TOWARDS PARTICULARISM

IN ETHICS AND EPISTEMOLOGY
2. On Moral Properties 21
3. Ethical Particularism and Morally Relevant Properties 37
4. Intuitionism in Meta-Epistemology 51
5. The Role of Imaginary Cases in Ethics 62
6. Externalism for Internalists 75
7. The Particularist’s Progress 91
8. Necessity, Universality, and the A Priori in Ethics 114
SECTION 2: MORAL METAPHYSICS
9. Two Conceptions of Moral Realism 129

10. Contemplating One’s Nagel 145
11. In Defence of Thick Concepts 160
12. McDowell, Williams, and Intuitionism 178
13. Practical Concepts 197
14. Should We Pass the Buck? 213
SECTION 3: ACTION AND REASONS

15. Arguments from Illusion 229
16. Why There Is Really No Such Thing as the Theory of Motivation 247
17. On How to Act—Disjunctively 261
vi 
18. Enticing Reasons 281

19. On Knowing One’s Reason 304
SECTION 4: LEARNING FROM THE INTUITIONISTS

20. Prichard on Duty and Ignorance of Fact 321
21. Moore’s Account of Vindictive Punishment: A Test Case
for Theories of Organic Unities 338
22. Has Anyone Ever Been a Non-Intuitionist? 355
23. More Right than Wrong 375
24. Prichard on Causing a Change 394
Places of First Publication 405

Works Cited 407
Index 413
Preface
This collection consists of papers all of which have been previously published.
I am grateful to Wiley-Blackwell, the Royal Institute of Philosophy and Ridgeway
Publishers for permission to reprint material of which they hold the copyright.
The papers are preceded by an Introduction which is intended as an intellectual
autobiography. I have found the attempt to put order into what was a pretty
random process both difficult and interesting. To the extent to which one can
discern a pattern, it was certainly not one determined in advance. Like most
people, I lurched from topic to topic as the mood struck me—stimulated of course
by invitations and other similar motives. But looking at things in retrospective, a
shape does emerge.
In selecting papers for this collection I had three criteria in mind. The first was
absolute quality (real or supposed). The second was to try to construct as much of
a narrative as possible. The third was to try to show the range of areas within
which I have worked. In the last respect, I was sad not to be able to find room for
two papers which are way outside my normal stamping ground, one on Proust’s
conception of the self (Dancy 1995d) and the other on Hume’s Dialogues
Concerning Natural Religion (Dancy 1995c).
I owe thanks to many, of course. In the two departments in which I taught in
the UK, I had enormous support from David McNaughton, Philip Stratton-Lake,
and Brad Hooker. David and I invented what I once (with great pride) heard
referred to as Keele Realism. Philip was supposedly a Kantian when we hired him
at Keele, but by the time he followed me to Reading he was as interested in Ross as
in Kant. Brad didn’t agree with anything I thought but did everything he could to
help me work it all out anyway.
Outside my own departments, the people with whom I have had the most
sustained philosophical interactions are, or were, Derek Parfit (on whom I wrote a
biographical piece for the British Academy: see Dancy 2020), Joseph Raz, Michael
Smith, and John Broome. To these people I owe a particular debt because of the
many ways in which they have influenced my work directly, and for that peculiar
mix of personal and philosophical friendship that so enriches academic life. But of
course, after fifty years in the profession there are many others whose friendship
and support, both philosophical and otherwise, have been very important to me.
With respect to this collection, I owe thanks to two people in particular for wise
advice: to Peter Momtchiloff, conversations with whom persuaded me that a
collection like this might make sense and that I should not dismiss the idea as
little better than vanity publishing, and to my wife Sarah, who has given me all
viii 
sorts of wise advice and who kindly agreed to do the copy-editing. I am also
grateful to the three anonymous advisers for OUP, and to Michael Smith and Jay
Wallace for advising me about my Introduction.
Actually, of course, I owe Sarah much more than this. One signal instance: in
the late 1980s I had been teaching at Keele for seventeen years and had begun to
publish in a way that I was quite satisfied with, but she thought I could do more, or
better. So she suggested that I should try to set up a sort of lecturing tour in the US,
so as to put myself about a bit. When I said that we could not possibly afford such
a thing, she pointed out that US departments give honoraria to speakers at their
colloquia, and pay expenses too, which meant that if I could get four or five
invitations the honoraria would cover the trans-Atlantic fare and the whole thing
would cost us nothing. So with many misgivings I wrote to a few people in the US
saying that I hoped to be flying over their institution on the way from one talk to
another and asking whether there would be any interest in my stopping off to give
a talk for them. And it worked. So I had this nice little tour set up. But the icing on
the cake came when I was writing to John McDowell about something else and
with some hesitation made the same suggestion to him. The completely unex-
pected result was a phone call from the then Chair at Pitt offering me a Visiting
Professorship for a year.
Introduction: No More Answers?
This book is a selection from papers published between 1977 and 2015. Most of
the earlier papers were published in journals, but then, as has happened to many,
nearly everything I wrote emerged in collections, either as records of conferences
or in celebration of the work of others, and so became less easily available.
The title that I originally intended to give to this collection as a whole was ‘No
More Answers’, which seemed appropriate to a collection published at the end of
one’s intellectual career. (Sadly, my publishers would not allow me this title, for
marketing reasons.¹) I owe that title to my son Hugh, but in a rather different
context. I was amusing myself when he was about six years old by asking him why
I should not send him to bed. His answer was, ‘I don’t want to go to bed’. So
I asked him why I should not make him do what he didn’t want to do. He said, ‘It
would make me unhappy’. I asked him why I shouldn’t do what would make him
unhappy. And his reply was, ‘There are no more answers, Daddy’. This seemed to
me to be the right reply. We can often enough explain why a consideration is a
reason, but eventually that desire for explanation will hit a brick wall, in the form
of a point at which no more can be said, without that doing anything to destabilize
the reason we were trying to explain.
My introduction to philosophy was peculiar. At the end of six terms reading
Greats at Oxford, with J. O. Urmson and C. C. W. Taylor as my tutors, I had read
Plato’s Republic and Aristotle’s Ethics (in Greek), Locke and Hume on the social
contract, Mill on utilitarianism, Russell on denoting, and then pretty much
everything written in Oxford since 1950 (but not including Anscombe’s
Intention). Not only was this rather limited fare, but I found the whole thing
very bewildering and had no real idea what was supposed to be going on. Luckily
for me, I experienced a kind of revelation one afternoon in the Spring vacation of
my final year as an undergraduate when on a reading week with some friends, a
moment at which things suddenly started to make sense. I returned to Oxford for
the final summer term full of new enthusiasm, and even applied to read for the
BPhil. In those days, doing the BPhil at Oxford was a truly hazardous procedure
because only around 25 per cent of the annual entry managed even to pass. But
I didn’t care whether I passed or not. All I wanted to do was to rectify some of the
most glaring gaps in my philosophical education, expecting after that to become a
¹ OUP was unwilling to allow this title on the grounds that the titles of academic books should send
clear signals about their content, which this certainly wouldn’t.
Practical Thought: Essays on Reason, Intuition, and Action. Jonathan Dancy, Oxford University Press.
© Jonathan Dancy 2021. DOI: 10.1093/oso/9780198865605.003.0001
2 
schoolmaster like my father, or perhaps enter the Civil Service. But somehow
I enjoyed my time doing the BPhil enough for me at least to apply for such
university lectureships as were advertised, and so ended up moving to Keele
University, where I spent the next twenty-five years. In my application I had
announced that I specialized in philosophical logic, without any very clear under-
standing of what that amounted to.
It was during my two years reading for the BPhil (1969–71) that Donald
Davidson hit Oxford, not in person (though he gave the Locke Lectures there in
1970), but in the form of his paper ‘Truth and Meaning’ (1967). I remember
reading this paper then and thinking what a difference it could make to moral
philosophy, which at that time, at least in Oxford, still seemed to be dominated by
R. M. Hare. It looked as if a Tarski-style theory of truth could be written for moral
sentences (apologies for this expression—perhaps I should have written ‘sentences
in the Language of Morals’) as for any others, and that there need be no misgivings
about whether the ‘if and only if ’s involved were other than straightforwardly
truth-functional—which they would have to be if Hare’s prescriptivism were to
survive, since prescriptive utterances, for him, were not capable of truth. In this
enthusiasm I was merely following the local flow (without yet knowing that I was
doing so). In the next few years, David Wiggins, John McDowell, and Mark Platts,
with Philippa Foot and Iris Murdoch in very different ways, contended for the
possibility of truth in ethics after a bleak period in which all such thought had been
banished into outer darkness. I found this renewal of cognitivism in a new form
exciting, and of course it opened the doors for a revival of interest in the history of
cognitivism in ethics, especially the local history of moral philosophy in Oxford
and Cambridge.
When I got to Keele, I found an environment in which there was no pressure to
publish at all, since publication was not then required for tenure—which was just
as well because as far as I knew I had absolutely nothing to say. I now think that
I could have published my BPhil thesis, which was on the theory of categories; but
I had such a low opinion of my abilities at that time that such a thing never
occurred to me (even though it was awarded a distinction in the BPhil, and I did
eventually publish part of it in a collection of papers on Gilbert Ryle (Dancy 2015).
So with no pressure I had time just to look around. As I said, I was, if anything, at
that stage officially a philosophical logician; at least that was the guise under which
Keele had hired me. But with no formal training in logic I was definitely at a
disadvantage. So I learnt up basic logic and worked through Hughes and
Cresswell’s textbook on modal logic, and this led to my first article, ‘The Logical
Conscience’, reprinted in this volume as Chapter 1, which applied modal distinc-
tions to moral philosophy, specifically to an argument of Ewing’s, and to the
theory of rationality. The general idea concerned the very plausible thought that
one ought not to do what one believes to be wrong. What if it is not wrong? Then
one’s belief that it is wrong would be false. But that would mean that it would not
   ? 3
be wrong to do what one believes to be wrong, so long as one is mistaken about

that. And this does not seem comfortable. My solution to this conundrum appeals
to a structural distinction between the thought ‘if you believe it to be wrong, you
should not do it’ and ‘you ought, if you believe it wrong, not to do it; the ‘ought
‘has been relocated in such a way that its scope has been changed. This is logic in
the service of ethics.
So I must already have been reading some of the moral philosophy written in
the 1930s. My father was amazed to hear that I had been reading Ross; he had
attended the last series of lectures that Ross ever gave, just after the war. (My first
copies of The Right and the Good and Foundations of Ethics belonged originally to
my father and my father-in-law, respectively.) When I first started The Right and
the Good, I could not believe that nobody had even mentioned this book to me
when I was at Oxford. All such moral philosophy had been blown away by Hare.
But it was coming back.
My first serious foray into moral philosophy was really into moral metaphysics
and the theory of reasons. ‘On Moral Properties’, Chapter 2 in this collection, was
driven by an interest in the distinction between supervenience and resultance.
Supervenience is a co-variance relation. The moral supervenes on the natural if
and only if, wherever the same pack of natural (subvenient) properties recurs, you
get the same moral (supervenient) property. And the ‘same pack’ here has to
consist of all the natural properties of the action, for any difference will undermine
the requirement that where these natural properties recur, the same moral prop-
erty must recur. Resultance is a different and much more selective relation—a
‘making relation’, one might say, or a form of ‘because’. The action’s wrongness
results from its uncharitableness, say, if it is made wrong by being so uncharitable.
It is perfectly possible for one action to resemble another in the respects that made
the latter wrong, but to differ in other respects so that what made the latter action
wrong does not have the same effect in the former. And the same applies to the
features that made the latter action uncharitable; they might not have that effect in
another case. As I began to put it, what is a reason in one case may not be the same
reason in another; if it isn’t, the second case will have some feature that prevents
what was a reason there from being the same reason here—a disabler, as I called it.
Such distinctions have a significant impact on Hare’s doctrine of the universaliz-
ability of moral judgement, to its detriment—or so I argued.
But as well as Hare, Ross was a target—more specifically so in my ‘Ethical
Particularism and Morally Relevant Properties’ (Chapter 3). It was in thinking
about Ross’s system of prima facie duties that it occurred to me that matters were
less regular in ethics than Ross supposed. Ross thought that a promise invariably
generated a prima facie duty, but it seemed to me that though most promises did
that, some promises did not. A promise might, and would, if the circumstances
were right (or should I say wrong?), generate no duty of any sort, prima facie or
otherwise. For the circumstances to have this effect, it must be possible for there to
4 
be considerations that would prevent what would otherwise ground a prima facie
duty from doing so—disablers, again. And this is really the same problem as the
one I had seen for Hare, because the logic of Hare’s system required you, once you
have approved of one action, to approve of any other that shared with the first the
features that were your reasons for that approval. But there was in fact no such
requirement, because of the ever-present possibility of disablers present in the
second case and not present in the first. So Ross’s system collapsed in the same
sort of way that Hare’s did, and so would any other system of exceptionless
principles. But principles have to be exceptionless, or they are not principles.
And so it looks as if principled ethics in general is in trouble.
As well as these questions about the structure of moral thought and reasoning,
there is also the metaphysical question what sort of property rightness and
wrongness are in the first place, the answer to which should be able to tell us
how we can recognize instances of these properties. We know we are dealing with
to-be-doneness and not-to-be-doneness, but what sort of properties are these?
And what is it to recognize instances of these properties? One answer to the
metaphysical question is that to be wrong is to be such as to merit disapproval, or
abstention. This is a normativized version of the dispositional conception of
secondary properties, e.g. colour, as the property of being such as to appear in a
certain way to beings equipped with the relevant faculties. I spent many happy
hours worrying about these metaphysical issues and their epistemological aspect.
One product was ‘Two Conceptions of Moral Realism’ (Chapter 9 here), but I was
also led into the theory of thick (rather than thin) ethical concepts, as appears
in ‘In Defence of Thick Concepts’ (Chapter 11) and ‘Practical Concepts’
(Chapter 13), spanning nearly thirty years.
Probably one should not poke one’s nose into the thickets of moral epistemol-
ogy without some competence in epistemology more generally. But it was no such
grand motivation that led me in the latter direction. At Keele I found myself
lecturing on all sorts of topics that were not officially my business. One especially
tough year was the one in which I gave all the first-year lectures myself (except for
the ones on political philosophy, a subject which consequently I have never
touched). This was for the simple reason that everyone else was on leave at the
wrong time. Still, I found myself giving the second-year lectures on epistemology,
starting in the late 1970s, which required a lot of work to get up. This was the
period when epistemology was beginning to release itself from the coils of Gettier,
and it was therefore a good time to plunge in. As I learnt more each year, the
lectures developed, but by 1983 I found that they had ceased to change. So it
seemed to me that I should either stop giving them, because giving the same
lectures year after year is not what universities are about, or write them down and
then stop giving them, since giving one’s own book as lectures is not what
universities are about either. So I opted for the latter course; An Introduction to
Contemporary Epistemology appeared in 1985, and I never lectured on
   ? 5
epistemology again—though I edited, with Ernest Sosa, the first edition of A

Companion to Epistemology in 1991, and continued to receive invitations to give
papers at conferences on epistemology for many years.
One of the positions for which I expressed support in that 1985 book was direct
realism in the theory of perception, which holds that in perception we are (or at
least can be) directly aware of the nature of our surroundings. My interest in this
theory no doubt contributed to the interpretation of Berkeley’s metaphysics which
I offered in Berkeley: An Introduction (1987). For Berkeley, the physical world
consists of ideas in the mind of God—ideas that God shares with us when we
perceive the nature of our surroundings. Perception is, therefore, a form of direct
(unmediated) awareness of aspects of that world. Berkeley emerges as a direct
realist in the theory of perception. But those perceived aspects of the world do not
consist of instantiations of primary qualities, understood as ways in which lumps
of inert matter can be. What we are aware of, for Berkeley, is always an idea, and
some ideas—namely the ones that God shares with us—are physical things. So the
physical world is a set of ideas—ideas that God is sharing with us. Since to share
one’s ideas intentionally is to communicate, physical events are, for Berkeley,
instances of divine communication. The world we perceive is really a set of divine
utterances; the world in its entirety is a sort of book, which we understand in the
sort of way in which we understand a conversation. In this way Berkeley arrives at
a traditional understanding of natural events as a sort of text. The natural world is
not an inert lump of meaningless matter. Natural events are a book written by God
for us to read.
Returning to my epistemology textbook: the final pages of that book suggest
that an influential sceptical argument against the possibility of empirical know-
ledge is effectively identical with an argument in ethics, and that both arguments
can be shown to be fallacious. In ethics the appeal is to a principle of universal-
izability: if in one case we pronounce an action right because it has certain
features, we must say the same of any relevantly similar action, where ‘relevantly
similar’ means ‘similar in all the respects that were our reasons for judging
the first action right’. I had already argued that this principle is false; it is false
because a second case, despite being similar to the first in all features that were
our reasons for judgement there, might have further features that prevent what
were reasons there from being reasons here. So similarity in the reasons-base, as
it were, cannot generate a rational requirement that one make the same judge-
ment the second time. What would generate such a requirement would be that
the second case is entirely indistinguishable from the first. But this appeal is to
supervenience, not to a more limited similarity. And supervenience has no teeth,
because there cannot be two cases that are similar in all natural respects (because
of the identity of indiscernibles). The most we could claim is similarity in all
relevant natural respects, but any requirement expressed in those terms would
be toothless.
6 
This appeal to an argument whose home ground is ethics, to address an issue in

epistemology, presupposes that ethics and epistemology have the same rational
base. The theory of reasons is the place where that base is to be found.
The connection between ethics and epistemology surfaces again in ‘Externalism
for Internalists’ (Chapter 6), which was written rather later. Here I try to delineate
fruitful similarities between a distinction between internalism and externalism in
the theory of justification and a similar distinction in moral theory. One of my
main moves is the scope distinction introduced in Chapter 1 in this collection. But
the hope evinced by these efforts is that normative theory will look pretty much
the same in its application to belief as it does in application to action, despite the
admitted differences between belief and action. (But all this leaves special issues
about knowledge out of account.) This hope remains with me today and drove my
last book, Practical Shape (2018).²
Section 1 of this collection contains philosophical reflections that are more
direct than subtle. The tripartite distinction between reasons, enablers/disablers
and intensifiers/attenuators, which first emerges in ‘On Moral Properties’
(Chapter 2), is fairly basic stuff and has, I think, been largely accepted by people
who work in the area. But it took me a long time to work out how to run, and then
how to present, a full theory of moral thought which does without principles
altogether. By 2000 I had achieved a sort of position statement for a collection
edited by Brad Hooker and Margaret Little, reprinted here as Chapter 7. That
paper, ‘The Particularist’s Progress’, is preceded by one written fifteen years
earlier, ‘The Role of Imaginary Cases in Ethics’ (Chapter 5). Appeal to imaginary
cases is often made in the search for a defensible set of principles, but that was not
my concern in that paper. My concern was rather to make sense of what I took to
be a perfectly standard practice of appealing to imaginary cases to help one
determine how to act in an actual, particular situation. The standard way of
understanding that practice is to suppose that the imaginary case reveals a
principle which we can then apply to the actual case that we are really struggling
with. Obviously, as a particularist, I could not run that story—but neither could
I simply denounce the appeal to imaginary cases as misguided. My response to
this difficulty is to show that everyone is in trouble here, which I still believe. The
challenge is to explain the force of the move from what we want to say about an
artificially foreshortened imaginary case, which has no properties other than those
specified in the (often very limited) description of that case, to what we want to say
in a real case which necessarily has a vast range of properties beyond those in
which it resembles the imaginary case. But I don’t think that the ‘solution’ I offer
on behalf of particularism is anything like fully satisfactory. My suggestion was
that an imaginary case acts as a sort of reminder of the importance that a
² I note here that I first introduced the notion of practical shape in a very early paper on
supererogation: Dancy (1988).
   ? 7
consideration can have, which therefore needs to be born in mind when one is
considering a complex case where that consideration is in play. But I have never
been very happy with this suggestion, and tend to agree with those who thought
that my puzzle was better than my solution.
Section 2 of this collection is concerned with far more complex issues, mainly in
moral metaphysics. I have already floated the question what sort of properties
rightness and wrongness could be, if they are to be properties at all. ‘Right’ and
‘ought’ are often treated similarly, but this should probably be avoided. I tend to
suppose that all ‘ought’s are attached to some agent; it cannot be true that
something ought to be done about this unless there is someone who ought to do
it—which there often won’t be. But rightness is not in that sort of way attached to
an agent, as perhaps is shown by the fact that ‘ought’ implies ‘can’ but ‘is right’
does not. The right response might sadly not be one that anyone is capable of
achieving.
One conception of deontic properties, already mentioned, is dispositional: to be
right is to be such as to elicit approval in a disinterested moral agent. (For this to be
informative, approving cannot be understood as ‘thinking right’.) I wondered
whether this dispositional conception can be got to run. It was an application to
moral metaphysics of a dispositional conception of colour: to be red is to be such
as to appear a certain way to creatures suitably equipped. But that conception of
colour seemed challengeable to me too, because it is at odds with the phenomen-
ology. The redness of a red object does not present itself to a perceiver as a
disposition to look that way. This of course raises the question what it would be
for a disposition to appear as such. We need to have an example to work on, of a
disposition that appears as such, and my suggestion was the sharpness of a knife.
We can experience that sharpness by running a finger along the blade without the
knife doing any cutting at all. What we sense is that the knife would cut easily. But
it did not seem to me that rightness is experienceable as a disposition in any
similar way.
In addition, there are many evaluative concepts that are not deontic, by which
I mean that they are not the same sort of concept as ‘right’ or ‘wrong’, but which
still raise many of the same metaphysical and epistemological issues. In particular,
there are the so-called ‘thick’ concepts. Most terms of aesthetic approval and
disapproval, praise and disparagement, are thick rather than thin. It is a debatable
question whether ‘beautiful’ and ‘ugly’ play the same role in aesthetic appraisal as
‘right’ and ‘wrong’ do in ethics, though my answer would be ‘no’, simply on the
grounds that beauty and ugliness are evaluative rather than deontic notions. There
may then be no thin aesthetic concepts, and perhaps that is what we should
expect. (But could ethics dispense with ‘right’ and ‘wrong’?)
The theory of thick concepts has interested me ever since I chaired a sympo-
sium on that topic between Allan Gibbard and Simon Blackburn in 1992. In many
debates these two were on the same side, and indeed their basic presuppositions
8 
were very similar on this occasion. But Gibbard was willing to accept current
presuppositions about the thick and work with them within his system, while
Blackburn wanted to shake things up much more. I had some comments about
their debate, which I was supposed to deliver before opening the discussion to
contributions from the audience, but in fact the two protagonists talked for so long
that I suppressed my comments entirely. Frustrated, I turned those comments into
a stand-alone paper (‘In Defence of Thick Concepts’, Chapter 11 in this volume).
Blackburn had argued that there are no thick concepts, and therefore the notion of
the thick could not generate any argument against non-cognitivism in ethics.
I wanted to accept many of his observations but still retain the possibility of thick
concepts, supposing that the thick is a fruitful area of enquiry, and one which non-
cognitivists may yet find it hard to accommodate. Here again I built on work by
McDowell and Wiggins. The question we were all worrying about was the
question what sort of fact the fact that a joke was funny or an action brave
could be.
By this stage of my intellectual progress I had firmly in place the two
motivations that have driven most of my philosophical thinking. These are,
first, the merits of disjunctivism, and second, the effects of holism in normative
theory. The latter I thought up for myself, but the former I learnt from others.
The main lesson of disjunctivism is that one should not let the nature of the
bad case (or cases) upset one’s account of the good case, and in particular
that one should not start with the bad case and then write one’s account of the
good case as the case that, though essentially resembling the bad case, has some
redeeming features.
Disjunctivism appeared on the philosophical scene in the early 1980s—
Snowdon (1981) and McDowell (1982)—though everyone acknowledged an ori-
gin in Hinton (1973). Its role at that point was restricted to the theory of
perception and the theory of knowledge. In the theory of perception (the original
primary example) the question was whether perception was a sort of illusion-plus,
or whether there might in fact be no common element between perception and
illusion. In the theory of knowledge more generally the question was whether
knowledge is some grand form of belief—a sort of belief-plus—or a different state
altogether, a sort of openness to a fact rather than a closed state dignified by its
having latched on to a fact. Once one sees the possibility of the latter view, one can
make a lot more sense of the talk of intuition that is common in Ross and
Prichard—and indeed in Oxford-based theory of knowledge more generally.
I had mentioned disjunctivism approvingly in my epistemology textbook, but
did not allow it to play a large part there because it was at the time only in its
infancy and did not seem to me to be appropriate fodder for undergraduates. But
the possibility of disjunctive conceptions more generally has been ever with me,
though, as will emerge, I have found it hard to follow the disjunctivists in the
philosophy of action.
   ? 9
Still, once one sees the possibility of disjunctive conceptions, one is ready to
refuse all attempts to understand good cases in terms taken from our account of
bad cases. And this lay at the bottom of my attempt in 1995 to write a new theory
of motivation (‘Why There is No Such Thing as the Theory of Motivation’, here
Chapter 16), one that made it possible for us to act for good reasons, while
allowing that some of the reasons for which we act are not good reasons at all.
This suggestion emerged in full dress in my Practical Reality (2000a). I reached it
as the final stage of a process of examination, and then rejection, of Humean
conceptions of motivation. In Moral Reasons (1993: chs.1–2) I started with an
examination of current responses to Humeanism, which largely consisted of
suggestions by Nagel and McDowell. I found their responses insufficiently radical,
and proposed instead a conception of motivation which was still a theory of
motivating states, but which conceived those states as entirely cognitive. Desire,
I suggested, though it must be found a place within a conception of motivation,
should be conceived not as what motivates, but as one’s being motivated—
motivated, therefore, by something other than desire. One could of course allow
this but still insist that what motivates one must be a state of oneself—a psycho-
logical state that one is in. And if one does so insist, the natural alternative to
Humeanism is the view that one is motivated solely by one’s beliefs; on this form
of anti-Humeanism, one would be motivated by the gap between one’s beliefs
about how things are and how they might be if one were to act in a certain way.
This non-Humean, cognitivist conception of motivation was still psychologis-
tic, in the sense that it maintains that motivation is the product of certain
psychological states in the agent. But I also suggested, as a step beyond pure
cognitivism, a non-psychologistic conception of motivation, under which one is
motivated, not by one’s believings, but by what one believes (1993: ch. 2.5). In this
volume, Chapter 16 pursues this idea, whose importance I don’t think I really saw
at the time, and which reached its fullest expression in Practical Reality. The main
argument in that book is that it must be possible to act for a good reason, and that
good reasons are not restricted to psychological states of the agent—in fact, most
good reasons are not states of the agent at all. I still think this is a really good
argument. But of course it was attacked by people who wanted to know about
acting for bad reasons and actions where one’s reason would be a good reason
were it the case, but where since it is not the case it isn’t a reason of any sort. The
argument continued between those who wanted to understand acting for a good
reason in terms that would also capture the nature of acting for a bad reason, and
those who wanted to understand acting for a bad reason in terms that would also
capture acting for a good one.
There are various possibilities here. The first is that the agent’s beliefs about the
situation that confronts him are true, he acts in that light, and what he believes is a
reason so to act. The second is that the agent’s beliefs about the situation that
confronts him are true, he acts in that light, but what he believes is no reason so to
10 
act. The third is that the agent’s beliefs about the situation that confronts him are
false, but he still acts in that light, and what he believes would be a reason so to act,
were it the case. The fourth is that the agent’s beliefs about the situation that
confronts him are false, he acts in that light, but what he believes would be no
reason so to act, even if it were the case. So there is the good case, and then various
less successful versions. My proposal was to understand all the less successful cases
in the light of the good case. But note that my good case was characterized in
terms of true belief, not in terms of knowledge. I left to one side any difference
there might be between acting out of true belief and acting out of knowledge.
As I said, I had one very simple argument, which was that it must be possible to
act for a good reason: otherwise put, it must be possible for a motivating reason to
be a good reason. Good reasons for action include such things as that he needs our
help, or that the sun is very hot today; and these are not psychological states of any
agent. If the agent’s psychological states are not the only good reasons for him to
do anything, the reasons for which agents do what they do cannot be restricted to
their own psychological states. It is true, of course, that when an agent acts for a
good reason, say, the reason that p, she will believe that p. But that she believes
that p is not the same reason as the reason that p, and her reason for doing what
she does should not be simply identified with her believing that p. Sometimes, of
course, the good reason to act will consist entirely of psychological states of the
agent—as, for instance, where she checks again because she is worried about
something, or because she has forgotten something. But this does nothing to
establish psychologism as a general theory of the reasons for which one acts.
I might check again because (or for the reason that) my wife is worried about
something.
If motivating reasons are not psychological states of the agent, we might still
suppose that they are not matters of fact, because one can be motivated by
something that is not the case. Something that is not the case is a false proposition,
let us allow. So perhaps we are motivated, not by our own believings, but by what
we believe, understood as a proposition. Again, we reach this position by thinking
about a case where things are not in perfect order—the case where things are not
as we take them to be but still we act in that light. But I maintained against this
that no proposition is capable of being a good reason, even though we express
reasons, good or bad, in propositional form. Propositions are just the wrong sort
of thing to be reasons of any sort.
I described my position here as realism in the theory of motivating reasons, as
opposed both to psychologism and to what one might call propositionalism. The
main opposition to my position—or at least the opposition that I took most
seriously—was some form of disjunctivism. This might seem odd, when I have
been heralding the force of disjunctive conceptions. But now we hit the fact that
disjunctivists do not agree among themselves, or perhaps I should say that there is
debate about what disjunctivism really amounts to. Having got embroiled in this
   ? 11
debate in Practical Reality (see esp. ch. 7.1), I returned to it in the papers presented
here as Chapters 17 and 19 (‘How to Act—Disjunctively’ and ‘On Knowing One’s
Reason’).
The main worry here is whether I should insist that the agent’s reason is
always of the same form, ‘that p’ rather than, say, ‘that I believe that p’. Now of
course we all know that these are different reasons—different normative reasons,
that is. But when we turn to the theory of the reasons for which people act, are
there two types of reason, ‘that p’ and ‘that I believe that p’? It is possible to hold
that when things are as the agent supposes, the reason for which she acts will be
the matter of fact, for instance that it is beginning to rain. But when it is not in
fact beginning to rain (perhaps she has just walked under her neighbour’s
sprinkler), can the reason for which she puts her coat on be that it is beginning
to rain? Must we not rather say, of such a case, not only that she put on her coat
because she thought it was beginning to rain, but that her reason for putting on
her coat was that she thought it was beginning to rain? And once we do say that,
should we not also allow that even when she is right about the rain, her reason
for putting on her coat will never be that it is beginning to rain, but always that
she so believes?
These may seem like comparatively incidental matters, but large decisions
about the nature of the explanation of intentional action hang on them. One
characteristic of disjunctivist approaches is that they do not let their account of the
good case be driven by what they are forced to say about the bad case, and that was
the feature I was hoping to exploit. But there are other ways of doing it. The most
forceful alternative to my position is the one that restricts acting for a reason that
is a fact to the case where the agent knows that fact rather than merely believes it.
For this to work, as the basis for an alternative picture, we are going to have to
think of knowledge as something other than a specially secure form of belief. And
there is a formidable band of philosophers whom I respect and who, taking such a
view, suppose that the only cases where our reason for action is a fact are cases
where we know that fact. I have in mind here especially John McDowell, Jennifer
Hornsby, and John Hyman.
It is important to be clear what the options are here. I maintain that what
motivates me is what I believe, not my believing it; and though I allow that
knowledge is not a form of belief, I say the same about cases in which what
motivates me is what I know. McDowell, Hornsby, and Hyman say, against this,
that the difference between belief and knowledge is stark enough to draw a
distinction in motivation: I am only motivated by a fact if I know that fact to be
so. McDowell writes, with characteristic force, that ‘if the truth of the relevant
belief is merely good luck, cognitively speaking, . . . it would be absurd to say the
agent’s reason for acting is the fact itself ’ (2013: 17). So we agree that the reasons
for which we act are never propositions, and that when things go best, the reason
for which we act is a fact. But we differ about cases in which we do not know that
12 
things are as we suppose when we act in that light. I pursue this debate in ‘On
Knowing One’s Reason’ (Chapter 19), and below.
It was around this time that it occurred to me that two views to which I was
already committed gave me a third view for free (see Dancy 2021). This third view
emerged in my Practical Shape (2018). The two views already on the table were (1)
holism in the theory of reasons and (2) the idea that the reasons for which we act
and believe are never propositions, but matters of fact (real or supposed). And it
seemed to me that this generated a much stronger form of Aristotelianism in the
theory of reasoning than any that had been suggested so far. Aristotle maintains in
various places that the conclusion of a practical syllogism is an action, and this
idea has been one of which nobody has been able to make much sense. But
whatever one says about the practical syllogism, we need not suppose that the
conclusion of practical reasoning need be anything other than an action. After all,
we act in the light of a simple reason; so why should we not also be able to act in
the light of the sort of complexity thrown up in practical reasoning, where we
consider reasons for and against, along with intensifiers, attenuators, enablers,
disablers and whatever other complications our reasoning may throw up?
Complexity in that from which we reason to action need not affect in any way
the directness of the relation between that from which we reason—matters of
fact, real or supposed—and that to which we reason, the action which we take in
that light.
This picture needs to confront views of reasoning which start from preconcep-
tions about theoretical reasoning. There is a tradition of supposing that in
reasoning to belief, we reason from propositions to propositions; and the study
of such reasoning has even been called ‘propositional logic’. Now of course, if
there are such things as propositions, there are going to be relations between
propositions. If one proposition is true, another might be made more probable
than it would otherwise be; and several propositions might together guarantee the
truth of another one. All this is undeniable. But I hold that no proposition (true or
otherwise) is a reason for anything. When we engage in serious (rather than
classroom) reasoning, we reason from—and only from—things that we take to
be so. Being so is not the same as being true. And even when the things we take to
be so are not so, that does not affect the light in which we are treating them when
we reason from them.
These thoughts apply with equal ease to practical reasoning—reasoning to
action—and to theoretical reasoning—reasoning to belief. The main difference
between practical and theoretical reasoning (other than their distinctive ‘conclu-
sions’) lies in the way we explain the force of the reasoning. In reasoning to belief
we mainly reason from considerations which raise the probability of our conclu-
sion, and the force of the reasoning is the extent to which that probability is raised;
in logical reasoning we reason from considerations which guarantee the truth of
our conclusion, and the force of the reasoning lies in that guarantee; in reasoning
   ? 13
to action we reason from considerations which raise the value of so acting, and the
force of the reasoning lies in the extent to which it reveals that course of action as
worthwhile, and as more worthwhile than others. But these differences in explan-
ation should not be allowed to obscure a fundamental sameness in the processes
we think of as reasoning.
At the end of his 2013 paper, however, McDowell makes a point that I have yet
to assimilate. A main purpose of Practical Shape was to contribute to making sense
of a remark of Kant’s at the beginning of the Groundwork, that practical and
theoretical rationality must at bottom be one. Kant makes very little progress
towards showing how this could be, and I hoped to do better. But McDowell
suggests that the only way to succeed in that direction is to award a certain
primacy to the case where the agent is reasoning from knowledge rather than
mere true belief. He writes:
But if the perfect manifestation of practical rationality is acting in the light of a

fact in the sense that goes missing in Dancy’s treatment, we cannot consider
practical rationality in abstraction from cognitive capacities, and we have to
recognize that practical rationality is more deeply embedded in rationality at
large than the highest common factor conception allows it to be. (2013: 28)
This stern rebuke deserves a careful response. The ‘highest common factor
conception’ holds that the distinction between knowledge and true belief is not
significant for the theory of practical rationality. And it is true that I saw no real
significance in the distinction between acting in the light of what one believes and
acting in the light of what one knows. My phrase ‘in the light of ’ applied equally to
either side of that distinction. Was that a mistake? One point on which we agree is
that theoretical rationality is more deeply embedded in practical rationality than
generations of theorists have been willing to allow. But is practical rationality
deeply embedded in theoretical rationality, or in ‘rationality at large’, on anyone’s
view? Isn’t it rather a one-way matter, as I just put it: theoretical rationality is more
deeply embedded in practical rationality than we have supposed?
I begin to wonder whether McDowell has as good a right to claim to express
Kant’s identity thesis as he claims. I had hoped to make progress on that point by
laying what I thought of as structural similarities between practical and theoretical
thought. In both, I suggested, we reason to the specific form of response best
suited to the situation as we conceive it. Can I stick to that picture while accepting
McDowell’s conception of the difference between knowledge and (mere) justified
true belief? It seems to me possible that I can. What needs to be done, to make
progress in that direction, is to accommodate the distinction between reasoning
from knowledge and reasoning from justified belief, so that this distinction makes
a significant difference in both practical and theoretical reasoning, while also
doing justice to points I have been at pains to make about cases that involve
14 
belief. McDowell himself does not quite meet this challenge, as becomes clear
when we look closely at his treatment of the less favoured cases, the ones that
involve reasoning in the absence of knowledge. Is he going to be able to avoid the
conclusion (which officially he rejects) that the only possible account of the case
where we are acting in the light of things as we believe but do not know them to be
is that we are acting in the light of our own mental states? Since we cannot be
acting in the light of the fact believed (because acting in the light of a fact is
restricted to the knowledge case), there may be nothing left for us to be ‘acting in
the light of ’ other than our own mental states—our beliefs. But the light in which
we act is not our grip on the world; it is the world as we present it to ourselves—
and these are different things. Officially McDowell applauds my insistence that
when we are wrong, and when we are luckily right (and so do not know), our
reason for doing what we do is not that we believe that p; it is still that p. But what
is this ‘that p’ in the less favoured case? It is not a proposition, and it is not a fact
either. What is there left for it to be?
My official answer to this question is that we are dealing with an intensional
context, and in such a context there is no need to take ‘that p’ as a referring clause.
As McDowell says, it is the analogue in English of the Latin ‘quod’ when followed
by a subjunctive clause.
So I still worry that McDowell may be forced into giving a distorted account of
the less favoured case. I don’t see the need to insist on McDowell’s reading of the
locution ‘acting in the light of a fact’. Much more important is the distinction that
underlies McDowell’s insistence, and the awarding of rational primacy to the case
of action in the light of knowledge. Here too I think we can agree. But clearly there
are answers still to come.
1
The Logical Conscience
On pp. 144–5 of his Ethics, A. C. Ewing writes:
It is a recognised principle of ethics that it is always our duty to do what we when

considering it sincerely think we ought to do, but suppose we are mistaken, then
we by this principle ought to do something which is wrong and which therefore
we ought not to do. Is not this a contradiction? It would be if we were not using
two different senses of ‘ought’.
The appeal to this contradiction is commonly cited as evidence that there must be
two senses of ‘ought’. Little, however, is said about what these two ‘senses’ are and
how, if at all, they are related.
Using the following crude notation:
Ba: a believes that

da: a does d
O: it ought to be the case that
we might try to represent Ewing as taking the line that the complex formula
‘(BaOda ! Oda) · BaOda · ~Oda’ is inconsistent, unless ‘O’ equivocates. And
there is no doubt at least that if all the terms in this formula are univocal, the
formula is inconsistent. But there is a much less awkward method of dealing with
the difficulty than that adopted by Ewing, a method which still retains his original
intuitions. We can instead suppose that he is mistaken about the logical form of
one of the conjuncts; specifically, the first. As in other modal contexts, questions of
scope arise, and with them in mind we notice the similarity between
(1) BaOda ! Oda

and (2) O(BaOda ! da).
If (2) improves on our original representation of Ewing’s original intuition that

‘it is always our duty to do what we when considering it sincerely think we ought
to do’, it looks as though Ewing was wrong in holding there to be a contradiction
here. For (2) appears to be consistent with
16   
(3) BaOda · ~Oda
in which case the revised Ewing formula
(4) O(BaOda!da) · BaOda · ~Oda
is unproblematic. But can we be sure that (2) and (3) are consistent? What we
want here is a proof of the invalidity of
(5) O(BaOda!da)!(BaOda!Oda).
Such a proof can be obtained within Hintikka’s system of deontic logic,¹ which
interprets the operator ‘O’ in terms of deontically possible worlds. Suppose the
consequent false; this tells us that BaOda · ~Oda. ‘~Oda’ is equivalent to ‘P~da’
(‘P’ = ‘it is permitted that’), which tells us that in some deontic alternatives² to our
world, ~da. Given the antecedent ‘O(BaOda!da)’, we deduce that in those
alternatives ~BaOda. Nothing in this leads to a contradiction, for the falsity of
the consequent makes ‘BaOda’ true in our world, not in any alternatives.
More formally, we can show (5) to be invalid by showing (4), its negation, to be
satisfied by the model system which consists of the following two model sets:
(α) BaOda, ~Oda, P~BaOda, O(~BaOda v da), da

(β) ~da, ~BaOda, (~BaOda v da), O(~BaOda v da)
where (β) is assumed to be a deontic alternative to (α). Inspection will reveal that
this set of sets of sentences satisfies Hintikka’s defining conditions for a model
system. We conclude, therefore, that (2) and (3) are consistent, while (1) and (3)
are not.
So if (2) can be taken as a reasonable version of the intentions behind (1),
Ewing’s ‘contradiction’ is resolved without appeal to different senses of ‘ought’.
How is (2) to be interpreted, and how far does its superficial resemblance to (1)
enable us to suppose them to have been confused? According to (2), in all deontic
alternatives to our world people do in fact do what, in those worlds, they believe
they ought to do. This differs markedly from (1), which tells us that in all deontic
alternatives to this world people do in fact do what, in this world, they believe they
ought to do. There is a further difference between (1) and (2). (1) tells us that in
this area everyone’s beliefs are true, (2) neither guarantees nor requires the truth of
¹ For this and other concepts of deontic logic, see, for instance, Hintikka (1969).
² Deontic alternatives to our world are, roughly, worlds in which people do in fact do what, in this
world, they ought to do; worlds, that is, in which all our duties are fulfilled. This is made much more
precise in Hintikka (1969).
   17
any beliefs, even of those beliefs only which are held in better worlds than this; for
the thing believed, Oda, is not identical with the thing true, da. We shall be unable,
therefore, to show (2) true by simple appeal to the doctrine that the inhabitants of
deontically acceptable worlds hold only true beliefs about what they ought to do.
For that is the further statement
(6) O(BaOda!Oda)
which may or not be true. If true, then together with the more probable doctrine
(7) O(Oda!da)
it will entail our (2). But we need not appeal to this dubious parentage in order to
be allowed to hold (2), which is after all only an attempt to capture the better
intuitions that lay behind (1). Our formula thus leaves undecided the question
whether in deontic alternatives to our world people still hold mistaken beliefs
about their duties.
It is here that the crucial difference between (1) and (2) is to be found. Both are
concerned, and in fairly similar ways, with getting people to do the things they
think they ought to do. Both have the consequence, on fairly natural assumptions,
that we are right to blame a man for failing to follow his conscience. But, crucially,
we do not need to stipulate that all the dictates of his conscience are true in order
to obtain this result. The distinction between (1) and (2) lies in the fact that, by
asserting (2), we require that people do what they believe they ought to do,
without committing ourselves to endorsing all those beliefs.
It is impossible, of course, to validate all the intuitions that led to the production
of (1) in that sense in which it conflicts with (3). Ewing’s expression of the point,
however, can be read as an interpretation either of (1) or of (2). Anyone therefore
who feels sympathy with Ewing’s account will know which to choose. In this case,
as in the case of other syntactical modal ambiguities, the ambiguity is cloaked in
natural language. This is a reason for supposing there to have been confusion
between (1) and (2).
It seems to me, therefore, that this way of resolving the initial contradiction has
great advantages over the resort to different ‘senses’ of ‘ought’. There are, of
course, other ways of achieving the same result. One is to reject the possibility
of (3). Since (3) is intended merely as the expression of a particular instance of a
general truth, this move amounts to the rejection of objective duties. Some of
those who have supported (1) or some similar doctrine have attempted to do so
while admitting the existence of objective duties; this is what led them to postulate
subjective duties as well, and it is these people with whom I am chiefly concerned.
My analysis offers them a way of admitting objective duties while requiring people
18   
to act according to their conscience, without validating everybody’s opinions and

thereby falling into contradiction.
Finally, a general point. The contradiction which I have tried to resolve is itself
only an instance of a group of similar paradoxes, all of which concern the relation
between a man’s beliefs, his actions, and the facts. We can ask whether it is rational
or sensible to act according to the facts or merely according to one’s beliefs about
the facts. We can ask whether it is wise/stupid to do what one thinks it is wise/
stupid to do. The more such examples are multiplied, the less plausible it is to
maintain that there are two senses of each of the key words ‘rational’, ‘sensible’,
‘wise’, ‘stupid’, etc. Any solution, however, to one of these paradoxes ought to be
generalizable to the rest. And this is in fact the case with the one offered here. For
it is, I think, entirely independent of the particular nature of the modal operator
used. The same result can be obtained if we use instead a ‘rationality’ operator
from epistemic logic. Using ‘R’ rather ungrammatically for ‘it is rational that’, we
derive the same distinction between ‘(BaRda!Rda)’, which is false, and ‘R
(BaRda!da)’, which is, I think, true.
SECTION 1
T O W A R D S PA R T I C U L A R I S M I N
ETHICS AND EPISTEMOLOGY
2
On Moral Properties
1. Resultance
We know a priori that if an action has a moral property, it has it in view of some
other properties which it has. I want to investigate this truism and explore its
relationship with two other truisms, but I shall not be offering an analysis of it.
Taken baldly as I have first expressed it, it does not go so far as some philosophers
would like. A more demanding version is:
We know a priori that if an action has a moral property it has it in virtue of

some non-moral properties which it has.
Both of these truisms are true enough. To express them reasonably simply, we can
use the term ‘resultant’ thus:
The F-ness of an object results from properties P¹ . . . Pn of that object iff the
object is F in virtue of its possession of P¹ . . . Pn, where F-ness is not identical
with P¹ . . . Pn nor with any truth function of members of P¹ . . . Pn.
And I shall allow myself to talk of one property resulting from a set of other
properties. So we have it that if an action has a moral property M, its M-ness
results from some of its non-moral properties. These non-moral properties may
be called the reasons why it is M, and typically they will form a rather small group
among the action’s non-moral properties. But when a man decides that the action
is M, his reasons will probably form an even smaller group; in a favourable case,
his reasons will be among the reasons why it is M, but in an unfavourable case,
they will not. In morals as elsewhere, a man can reach the right conclusion for the
wrong reasons, and even where he reaches the right conclusion for sound reasons,
he need not have exhausted all the available reasons.
All this might seem, from a certain point of view, uncontentious enough. Sir
David Ross expresses it this way:
It is clear that it is in virtue of my thinking the act to have some other

characteristic that I think that I ought to do it. Rightness is always a resultant
attribute—an attribute that an action has because it has another attribute.
(1939: 168)
22     
But he also writes:
An act that is right is right in virtue of its whole intrinsic nature and not of any
part of it. In respect of certain elements in its nature it may be prima facie right
and in respect of others prima facie wrong; whether it is actually right or wrong,
and if it is wrong the degree of its wrongness, are determined only by its whole
nature. (1930: 123)
Is this simply another conception which also needs to be spelled out, or is it a new
conception of resultance which competes with our previous one? In The Right and
the Good, his earlier work, Ross distinguishes carefully between the type of
resultance which I have already mentioned, and which he calls parti-resultance,
and the new type of resultance, called toti-resultance. We might take it that moral
properties are both parti-resultant and toti-resultant, it being true both that they
exist in virtue of the sum total of all the non-moral properties, and that some of
the non-moral properties of an action still play a special role in the determination
of its moral properties. But there can be objections to this attempt to have it both
ways—objections from both sides. Ross’s view, at least in The Right and the Good,
was quite clear: moral properties are toti-resultant and not parti-resultant. What
arguments can be produced in favour of this position?
I think it possible to discern two arguments in Ross. The first is as follows:
The equality of the two angles is a parti-resultant attribute. And the same is true
of all mathematical attributes. It is true, I may add, of prima facie rightness. But
no act is ever, in virtue of falling under some general description, necessarily
actually right; its rightness depends on its whole nature and not on any element
in it. (1930: 33)
If this is an argument, Ross is moving from the anti-naturalistic view that no

natural character is sufficient to guarantee rightness, to the view that rightness is
toti-resultant. Is this move valid? It is interesting that in a footnote to the passage
cited, Ross tells us that he has rather overstated his view in the interest of
simplicity:
Any act is the origination of a great variety of things many of which make no
difference to its rightness or wrongness. But there are always many elements in
its nature . . . that make a difference to its rightness or wrongness, and no element
in its nature can be dismissed without consideration as indifferent. (1930: 33)
But if the above argument is valid, it is valid against the position sketched in this
footnote. For the converse of the inference is that if rightness is not toti-resultant,
then some natural character is sufficient to guarantee rightness. Since Ross’s ‘true’
   23
view is that rightness is not toti-resultant, he should by his own reasoning

conclude what he believes false, that some natural character is sufficient to
guarantee rightness. This conclusion is not made more palatable by the qualifica-
tion that the natural character concerned is a complex one consisting of many
different qualities. However, as well as there being this ad hominem argument
against Ross, it is clear that the inference is invalid. Perhaps the way to show this is
to distinguish between two views:
1. Some limited natural character guarantees the rightness of any action which
has it.
2. Some limited natural character makes this particular action right.
Ross is arguing from the falsehood of the first to the falsehood of the second. But it
seems clear that this action can be right because it is the returning of an axe to its
owner, without such actions always being ‘necessarily actually right’. I shall return
to this point later.
The second argument discernible in Ross in favour of toti-resultance is based
on the agreed fact that ‘no element in the nature of an action can be dismissed
without consideration as indifferent’. This is the point that although an action
may have several properties in virtue of which it is prima facie right, there still may
be some overwhelming characteristic in virtue of which it is prima facie wrong and
which is strong enough to tip the scales so that the action is actually wrong. And
we cannot know whether the action is right or wrong until we have seen the effect
of each property on the sum total of its values.
The reply to this argument is as before. First, Ross’s own backsliding from true
toti-resultance indicates that he thinks the argument unsound. Second, he is right;
the argument confuses the truth that each property may make a difference with
the falsehood that each property does make a difference. The consideration
offered here in favour of toti-resultance is an epistemological one; but the con-
clusion is, in a broad sense, logical. The intuitionists seem agreed that in order to
discover the moral properties of an action or object, one needs first to discover all
its non-moral properties. Why should this be so? For no other reason, it seemed,
than that the action’s moral properties are toti-resultant; and so any moral
conclusion we come to before all the non-moral facts are in can at best have the
status of a well-ordered guess and might best be expressed in terms of the prima
facie. There are two complaints here: the first is that the notion of the prima facie
being used is not Ross’s technical one (for which, see below and Ross’s second
apology on page 20 of The Right and the Good) but the normal one, incompatible
with Ross’s, of an appearance presented at first sight but which later may turn out
to be illusory. Second, all the epistemological facts can be accounted for using just
the notion of parti-resultance. The crucial point here is that we can never know in
advance whether a given non-moral property is or is not relevant to the moral
24     
properties of the action. This is why we need to look at all the non-moral facts
before coming to our moral decision, and does not at all weight the balance in
favour of the notion of toti-resultance. It can still be the case that our final decision
is that the action is right in virtue of some of its non-moral properties, for its
properties in general are such that the action is right and that its rightness does
result from a limited number of them.
This being so, what can be said in favour of parti-resultance and against toti-
resultance? Again, there seem to be two points to be made here. The first is made
by Ross: ‘Any act is the origination of a great variety of things which make no
difference to its rightness or wrongness’ (1930: 33). Assuming that parti-
resultance and toti-resultance are competitors (on the basis that both are expli-
cated using the phrase ‘in virtue of ’), this point must create a difficulty for toti-
resultance. Moore’s doctrine of organic wholes may be an attempt to solve it. But a
worse difficulty for toti-resultance is created by its reliance on Ross’s notion of a
prima facie duty. This notion is subject to grave doubts.
The notion of a prima facie duty is introduced for two purposes. First, it is
introduced in order to increase the consistency of any complex moral position.
Suppose that I have as moral principles the following:
Murder is wrong
Breaking a promise is wrong
Lying is wrong
Helping others in distress is right (and so not wrong)
and so on. I will soon find a case arising of an action which comes under more
than one of these principles and is said by one to be right and by another to be
wrong. If I read my principles as universally quantified implications, I find myself
in self-contradiction and must adjust my moral scheme somewhere. But this, Ross
thinks, distorts the case; the validity of moral principles is not impugned by their
coming into conflict. Ergo, they must be of some other form. Ross therefore finds a
form under which they cannot conflict.
Murder is prima facie wrong

Lying is prima facie wrong
Helping others is prima facie right
In the case of a conflict of principles all we can do is to consider the case and
decide where the balance of wrongness or rightness lies.
Second, the prima facie is used to explain the fact that often, if an act that is a
duty is so in virtue of all its non-moral properties, among the properties in virtue
of which it is a duty will be some in virtue of which it is wrong. This paradoxical
situation can be resolved by deft use of the prima facie. The sentence ‘Among the
   25
properties in virtue of which it is a duty proper are some in virtue of which it is

prima facie wrong’ lacks any appearance of contradiction (cf. Ross 1930: 123).
There is a lot to be said about this notion of the prima facie, but I shall restrict
myself to two points in order to indicate where it is unsatisfactory. The first
concerns the nature of a conflict of duties. Where such a conflict occurs, one
may predominate; but the other is not for that reason abolished; it may create
residual obligations such as the duty to apologize for failing to keep a promise even
though it was in the circumstances one’s duty to break it, or may give rise to
feelings of regret for the action that one was unable to perform. Ross was aware of
this and tried to account for it by saying that prima facie duties are of the nature of
a claim on one, which is not abolished but overridden in certain circumstances.
But his explicit accounts of a prima facie duty leave no room for this sort of
remark. He offers us two:
(A) An action is a prima facie duty if it is an action which tends to be a duty

(1930: 28).
(B) An action is a prima facie duty if it is of such a kind that it would be a duty
proper if it were not at the same time of another kind which is morally
significant (1930: 19). (Ross also suggests the phrase ‘conditional duty’ in the
place of ‘prima facie duty’.)
The first of these accounts, in terms of tendencies, will not touch the particular
case; a particular action, even as an action of a certain type, does not tend to be a
duty but either is one or is not one. The second account cannot be faulted on that
score, but fails elsewhere. What it tells us is that where an action is prima facie
right in virtue of being the keeping of a promise, prima facie wrong in virtue of
being the neglect of the duty of self-preservation, and actually wrong since the
prima facie wrongness predominates over the prima facie rightness, the prima
facie rightness is the characteristic that the action would have been a duty had it
not been an action of some other morally significant kind. But why should this
characteristic create residual duties or a feeling of regret? Despite Ross’s good
intentions here, the notion of the prima facie seems quite unable to produce what
is wanted.
I now turn to the second, and for present purposes more important, area in
which we might look (without success) for help to the prima facie. Linked with
toti-resultance, we learn that an action is a prima facie duty in virtue of being a
promise-keeping if, if it had no other morally relevant characteristics, it would be a
duty in virtue of all its characteristics. This position attempts to account for the
peculiar place of promise-keeping among those characteristics by putting it
completely at the level of the prima facie. But it surely distorts the matter to
suppose that among the characteristics in virtue of which it is a duty proper,
promise-keeping has not some special place; particularly when we remember that
26     
in a case of conflict our action will have some characteristics in virtue of which it is
prima facie wrong but which still count, equally with promise-keeping, among
those in virtue of which it is actually right. Something must be said to pick out a
special relationship between the rightness of the act and some among its qualities,
and the notion of the prima facie does not provide what we need.
I conclude that the claim that moral properties are toti-resultant is not satis-
factory until either the notion of the prima facie is improved or some other
method is found to express the special place that some qualities have in deter-
mining the moral property of an act.
2. Supervenience
Even if the notion of toti-resultance gives an unsatisfactory account of the relation

between a moral property of an action and all its non-moral or natural properties,
there remains another and much more popular account of that relation. This is the
notion of supervenience. To say that moral properties supervene on the non-moral
is to say that if an action satisfies N₁, N₂ . . . Nn, where N₁, N₂ . . . Nn is a complete
natural description, and also has the moral property M, then any other action
satisfying N₁, N₂ . . . Nn must also have M, and the first action cannot change in
respect of M without changing in respect of N₁, N₂ . . . Nn, even though the
possession of N₁, N₂ . . . Nn does not entail the possession of M. This notion of
supervenience clearly falls into two parts, which are worth separating. Here is an
attempt¹ at a bipartite definition:
(S) If a property M is supervenient upon a class of properties C, then: M is not

identical with any member of C or with any truth function of members of C,
and if an object has M and C¹ . . . Cn (where C¹ . . . Cn are all the members of
C which it has), then it is impossible that it should cease to be M or become
more or less M than before without changing in respect of some member of C.
(S¹) If a property M is supervenient on a class of properties C, then: M is not

identical with any member of C nor with any truth-function of members of C,
and if an object has M and C¹ . . . Cn (where C¹ . . . Cn are all the members of
C which it has), then necessarily any other object which possesses C¹ . . . Cn to
the same degree also possesses M to the same degree.
Let us restrict ourselves for the moment, to considering the supervenience of

the moral on the non-moral, that is the case where M is goodness and C is the class
¹ (S) and (S¹) are taken from Blackburn (1971: 106).

   27
of non-moral properties. These definitions bring to light several difficulties with

the notion of moral supervenience. Apart from general problems concerning the
possibility of a sharp distinction between the moral and the non-moral, and any
doubt there may be about conceiving of all the non-moral qualities, the main
difficulty here is to prevent the right-hand sides of (S) and (S¹) from being too true
(where M is goodness and C is the class of all the non-moral properties) and so
proving the supervenience of the moral on the non-moral at the cost of making it
trivial. The right-hand side of (S) is too true in this case, since no person can
change in any respect without becoming older, and I am not sure that an action
can be said to change at all. The right-hand side of (S¹) is too true in this case, since
the identity of indiscernibles guarantees it, it being impossible for two things both
to possess all the same non-moral properties to the same degree. Assuming for the
moment that we are trying to preserve both (S) and (S¹), there seem to be two
methods of improving them. One is to look to the literature on the identity of
indiscernibles to provide some satisfactory minimal restriction on the notion of all
the non-moral properties. The other is to make a more radical change in our
version of C, the class of properties on which the moral properties supervene. The
natural candidate here is that of all the morally relevant properties: the moral
properties would supervene not on all the non-moral properties, but only on those
of them that are morally relevant. Of these two options, I very much favour the
first. Considerations here are twofold. First, can we produce a sound notion of the
morally relevant properties under which it will be true that the moral supervenes
on them? I shall provide reasons in the next section for doubting this. Second, it is
desirable at this stage to make some effort to distinguish between the super-
venience of moral properties on the non-moral and the universalizability of
moral judgements as propounded by R. M. Hare. Narrowing the subvenient
class in the way suggested simply collapses the two notions into one. But there
would be room on the alternative proposal to reject Hare’s doctrine of universal-
izability without rejecting the doctrine of moral supervenience, and this is an
option which ought to be left open as long as possible.
In the hope that the supervenience of the moral on the non-moral can become a
non-trival matter by an alteration in our account of supervenience along the lines
of the first alternative, I shall continue to speak as if it were a moral principle of
substance that two actions that share all their non-moral properties to the same
degree must share all their moral properties to the same degree, and that no object
can change its moral properties without changing its non-moral properties. This
gives us our third truism. And it gives us, so far, three relations between the moral
property M-ness of an object and its non-moral properties. The first is between
M-ness and the speaker’s reasons for asserting M-ness of the object; the second is
between M-ness and those properties from which the M-ness results; the third is
between M-ness and all the properties together. The possibility that some non-
moral properties are in some special way morally relevant suggests further such
28     
relations. But in order to introduce these, I need to return to my starting point and
look again at the relation of resultance.
A few further difficulties in the definition of supervenience emerge in Section 4.
3. Resultance and universalizability
The truism that actions have their moral properties in virtue of their non-moral
properties should not blind us to the fact that sometimes an action has a moral
property in virtue of its possession of another moral property. An action can be
good because it is generous. In the terminology of resultance, although a moral
property of an action always results from some of its non-moral properties,
sometimes a moral property results, on the way, so to speak, from another
moral property. The goodness of this action results from its generosity, which
in its turn results from some non-moral properties of the action, for instance its
being a substantial donation to a needy cause. And if this is true, one thing that
we might otherwise feel inclined to say is ruled out. We might, that is, want to
suppose that there is some short list of qualities from which moral qualities
result; a list, that is, of qualities in virtue of which actions get to be good or
bad. But this supposition is ruled out. There is no short list of qualities
from which all moral properties result. And we know this because we know
that many of the putatively exclusive qualities that are specially reasons for
moral assessment are themselves resultant. The quality of being a human being
is resultant, because if it is true that someone is a human being, it is true in
virtue of some other properties which he has. And we know that if a moral
property M results from a non-moral property A which itself results from
another non-moral property B, not only is it the case that M results from A,
but also that M results from B. For resultance is transitive. If it were not
transitive, i.e. if it were not true that where A results from B, and B results
from C, A results from C, then we would have been unable to assert our second
truism, viz. that moral properties always result from non-moral ones. They do,
but sometimes they only do so via intermediate moral qualities, as in the case of
the action which is good because it is generous.
What we know so far is that we cannot insist that the qualities from which
moral qualities in particular cases result should be general criteria for the ascrip-
tion of moral properties, for there is no shortlist of properties from which moral
qualities result. Nor can we in general say, where an action has a moral property
M in virtue of having non-moral properties, A, B, and C, that the presence of A, B,
and C count generally as reasons for calling actions M. The transitivity of
resultance makes it the case that there are many abstruse properties from which
sometimes moral properties result; we are not going to insist on that account that
the presence of those properties counts as a reason for moral assessment. It is not,
that is to say, the case that noticing the presence of these qualities ought to
   29
increase one’s confidence that the action is good. As we continue down the train of
resultance, we reach properties that are very unlike such qualities as being
conducive to human happiness, being commanded by God or whatever else
may be thought of as being intrinsically relevant to moral assessment. And so
we cannot say in general that where an action has a moral property M in virtue of
having non-moral properties, A, B, and C, then the presence of A, B, and C count
generally as reasons for calling actions M.
There is another and perhaps a more conclusive consideration which defeats
the identification of the properties in virtue of which a particular action is good
with reasons in general for calling actions good. When I decide that an action is
good, I see it as being good in virtue of my recognition of other non-moral
properties which the action has. These properties, perhaps, count as my reasons
for calling this action good, and in making my decision I shall probably focus on a
few special properties among all those from which, in the light of the transitivity of
resultance, the goodness of this action results. My reasons for calling it good, then,
are standardly to be found among the qualities from which the goodness results.
Can we say that my reasons for calling this action good must be generalizable, i.e.
that I must also call good any action similar to it in those respects? In general we
cannot say this; but it all depends on how similar the second action is to the
original one. If my reasons for calling the first action good are that it had qualities
ABC, I am not committed to calling any action similar to it in those respects, i.e.
which also has qualities ABC, good. For the second action might have some
further quality D which defeats any tendency that ABC had to make the action
good. Thus if my reasons for calling this girl good are that she is chaste and pious,
I am not committed to calling another girl good if, for example, besides being
chaste and pious she is also cruel. The natural tendency at this point is to revert to
the first girl and insist that her lack of cruelty must have been among my reasons
for calling her good. But I do not think this is so. I do not call someone good
because he is not cruel, though I may refrain from calling someone good on the
grounds that he is cruel. An argument of the same structure can be used also to
defeat the belief that if an action has a moral property M-ness which results from
its non-moral properties ABC, then any action which is ABC must have M-ness as
well. The properties ABC that go to make the action M may only do so in the
absence of some defeating property D-ness. If D-ness is present in the second
action, that action will not be M. But we cannot insist that the absence of D-ness
must have been one of the properties that made the first action M, just as (and this
is an analogy only) the facts that cause me to be writing now can only do so in the
absence of a nuclear explosion, but such an absence is not for that reason among
the causes of my writing. So it is not the case that where a moral property results
from a particular set of non-moral properties, any action that has that set of non-
moral properties will also have the moral quality. And it is not the case that where
my reasons for calling an action good are that it has properties ABC I am
committed to calling any other action which has the properties ABC good.
30     
It is not difficult to find people who have fallen into error on this point. Here are
some examples:
Hence to give a reason in support of the judgment that a certain individual, A,

ought or has the right to do some act, presupposes that anyone with the
characteristics specified in the statement of the reason ought or has the right to
do the same kind of act in a situation of the kind specified. (Singer 1967: 24)
Now if it is the possession of certain natural features which makes the action
wrong, then any other action which had just those natural features would also be
wrong. In this sense all moral judgments are universalisable.
(Swinburne 1976: 8)
If I say ‘x is good’ I must be able to say that any x which has the same relevant
qualities or properties as x is also good, because we judge x to be good for having
those qualities or properties. (Kovesi 1967: 158)
Sometimes, as in Swinburne’s case, the view I am denying is called the universal-

izability of moral judgements. If this is what universalizability is, it is false. But
Professor Hare’s way of formulating the doctrine is apparently different:
[A]ny singular descriptive judgment is universalisable . . . in the sense that it

commits the speaker to the further proposition that anything exactly like the
subject of the first judgment, or like it in the relevant respects, possesses the
quality attributed to it in the first judgment. (1963: 12)
The phrase ‘exactly like . . . or like it in the relevant respects’ looks like an attempt
to have two doctrines in one. Admitting that no two objects can be exactly alike,
let us take it that Hare’s true intention is that the speaker is committed to a similar
description of anything like the first object in the relevant respects. What does
Hare mean by this phrase, and does it offer a better notion of universalizability?
Unfortunately, Hare’s practice reveals that he, at least, means no more by this
phrase than the position we have already considered and rejected:
If a person says that a thing is red, he is committed to the view that anything
which was like it in the relevant respects would likewise be red. The relevant
respects are those which, he thought, entitled him to call the first thing red.
(1963: 11)
If he makes a moral judgment about one object, this must be in virtue of the
possession by the object of certain non-moral features . . . and therefore any other
object which possesses those features must have the same moral judgment made
about it. (1963: 20; cf. also p. 21)
   31
I think, however, that it is possible to find in the notion of a relevant respect

something more promising than what has so far been unearthed, and something
which Hare’s use of the phrase ‘exactly like’ makes one wonder whether he was not
implicitly considering. The purpose of introducing the phrase ‘relevant respects’
was to broaden the basis for comparison between the first action and subsequent
ones, in such a way as to escape the possibility of a subsequent action having a
non-moral quality which defeats in its case an attribution of the same moral
quality (viz. a quality such as D-ness in the previous example). It seems to me that
the phrase ‘similar in relevant respects’ has two promising interpretations. The
first is ‘similar in respects relevant to the moral properties of the first action’
where, in order to make such respects more wide-ranging than are the speaker’s
reasons or even the properties from which the moral properties result, we define
the notion of a relevant respect in this way: the qualities relevant to the ϕ-ness of a
ϕ object are those in virtue of which it is ϕ or which if it had had, or lacked, it
would not be ϕ; or which if it had had or lacked, it would have been more or less ϕ
than it is. Using the previous example of the girl who is good because she is chaste
and pious, the fact that she is not cruel is relevant to her goodness though it does
not count among the reasons why we call her good or among the qualities from
which her goodness results. So we are at least doing better than before. But
nonetheless, the doctrine of universalizability constructed using this sense of
‘relevantly similar’ is also false. For it may still be the case that the second action
is similar to the first in all respects relevant to the goodness of the first, but is still
so specially circumstanced that in its case those qualities do not suffice to make it
good. To see this, suppose that two subjects S¹ and S² share the qualities ABC in
virtue of which S¹ is good. Then S² may have a quality D which is a defeater. Now
suppose that S¹ and S³ share all qualities relevant to S¹’s goodness. D must be
among these, and in general the list of shared qualities will be much longer,
though still containing only some of S¹’s qualities. Suppose we have this
situation then:
S¹ S³
qualities
ABC A¹B¹C¹ . . . D
ABC A¹B¹C¹ . . . D

E, F, G, H, I, J
Now, though no single quality in the list E–J can be a defeater by itself, and so
prevent S³ from being good, the combination of E and F may do so without the
absence of E or the absence of F being by itself relevant to S¹’s goodness, on the
present definition of relevance. It seems to me that any relaxation of that defin-
ition to exclude the possibility of two qualities combining to make a defeater will
turn our present conception of qualities relevant to S¹’s goodness into a concep-
tion too broad to exclude any of S¹’s qualities.
32     
The other interpretation of the phrase ‘relevantly similar’ is to be found in the

principle that where one action resembles another in all respects relevant to moral
assessment in general, both actions must be given the same assessment. I have
given reasons already for supposing that there is no shortlist of respects that are
relevant to moral assessment. On the account of relevance given earlier, where the
notion of relevance is built on the notion of resultance but rather wider than it, we
cannot know in advance whether a given property may or may not be relevant to
our moral assessment of some object which has it. So there is not too much
difference between a principle that talks of any property relevant to moral
assessment in general and one which talks of any property whatever, since any
property may, in the sense defined, be relevant to the moral property of some
action. If this is so, the requirement of universalizability amounts to no more, in
the only defensible version, than the requirement that the speaker recognize the
supervenience of the property he is attributing to the action, and even this
requirement is trivial unless we can devise a non-trivial concept of supervenience.
What is more, we can now see a further reason why an attempt to do this by
restricting the subvenient class to that of all morally relevant properties is unlikely
to be successful, since on one interpretation of this restriction we are sinking
supervenience into a false doctrine of universalizability, and on another we are not
in fact making any progress at all, since there is no difference between all the
properties of an action and all its morally relevant qualities.
4. Moral resultance and supervenience contrasted
I hope that the previous section may have helped to emphasize the difference
between resultance and supervenience. Admittedly, both in some sense concern
the relation between the moral and the non-moral properties of an action. Both
seem to provide us with a sense in which the moral properties depend on or are
‘fixed’ by the non-moral properties. Both in some way encourage us to look for an
account in which we can say that, case by case, the moral properties of actions
consist in their non-moral properties. But the dissimilarity between the two
notions is far more impressive than their similarities. The crucial causes of
dissimilarity are first that resultance is a relation between a moral property of an
action and only some of its non-moral properties, whereas supervenience con-
cerns itself with all the non-moral properties; and, second, that in talking of
resultance we are talking about the properties of the particular action at hand
without necessarily comparing it to any other action, whereas in talk of super-
venience we are always working from the moral property of one action (already
determined) to the moral property of some other action (or that action itself at a
later date?). This is why resultance offers much more convincing answers in the
three areas I mentioned, viz.:
   33
1. Supervenience is not really a relation between the moral and the non-moral
properties of an action, but is either a relation between moral and non-moral
properties in general or a consequence of whatever relation holds in the particular
case between moral and non-moral properties. It is dubious to talk of those
qualities on which this action’s goodness supervenes, just as it is dubious to talk
of those qualities from which in general goodness results.
2. Only resultance provides a genuine sense in which the non-moral proper-
ties fix the moral properties. Supervenience only tells us that once they are
fixed, they cannot change without a change in the non-moral properties and
that if all (?) those properties recur in another action they would ‘fix’ the same
moral property.
3. The intuition that there is nothing more to the moral properties of an action
than its non-moral properties (a sort of token identity theory in ethics) is one
which is only plausible if expressed in terms of resultance. Stated in terms of
supervenience it is simply unattractive. For the intuition
(I¹) In a given case, the action’s goodness just consists in all its non-moral
properties
cannot hold a candle to
(I²) In a given case, the action’s goodness just consists in those non-moral
properties in virtue of which it is good
because whatever the relation of ‘consisting in’, it is not one which holds between
supervenient and subvenient in a given case. In fact the attraction of that relation,
for semi-reductive purposes, is not one which we need supervenience to formu-
late. As a fundamental element in an identity theory, therefore, supervenience is
distinctly disappointing. And this should not surprise us, so long as we see reason
to adhere to our original definition of supervenience. For according to that
definition, it is possible where M is supervenient on C for an object to have
M without having any properties in C at all. This is unworrying in the special
case of moral properties, because of the independently probable absence of objects
with a total lack of non-moral properties. It is more contentious when we turn to
other potential instances of supervenience, for instance that of mental states on
physical states. But it is difficult to discover from the supporters of this super-
venience whether they intend it to include the thesis that no object can have a
mental state without having a physical state. If they do, then the supervenience of
the mental will be very much harder to establish, and not, I think, in any way
necessary to the purposes of the supporters of supervenience, whose main interest
in it is that it provides a further way in which the mental and the physical cannot
come apart. The non-existence of mental states in a non-physical object is
34     
established separately by the identity theory; supervenience is an addition to this

and does not entail it.
Are resultance and supervenience logically independent of each other?
Supervenience does not entail resultance, for we could know (in whatever way
such things are known) that one property supervenes on a class of others without
knowing that in each case some of those others will play a special part. But, as
I hinted above, I think that there is a sense in which resultance entails super-
venience, since supervenience appears as a consequence of whatever relation holds
between moral and non-moral properties of particular actions. Given a non-trivial
formulation of supervenience, we know that the relationship between a moral
property and those from which it is parti-resultant is necessarily repeated in
another object which resembles the first in all (?) non-moral respects. Not only
must the second object have the same moral property; it must have it for the same
reasons, since there is ex hypothesi no question of a defeater.
5. Epistemology
It has been held that the fact that moral properties are resultant (whether toti-
resultant or parti-resultant), has consequences for our epistemology. Thus Ross in
his later book (in which he supports parti-resultance) writes:
Now it is clear that it is in virtue of my thinking the act to have some other
character that I think I ought to do it. Rightness is always a resultant attribute, an
attribute that an act has because it has another attribute. (1939: 168)
He continues:
It is not an attribute that its subject is just directly perceived in experience to

have, as I perceive a particular extended patch to be yellow or a particular noise to
be loud. No doubt there are causes which cause this patch to be yellow, or that
noise to be loud; but I can perceive the one to be yellow, or the other to be loud,
without knowing anything of the causes that account for this. I see the attributes
in question to attach to the subjects merely as these subjects, not as subjects of
such and such a character. On the other hand, it is only by knowing or thinking
my act to have a particular character, out of the many that it in fact has, that
I know or think it to be right. (1939: 168)
It is not very clear which of the two aspects of rightness mentioned, resultance or
inability to be directly perceived, is leading here. Is Ross contending that since
rightness is resultant, it cannot be directly perceived, or that since it cannot be
directly perceived, it is resultant? The first is the stronger; but neither line is
   35
acceptable. One reason for a bogus distinction between colours and moral qual-
ities with respect to resultance is the view that if a quality is resultant it can never
be perceived entirely on its own account, but only via awareness of the qualities
from which, in this instance, it results; since colours can be directly perceived, they
cannot be resultant. And another similar reason derives from the fear that if moral
qualities are truly perceivable, we should be able to learn the meanings of moral
terms by pure observation and ostension, which would remove from them the
elements of personal decision, action-relevance, sensitivity to conscience, etc.
Observability itself has no such consequences.
It is however a simple matter to show that a quality may be resultant but be
perceivable without any awareness of the qualities from which, in this instance, it
results. The quality of weakness is resultant. A chess position cannot be simply
weak; it must be weak because of some other characteristic. But a chess expert can
see weakness in a position without yet being able to see the particular failing in
which its weakness consists. Maybe he cannot see the weakness directly in the
sense that he sees from the shape of the position that it is weak. But this is
irrelevant to the question whether resultance is compatible with direct perceiva-
bility, for that question gives a different sense to the notion of direct perception.
For our purposes, what we want to know is whether a resultant property can ever
be perceived without some awareness of the qualities from which it results. And
the answer is clearly that it can. In the case of moral properties, then, there seems
to be no reason to argue from their resultance to the conclusion that they cannot
be perceived in their own right. We may of course still wish to accept this
conclusion for other reasons, although probably in that case we will give a
different sense to the phrase ‘in their own right’, i.e. we will make different
exclusions.
As far as resultance can take us, then, we may talk of seeing that the position is
weak, or seeing that the action is right. A quality can be resultant and directly
perceptible, or resultant and only indirectly perceptible, or non-resultant and
directly perceptible. So are there no relations between the epistemology of moral
concepts and their ‘logic’? To point to what is possible is not to reduce the
importance of what is typical. It may typically be the case that colours are
perceived in their own right, though this will not prevent colours from being
resultant. We know empirically that colours are parti-resultant from other phys-
ical properties, and supervenient on a certain sub-class of physical properties. The
contrast between resultances known a priori and those known empirically does
not seem to me to be crucial; on general Quinean grounds, I take it to be based on
a difference of degree rather than of type, although of course a difference of degree
is still a difference. Again, it may typically be the case that in moral judgement we
consider properties that are non-moral on which to base our evaluations. Is it
because of this that we know that moral properties result from non-moral
properties? It is important to see here that the fact adduced, namely that our
36     
reasons for moral judgements are standardly the presence of non-moral proper-
ties, falls short of establishing between the moral and the non-moral a transitive
relation of resultance, since the resultance-relation is much broader than the
reasons-relation. How then do we know that moral properties result from non-
moral ones if we cannot infer this from the fact that non-moral properties count as
reasons for moral judgements or that we cannot directly (in some sense) observe
the moral? It looks as if we shall have to make a bald claim to a priori knowledge of
moral resultance. With such knowledge, we can prove the supervenience of moral
properties, if we want. Alternatively, we can claim independent a priori knowledge
of supervenience. Neither approach seems very objectionable in the case of moral
resultance and supervenience.
6. Summary
The main aim of this paper has been to separate the notions of resultance and
supervenience; perhaps the central element of that separation is the view that
supervenience is not a relation between moral and non-moral properties in a
given case, but rather a consequence of that relation, whatever it may be. And
I gave reasons in Section 4 for finding resultance a much more profitable notion
from the point of view of moral theory; here and elsewhere I have had in mind the
attractions of an intuitionist theory of ethics, but I hope that the points I make
would be acceptable regardless of one’s general ethical standpoint. Theorists of
different standpoints should be able to agree on whether moral properties are
parti-resultant or toti-resultant, whether and in what sense they are supervenient
or universalizable etc., despite differences in the ways in which these claims are
interpreted in the different classical ethical theories.
A secondary theme in the paper has been the attempt to keep supervenience
separate from universalizability and so open the prospect of accepting the former
while rejecting the latter. Confusion here seems to me to have earned for univer-
salizability an unjustifiably easy ride.²
² I thank Harry Lewis and David McNaughton for their attempts to improve this paper.
3
Ethical Particularism and Morally Relevant
Properties
This paper is about the non-existence of moral principles. Its conclusion is a

thorough particularism, according to which our ethical decisions are made case by
case, without the comforting support or awkward demands of moral principles.
Defence of this position occurs at the end, unless it be a defence to show how a
position naturally arises from the defects of its competitors.
There have been ethical particularists in the past, but their names are hardly
legion. E. F. Carritt cast doubt on the need for and possibility of moral rules. Sartre
is familiar for the doctrine that each choice is a new one which must be made
without insincere appeal to the authority of former choices. Some remarks of
Prichard’s in his ‘Does Moral Philosophy Rest on a Mistake?’ (1912) make it
appear that he was sensitive to the attractions of the sort of position I shall outline.
It is even possible, here and there, to appeal to the numinous shade of
Wittgenstein.¹ And John McDowell’s distrust of the role of principles in moral
theory is at last beginning to be articulated. But by and large particularism has
suffered neglect; and failing neglect, abuse. Sidgwick mentions it as a form of
intuitionism, and quickly passes on. Rashdall styles it ‘Unphilosophical
Intuitionism’ and says that it can hardly claim serious refutation; but as is
common when a writer is faced with a view he thinks derisory, the refutation
which Rashdall then provides is hardly serious. Frankena calls it ‘Act-Deontology’;
but most of his criticisms do no more than assume its falsehood.² All in all, the
history of ethical particularism is far from edifying. That things should have been
better, I hope to show by a sort of progress which starts from utilitarianism and
eventually issues in full-blooded particularism.
Utilitarianism could be conceived as compatible with intuitionism, on the
grounds that the former offers a criterion for truth in moral judgement and the
latter offers an account of the origin of that criterion. But in the original
battle between the two views, the crucial difference was about the number of
unreducible ethical principles. The Utilitarian was a monist, asserting the exist-
ence of only one principle. The Intuitionist was a pluralist, asserting the existence
of many.
¹ Cf. Carritt (1928: 13); Prichard (2002: 18–20); Wittgenstein (1969: §26).
² Cf. Sidgwick (1907: 99–100); Rashdall (1907: 80–3); Frankena (1963: 15).
38     
The utilitarian monist enjoys what is in many ways a healthy philosophical

position. From the perspective with which this paper is concerned, his position
has three advantages; though the importance and indeed the relevance of these
advantages may not become clear until we have seen how his competitors get on
without them. First, his approach is consistent, in the following sense: it will never
yield contrary recommendations in a single case. For if there is only one principle,
then perhaps only one property needs to be thought of as morally relevant; the
question whether an action is our duty will depend only on the degree of presence
or absence of that property at the end of the day. That other properties may affect
the degree of presence of the crucial property is beside the point; we may if we
wish think of these as secondarily relevant in some way, but we do not need to. The
monist can get by with just one morally relevant property. Second, and conse-
quently, this approach gives a possible account of what it is for a property to be
morally relevant; a property is morally relevant iff the degree to which an action is
right is determined by the degree to which it bears that property. Third, this
approach offers room for various epistemological positions. We can claim direct
knowledge of the Great Principle as a self-evident truth; we can claim to discover
the truth of the Principle by intuitive induction from particular cases; or we can
claim to know it as a contingent generalization from past instances. So in these
respects, at least, the utilitarian monist’s position is healthy.
But monism, of the utilitarian variety or otherwise, suffers from one great
disadvantage, which here I assert without argument. It is false. It just is not the
case that there is only one morally relevant property, nor is there only one Great
Principle in ethics.
Are the two bald assertions in the last sentence, for which in the present context
I make no apology, assertions of the same form of pluralism? I shall take it here as
I think all pluralists would, that there is a one–one correspondence between moral
principles and morally relevant properties. Pluralists generally assume that if a
property tells in favour of an action being a duty, it will tell in favour of any action
that bears it. For such a property ϕ, there will be a moral principle ‘Promote ϕ
acts’, and any such principle will mention a property which is morally relevant.
If monism is false, whether in the utilitarian version or otherwise, pluralism
must, it seems, be the answer. But ordinary pluralism seems to have none of the
advantages of monism, while its disadvantages are equally severe. First is the
problem of consistency. How is it possible for an agent with two ethical principles
whose recommendations conflict in a particular case to retain both of them after
the struggle? On the face of it, where one holds two principles which say respect-
ively ‘Promote ϕ acts!’ and ‘Eschew ψ acts!’, when we have an act which, as is
possible, is both ϕ and ψ, something has to give. (If one’s principles were of
indicative mood, an explicit contradiction would be generated, though this may
not be the best way to see the problem.) How can the pluralist make sense, then, of
the co-tenability of two principles which suggest conflicting answers in an
  39
awkward case? Something must be said about the nature or logical content of
moral principles to make it possible for him to surmount this obvious challenge.
The second difficulty for pluralism is found in the list of moral principles. If this
list is indefinitely long, then quite apart from other problems we may have in
understanding its nature we are immediately saddled with an epistemological
choice. How are the principles known? The pluralist may adopt either a generalist
or a particularist epistemology; he may say that the principles are directly known,
or that they are ‘seen in’ particular cases.
Generalist epistemology has often taken the view that the principles are self-
evident, but in my view the longer the list of principles, the less the likelihood that
all of them are self-evident. (Remember that Sidgwick’s attempt at pluralism
foundered on the twin demands that all principles be both self-evident and
consistent.) Suppose we start with a short list of extremely simple principles
such as ‘Actions that cause pleasure are right’. We can immediately recognize
that some actions that cause pleasure are none the less wrong, and this leads us, as
pluralists, to complicate our principles in the hope of catering for the recalcitrant
cases. But quite apart from the difficulty of being reasonably sure that we have
reached a formulation which will prove invulnerable in the future, there seems to
come a point where it is no longer plausible to suppose that the elaborate
principles to which we have come are self-evident. We might agree, after consid-
eration, that they are true; but this is nothing to the point. Still, perhaps we can
retain the generalist view that these principles are directly known rather than
derived from knowledge of particular cases, without asserting that they are self-
evident, in any sense in which that term has been properly understood (e.g., they
are not assented to as soon as proposed, nor assented to by all those who
understand them).
Let us suppose that some form of generalist epistemology can survive the lack of
self-evidence. The problem is to relate our general knowledge of principles to what
we want to say about the moral properties of particular cases. The particular cases
here cannot be seen as tests for the evaluation of principles, without abandoning
generalism in epistemology. Principles are not like theories, for theories explain
what is true in particular cases without determining it, while principles determine
what is true in particular cases and explain it. So we discover what our particular
duties are by relating our general knowledge to the nature of the particular case. In
which case it is difficult to suppose that the nature of our particular duties, as
revealed by this procedure, could ever cause us to reassess the principles from
which those particular duties flow.
Those who, with me, find difficulties in such generalist epistemologies even
after they have been shorn of claims for self-evidence may do so because they
cannot imagine a situation in which particular cases are not tests for principles, as
on a generalist approach they cannot be. All other accounts start from our
knowledge of the particular case, saying that we ‘see’ the principle in the particular
40     
case by intuitive induction, or that we derive it by ordinary induction from a

number of particular cases. This attempt to move from the particular to the
general raises the same problem as that raised by generalism—a problem which
is crucial to the argument of this paper. How are we to match what we want to say
about what it is for a property to be morally relevant in a particular case to what
we want to say about what it is for it to be generally morally relevant? To rely on
intuitive induction is to suppose that we can see in a particular case that a property
is generally morally relevant. How this is to be done, unless particular relevance is
like general relevance, is a matter for speculation.
So the third and most crucial difficulty for pluralists is the account they are to
give of moral relevance. The previous monistic definition will not answer now, for
if ϕ is morally relevant so is ψ (in the above example) and they cannot both
determine the degree to which the action is right; if ϕ does so ψ does not. Nor is it
possible to define morally relevant properties in terms of the contents of moral
principles; this would achieve nothing. Still, our epistemological generalist prob-
ably thinks of morally relevant properties as those relevant to the moral value of
any act to which they belong. And the epistemological particularist will feel that if
a property is morally relevant in a particular case, this can only be because it is
generally relevant. The question is how to understand one or both of these
remarks so that they come out true. Most pluralists simply assume without
argument that there is a sense of ‘general relevance’ which will do the trick. But
if a property which is relevant to the moral value of a particular act is one which
actually affects the value of that act, so that somehow the act would have had less/
more value without it, what guarantee is there that all properties that we think of
as generally morally relevant will ‘make a difference’ in every case in which they
occur? For instance, it may be thought of as generally morally relevant whether
anyone derives any pleasure from the act. But quite apart from cases where the act
is the worse when the agent derives pleasure from it, are there not some where the
fact that someone derived some pleasure makes no difference whatever to the
moral worth of the act? Similarly, why should we admit that if a property ‘makes a
difference’ in a particular case, then it generally ‘makes a difference’? Isn’t it
possible that circumstances in a later case have the effect that the presence of
this property does not make a difference there, though it does here? I conclude
that pluralists have difficulty in providing a smooth account of the relation
between general relevance and relevance in a particular case.
I have presented these problems for pluralists as if at least the first is separate
from the others. The problem about consistency of principles might seem different
from that of the smoothness of our account of general and particular relevance
and from that of epistemology. But it seems to me that at least from the particu-
larist point of view there are clear links between these areas, even if to some degree
they have to be taken separately. The problem of consistency of principles really
arises because of conflict in particular cases. Particularist epistemology tells us that
  41
moral knowledge comes from our knowledge of cases. One relevant thing we can
observe in a case is that two properties are militating against each other there.
Now what is it for two properties so to militate? One answer is to explain this in
terms of different ethical principles. It is conflict between principles that lies
behind conflict between properties in a given case. But this answer is uncongenial
to particularist epistemology; it is conflict between principles that needs to be
explained by conflict between properties relevant to particular cases, not vice
versa. A mistake here leads one to see the difficulty about the particular case
misleadingly in terms of a truth-functional contradiction between the conse-
quences of moral principles in particular cases. I only want to stress for the
moment what a surprising attempt this is; it is not really at all obvious that appeal
to conflict between principles explains conflict between properties in a given case.
This, of course, is a particularist’s complaint. The generalist will see the matter the
other way round. For him, inconsistency between principles is something that we
could have discerned before it was revealed in the particular case, for what the case
reveals is that something can be both ϕ and ψ—a possibility which we should and
could perhaps have noticed before. The co-instantiation of ϕ and ψ only creates
conflict because of the principles into which they are keyed.
Still, these remarks show that our three problems are not unrelated; and the
purpose of doing that is to make more sense of W. D. Ross’s attempt to revive
pluralism in the face of the problems (1930; 1939). The main characteristics of
Ross’s attempt are an intransigent particularist epistemology and a novel account
of the nature of moral principles in terms of the prima facie. In virtue of his
emphasis on the prima facie we could call Ross’s theory PF-pluralism; but it should
be remembered that all Ross is doing is defend pluralism against objections by
saying what moral principles really are. PF-pluralism holds that pluralism, properly
understood, is sound. The only real difference between pluralism and PF-pluralism
is Ross’s notion of the prima facie; and this notion is obviously designed primarily
to account for the problem of co-tenability of conflicting principles. But Ross
would hold, I think, that it also provides acceptable pluralist accounts in the
other two areas; in epistemology, it shows how we can start from the particular
case and shows what we can see there for our intuitive induction to get to work on,
while it also purports to provide a suitable general account of general and particular
relevance. And I have tried to show that the three areas are in principle related
closely enough for it to be possible for one radical shift to solve problems in all
three places. This is possible; but I shall be arguing that Ross does not achieve it.³
³ McDowell’s attack on principles fails to consider Ross’s work; which is a pity, since the notion of
uncodifiability on which he relies is only relevant to a conception of principles as true universal
propositions. Under Ross’s interpretation of principles, McDowell’s Aristotelian argument that par-
ticular cases will always escape the codifier’s attentions becomes, apparently, powerless. Cf. McDowell
(1979). Of course, Ross was only putting into order an approach to pluralism made by Prichard and,
long before, by Richard Price.
42     
How does the notion of the prima facie provide new pluralist answers to our
three problems? It accounts for co-tenability of moral principles after a conflict by
providing a new view of what a moral principle says (and this seems just what is
needed). Reading a principle ‘Promote ϕ acts’ as saying
(1) ϕ acts are right
lays one open to the charge of contradiction in a particular case if one also
holds that
(2) ψ acts are wrong
and that an action can be both ϕ and ψ.

But Ross suggests that moral principles are really of the form
(10 ) ϕ acts are prima facie right

(20 ) ψ acts are prima facie wrong
and that 10 and 20 can be so understood that an action which is ϕ and ψ can be prima
facie right and prima facie wrong, without incoherence. Whether this is so will depend
on the exact account given of the prima facie; but one might feel that something like
this is so obviously right that we won’t be too dismayed if our exact account needs
revision. So, given that things are looking promising in our first problem area, let us
turn to the other two. In these, again, it is quite easy to see what Ross is after. He wants
to say that in particular cases individual properties make the actions that bear them
prima facie right (wrong etc.), that any such property is generally a prima facie right-
making property, and that it is possible to see in a given case that here (and hence that
generally) this property makes its bearer prima facie right.
Ross adds to this the sort of particularist epistemology which I recommended
earlier for the pluralist. He holds that we come to know moral principles by
intuitive induction from the nature of a particular case, and that in determining
the nature of that case we can make no appeal to principles at all (except in a few
degenerate instances). According to him, the answer to the question what one
ought to do now is irredeemably a matter for particular judgement in which one’s
knowledge of principles plays no role.
But when I reflect on my own attitude towards particular acts, I seem to find that
it is not by deduction but by direct insight that I see them to be right or wrong.
I never seem to be in the position of not seeing directly the rightness of a
particular act of kindness, for instance, and of having to read this off from a
general principle—‘all acts of kindness are right, and therefore this must be,
though I cannot see its rightness directly’. (1939: 171)
  43
And knowledge of principles comes from knowledge of particular cases:
Their rightness was not deduced from any general principle; rather the general
principle was later recognised by intuitive induction as being implied in the
judgements already passed on particular acts. (1939: 170)
It is worth pausing to contrast this appeal to intuitive induction with the other
option open to the epistemological particularist, which is that we derive principles
from instances by empirical generalization, in a way that is supposedly familiar
elsewhere. Broad has well characterized this option (1930: 271); since many
instances are required, we can suppose a particular case offers in itself no reason
to suppose that the pairing of natural and moral properties would recur. We
therefore think that on this approach all that we can observe in a particular case is
mere coexistence of natural and moral properties. Now it would be wrong to
object to this that we know a priori that moral properties are supervenient upon
natural ones. For supervenience is a relation between all the natural properties and
the moral ones, and the pairings we were speaking of were pairings between moral
properties and some only of the natural ones (those from which they result). Of
course supervenience does not by itself establish anything resembling what we
normally think of as a moral principle. The true objection to this epistemological
approach is that moral principles are not empirical generalizations. This is so
despite the fact that generalizations can survive counterexamples without being
any the worse for wear, which would be useful for dealing with the problem of
conflict between principles. Thus it can be true that tigers have tails in the face of a
tailless tiger; this is not quite the same as two principles conflicting, but is one
aspect of it. If two moral principles conflict, there will be an object which has a
moral property that a moral principle says it does not have, without the moral
principle being rejected for that reason. Unfortunately, though, no matter how
close the analogy between moral principles and generalizations, we should not
take it that moral principles are generalizations.
The argument that moral principles are not generalizations runs as follows.
Suppose that we have a moral principle ‘Eschew ψ acts’, and a generalization that
tigers have tails. Some acts which are ψ may be right despite their ψ-ness, and in
some of these cases (though not necessarily in all) the rightness will be reduced by
their being ψ; they will be wrong qua ψ act, or the worse for being ψ, but still right
overall. Moral principles (and morally relevant properties too) are able, even when
defeated in a particular case by countervailing considerations, to linger or have
residual effects. But generalizations are not able to do this. A tiger that has no tail
is not somehow one which has a tail qua tiger, or one which has more of a tail for
its being a tiger, even though for other reasons it has not got one.
The alternative option, according to Broad, is that we derive a moral principle
by intuitive induction from a single case. The problem is to see how this is
44     
possible. For it to be possible it must be possible to discern a relation between

rightness and being ϕ in the particular case. This relation is somehow able to
create a modal truth, for in virtue of it we know that necessarily any object which is
ϕ is prima facie right. (Empirical generalization would only be able to establish
moral principles as contingent truths.) Any intuitionist then who believes that one
case is sufficient to establish a principle, or that principles are necessary truths, is
bound to opt for intuitive induction in his epistemology. The problem that faces
Ross is how to make this option stick, by giving an account of what is discernible
in the particular case that could enable a principle to be derived from it. So after all
these preliminaries I now turn to Ross’s explicit account of the prima facie, to see
whether it is able to provide what is needed. The two questions to be asked are:
(1) Does any notion of the prima facie make possible the view that we
discover moral principles in/from what we discern in particular cases?
(2) Does any notion of the prima facie make true the views that
a. If one action is prima facie right in virtue of being ϕ, then any ϕ action is
prima facie right in virtue of being ϕ.
b. If ϕ actions are prima facie right, then each ϕ action is prima facie right
in virtue of being ϕ?
For any conclusion as to whether Ross makes good sense of conflict between
principles ultimately depends on the answer to these questions.
Ross offers two accounts of a prima facie duty. The first is in terms of
tendencies; an action is a prima facie duty in virtue of being ϕ iff ϕ actions tend
to be duties proper. It is obvious, I think, that talk about tendencies can only be
cashed at the general level. For particular acts do not have tendencies to be duties
proper; they either are so or not. But for neatness’ sake I shall ask of this account
the two questions mentioned above.
(1) Epistemology. How could one perceive from a particular case that ϕ actions
tend to be duties proper? Since talk about tendencies cannot be cashed at the
particular level, nothing is said about what one observes when one observes that
an act is prima facie right for being ϕ. For the account mentions nothing that a
particular case could reveal.
(2) Relevance. Is there any reason to suppose that where ϕ actions tend to be
duties proper, something like this will be the case—that wherever an action is ϕ, it
will be the better for it? The moral principle ‘Do not forget your obligations’ is
consistent with an action’s being the better for its agent having forgotten his
obligations. For instance, if I promise to help you move house and fail to turn up,
it is better if I have forgotten my obligations than if I have not.⁴ In fact it seems
⁴ It may be that in this sort of case we first notice a moral principle which does not specify a property
which is generally morally relevant.
  45
obvious that this supposition is false. Nor can we be certain that the opposite move
is any better. If in a particular case some action is the better for being ϕ, does this
show that actions that are ϕ tend to be duties proper?
So I conclude that this first attempt is an obvious failure in the two crucial areas.
And by way of quick comment, I would say that it only seems to make sense of a
conflict of moral principles at the cost of making such principles too similar to
empirical generalizations.
But Ross’s second account of a prima facie duty (perhaps his official one) is
better:
I suggest ‘prima facie duty’ or ‘conditional duty’ as a brief way of referring to the
characteristic . . . which an act has, in virtue of being of a certain kind, . . . of being
an act which would be a duty proper if it were not at the same time of another
kind which is morally significant. (1930: 19)
Let us approach this account with our two questions.

(1) Epistemology. I take it that ‘morally significant’ is not significantly or rele-
vantly different from ‘morally relevant’. That being the case, how am I to see, in or
from a particular case, that if the action had no other morally relevant characteristic
it would be a duty proper? We need persuading that this is possible. If morally
relevant means ‘generally morally relevant’, Ross will certainly assert that if an
action is a prima facie duty in virtue of being ϕ, no property ψ can prevent that
action from being a duty proper unless ψ be a generally morally relevant property.
But on what account of general moral relevance will this be true? If a property which
is generally morally relevant is one which always ‘makes a difference’, why shouldn’t
a property which ‘makes a difference’ here fail to do so elsewhere, since other
circumstances deprive it of its influence? However, if ‘morally relevant’ means
‘relevant to the moral worth of the particular case’, it is easy to see how I am to
discern in that case that where ϕ ‘makes a difference’ then if nothing else ‘makes a
difference’ the only difference will be the one that ϕ makes. So Ross does here
provide an account of the prima facie under which we can discern a prima facie
duty in a particular case. And the account provided can be generalized. For it seems
inescapable that if a property would decide the issue if it were the only one that
mattered here, then it would do the same wherever that unlikely circumstance
recurs. For such a thought is insulated against the interfering effects of different
contexts. Thus, epistemologically at least, Ross seems to have succeeded; in his
official sense of ‘prima facie’, if one act of a certain type is discovered to be a prima
facie duty, we know that any act of that type is a prima facie duty.
But two questions remain, under the same general heading. Ross has provided
an epistemologically possible account, but is it a correct one? First, can we really
explain the behaviour of different properties in the more normal case where there
are several morally relevant properties by appeal to a case, which never exists,
46     
when there is only one? Why should a property which would decide the issue if it
were the only one that mattered be one which I should care about when it is not
the only one? Second, do we capture what is noticed when we notice a quality as
one which makes the act right, by saying that we notice that if no other quality
made any difference this one would decide the issue? Isn’t there still the danger
that a property could be such that, though if it were the only relevant one it would
decide the issue, it is not in fact among those which have any effect on the
outcome of all? For instance, it might be that if per impossibile, an action’s only
relevant property is that it causes pleasure to someone, that property would
determine its moral worth; but that in another, normal, complicated case that
property does not affect the issue either way. If this is possible, Ross’s epistemol-
ogy, though possible, is not correct. For what we notice when we notice that a
property does ‘make a difference’ is not identical with what we notice when we
notice that a property, if it were the only relevant one, would determine the issue.
(2) Relevance. It is an important and not merely verbal point that if ‘morally
significant’ is equivalent to ‘morally relevant’, Ross’s account of ‘prima facie’ does
not provide any independent understanding of moral relevance, since the term we
seek to understand occurs unexplained in the account, and cannot be removed
from it. But it is not specially because of this that Ross fails here to show that if a
property is relevant in one case, it must be relevant wherever it occurs. That failure
is really due more to what was pointed out in the immediately preceding para-
graph, that to say that a property would decide the issue if it were the only morally
relevant one is not to say that it is relevant even in the present case, let alone in
others. The essential generalist move from initial case to other instances thus goes
unsupported. And similar remarks can be made about the view that if a property is
generally relevant it must be relevant wherever it occurs.
So I conclude that Ross fails to show how a generally relevant property gets a
grip on a particular case; he fails to show how a particular case creates a generally
relevant property (and hence a principle, in favourable cases) and he fails to show
how I can come to know that a property is generally morally relevant from what
I can see ‘in’ a particular case.
I have gone into detail about Ross because I feel that his failure is instructive. It
is not just that it is the failure of the only detailed attempt to make sense of
pluralism. The difficulties he faces are ones which face any such attempt. What has
happened is that Ross has two independent inputs, which he attempts to reconcile
but which are essentially at odds with each other. The first of these is the
particularist epistemology about which he is so emphatic. The second is the
generalist view that what we learn from particular cases constructs general prin-
ciples, or that if a property is relevant anywhere it is relevant everywhere. The latter
requires that what is observable in a particular case must be shown as essentially
generalizable. But the more Ross leans towards generalism, as in the generalist
account of the prima facie in terms of tendencies, the harder it is for him to work
  47
his way back to the particular case so as to suit his epistemology. The more he turns
his attention to the particular case, the harder it is to find anything which we could
discern at that level and which would reveal any properties as generally relevant.
And this tension will recur in any pluralist view which adopts a particularist
epistemology. But pluralism cannot avoid this epistemology; the more it finds itself
asserting that we know a large number of ill-assorted principles, the less plausible is
the view that the principles are self-evident or in some other way discoverable
independently of particular cases. So I conclude that a particularist epistemology is
in the end inconsistent with generalism in the matter of moral relevance.
Even if the pluralist abandoned epistemological particularism, he would still
have to face the problem of how his moral principles are relevant to particular
cases. It is Ross’s merit that he attempts to meet this problem head on, even
though he does not solve it. The generalist’s problem is the same as Ross’s, though
it is viewed from the other side. Both want a smooth account of relevance, Ross so
that he can show how a particular case somehow creates a principle, the generalist
so that he can show how a principle manages to have any effect on its instances.
But in the end the reason why one cannot move from particular to general is the
same as the reason why one cannot move from general to particular.
The right solution, it seems to me, is to cling to the particularist epistemology
and abandon the generalist tendencies that are unable to be made consistent with
it. The position is eventually forced on us because after the discovery that more
than one property is morally relevant, we begin to admit a plethora of such
properties without there being any way of ordering them. When we face this
plethora honestly, we have to adopt a particularist epistemology, and thus reject
the view from which we came to it. As particularists, we give no sense to the notion
of a property being generally morally relevant, since we cannot relate this satis-
factorily to our epistemology; and hence we fail to understand the possibility of
moral principles. So the progress is from monism, the view that there is only one
moral principle, through pluralism, the view that there are many, to particularism,
the view that there are none.
What extra oddities does this particularism add to ethical intuitionism, a
doctrine widely held to be odd enough already? It should be noticed that our
epistemology is not significantly different from Ross’s; we discern directly that
individual acts are right, without needing any detour through principles. In Ross,
the drive to principles is not epistemological but metaphysical. He feels that an
individual act cannot be right without there being some principle behind it, as it
were. I am suggesting that we accept Ross’s epistemology and abandon his
metaphysics. Does this make the epistemology significantly odder? I cannot see
that it does. But I recognize that one activity in particular is commonly seen to
conflict with the particularism I have defended, and that is the giving of reasons
for moral judgements. Surely, it will be said, the giving of reasons is essentially an
appeal to moral principles.
48     
Before giving some idea of how, in my view, the particularist should reply to
this, I want to approach it by considering other criticisms of particularism which
I feel are misconceived.
Frankena argues as follows:
Now it is true that each situation has something new or unique about it, but it
does not follow that it is unique in all respects, or that it cannot be like other
situations in morally relevant respects. After all, events and situations are alike in
some important respects, otherwise we could not make true general statements of
a factual kind, as we do in ordinary life and science. Therefore, there is no reason
for thinking that we cannot similarly make general statements of a moral kind.
(1963: 21)
All that needs to be pointed out here is that Frankena assumes that there are such
things as ‘morally relevant respects’, which the particularist denies, and also takes
it that moral principles are reached by empirical generalization rather than by
intuitive induction.
A second criticism raised by Frankena is more common, but still fruitless. He
suggests that ‘it is impossible for us to do without rules’. By this, he does not mean
the unimpressive remark, so common in discussions of utilitarianism, that some-
times we don’t have time to work out the right answer and so have to rely on
principles as a short cut. Instead he appeals to a remark of Hare’s: ‘Without
principles we could not learn anything from our elders’ (1952: 61). This is the
more promising view that without principles moral education is impossible; in
fact, as Hare sees, it is not just teaching but also self-teaching that seem odd from
the particularist point of view. But are they odd for the same reason?
Rashdall claims that if there are no rough rules or principles of ethical judge-
ment, ‘moral instruction must be treated as absolutely impossible’. He elaborates:
We do not say to a child who asks whether he may pick a flower in somebody
else’s garden, ‘My good child, that depends entirely upon the circumstances of the
particular case: to lay down any general rule on the subject would be a piece of
unwarrantable dogmatism on my part: consult your own conscience, as each case
arises, and all will be well.’ On the contrary, we say at once: ‘You must not pick the
flower, because that would be stealing, and stealing is wrong.’ (1907: 82–3)
It seems to me that there are three sorts of anti-particularist attack in all this. The
first attack is some version of Hare’s doctrine of universalizability. The thought is
that particular cases create general principles; if Jones ought to do Z in this
situation, he ought to do Z in any similar situation. Here I cannot do more than
say that either this thought is straightforwardly false, or it amounts to conflating
universalizability and supervenience (cf. Dancy 1981; repr. as Chapter 2 in this
  49
volume); but supervenience does not create anything recognizable as a moral

principle (e.g., the universal propositions created by supervenience could not
conflict). The attack ‘If you see this in that way you must see that in the same
way too’ becomes for the particularist the question ‘How can you see those two so
differently when they are so similar?’. Perhaps this is a question which the
particularist may at any time be called upon to answer (yet why should he answer
it?); but it is also a question which it is always possible for him to answer.
The second attack is that particularism is unable to say anything about how past
experience can be called on to help us reach a decision in a new case. ‘Self-
teaching, like all other teaching, is the teaching of principles’ (Hare 1952: 61).
The idea here is that past experience is relevant because it produces an armoury of
pairings of natural and moral properties by which a decision in a new case can be
guided. How is it that past instances teach us what is right unless they offer a wide
selection of natural circumstances which I can be certain that rightness accom-
panies? And surely they do this by the gradual construction of principles, to which
I can appeal when I want help in coming to a decision.
I think that the particularist’s response here is that it distorts what is really a
very complex matter. If, as is suggested, we have been able to discern pairings of
natural and moral properties, why shouldn’t we simply do that this time? Why is
there any need for the detour through past cases and the principles they create? If
the answer is that we need help in this situation because we find the moral
property hard to discern here, the particularist can say that this itself is evidence
that the new case is not relevantly similar to the old ones; in moving from past to
present we come up against all the difficulties about moral relevance.
But still, isn’t past experience even able to be relevant to new decisions? To deny
this would put the particularist beyond the pale; he must accept it but offer a rival
explanation.
One such explanation appeals to Wittgensteinian thoughts about what counts as
going on in the same way, and the kind of necessity that binds previous instances to
a new one. Competence with a moral concept (e.g., generosity) is knowledge of a
rule, not a moral rule but a rule whose grasp is simply the ability to carry on using
the word ‘generous’ correctly in new instances. Someone who comes to a new
situation knowing what generosity is is someone who has learnt a rule (here the
importance of experience) and his knowledge of the rule is manifested now in his
decision that this situation is another of the same sort (here again the importance
of past cases). But there need be nothing one can point to in the past cases which
can determine or even guide his choice; what makes his choice right is not that it is
dictated or even made probable by principles created by the past instances, but
simply our acceptance of the choice as an instance of carrying on as before.⁵
⁵ This answer, like that to the next objection, needs a lot of filling out which space here precludes.
(But cf. McDowell 1979 and 1981.)
50     
The third attack is the view that the giving of reasons is essentially an activity of
generalization. Frankena says, ‘If Jones answers your question “Why?” by saying
“Because you promised to” . . . he presupposes that it is right to keep promises’
(1963: 23).
The force of Rashdall’s last sentence is not that we always do add a moral
principle, but that if we had merely said ‘You must not pick the flower, because
that would be stealing’, we would be committed to the principle that stealing is
wrong. Now the particularist denies this, clearly. So it will not do just to assert it
against him. But there is a genuine question for the particularist here. Since we do
go in for giving reasons, then if this common activity is not an activity of
generalization, but something restricted to the confines of the particular case,
what is it? What is going on when I say, to others or myself, ‘This action is wrong
because it is ϕ’? Here I am not just reacting to the particular situation as a whole,
but picking out some features of it as particularly significant (here). What is this
activity, and can the particularist explain it without abandoning his position?
The direction in which I think the particularist should move is to compare the
activity of choosing some features of the particular situation as especially salient⁶
(significant) with the activity of the aesthetic description of a complex object such
as a building. In such a description, certain features will be mentioned as salient
within the context of the building as a whole. There is no thought that such
features will be generally salient; they matter here and that is enough. Someone
offering the description is telling his audience how to see the building; he is doing
this by selecting for stress just those features which must be salient if someone is to
see the building the right way. He picks these features out, but knows that their
importance cannot be assessed or even discerned by someone who cannot see the
whole building. One could not (and here is one important feature of the analogy)
discover how the building was just by considering its salient features; salient
features are not epistemological clues, and by this analogy reasons are not clues
either. The man who provides reasons is not so much providing evidence for his
ethical judgement as trying to show his audience how he sees the situation. He
supposes that to see it his way is to join in with his judgement about what is right
and wrong; so if you do come to see it his way you will agree with his ethical
judgement, but by giving his reasons he is not arguing for that judgement, in the
way in which adherents of moral principles might suppose.
I hope that this brief discussion of the difficulties for particularism helps further
to characterize the sort of view I would recommend.⁷
⁶ This notion of salience first occurs, as far as I know, in Wiggins (1978).

⁷ I would like to thank David McNaughton for helpful comments on earlier drafts.
4
Intuitionism in Meta-Epistemology
This paper is a response to William P. Alston’s ‘Meta-Ethics and Meta-

Epistemology’ (1978), in which he investigates various ways in which distinctions
drawn and positions outlined in meta-ethics may be rediscovered in more or less
analogous form in meta-epistemology. Alston takes it that meta-ethics is in a more
advanced state than is meta-epistemology, and in particular that its practitioners
are more self-conscious about the nature of their utterances, theories and meth-
odologies. I agree with him in finding the relationship between the two areas an
attractive and promising subject, as I hope that this paper will show. But I think
that his remarks can be improved upon by a more attentive look at ethical
intuitionism; in particular, I think that there are forms of intuitionism which
Alston has neglected, and that the forms he does offer are not clearly delineated.
Alston rules out ethical non-cognitivism as being without serious parallel in
epistemology, although he does cite J. L. Austin as a possible non-cognitivist (and
I think that J. O. Urmson ought also to be mentioned). He therefore finds the main
division in ethics to be between the cognitivist approaches of the naturalist and the
intuitionist. Now there have been various areas of contention between these two,
and it is important before going on to identify these and to establish which is the
most important. One area—the one on which Alston concentrates—is the ques-
tion of the definability of ethical predicates in terms of non-ethical predicates. The
naturalist asserts this definability and the intuitionist, it seems, rejects it. In fact,
ever since Moore’s Principia Ethica (1903) the naturalist has been thought of in
terms of a hopeless attempt to cross the fact–value gap (or some similar error). But
there is a second, and in fact more important, area of dispute. This concerns the
question of the number of irreducible ethical principles. The naturalist who
identifies the moral property of rightness with some, albeit complex, non-moral
property ϕ finds himself asserting, in effect, that there is only one moral principle:
promote ϕ acts! But the intuitionist is known for the view that there are a number
of irreducible principles, and not necessarily a small number either. Sir David Ross
argues at length against the utilitarian on these grounds, and it is in this sense of
intuitionism that J. O. Urmson is willing to support the intuitionist. So here we
have a second area of dispute between the intuitionist and the naturalist.
Which is the more important? Unfortunately, as it turns out for Alston, it is the
second, for it is the intuitionist’s pluralism in the matter of the number of ethical
principles which determines his position on the question of the definability of
ethical predicates. Alston draws his contrast thus:
52     
[T]he Naturalist holds that ethical terms can be defined, explicated, or analyzed
in ‘factual’ terms, that they at bottom are factual terms . . . The Intuitionist, on the
other hand, maintains that ethical concepts are sui generis, that they are of a
distinctively and irreducibly normative or evaluative sort, not to be reduced to
matters of fact, however complex. (1978: 276)
In a way this is right enough; but why does the intuitionist take up this position?
Alston presents him as impressed by, and in fact as wielding, the distinction
between fact and value; in his view the intuitionist’s complaint against the
naturalist is the same as the non-cognitivist’s, for they both accuse the naturalist
of attempting to cross the uncrossable. But this is quite the wrong way to see the
intuitionist’s denial of the definability of ethical terms. It ought to be possible to
characterize the intuitionist’s stance here without appeal to the fact–value gap,
since many modern intuitionists reject that gap as misconceived. The intuitionist’s
argument against the naturalist on the question of definability is not that he
crosses the fact–value gap, but that he creates order where there is none, by
finding some single (maybe complex) property in virtue of which all right actions
are right, when in fact there are at the least several different properties which make
right actions right, and hence several different moral principles. It is no use for the
naturalist to retort that even if this is so the disjunction of such properties will be
identifiable with rightness. For the intuitionist need not hold that the disjunction
is in any manageable sense finite; and even if he admits that it is finite, he can
claim that rightness is in some intuitive sense a single property, whereas the long
disjunction of properties by virtue of which actions get to be right cannot be seen
as a single property at all. The natural bases from which different actions derive
their rightness bear no generalizable relations of similarity to each other. Holding
therefore in this sense that moral properties are naturally shapeless, the intuition-
ist argues against the definability of ethical terms without admitting the existence
of a fact–value gap.
What is more, it would be wrong to use the fact–value gap so as to characterize
the naturalist as holding that moral properties are not evaluative but factual: this
position can only be wished on those who see, as the naturalist does not, the sort of
contrast between the evaluative and the factual that is sustained in the fact–value
gap. The naturalist’s position is rather that moral properties are evaluative and
factual.
So on both these counts it seems to me that Alston is wrong to take the fact–
value gap as the most revealing criterion in the dispute between naturalist and
intuitionist. In fact the crucial dispute is between the intuitionist’s pluralism and
the naturalist’s monism; positions on the issue of definability are a consequence of
positions on the monism/pluralism debate.
With this under our belts, let us turn to the search for epistemological coun-
terparts to ethical theories. Alston’s candidate for an epistemological counterpart
  - 53
of intuitionism is the justified true belief (JTB) conception of knowledge; accord-

ing to this view the fact that S knows that p consists of at least three components:
(1) it is true that p

(2) S believes that p
(3) S is justified in believing that p.
Alston’s reason for taking this to be an analogue of intuitionism is that

‘justified’ in (3) is ‘naturally construed as an evaluative term’. I hope that my
remarks in the previous paragraphs raise some doubts about this reason.
A naturalist might admit that ‘justified’ is an evaluative term, but claim that it is
also factual. Admittedly he does not say that it is irreducibly evaluative; but neither
can we say straight off that JTB theorists take it to be irreducibly evaluative. Some
JTB theorists want to give a causal account of justification, and indeed Alston
wants to see them as identifying justification with the satisfaction of the conditions
specified in their causal account (see below). So the fact that ‘justified’ is evaluative
does not warrant Alston in classifying JTB theorists as intuitionists yet. Indeed, if
the distinction between intuitionism and naturalism is rather to be drawn in terms
of the shapelessness or otherwise of the natural base of ethical properties, the
question of whether JTB theorists are epistemological intuitionists is to be decided
by whether they take the natural base for justification to be shapeless, i.e., by
whether they take there to be natural similarities between all justified beliefs or
whether they think that there is no way of capturing, in a single factual phrase,
necessary and sufficient conditions for justification. In short, whether they are
monists or pluralists.
Now in fact R. M. Chisholm, a leading JTB theorist, looks like a pluralist. He
offers a large number of different epistemic principles (principles of justification)
without attempting to derive them from a common source (see esp. Chisholm
1977). So I agree with Alston’s conclusion here, that Chisholm is a paradigm
epistemological intuitionist.
Taking JTB theorists as intuitionists, Alston offers as naturalists the proponents
of the causal theory of knowledge (CTK). But the difficulty here is that Alston
takes the contrast between naturalist and intuitionist to be pretty stark; the
weakest version of the contrast in ethics is that one takes ‘good’ to be evaluative
and the other takes it to be descriptive. And it is very surprising to find it suggested
that there is a similar contrast between JTB theory and CTK. What about those
writers who take CTK to be a definition of ‘justified’ as it occurs in (3), hoping by
this causal account to bolster the JTB conception up against Gettier-type objec-
tions? It is hardly plausible to suggest that they are involved in some radical
confusion. If on the other hand we weaken the contrast between naturalist and
intuitionist, as I have been suggesting against Alston, we can begin to make more
sense of his proposal to see proponents of CTK as naturalists; in fact the nearer
54     
naturalism comes to being a special case of intuitionism, the more plausible this
attribution becomes. The contrast between monism and pluralism is just the sort
of contrast required here, for JTB theorists who reject CTK may do so on the
grounds that although beliefs caused in the specified way are indeed justified,
beliefs can be justified in other ways as well.
So again I agree with Alston’s conclusion here, but for a different reason.
Mine is that proponents of CTK find a common pattern or shape in all justified
beliefs, while Alston’s is that they offer definitional accounts of justification
which are of a factual type (by holding a belief justified iff caused in a certain
way). It is worth remarking in this connection that some proponents of CTK
take themselves to be offering not a definitional account of ‘justified’ at all,
but a substantial normative theory about the conditions which make a belief
justified. A. I. Goldman takes this position, for example (1979: 1). So Alston’s
remark (1978: 227) that the CTK is ‘best presented’ as a definitional theory
seems to me not to reflect any improved insight into the essential nature of
CTK, but merely to persuade us of the neatness of his classification of epistemo-
logical theories.
Alston goes on to present some results of the application of these analogies to
meta-epistemology; which are themselves a partial vindication of his method of
dividing up the subject, even if I am right in holding that method to be mistaken.
But I want now to divert attention to a particular issue in this area. Since, for the
purposes of instructive analogies between ethics and epistemology, it is of the first
importance that the distinctions within ethics be drawn correctly, and since
I think that Alston has failed to notice some forms of intuitionism which might
be of help here, I turn to further examination of the crucial issue, that of the
contrast between ethical monism and ethical pluralism.
What are the different forms of ethical pluralism? Up to now I have offered two
descriptions of the pluralist, as believing that there is more than one moral
principle, and as believing that moral properties are naturally shapeless. Let us
work for the moment with the former characterization, and reintroduce the latter
at a suitable point later on.
Alston discerns two forms of pluralism, distinguished by their epistemology.
The first he calls the ‘self-evident first principles’ form, and the second the ‘moral
sense’ form (1978: 283). According to the latter we ‘start with immediate know-
ledge of particular instances . . . and then determine what general principles will
cover those cases’. According to the former, general principles are presumably
apprehended directly, as self-evident. Now which ethical writers can we saddle
with these positions? If these are the only two options, it is easy to see some
candidates for moral-sense theory, but not so easy to find any self-evident first
principles theorists. Perhaps Sidgwick and Moore were impressed enough by self-
evidence to count in the latter class. Ross, Prichard, McDowell and Wiggins will all
count as moral-sense theorists.
  - 55
Let us stay with this dichotomy for a short while, before looking deeper; some
results already emerge. It is characteristic of writers who feel that we work from a
knowledge of self-evident first principles to a decision in particular cases to hold
that the right answer in the particular case is never self-evident, nor even some-
thing that we can know. Moore holds that ‘fundamental principles of Ethics must
be self-evident’ (1903: 143 [1993: 193]), but that when we attempt to move from
these principles to decide whether a particular action is our duty, ‘we never have
any reason to suppose that an action is our duty: we can never be sure that any
action will produce the greatest value possible’ (1903: 149 [1993: 199]). Ross
agrees that, though the general principles of prima facie duty are self-evident,
from these ‘we come to believe something not self-evident at all, but an object of
probable opinion, viz that this particular act is (not prima facie but) actually right’
(1930: 33).
This unwillingness to hold that self-evidence at the level of principles does
anything to make our decisions in particular cases firmer is interesting when we
attempt to draw the analogy between ethics and epistemology. There seems to be
present among epistemologists the hope that if we could only establish the general
epistemological principles that lie behind particular cases, this would give us
operable criteria in the sense of necessary and sufficient conditions which we
can use to determine e.g. whether and in what circumstances we have knowledge
of the future. Chisholm says, in the middle of a very revealing passage, that ‘we
suppose that by investigating . . . instances we can formulate criteria which any
instance must satisfy if it is to be accepted or permitted, as well as criteria which
any instance must satisfy if it is to be rejected or forbidden’ (1977: 17). But it seems
that if the analogy between ethics and epistemology is to be trusted, the establish-
ing of such criteria (which might be thought of as the discovery of epistemological
principles) will not have any very striking effect; in particular it will not succeed in
producing any accredited instances of knowledge. And this conclusion is import-
ant if, for instance, we view the production of such instances as a way of dealing
with the sceptic.
But, we might ask, is the analogy very strong here? For Moore’s hesitation to say
that our beliefs about our duty in a particular case can ever count as knowledge
stems from his view that in order to establish what is our duty we have to follow an
infinitely long chain of effects in order to find out which of the available actions
creates a preponderance of things good in themselves. But no such consideration
is relevant, or likely to be relevant, to the epistemological case, since whether a
belief is to count as knowledge is surely independent of its effects, but related if
anything rather to its origin. So the analogy is weak, and epistemological prin-
ciples might still produce accredited instances of knowledge.
I think that this point is sound. But the point is not yet decided until we succeed
in making similar remarks about Ross; which is more difficult. For Ross’s position
is that if we have a number of principles which can conflict in a given case without
56     
any of them being withdrawn for that reason, such principles must express what
we might call prima facie reasons rather than conclusive ones. That being so,
different prima facie reasons may militate in different directions in a given
case, and a decision as to where the balance comes down is inevitably a decision
and in no way secure enough to count as knowledge. Now these remarks are,
as I have expressed them, as applicable to epistemology as to ethics, and hence
the conclusion we draw on one side should, it seems, be drawn on the other side
as well.
There is further profit to be drawn from Ross. The effect of Ross’s view that
principles express prima facie reasons, was, as intended, to reduce their vulner-
ability to disproof by counterexample. In fact, it is not clear what would count as
disproof of a suggested ‘general principle of prima facie duty’. What is needed,
where the principle to be disproved is ‘ϕ actions are a prima facie duty’, is not an
instance of an action which is ϕ but not a duty; such an instance would be
powerless. What we need is instead an instance where an action’s being ϕ is
simply irrelevant to whether it is a duty or not, or an instance where its being ϕ
militates against its being a duty. Now if these remarks are, as I think they may be,
applicable equally to epistemology, we begin to wonder why epistemologists have
not made more use of the concept of the prima facie. For although it has the
consequence that accredited instances of knowledge cannot be produced, it
reduces the vulnerability of epistemological principles to counterexample, tells
us what such a counterexample should be like, and alters our conception of what
it is like to move from a consideration of prima facie reasons for and against this
beliefs being a case of knowledge (he had a certain amount of luck, but on the
other hand he used a sound method of discovery), to a decision in the particular
case. Although a comparison between prima facie duties and defeasible justifi-
cation is often made, with allusions to Ross (cf. Chisholm 1964; Swain 1978;
Pollock 1971), I think that the Rossian moral that all principles of justification
are prima facie principles is rarely drawn. And yet it seems that the attractions
(and the disadvantages) of this conclusion are the same as those of its analogue
in ethics.
I now return to Alston’s distinction between moral-sense theorists and self-
evident first principles theorists. This distinction has proved initially useful, but is
not really sound. The first cracks in it appear when we notice the context of Ross’s
remarks quoted earlier. I now give the passage in full (or fuller, anyway):
What comes first in time is the apprehension of the self-evident prima facie
rightness of an individual act of a particular type. From this we come by
reflection to appreciate the self-evident general principle of prima facie duty.
From this too . . . we come to believe something not self-evident at all, but an
object of probable opinion, viz that this particular act is (not prima facie but)
actually right. (1930: 33)
  - 57
If this passage is representative, Ross, at the time at which he wrote The Right and
the Good, was both a moral-sense theorist and a self-evident first principles
theorist. We do not need to puzzle ourselves so as to decide whether this is really
so; whether it is so will depend on what Ross meant by ‘From this we come by
reflection to appreciate’ and what is meant by ‘first’ in ‘first principles’. But it does
seem that there is room for someone to say that we start, epistemologically, from
scrutiny of particular cases and that we see in those cases principles which we can
recognize as self-evident, though we would not have done so if the particular case
had not brought them to our attention (which would explain the advantage for a
moralist in having a very wide range of examples at his disposal, an advantage
which most English moral philosophers have conspicuously lacked). This does in
fact seem close to Chisholm’s position in the passage quoted above.
Let us try abandoning, for the moment, the perhaps unprofitable question
raised by Alston’s distinction between moral sense and self-evident principles as
to whether knowledge of particular cases precedes knowledge of principles or vice
versa. (One reason why it seems unprofitable is that for the purposes of the
discovery of principles imaginary cases seem to be just as effective as real ones.)
Instead I want to remap the general area of intuitionism in my own way, using not
one but two distinct criteria. This should, and does, yield not two but four types of
intuitionism.
The impetus for this revision comes from a difference between the views
expressed by Ross in The Right and the Good (1930) and the views in his
Foundations of Ethics (1939). In both places Ross gives the epistemological
priority to particular cases: ‘Their rightness was not deduced from any general
principle; rather the general principle was later recognised by intuitive induction
as being implied in the judgements already passed on particular acts’ (1939: 170).
But in the later work the epistemological role of principles is very much reduced.
Whereas in The Right and the Good he says that I discover principles from
particular cases and then work from them to make decisions in new cases, in
Foundations of Ethics we read: ‘I never seem to be in the position of not seeing
directly the rightness of a particular act of kindness, for instance, and of having to
read this off from a general principle’ (1939: 171). Here Ross is willing to give an
account of how we come to know principles, but unwilling to allow those
principles any epistemological role in the making of new decisions. And we
might therefore ask why he is so convinced that we do have knowledge of moral
principles, and why they are allowed to creep into his ethical theory at all, when
they thus appear to be of no use. The answer is that Ross thinks that particular
cases create general principles; there could not be a particular case unless there
was a general principle behind it, as it were. But this conviction does not suit his
particularist epistemology in the slightest. It is an independent input and hence
presumably a question for independent determination. There are two
questions, then:
58     
(1) Are moral principles epistemologically effective?

(2) Are moral principles immanent in particular cases?
To which Early Ross replies (1) yes, (2) yes, while Later Ross replies (1) no, (2) yes.
(I am reminded of Broad’s remark that the Foundations of Ethics could well have
been called The Righter and the Better.) Can we find candidates for the other two
possibilities? Someone who answers (1) no, (2) no, would see no role for principles
in ethical theory—a thoroughgoing particularist. Probable candidates here are
McDowell and Wiggins. Could there be anyone answering (1) yes, (2) no? The
map is so far like this:
(1) (2)
Yes Yes Early Ross

Yes No ?
No Yes Later Ross
No No McDowell, Wiggins
It seems to me most likely that the person we are looking for here is someone
like Moore who would conceive of it as impossible to work from knowledge of a
particular case to knowledge of a principle. But this will only be so if the thought is
not that there is a route from case to principle which we cannot follow, but that
there is no such route for us to follow at all. Such a thought will appeal to those
who, like the particularist, fail to see how one can hope to work from the special
nature of a particular case to something independent of that special nature. And it
is easy to see how that view can be combined with the view that there is a route
from principle to case, although the traffic is all one way. It is possible that Moore
should be cited here, and maybe also Sidgwick; in which case we get the surprising
result that those offered as self-evident first principles theorists are not those who
give the widest role to moral principles.
Perhaps the most important difference between Alston’s map of the territory of
pluralism and my own is that, as I see it, Alston does not see the possibility of the
particularist; for him the role of the particular cases on which moral sense operates
is restricted to the establishing of principles, an activity in which the particularist
is not interested. The other main difference is that my map offers various
alternative conceptions of the nature and importance of moral principles, which
we may hope to apply to the analogous case of epistemology. What account
should we give of epistemological principles? Are the arguments designed to
support this or that form of pluralism in ethics applicable to epistemology?
The central question on which these four forms of intuitionism differ is that of
how moral principles are (or in the case of the particularist, are not) related to
particular cases. I start with a comparatively uncontroversial point.
  - 59
The view of Later Ross that we never seem to be in a position of reading off an
answer for a particular case from a principle seems to be more patently true in
epistemology than in ethics. Although people are prone to assert that, sometimes
at least, they make ethical decisions on principle, and some would even go so far as
to say that all ethical decisions ought to be made on principle, as a matter of one’s
consciousness of how one goes about making epistemological decisions, it would
not be too bold to say that they are never made on principles. And of course this is
partly because though there is a plethora of common candidates for moral
principles, there seems to be a lack of formulated candidates for epistemological
principles. So we can conclude that epistemological principles are not epistemo-
logically effective, leaving till later the question whether they ought to be, i.e.,
whether there is something dubious about making one’s decisions case by case.
This decision in favour of Later Ross on the epistemological front leaves us with
only two views, and only one disagreement: are principles immanent in particular
cases? Later Ross holds that they are, against the McDowell–Wiggins view that
they are not. What arguments have been employed in this confrontation? Some at
least have concerned the notorious properties of supervenience and universaliz-
ability. I take supervenience first. Here is a definition of supervenience:
A property ϕ is supervenient on a class of properties C iff ϕ is not identical with

any member of C nor with any truth function of members of C, and if an object
has ϕ and C0 . . . Cn (where C0 . . . Cn are all the members of C which it has) then it
is impossible that it should cease to be ϕ or become more or less ϕ than before
without changing in respect of some member of C, and necessary that any other
object possessing C0 . . . Cn to the same degree also possesses ϕ to the same degree.¹
It is traditional to assert in this sense of ‘supervenient’ that moral properties are

supervenient on non-moral ones. And it seems possible to say also that epistemic
properties are supervenient on non-epistemic ones; this is quite compatible with
their being irreducibly epistemic, since supervenience threatens no reduction.
Does supervenience express a sound sense in which a particular case creates a
general principle? In one way it does, for if in a certain case we have it that a
gardener’s belief that it will rain is justified, we know that any case which is non-
epistemically indistinguishable from the first will be a case of a justified belief—
indeed of a justified belief by a gardener that it will rain. So there is a sense in
which any particular case creates a general truth (and our knowledge of such
general truths derives from our knowledge of particular cases). But the general
truths so created are not in any way like general principles which we can apply to
cases that are in some respects dissimilar to the first. Any difference between the
¹ This definition is taken from Blackburn (1971:106).

60     
first case and the second is sufficient to make the general truth created by the first
irrelevant to the nature of the second. So supervenience does not provide anything
that an epistemologist might be tempted to call an epistemological principle.
The matter is different with universalizability. Hare expresses his understand-
ing of this term thus:
If he makes a moral judgement about one object, this must be in virtue of the
possession by the object of certain non-moral features . . . and therefore any other
object which possesses those features must have the same moral judgment made
about it. (1963: 20)
And it is in virtue of remarks like this that the doctrine of universalizability is said
to be merely the doctrine that all reasons are general reasons. If we write
‘epistemic’ and ‘non-epistemic’ for ‘moral’ and ‘non-moral’ in the above quota-
tion, we get a doctrine to which Later Ross would gladly assent. Such universal-
izability will create moral and epistemic principles; for instance, a belief judged to
be justified by virtue of certain features will create the general principle that beliefs
with those features are justified.
This doctrine of universalizability is far from trivial. For present purposes, its
relevance is that one form of scepticism uses it as an essential prop. The form
concerned is that which, taking a particular claim to knowledge, demands that the
claimant extract from it a principle which is successful in all other cases to which it
applies. This demand can be made more urgent if the sceptic can produce a case
which is apparently relevantly similar to the first but where there is no knowledge.
At its most general, the argument can be seen in the following way:
Claimant: I have applied a sound procedure and reached the truth that p. Hence
I know that p.
Sceptic (Gettier?): Some cases of the application of a sound procedure and the
production of a truth are not knowledge. Hence you do not know that p.
The sceptic’s move at each point is to demand from the claimant a principle
distinguishing successful from unsuccessful claims, and to complain of any
principle produced that it is subject to counterexample. A claimant who is
impressed by universalizability in epistemology must feel the sceptic’s demand
to be germane to his conception of what is going on, and therefore agrees that he
must meet it or be confuted. But the obvious escape route, if it is available, is to
deny the universalizability of epistemic judgements and hence refuse to submit to
the demand for a principle.
Can this be done? I have tried to show elsewhere how it can be done for ethics.
Although there is not space here to give the matter the attention it needs, the
general line can be seen with reference to the remark by Hare quoted above. If the
  - 61
properties by virtue of which someone judges an object to be ϕ are properties A

B C, this need not mean that he must judge every object which has A B C to be ϕ,
for another object might possess as well as A B C a property D which defeats the
propensity of A B C to make the action ϕ. And we cannot conclude from this that
the absence of D was among the properties in virtue of which the original object
was (judged to be) ϕ. The behaviour of ‘background’ properties is too complex for
this simplistic solution. (I mention here that it is in the discussion of the relation
between ‘background’ and other properties that the notion of shapelessness
referred to earlier begins to play an important role. For if the background
properties are not inert (far from it), their activity may be seen as affecting, in a
particular case, the ability of ‘general’ reasons to be reasons there, and as prevent-
ing what are reasons there from being general reasons.)
This argument, if sound, seems as applicable to the epistemic as to the moral
case. It is, of course, part of the defence of particularism, and as such reveals that
the particularist is in a much stronger position than his rivals to rebut a certain
important form of scepticism.
What about the argument for particularism? In ethics, this is the attempt to
show that there is no sound conception of a ‘morally relevant’ property that can
support the attempt to move from particular case to principle or from principle to
particular case. Again, the details cannot be given here; but it should be noted that
whereas there is a strong inclination to think of some properties (e.g., causing a
diminution in happiness or welfare, or being commanded by God) as generally
morally relevant, on the grounds perhaps that they always make a difference to the
moral worth of actions that have them, there is no such strong inclination to think
of some properties as generally epistemically relevant. In morals one can hold that
the discovery of morally relevant properties (maybe they are self-evidently so, or
maybe one only discovers them to be so from attention to a particular case) leads
to the establishment of moral principles. But the attraction of this route to
principles seems sensibly diminished in the case of epistemology. Have there
been properties with which it has been plausible to hold that epistemological
questions are centrally or essentially concerned, in the way that it has been held
possible even to define the scope of ethics as the field of human welfare?
I conclude with a summary. The aim of this paper has been to examine
intuitionism in greater detail than did Alston, in order to expose some at least
of the instructive analogies to be found there. Four forms of intuitionism emerged,
differing on the question of the relation of particular cases to principles. It was
held that this question, the answer to which affects crucially our conception of the
nature and role of epistemological principles, could perhaps be decided in favour
of the particularist. One advantage of this decision would be that we could show a
certain sort of sceptical challenge to be irrelevant. But whichever way the decision
goes, it is surely helped by the observation of essentially similar styles of argument
in the field of ethics.
5
The Role of Imaginary Cases in Ethics
C. D. Broad ends the Preface to his Five Types of Ethical Theory by admitting that
‘my range of experience, both practical and emotional, is rather exceptionally
narrow even for a don. Fellows of Colleges, in Cambridge at any rate, have few
temptations to heroic virtue or spectacular vice; and I could wish that the rest of
mankind were as fortunately situated’ (1930: xxiv). Broad’s excuse was that for the
moral philosopher, whose job is the analysis of ethical concepts, little in the way of
moral experience is necessary. This however seems to me to depend upon a
dubious conception of the role of philosophy, including moral philosophy, and
its relation to experience.
Be that as it may, in the present paper I examine one question about the
dispensability of moral experience. Someone like Broad who proclaims a paucity
of moral experience may suppose that his knowledge of the distinction between
right and wrong (his moral outlook), derived though it is from such moral
experience as he may have enjoyed (or suffered), is supplemented by the activity
of what we may call the moral imagination. If the diet of actual cases is insufficient
to support a full moral view, it can be enriched by attention to some imaginary
cases. For imaginary cases are as good, for the purposes of moral epistemology, as
actual ones.
In a similar vein we may admit that our knowledge of the moral properties of a
case which now confronts us is in some way dependent on or informed by our
moral experience. Like other forms of cognition, moral cognition is in part the
attempt to reach a decision in the present case in the light of decisions in previous
cases. But if on some occasion we suppose that we have experienced no relevantly
similar previous cases, we are not left to make the present decision in a void. For
the role of the previous cases can be played equally well by imaginary cases, which
we can supply at will. If we find that our past experience is here no guide or
insufficient guide, we can let ourselves be guided instead by cases made up for the
purpose.
I want to examine in more detail how it may be supposed that imaginary cases
can perform this heuristic role. I shall first suggest some difficulties in the idea that
imaginary cases can be any guide at all. Then I shall argue that a common (even
universal) conception of ethics is incapable of resolving these difficulties, if they
       63
are real. Finally I shall suggest an alternative resolution of the difficulties, designed
to suit the sort of moral theory that I find persuasive.
There are of course other questions that could be asked about the role of
imaginary cases in ethics. For instance, one could ask about the relative import-
ance of imagination and of real experience. Why is there some premium on
experience? Is it that imagination needs some input from experience, as it were,
to get a start on? Or is it that our judgement about an actual case, particularly one
in which we are involved, is more likely to be truly appreciative of the nature of the
case and especially more directly concerned with what to do than is our judgement
about an imaginary case where the question what to do seems less pressing? One
could dispute the comparative advantages of involvement and dispassionate
judgement (I know which side Broad would be on); but this question is not at
issue here, except indirectly.
So the question is how, if at all, imaginary cases can substitute for experience of
actual cases as a guide to moral decision. In speaking of the appeal to imaginary
cases I am not here thinking of the sorts of cases once stigmatized as ‘desert-island
morality’. It used to be suggested that the utilitarian gives a counterintuitive
account of our obligation to obey our promises, on the grounds that a promise,
made to a man dying on a desert island, to spend his accumulated millions on the
foundation of homes for retired steeplechasers gave the promiser no reason at all
to keep his promise when he returned to civilization. But he had a reason; the fact
that he had promised is a reason, but it is not one which the utilitarian, qua
consequentialist, can countenance.
There are various complaints which one could make, and which were made,
about the attempt to conduct moral philosophy by this sort of appeal to bizarre or
extreme examples. And one could make similar complaints about this use of
thought-experiments elsewhere, e.g. in the theory of personal identity. But the
sorts of examples I shall be concerned with will not be bizarre or extreme. I think
that in most instances the imaginary cases we appeal to are comparatively
ordinary, with good reason.
Nor shall I be considering directly attempts to help us decide cases which start
from the question ‘Would it make a moral difference if the case were a bit different
from the way it actually is?’. For instance, we might ask ‘Would it make any
difference if the person concerned were a friend of yours?’ as part of an attempt to
come to a decision in a particular case. But I think that the problems about moving
from actual to possible cases are of a different nature from that of the move from
possible to actual. Here I want to focus on rather different cases.
The sort of appeal to imaginary cases I have primarily in mind has instances
varying from some of the parables of the New Testament and Aesop’s fables to
more direct attempts to determine our attitudes in the following sort of way. There
is at present a controversy about whether Britain ought to return the Elgin
Marbles to Greece. An argument might run as follows. Suppose that an orphan’s
64     
estate is under the control of a not particularly conscientious trustee, who for
reasons of his own allows a collector to remove part of that estate in exchange for a
(not really sufficient) payment which the trustee then absorbs for his own
purposes. The orphan eventually comes into his own estate and attempts to
recover what was his property. Whatever be the legal situation, must we not
admit that the collector has some moral obligation to return the property? And
surely that obligation is not reduced or in any way counterbalanced by the fact
that the property is now on view in a collection of objects many of which were
acquired in similar dubious circumstances. So the collector ought to return the
property. And similarly Britain ought to return the Elgin Marbles.
This seems to be an instance of a recognizable form of moral argument. In fact,
however, there are two distinct sorts of arguments like this. The first attempts to
establish what we actually ought to do, as in the case above. We suppose that our
duty is not clear, but that it can be made clear by the consideration of a relevantly
similar but imaginary case. The second attempts only to establish that a consid-
eration is at least relevant and ought not to be left out of account. If I am unclear
about whether a consideration is morally relevant in the present case, we might
hope to help by providing an imaginary case in which that consideration is clearly
relevant. It matters there; so it must matter here. In Ross’s terms, the distinction
here is between the use of an imaginary case to show that an action is a duty
proper, or that it is a prima facie duty in virtue of having a certain property.
I mention this distinction now; it will come into play later.
I think that it would be misguided to object to an argument like that given
above that it is inconclusive. We might say this because we feel that a similar
argument can be given for the opposite conclusion. But if we were going to take
this as an objection to a style of argument, we would leave ourselves rather short of
resources. Our objection would only be effective if we felt that the opposite
conclusion could equally well be drawn from the very same considerations. But
this is not normally the situation with arguments from imaginary cases. Instead,
one side argues from one imaginary case and the other from another. This does
not seem to be in itself grounds for suspicion.
I want to suggest instead three or maybe four problems in this appeal to an
imaginary case. These problems, or arguments, are attempts to flesh out the
intuition with which I think one should approach this area, that there is something
very odd about the idea that we can learn important truths about the way we
should behave from an examination of cases which are the creatures of our own
imagination.
The first problem is how one can reasonably come to a view about what it
would be right or wrong to do in an imaginary case. Now of course some
imaginary cases are so sketchily described that the only possible reaction is that
one would need to know more before deciding. But there seems to come a point at
which that reaction becomes less convincing. As our description of the imaginary
       65
case becomes fuller, the temptation to form a view becomes more compelling.
Should it be resisted? There is an obvious but crucial difference between imaginary
cases and real ones, which is that an imaginary case has no nature beyond that
specified in the description. Beyond what is explicitly offered, the imaginary case is
indeterminate; there are no further ‘facts of the matter’. One might try to use this
feature in order to insist that we cannot come to a view about an imaginary case.
But this will not be straightforwardly true. We do, after all, manage to reach
reasonable views about actual cases, though our knowledge of them is less than
complete. Still, there is a difference. The only sensible way of saying that the
incompleteness of an imaginary case is no bar to our coming to a view there is by
stipulating that the case as described has no further morally relevant properties.
Imaginary cases can always be filled out further, of course. But we are expected to
take it that there is at least one way of filling out an imaginary case under which
the properties so far mentioned are (or include) the only morally relevant prop-
erties of the case, and that the way in which they are relevant is not affected by the
features of the further description. This last point is that the further description
not only does not provide further properties that matter, but does not provide
further properties which affect the questions whether the first ones matter, how
they matter and how much they matter.
The moral properties of a right action are parti-resultant, in Ross’s sense (1930:
28).¹ The action has an enormous number of non-moral properties, only some of
which are the reasons why the action is right. So in this sense the action’s rightness
emerges or results from only a part of its non-moral properties; rightness is a parti-
resultant property. But beyond the preferential group of right-making properties,
the remaining properties need not be completely inert; some at least will play a
supportive role. For suppose that the action is right because it is generous,
thoughtful and kind; we can say that it is in virtue of its generosity, thoughtfulness
and kindness that it is right. But each of these three will in turn result from or exist
in virtue of further properties. The generous action may be generous in virtue of
being a substantial donation to a needy cause, or whatever; but its generosity must
result from something. One property results from another when the first exists in
virtue of the existence of the second. So the relation of resultance, which holds
between moral properties and reasons, continues down what I call the resultance-
tree. Beyond the limited group of properties which are ‘reasons why’ there will be
properties which do affect the rightness of the action, but do not count among the
reasons why the action is right. These are properties whose absence would tend to
cause the action not to be right, or at least not to be right in that way. Beyond the
select group of properties from which the action’s rightness is parti-resultant,
¹ In The Right and the Good, Ross held that moral properties, including crucially the property of
being a duty proper, are toti-resultant, not parti-resultant. I argue that this is a mistake, in ‘On Moral
Properties’ (1981; repr. as Chapter 2 in this volume).
66     
then, there lies a further group which, with the first group, form the class of
properties morally relevant to the case in this extended sense.
The question then is this. If we have been told the properties which de facto
constitute the reasons why the action is right, and nothing much more than these,
can we reasonably be expected to form a sound view on the question whether the
action is right or wrong?
The problem lies in the idea that the moral relevance of the properties so far
described is not affected by any further properties; that there is a way of filling out
the description which leaves untouched that moral relevance as it presently stands.
But in the absence of further description there is no such thing as the moral
relevance of the properties so far given. It is not that they have a relevance which
will or may not be overturned; in the foreshortened state of the case the question
of their relevance remains indeterminate because it is crucially dependent upon
other matters which are so far indeterminate. Thus, how important is it that the
trustee demanded insufficient payment for the property, and how important is it
that he absorbed that payment for his own ends? There cannot be an answer to
this question in the absence of further description. And if this is so, in responding
to the case as so far described we are pretending (claiming) to make a decision
about a matter which is indeterminate. But such a pretence does not make sense.
It would be misguided to suppose that, if there is a difficulty here, it could be
resolved by the provision of a bit more ‘information’ about the imaginary case.
The problem will remain: we will still be in the business of determining a matter
which is indeterminate. The remaining indeterminacy will remain infectious; and
what it infects is not our description of the case as so far given, but the suggestion
that there be such a thing as the moral relevance of the properties contained in
that description, as so far revealed. If there is such a thing, we can hope to be able
to continue that description in a way which does not alter the relevance. But if
there is not, a moral judgement about the imaginary case as so far revealed
attempts the impossible. There is no such thing as the moral relevance of the
properties revealed, about which we could come to have a more or a less
sound view.
So there is a problem in supposing that we can reasonably come to a view about
the moral properties of an imaginary action, stemming from the indeterminacy
which affects imaginary cases. But even if there were no such problem, there
would still be a difficulty concerning the move from the imaginary case, so
determined, to the actual case. Suppose, first, that the move is one designed to
take us from the thought that the imaginary action is a duty proper to the thought
that the real action is a duty proper. We are taking it now that we can discern the
moral property of the imaginary case because there is a way of continuing the
description of the imaginary case which will not alter the moral relevance of its
properties, as so far revealed. This means that there is an important difference
between the imaginary case and the actual case. In the one, we know that a fuller
       67
description would not upset the moral relevance of the properties we know. In the
other, we do not know this, and we would be quite unjustified in supposing it to be
true. The fact that there is a way of continuing the description in the imaginary
case is no reason whatever to suppose that that sort of way is the way in which the
description of the actual case is filled out. So even if we can reach some deter-
mination of the moral relevance of properties in the one case, we cannot use it to
help us in the other.
There is a further reason preventing the move from an imaginary case to an
actual one, stemming from the reasons why we are tempted to make such moves
in the first place. The original idea was that we found the actual case difficult, and
so we derived help from a case which was easier. So there is a further difference
between the two cases; in one we find it hard to discern what we ought to do, and
in the other we find it easy. But this could only be so if there were other relevant
differences between the two cases, in virtue of which one is hard and the other
easy. The one that is easy can only be so because it does not contain factors which
complicate the issue in the hard case. And, this being so, it is odd to suppose that
our decision in the easy case should be of help when we come to the actual,
hard case.
On the other hand, if the imaginary case is no easier for us, it is not obvious
what the point of turning our attention away from the problem actually facing us
could be. Either it would just put the problem further back, or it would give us two
problems rather than one.
There is a natural reaction to this insistence on differences between imaginary
and actual cases. This is that though the cases are different, they are not relevantly
different. Arguments from imaginary to actual cases depend upon, not total
similarity, but relevant similarity. Relevant similarity is all that can be required,
and it can (sometimes at least) be produced.
I would not be willing to admit that the two sorts of differences mentioned
above are not relevant. What makes us justified in supposing that the respects in
virtue of which the actual case is difficult are not morally relevant, and that the
difficulty is an isolated—or isolable—epistemic feature? But let us nonetheless go
along with this appeal to relevant similarity for a little. Our consideration of it will
reveal the last reason why it is hard to see how arguments from imaginary cases
could ever be effective.
The suggestion is in fact familiar in various areas of philosophy. We have two
cases, A and B, which are different in many respects. However, A has a property F,
and only some of its properties are relevant to its being F or the degree to which it
is F. B is similar to A in all those respects, and hence it must be F too, and be F to
the same degree.
What could possibly be wrong with this simple procedure? One might com-
plain, as I have on earlier occasions, that there is no shortlist of morally relevant
properties, and therefore that though the procedure may work well in some other
68     
cases, such as where F-ness is squareness or heat, it will not work in the moral
cases which are our present concern. But I think that this complaint would be too
kind. The suggestion here is that we are to determine the similarity of A and B in
respects relevant to F-ness before we have determined their similarity (or other-
wise) in respect of F-ness; and that we should work from the first decision to the
second. Does this make good sense? The first point to make is that if we are to
work in this way, we need to determine not just that A and B are similar in
respects R¹–Rn, but that R¹–Rn are all the properties relevant to F-ness (or relevant
to the F-ness of these two cases). Can we do this in advance of determining
whether B is F or the degree to which it is F?
There may be some properties for which this sort of procedure will work, but
I cannot see that it is going to work for many properties, and particularly not for
moral properties. It seems to me that in order to determine whether A and B are
similar in all morally relevant respects, or in all respects relevant to their moral
properties, we would need first to decide on the moral properties of each. It is only
if we know the moral properties of each that we can have a reasonably firm view
on which of their properties are morally relevant there, and hence come eventually
to a view about whether A and B are similar in all respects relevant to their moral
properties. So the procedure we are considering puts first what in my view could
only come last. It is true that if we could discern independently that A and B are
similar in all morally relevant respects we would know that they must have the
same moral properties. But this is of no practical use to us as the ground for moral
inference, because we cannot discern that A and B are similar in morally relevant
respects unless we already know the moral properties of A and B.
The suspicion of the notion of relevant similarity which is expressed in previous
paragraphs is a general suspicion which, if sound, reappears in many other areas
of philosophy. I have expressed it in comparatively mundane, though formal,
terms. But the point is the one made by Wittgenstein in his discussion of rule-
following (1969: esp. §§139–242).² There is nothing available to drive us, as it were
on rails, from what we want to say in one case to what we want to say in the next.
The feeling that we have to go on this way, or that if A is F, B must be F, is a feeling
which needs explanation; but the explanation cannot be in terms of a new decision
being rendered inevitable by a previous one. The appeal to relevant similarity is an
instance of the sort of explanation here ruled out. It attempts to persuade us, in the
wrong way, that our description of the present case is already determined or fixed
by what we have said about a previous case.
To sum up: in this part of the paper I have provided four arguments of
decreasing strength and increasing plausibility. Each is independent of its prede-
cessors, and can be accepted even if they are not. But even the last weakest
² For representative comment, see McDowell (1981).

       69
argument is sufficient to raise a genuine question about the nature and efficacy of
the appeal to imaginary cases.
One final comment on the four arguments is that all but the first apply equally
to all attempts to argue from the nature of one case to that of another. It is only the
first that hinges specifically on a feature of imaginary rather than actual cases, their
non-epistemic indeterminacy. Nonetheless, the combination of arguments seems
to me to make a strong case, and I welcome the tendency they have also to
undermine arguments from actual cases.
II
These, then, are the reasons for holding that an appeal to imaginary cases does not
obviously make sense. If we agree that these reasons have some force, we naturally
look for a theory of moral reasoning which will make good sense of the appeal to
imaginary cases. A candidate theory is the generalist’s claim that imaginary cases
reveal moral principles to us which we can then use to determine our attitude in a
new case.
I want to argue that if we have a problem about imaginary cases, the appeal to
moral principles can provide no answer, and that the use of moral principles in the
way suggested would seriously distort our appreciation of the present actual case.
(In fact, I hold that the sort of reasoning recommended by the generalist is little
short of pernicious.)
I pointed out earlier that there are two ways of using imaginary cases, first to
determine what we ought to do in an actual case (our ‘duty proper’) and second to
determine what is morally relevant in the actual case (which properties create
‘prima facie duties’). We will consider the first way first. The suggestion here is
that the imaginary case reveals a moral principle. It seems to go this way: here is a
case of a right action. It has properties F¹–Fn. So all actions that have F¹–Fn are
right. If our present action has those properties, then, regardless of any other it
may have, it is right.
I think it would be generally agreed that this approach at least is completely
unconvincing, for the sorts of reasons given earlier. Remember that the imaginary
case we are working from is necessarily incomplete; the actual case is far richer in
properties. This means that there is no way whatever in which a foreshortened
imaginary case can drive us to a decision in an actual case, because the further
properties of the actual case may conspire to defeat whatever we learnt from the
imaginary one. And if we weaken the approach so as merely to say that if the
imaginary action is a duty proper, the actual action must be a duty proper unless it
differs from the first in some morally relevant way, the approach becomes so weak
as to be useless. We had hoped, after all, to use the imaginary case as a guide to the
70     
actual one. But that would be impossible if, in effect, our first decision is whether
the imaginary case is a reliable guide to the actual one.
We must therefore suppose the generalist to be suggesting that imaginary cases
reveal to us moral principles in this sense: the imaginary case reveals to us a
collection of properties which are individually morally relevant there, and we
suppose that those properties are morally relevant here in the actual case (and
relevant to the same degree, in the same way etc.).
This seems to me marginally better as an answer, but it is vulnerable to the same
complaints. It is not obvious that what we are said to learn in the imaginary case is
going to provide the help we were seeking. For even if we do learn from the
imaginary case the importance that a property has there, this will only help us to
reach a decision here if we decide that the actual case has no further morally
relevant properties. The imaginary case cannot help us in this decision at all.
Essentially we have to make up our minds about the moral make-up of the actual
case before we can come to a view about whether the imaginary case is after all a
reliable guide.
So the attempt to resolve our problems by appeal to the epistemological role of
moral principles seems to me to fail. And the reasons for which it fails are among
the reasons which persuade me to reject the conception of moral principles offered
by generalism. For if moral principles, as there conceived, are of no use, we should
be tempted to suggest that they are dispensable. There are further arguments that
they are worse than useless, that their failings are not wholly epistemological; but
we can pass these by here. Let us take it, then, that the problems of generalism and
its notion of moral principles tempt us into particularism, the view that ‘ethical
decisions are made case by case, without the comforting support or awkward
demands of moral principles’ (Dancy 1983: 530; repr. as Chapter 3 in this
volume).
An immediate objection to such particularism is that it can make no sense of
the appeal to imaginary cases. For the particularist, it seems, an imaginary case is
just another case; there is nothing in the imaginary case which should or could
help to determine our attitude to the present, actual case. Now it seems that if the
generalist’s appeal to moral principles, whether thought of as expressing prima
facie duties or not, does not show how appeal to imaginary cases can be helpful in
making up our minds about actual cases, we cannot really hold it against the
particularist that he too has difficulty in solving the problem. However, we cannot
simply wish the problem away, either, by announcing that the appeal to imaginary
cases is misconceived and should be abandoned. This is definitely a last resort.
Our duty as philosophers of ethics is to make sense of the discoverable patterns of
moral reasoning; if we cannot do this it is a fault on our side, a fault in the
philosophy rather than in the reasoning. In similar vein, the particularist in ethics
has two possible stances with respect to moral principles. The first is the aggressive
position, holding that the appeal to principles is simply misconceived. The second,
       71
more conciliatory, attempts to provide some account, within the constraints it

accepts, of what is a very common practice of somehow appealing to general
truths and previous cases in the course of reaching a moral choice or judgement,
and in the justification of one when reached. I think that there is a way in which
the particularist can make his own sense of this practice, which also provides some
sort of an answer to the problem about imaginary cases.
The suggestion I want to make is that a moral principle amounts to a reminder
of the sort of importance that a property can have in suitable circumstances. It
seems to me that this suggestion makes good sense of a number of puzzles in
moral philosophy, including the ones with which we are presently concerned; and
that it does this in a way perfectly compatible with the spirit of particularist
arguments against generalism in ethics. So my suggestion, as I shall argue, will
not be a new form of generalism in ethics; it is an account of what moral principles
tell us, which is not available to the generalist. The generalist’s principles have
already been shown useless at best. The particularist’s account of principles will
show them to be of some use, though not the use that the generalist wanted to put
them to. First, however, I turn to the puzzles which this account of moral
principles will solve.
First, someone in possession of a large list of principles, so long as he does not
misunderstand their proper role, is at an advantage when coming to a decision in a
particular case. He wants to be sure that he does not miss the importance or
relevance of any relevant property. A panoply of moral principles, understood in
the way suggested, can function as a sort of checklist for this purpose. It will not be
a complete checklist, of course, for we can give no sense to the idea that we might
now have finished the list of moral principles (or of morally relevant properties).
But our approach allows for this and explains it. There is no limit to the number of
properties which can on occasion be important. But of course some properties are
much more commonly important than others (some, perhaps, are always relevant,
regardless of other circumstances) and therefore some moral principles should be
viewed as more ‘central’ than others. This centrality does not entail that the
properties mentioned in the ‘central’ principles are always more important than
those mentioned elsewhere. But it does create some sort of structure in what
would otherwise be a bewilderingly random list of properties which can matter in
suitable circumstances. So the account offered makes its own sense of the use of
moral principles in reaching decisions in a new case.
It can also give an account of the appeal to the importance of a property in
justification of a choice or judgement already made. In answer to the question
‘Why did you do that?’ one may mention a property which one took one’s action
to have, by saying perhaps ‘It was the only honest course’. This is not an explicit
appeal to a principle, and I have suggested elsewhere how a particularist should
account for such a reply (Dancy 1983). But if we suppose that one were to add ‘and
it is important to be honest’, we have the sort of explicit appeal to a moral
72     
principle which would be predicted on our present account of principles. It is not

suggested, of course, that honesty is the only important quality, only that honesty
can be important and that it was here.
Second, our account seems to make sense of the appeal to imaginary cases.
Although there remain problems about the indeterminacy of such cases beyond
the thin description offered, we can hope that an imaginary case is an abbreviated
sketch of a case where a property can be seen to be important; where the
importance it can have is revealed. This is especially so in the case of some of
the parables of the New Testament, or the morals attached to Aesop’s fables. And
if this account is right, it makes good sense of the idea that an imaginary case is
one in which whatever one is looking for is somehow easier to see, without this
being a stumbling-block in the way of a move from that case to an actual one.
Third, our account gives an answer to the main question in moral epistemol-
ogy. I assume here that we must find a way of saying how our knowledge of a
moral principle can be derived from what we can see in a particular case. I take
this form of empiricism in ethics to be the only one which can make sense of our
feeling that particular cases must be able to function as some sort of a test for
moral principles. This need not be a very direct test, any more than the sense in
which scientific theories are testable by particular cases is direct; but only an
empiricism of this sort leaves us with a chance of showing that moral principles
are not immune to the behaviour of particular cases. It seems that a particular case
can reveal the importance that a property can have. It may bring this home to us
and force us to recognize it when before we denied the possibility of such a thing.
I have in mind here the example of an adolescent who maintains that good
manners are hypocrisy; this position is vulnerable to the occurrence of a case
which brings it forcibly home how important good manners can be. So on our
account ethics is empirical. Moral principles are learnt in and from particular
cases.
Fourth, our account gives a good sense to the idea that moral principles, if true,
are necessarily true. Traditionally this idea amounts to one of three thoughts. The
first is the useless truth that if this action is right, then any action similar to it in all
non-moral respects must also be right. There is necessity here; supervenience,
which is what is at issue, is to be expressed in modal terms.³ But it is, I think, the
wrong sort of necessity; it seems to offer a modal inference rather than a neces-
sarily true moral principle. And anyway, supervenience does not provide us with
useful moral principles. The second thought is not uselessly true, but grotesquely
false. This is that if this action is right because of its F-ness, then any action that is
F is necessarily right. Whether the necessity here lies in the inference rather than
in the conclusion, this thought is false for the sorts of reasons with which we are
³ See Blackburn (1971: 105–6).

       73
now familiar. Whatever may be the case here, other properties may conspire to
interfere elsewhere. (This is simply the thought that moral principles are defeas-
ible.) The third thought is that if a property is morally relevant in one case, it
necessarily has that same relevance in every case in which it occurs. I am not
convinced that the necessity here is other than inferential. Mightn’t there never
have been a case in which the property was morally relevant? But, waiving this,
still the particularist cannot accept the thought that a property’s relevance in one
case is necessarily repeated on its every reappearance. And a main reason for the
particularist’s view here is the difficulty of making epistemological sense of that
thought. How could we learn from what we can discern in a particular case that a
property has the sort of relevance here which is necessarily repeated elsewhere?
I have examined this question, and the defects of standard answers to it, else-
where.⁴ The problem is to show both how the truth of a moral principle can be
discerned in a particular case, and how what we are there observing is a necessary
truth. And our present account can do this, without falling into the trap of making
the necessity lie in an inference from what is observed to some more general truths
rather than in the truth observed. For our suggestion is that we can see in a
particular case the sort of importance that a property can have, in suitable
circumstances. What we are observing is already modal, and if our observation
is correct there cannot be a situation in which our property could not have that
importance if the circumstances were suitable. Hence what we observe, if true, is
necessarily true. (This is just an example of the inference from ◊ p to ⃞ ◊ p in S5;
but I cannot claim to be competent to judge whether S5 is a suitable system for the
formalization and evaluation of ethical inference.)
Finally, the account of moral principles here suggested is compatible with the
constraints of particularism, though of course it represents particularism in a
conciliatory rather than an aggressive mood. Although we are able to observe, in a
given case (imaginary or actual), the importance that a property can have in
suitable circumstances, the particularist can still insist that no notion is available
of a sort of circumstance in which it must have that importance. The particularist’s
strictures on the possibility of inference to the nature of a second case from what
we see in the first are not violated by our account. This is the main reason why the
epistemological problem is easier for the particularist than for the generalist, since
the particularist wants to insist that the results of observation are less powerful
than the generalist needs them to be. There is no substitute for the sort of detailed
attention to each new case which an appeal to principles might lead us to shirk. If
there are moral principles, on this account, their role is only to act as reminders.
They will not solve our epistemological problems for us nor serve to create the
moral properties of the new case.
⁴ In my ‘Ethical Particularism and Morally Relevant Properties’ (1983; repr. as Chapter 3 in this
volume).
74     
I end with a couple of remarks about why I think that the appeal to moral
principles, and the attempt to use them to guide decision, is pernicious. The first is
that the question which principle(s) is (are) applicable tends to become a substi-
tute for a genuine and unbiased examination of the new case as it stands. Instead
of asking what the new case is like, we ask instead which other cases it resembles
sufficiently for us not to have to bother to look further. Thus, from one point of
view, the use of principles as traditionally conceived amounts to a sort of moral
sloth. From another point of view, reliance on principles tends to prevent people
from making the decision they would otherwise make because they cannot see
how to make it consistent (in the wrong, generalist, sense) with previous choices.
In this sense, the reliance on principles makes moral life unnecessarily harder, and
distorts people’s perceptions of the nature of the problems which face them.⁵
⁵ I am grateful to colleagues at Keele and to Michael Smith for helpful comments on previous drafts.
6
Externalism for Internalists
The current popularity of the distinction between internalist and externalist

approaches to epistemology has not led to any great increase in our understanding
of the area. The only thing on which most writers on the topic agree is that most
other writers do not understand the distinction at all well. I am no exception to
this rule, and in this paper I try to lay out what I think the basic structure of a
theory of justification of belief should be in a way that is sensitive to the issues
debated between internalists and externalists. At the same time I attempt to relate
what I have to say to ethics. It has been common for proponents of both sides of
the epistemological argument to make appeal to the appearance of a view like
theirs in ethics, in ways of which I do not wholly approve.¹ Not that I mind the
appeal to ethics itself; this seems to me quite in place. I just don’t think that much
is gained by finding a view similar to one’s own in a cognate area which has had a
good run for its money there. This might achieve respectability for one’s own view
(were this in danger of being denied) but could hardly count as evidence for its
truth. And the problem is that even if the various analogies were sound, the
positions they appeal to in ethics are so contentious that not a lot is gained. My
general aim here is to discover the truth on both sides. In my view the two areas
are strongly analogous, and the truth in one will probably be pretty like the truth
in the other. However, since I do not think that the truth of the matter has yet been
discovered in ethics, I shall not be arguing from ethics to epistemology. Instead
I will be trying to move forward on both fronts at once.²
I start with a standard characterization of extreme versions of the two views in
epistemology. The externalist holds that a belief may be fully justified if it has
certain properties which we may call truth-effective; to fix ideas, think of the
leading such property as that of being the product of a reliable method or process.
So our externalist holds that a belief is fully justified if it is the product of a reliable
method or process. No more than this is necessary for justification, nor is there
any other notion of justification than this one. So facts that are not believed by the
agent may suffice for justification, and the agent’s believing them adds nothing.
(Of course, among the facts that contribute to justification may be the fact that the
agent believes something.) The theory offers no room for differing accounts of
¹ See, among many others, Goldman (1980); Bonjour (1980).

² In pursuit of this aim, I will be speaking indifferently of agent and of believer, of act and of belief.
My epistemology is explicitly normative.
76     
belief and of believer. We can say ‘S’s belief is not justified, but if things were in
general as S believes them to be, S’s belief would be justified’. But we don’t mean by
this to suggest that S (the agent) is here impeccable though his belief is flawed;
S inherits the faults of his acts (beliefs).
The internalist position is, to start with, equally trenchant. The standard
characterization of internalism is the claim that justification can only be achieved
by appeal to elements that are internal to the agent’s perspective. This is not yet
much help, and I shall later argue that it suffers from ambiguity. One way of
making it less metaphorical is to read it as saying that only S’s other beliefs are
relevant to the question whether S is justified in the belief that p. Once this sort of
internal justification has been achieved, there are no further questions about
justification. In particular, that the agent’s beliefs are true adds nothing, and
nothing is subtracted if they are false. This theory too has no room for differing
accounts of belief and of believer. We can say ‘S is not justified in believing that p,
though there are facts which, if S had believed them, would have justified him in
believing that p’; but we do not mean by this to suggest that S’s belief that p is
somehow sound as things stand. S’s belief that p inherits S’s faults as a cognitive
agent.
So both my extreme internalist and my extreme externalist speak indifferently
of the justification of belief and of believer; both take it that if the belief is justified
the believer is and vice versa. Neither of these extreme positions admits the
possibility that the belief be justified when the believer is not, nor that the believer
be justified when the belief is not. Two thoughts persuade us to make room for
these possibilities. The first is that even if things are going badly in externalist
terms, we are not always inclined to attach much epistemic blame to the agent; the
second is that even if the agent’s epistemic behaviour is irresponsible, we may find
something to approve in the end product. But I want to pause to look for ethical
analogies for the positions we already have, before moving on to more complex
theories.
Extreme externalism will appeal to the sort of moral theory that focuses
primarily on the value of acts, and tries to write its account of the worth of agents
directly from what it finds there. The fact is, however, that this sort of moral
theory cannot cope with everything we want to say in ethics. In particular, we
want at this stage to make room for the idea that agents sometimes derive moral
worth from the attempt to do the right thing, or from their belief that what they
were doing was right, or from the fact that if the circumstances had been as they
thought they would have been doing the right thing. This is just the sort of idea
that the extreme externalist is trying to get away from. So the sort of moral theory
that extreme externalism can appeal to is not much support. The same is true for
extreme internalism. It is likely to appeal to Kant in some way. The first chapter of
his Groundwork (1785) is intended to undermine the apparently common-sense
idea that actions can have moral properties which they derive somehow from the
   77
sort of difference they make to the world. We could take this to mean that the
distinction between act and agent has no moral relevance for Kant. Moral worth
attaches itself primarily to agents, and only derivatively to acts, in the sort of way
that extreme internalism suggests. My own view, however, is that Kant cannot be
used like this. It is true that given the choice between act and agent Kant will
always focus on the agent, but I think that rather than accepting the distinction
Kant has really left it behind. We can test this by asking what Kant would say
about the agent who obeyed the Categorical Imperative in this sense, that he acted
out of respect for the universalizability of his maxim, but in a case where the
maxim was not in fact universalizable. Nothing in Kantian scholarship helps me a
lot with this question. Its purpose here is just to show that there is the same
pressure for Kant to admit that there are two forms of evaluation going on at once,
according to one of which things are going on well enough and according to the
other of which they are not. This need not be the act/agent distinction itself, but it
is still enough to make the appeal to Kant a less than complete defence for the
extreme internalist.
So the ethical analogies to which these two approaches appeal do not really offer
much in the way of support. They are weak on the very point on which they were
being looked to for help, namely that there is no need for anything other than a
trivial distinction between act and agent. And there are other reasons for being
unhappy with the choice between the two approaches. It is agreed that externalism
probably makes our epistemic life very easy. Given that human cognitive methods
and processes are genuinely truth-effective, most of our beliefs will be justified.
Scepticism is not an endemic problem for externalism, and this is both one of its
chief attractions and one of its main weaknesses. Internalism, on the other hand,
though it has considerable intuitive appeal, leaves us very vulnerable to the sceptic.
We would like to be able to say that the fact that human cognitive processes and
methods are truth-effective does somehow contribute to the justification of the
beliefs they generate, but on the internalist perspective this will only be true of
those agents who have the relevant beliefs, and most of us are not in that fortunate
position. Given then the difficulty of saying that everything necessary to justifica-
tion is believed by every believer, it looks as if justification is going to be a fairly
rare occurrence, by the internalist’s lights. Internalism may have the support of
intuition, as I believe it does, but it is a gift for the sceptic.
This leads one to suppose that the two views I started with are too extreme, and
that some form of intermediate position must be found. Something has gone
wrong here. The choice we are being offered is too stark. Neither side is willing to
admit the relevance of anything that the other side wants to see as important, and
yet underlying each view there seems to be something intuitively plausible. What
seems to be needed is a way of admitting that facts that are not believed can be
relevant to the epistemic status of an agent or of a belief, while accepting that the
only things that can justify are things internal to the agent’s perspective. I want to
78     
suggest a way of doing this, and I shall work towards it by considering other
compromise positions to be found in the literature.
The first of these is Alston’s internalistic externalism, which, in Audi’s formu-
lation, comes out like this:³
S’s belief that p is epistemically justified iff:

1. S’s belief is based upon grounds G such that
(a) the members of G are the sorts of things that are typically accessible to
normal humans (e.g., experiences, beliefs)
(b) the members of G are fairly directly accessible to S upon reflection
(c) G reliably indicates the truth of p
and
2. S has no additional undermining beliefs.
As can be seen, this position has both internalistic and externalistic elements. 1
(a) and 1(b) are internalistic; experiences and beliefs are just the sorts of things
that are within an agent’s perspective, in a way that is merely stressed by the
demand that they be accessible to this particular agent. 1(c) is intended by
Alston to be externalistic; there is no suggestion that this fact need be one on
which S has any grip at all. The status of (2) in this respect is, I think,
indeterminate.
What I want to say in criticism of Alston here is that his position is one with
which nobody will agree. Our externalist will not see the relevance of requiring 1
(a)–(b), and our internalist will not see that 1(c) adds anything at all. Alston’s view
is in a sense a compromise, but it is not the sort of compromise I am looking for—
it is not one which either party is going to be able to accept. A plausible com-
promise position will be one which in some way respects the intuitions which
drove the two originally opposing views, so that each can accept the compromise
without abandoning too much of its own motivation. Alston seems to give us none
of this.
Better would be to attempt to meet directly the defects we found in the extreme
theories. What seemed to emerge from them is that we need
(1) a distinction between act (belief) and agent (believer)
which
(2) enables our assessment of them to diverge.
³ See Alston (1986: 179–221). This formulation of Alston’s position, which I use merely for an
example, comes from Audi (1988: 407–42).
   79
Something very much on these lines is supposed to be offered by the distinction

commonly drawn between objective and subjective justification. This promises to
provide a good sense to the act/agent distinction, objective justification being
justification of act and subjective justification being justification of agent. One
might worry that the two sorts of justification are not very clearly related to each
other, but that need not be so. There are various ways of running the objective/
subjective distinction, but a standard way would be this:
S’s belief that p is objectively justified iff it has properties F¹–Fn.

S is subjectively justified in believing that p iff S believes that his belief that p has
properties F¹–Fn.
Here it is clear that subjective justification is defined in terms of objective

justification, since (ignoring problems of opacity) we could write the second
clause thus:
S is subjectively justified in believing that p if S believes that his belief that p is

objectively justified.
What we purport to have here is a way of thinking of subjective justification as

concerned with the agent, and objective justification as concerned with the act.
And this is how a compromise between internalism and externalism might be
run. As a compromise, it is not a complete success, since it requires both sides to
give up something of which they were fond. But all they lose is the claim that the
sort of justification they were focusing on is not the only sort of justification.
The externalist agrees that as well as justification of act there is justification of
agent; the internalist accepts that as well as justification of agent there is
justification of act. Neither loses the story they wanted to give about the place
they started from.
But despite this, we have still not reached a genuinely intermediate position
between internalism and externalism. So far we have merely moved away from the
insistence of the two extreme positions that where the act is justified, so is the
agent. We now have room for the idea that our assessments of act and agent can
come apart; we can say that the agent may be subjectively justified in holding a
belief which is not objectively justified, or that the agent may not be subjectively
justified in holding a belief which is in fact objectively justified. But we have not yet
made the final move. We make this move when we accept that these are genuinely
independent assessments, in a way that enables us to approve fully of the agent for
holding a belief which is in fact unjustified. The objective/subjective distinction
begins to have real bite with this addition, for with this we begin to persuade
ourselves that there are two independent elements in a full epistemic story, each of
80     
which can be fully satisfied without the other. The agent story is the internalist
story; the act (belief) story is the externalist story.
The main problem that I see as facing this position is that of coping with the gap
between the two stories that it wants to keep apart. The two stories need to be
separate but not too separate. The danger is that the gap that is opening up
between objective and subjective justification will be so great that we are forced
to hold that there are really two senses of the word ‘justified’; subjective justifica-
tion is not the same sort of thing as objective justification at all, and there is no
point in pretending that it is. It is not that if we add subjective justification to
objective justification we have made things better in the same direction; rather we
have told a completely separate story. So justification is not one thing but two.
Some philosophers are not frightened by the idea that there are two senses of
‘justified’. They notice the very same move being made in ethics, associated with
the distinction between objective and subjective rightness. Again, this distinction
can be run in several ways, but a classic way of doing it would have it that:
An act is objectively right if its consequences are the best available.

S acts subjectively rightly if S believes his act to be the one whose consequences
are the best available.
We could express this using the language of ‘ought’ rather than of ‘right’ thus:
S ought (objectively) to do the action whose consequences are the best available.
S ought (subjectively) to do that action whose consequences S believes will be
the best available.
Given the obvious truth that the action whose consequences are the best available
may not be the one of which S believes this, we have it that the same action may be
both objectively right and subjectively wrong. This is in danger of meaning that
S both ought and ought not to do it; the contradiction which threatens here is
apparently only relieved by admitting that the word ‘ought’ has two distinct
(though of course related) senses. We have the very same move in ethics as we
have in epistemology.
So there is a strong pressure to accept that, once we make explicit the fact that
justification of act and justification of agent can come apart in this way, so that
each can be fully present without the other, we have in effect admitted the
existence of two senses of ‘justified’. The contradiction that drives us is the
admission that S may be justified in believing something which is not justified.
Epistemically speaking, S both ought and ought not to believe it.
By admitting that there are two senses of ‘justified’ we have implicitly aban-
doned our earlier claim to have retained a comprehensible link between subjective
   81
and objective justification. The claim that agent-assessment is independent of act-

assessment has pulled the two sorts of evaluation too far apart. So long as we
simply understood subjective justification in terms of objective justification, so
that S was subjectively justified in believing that p iff S believed himself objectively
justified in believing that p, all was well. It is when we suggest that in such a case
S is blameless that trouble arises, for we now say that S can be justified in believing
something unjustified. And to say this we have to give up the claim of harmless
interdefinability.
But perhaps things are not as bad as this makes out. The real worry was that it is
impossible to keep assessment of act and agent apart without generating two
distinct senses of the relevant evaluative terms. This worry is led by the ‘discovery’
of a contradiction, but I myself am not convinced that the discovered contradic-
tion is a genuine contradiction at all. It seems to me that anything we have yet
unearthed can be handled using a single sense of ‘ought’. We can assert that agents
are right to do what they think right, while accepting that they are sometimes
mistaken in what they take to be right, without admitting more than one sense of
‘right’—and similarly for ‘ought’. This is because the crucial remark that S is right
to do what he thinks right is structurally ambiguous. We can express the ambi-
guity at issue here using a simplified form of Hintikka’s deontic logic.⁴ Take the
following manual:
BSp: S believes that p

Sda: S does a
Rp: it is right that p.
The ‘R’ in ‘Rp’ is a modal operator, and ‘Rp’ is interpreted as saying that in all
deontically perfect worlds, p. The sentence ‘S is right to do what he thinks right’
can mean either
(1) (BS (RSda) ! RSda),
which makes S infallible on such matters, or
(2) R(BS (RSda) ! Sda),
which does not. If we read our English sentence as being of the first form, and
want to say that some action which S believes right is in fact wrong, we have to
admit two senses of ‘right’. For there is a contradiction in:
⁴ I first presented this idea in ‘The Logical Conscience’ (1977; repr. as Chapter 1 in this volume).
82     
(1a) (BS(RSda) ! RSda) & BS(RSda) & –RSda.
If we use the second form, we can make do with a single sense. There is no
contradiction in:
(2a) R(BS(RSda) ! Sda) & BS(RSda) & –RSda
Note that we cannot infer (1) from (2) directly without committing the standard
modal shift fallacy of moving from
(3) N (p ! q)
to
(4) p ! Nq.
We might hope to support the inference in the special case of (1) and (2) by
introducing a separate principle
(5) R(Sda ! RSda)
but I do not see this as a reliable move, for the reason that I do not imagine that
every action performed in a perfect world is itself going to be morally perfect or
right. It seems to me that at least some actions done there will be morally
indifferent. A morally perfect world is surely to be thought of as one in which
every action right in the actual world is done, not as a world in which everything
done is morally right.
So I do not think that the distinction between objective and subjective justifi-
cation is necessarily associated with the discovery of a contradiction of the sort
contained in the thought that conscience is both paramount and liable to error, in
a way which can only be resolved by accepting that there are two senses of ‘right’,
‘ought’ or ‘justified’. What is more, if we were to allow the existence of two senses
in these cases, we would have to allow them in every structurally similar case—and
there are too many such cases for us to feel at all comfortable here. For instance, it
is silly to do what one thinks it is silly to do; it is rational to do what one thinks it
rational to do; it is cruel to do what one thinks it is cruel to do; it is pointless to do
what one thinks it is pointless to do: and so on. Are we to announce similar
ambiguity in each such term, or are we to allow the existence of the same
structural ambiguity case by case? Simplicity argues strongly for the latter.
So it may be possible to keep the assessment of act and agent far enough apart
to satisfy the purposes of the compromise position, without pulling them so far
apart that we are left with two different senses of the relevant evaluative terms. But
   83
we should not forget that the ‘two senses’ theory had a purpose. This was that it
made it possible for us to live with an otherwise awkward fact, that our com-
promise position can recommend an act to an agent which it then says it would be
wrong of him to do. It seems undeniable that this is what is going on within the
full-blown version of the subjective/objective distinction. The belief is recom-
mended by the assertion that it is justified, and then the agent is told that, given
his circumstances, it would be wrong of him to do it.
Can we escape this result by appeal to the structural distinction I drew between
(1) BS(RSda) ! RSda
and
(2) R(BS(RSda) ! Sda)?
I doubt it. It seems to me that we are still left with an incoherence. Take any case
where S believes that it is right for him to do a, when in fact it is wrong (not right)
for him to do a. (Analogy: S believes that his evidence supports p when in fact it
supports not-p.) Does the theory recommend his doing a? Not exactly, but it
recommends his either doing a or ceasing to believe that it is right for him to do a.
But of this disjunction it recommends that S not do the first disjunct, and it seems
to me that the whole thrust of the theory is that it can approve (and so recom-
mend) one’s believing that it is right to do an action which in fact it is wrong to do.
So in the sort of case we are considering it recommends that S not do the second
disjunct either. But a theory which recommends a disjunction while recommend-
ing that we do neither disjunct is incoherent. (Of course it would be coherent if it
merely failed to recommend either disjunct.) So we are still stuck for an account of
how our compromise theory can keep its evaluations of act (belief) and of agent
coherent.
This is where the compromise position makes its real appeal to ethics. Surely we
are all familiar with the suggestion that there can be an action which is itself right,
but which the agent derives no moral worth from doing. And we can cope equally
well with the idea that an agent can acquire moral worth from an action which,
through no fault of his own, turns out badly. The action makes the world worse,
but the agent was right to do it.
I confess, however, that I find these well-known views hard to accept in ethics.
Take the classic example of the Pharisee who gives alms to charity ostentatiously.
Are we really to allow that the action is right here? This is to suppose that the
motive is no part of the action, so that what is done is not affected in itself by why
it was done. But I can see no rationale for carving things up in this way. The
motive is not a peculiar kind of cause, which is all over with by the time the action
starts. I want to say that the motive infects the action, because it plays a large part
84     
in determining which action was done. A bad motive will mean a bad action, no
matter how well things turn out. In the contrary case, where the action is supposed
to be bad but the agent right to do it, it is not clear to me that the right way to cope
with this is to cut the action off from the agent so that the two are assessed
separately. I suggest instead that we could distinguish the worth of the agent-
in-the-action from any value the consequences may have.
So I am not much impressed by the ethical analogy to which we were appealing
to show how it can be coherent to hold a theory that recommends an action
(belief) which it then says the agent is wrong to do (believe). It may be that a
theory that does this is not in direct contradiction with itself, but it does seem to be
issuing contrary recommendations, and this is bad enough.
One might fail to see the problem here for either of two reasons. First, one
might suppose that the belief is called justified because it is probable given the
evidence that there is (the evidence available), rather than because it is probable
given the evidence that S has. If this were so, calling the belief justified would not
yet be to recommend it to anyone; all epistemic recommendation would occur at
the level of the agent, and a belief which is not probable on the evidence that there
is might still be one that given the evidence we have we are all right to believe.
Here there are not two recommendations, but only one. But this would be to
confuse the theory of justification of belief with the theory of probability. One
theory may of course build on the other, but they are not identical. It is not the
proposition believed that we are concerned with in the theory of justification, but
the belief in that proposition. Equally, the belief we are concerned with is not an
ownerless belief which is somehow justified in abstraction from any owner. It is
justified as a belief of S’s, though unfortunately this does not seem to mean that
S is to be justified in believing it.
The second way of failing to see the problem is similar. We might make the
mistake of supposing that our account of what it is for a belief to be justified is
non-normative in some strong sense. Any such view is hopeless, I think. We may
indeed specify non-normative conditions under which we accept a belief as
justified (e.g., being the product of a reliable method), but this should not obscure
from us that we are setting ourselves an aim in the adoption of the theory, namely
the aim of acquiring beliefs like this. I do not see how a theory of justification can
fail to be in the business of specifying a cognitive aim in this sense; it is making
recommendations. But, given that, the problem is how to give a coherent story
about the two sorts of recommendation it makes, especially since we admit, or
rather insist, that the very same belief that in one voice it recommends it may also
in another forbid. Effectively, the theory will on occasions specify an aim for us
and then tell us not to aim for it. I take this to mean that the theory is self-
defeating in a damaging way.⁵
⁵ I do not have space to show this here, but I attempt to rebut current claims that being self-defeating
in this way does not matter in my ‘Parfit and Self-Defeating Theories’ (1997a).
   85
The thrust of my criticism of the compromise theory has been to cast doubt on
the act/agent distinction which it uses. One might think that this is clearly wrong.
After all, we obviously need some distinction between objective and subjective
justification (as we do between objective and subjective rightness), and surely this
is all the act/agent distinction amounts to. I think, however, that this reply is a
mistake. In my view, the distinction between act and agent is just a different
distinction from the distinction between objective and subjective anything. We
originally held that what was necessary was a way of distinguishing what we said
about the belief (action) from what we said about the believer (agent). It was a
simple assumption that the right way to do this was to introduce the distinction
between objective and subjective justification. But one can distinguish between the
objective and the subjective properties of an action, without running anything
much in the way of a distinction between agent and action. Equally one can
distinguish between properties of agent and properties of action without concern-
ing oneself about the difference between objective and subjective properties. To
give an example of the sort of thing I mean here, consider the opening chapter of
Derek Parfit’s Reasons and Persons (1984). He offers four (not two) clauses.
Subjective and objective rightness for acts are defined as follows (1984: 24):
An act is objectively right if its outcome is the best possible.

An act is subjectively right if the agent believes that its outcome will be the best
possible.
The worth of agents is defined differently, in terms of that of motives or sets of

motives. Parfit is discussing a theory C (consequentialism) which defines rightness
of act in terms of value of outcomes. A motive is a member of a set of motives
which C can recommend if having that set of motives would lead to outcomes at
least as good as the outcomes of any alternative set of motives. We can call such a
set of motives ‘C-approved’, and define something called ‘agent-rightness’ as
follows:⁶
An agent acts objectively rightly (here) if his motive is a member of a C-

approved set of motives.
An agent acts subjectively rightly (here) if he believes his motive to be a member
of a C-approved set of motives.
An epistemological theory T could have the same structure as Parfit’s C has

here. We could hold:
⁶ Notice that this notion offers an evaluation of the agent in doing a particular act; it purports to be
atomistic.
86     
A belief of S’s is objectively justified if S’s evidence supports it.

A belief of S’s is subjectively justified if S believes that his evidence supports it.
S is objectively justified in believing if S is using a T-approved method.
S is subjectively justified in believing if S believes his method to be T-approved.⁷
What are we to make of this? One’s first thought is that for the purposes of moving
away from the extreme theories with which I started, all we need is the objective/
subjective distinction. I am encouraged in this idea by the fact that the original
distinction between internalism and externalism can be written separately for act
and for agent (for belief and for believer). Thus we could take the externalist view
that an act is right if it has the best outcomes or the internalist view that acts are
right if they are expected to have the best outcomes (the criterion for rightness of
act is not the actual but the expected consequences). And one could run the same
distinction for agent-rightness, and for agent-justifiedness, taking the externalist
line that an agent is justified if using a truth-effective method, or the internalist
one that the agent’s justifiedness requires that the agent believe himself to be using
a truth-effective method. So it seems odd, from this perspective, to resolve the
internalism/externalism debate by appeal to a distinction between action and
agent. It genuinely is the objective/subjective distinction that we should be
using, and all the talk about act and agent was an irrelevance.
For myself, I have no trouble with this. I accept the need for an objective/
subjective distinction; I only want to reject the sort of act/agent distinction that the
compromise theory relies on. But the fact is that the thrust behind the comprom-
ise position does require of it both distinctions rather than merely the one. The
admission that matters is that a believer may be completely blameless in believing
even though his belief is not justified. This stress on the epistemic status of the
believer as opposed to that of the belief (the independence thesis) seems to me to
require the addition of the act/agent distinction to the objective/subjective one.
We want to be in the position of having nothing to say against the believer but
much to say against the belief, and the distinction between objective and subjective
⁷ There is a complexity here which it is worth keeping track of. Parfit mentions later (1984: 153) a
different distinction between the objective and the subjective which, written out in the terms I have
used above, would emerge like this:
An act is objectively right if its results are best.
An act is subjectively right if, were things to be as the agent generally believes them to be, its results
would be best.
The analogue in epistemology would look like this:
A belief is objectively justified if the evidence supports it.
A belief of S’s is subjectively justified if, were things to be as S generally believes them to be, the
evidence would support it.
This creates a more convincing epistemological theory than the one I have laid out in the main text. The
notion of the agent’s general beliefs is intended to draw the focus away from the belief that the outcome
of this action will be best or that the evidence favours this belief.
   87
properties of the belief is not calculated to help us much in that direction. If we

want to distribute approval in the way the compromise theory does, what we want
is not two ways of approving of one thing (why should we approve of an act just
because its agent is satisfied with it?) but something like one way of approving of
two different things. With this motivation behind us, our distinction between act
and agent is going to be a serious one rather than the lesser one capturable by
separating objective from subjective properties of acts.
Is it possible then that the act/agent distinction will do the job for the com-
promise position all by itself? I don’t think so. The act/agent distinction does not
concern itself with what happens in cases where the agent is ignorant or in error.
This is what the subjective/objective distinction is concerned with. So if we want to
approve fully of the agent in certain cases of error, we will need both distinctions,
not just one.
I take it then that a new way needs to be found of reconciling the opposing
demands of internalism and externalism, which makes room for the objective/
subjective distinction but has no serious use for the act/agent distinction. And this
is what I now attempt to provide. I start from my original expression of the
intuition underlying internalism. This was that justification must be achieved by
appeal to elements that are internal to the agent’s perspective (the believer’s
perspective). As I see it, this intuition can be more or less captured by any of
three different approaches, each of which can therefore be called internalist. The
first and most extreme of these we have already seen. It is the proposition that:
(Extreme internalism) Only beliefs of S’s are relevant to the determination of S’s
epistemic status.
A weaker but still strong form of internalism could hold this less demanding
position:
(Strong internalism) A set of justifiers which has some members not starting
BS . . . must have, for each such member p, a member BSp. BSp will be among
the justifiers.
The weakest version will hold only that:
(Weak internalism) If a set of justifiers has some members not starting BS . . .

then for each such member p it must be true that BSp, though BSp need not be
among the justifiers. In such a case, BSp will be not a justifier but an enabler.⁸ It
enables p to contribute to the justification, without acting as a justifier itself.
⁸ Steven Strange suggested this term ‘enabler’. When I first mooted the idea, on the final page of my
Introduction to Contemporary Epistemology (1985), I spoke more clumsily of ‘allowers’.
88     
Each of these positions could be an expression of the intuition that only what is in
S’s perspective can serve to justify a belief of S’s. In that sense each of them seems
to me to be internalist enough. Each of them is in opposition to the externalist
view that a fact can justify even when it is not believed by the agent. My preference
is for the third, weaker form. And the reason for this is that this form can be
shown also to capture something at least of the intuition underlying externalism.
This is in two ways. First, most of the things that occupy the justifiers box will be
facts rather than beliefs; they will be things believed rather than believings of them.
Second, many enablers will be brute facts, facts which are not believed by the
agent. So a fact does not need to be believed in order to contribute to justification,
only it cannot in that case do so by functioning as a reason.
The contrast I intend to draw between justifiers and enablers is a contrast
between considerations which are reasons and considerations which are not
themselves reasons but which make it possible for the things that are reasons to
function in that way. To give a simple example: that my nervous system is in good
working order is a truth which is not among my reasons for believing that
tomorrow is Friday, though it needs to be true for the things that are my reasons
to be my reasons. However, I do not need to believe it. In fact my reason is
probably just this, that today is Thursday. (I need to believe this, of course, or else
it could hardly be a reason for me and so justify my belief.) Not all enablers are of
this sort, however. Sometimes the fact that I do not believe something functions as
an enabler. Here I am talking about the absence of undermining beliefs. That
undermining beliefs are absent enables the things that are my reasons to be the
reasons they are. Further enablers might be beliefs which do not themselves count
as reasons, but without which one could have no reasons at all. Here I am thinking
of the belief that there are solid objects, that other people can understand what
I say, that things do not happen just because one wants them to, and so on.
Replacing talk about justifiers by talk about reasons should make one external-
ist aspect of this approach clearer, namely that it holds that facts are reasons. So
I want to claim that my approach captures, if not all the intuitions that underlie
the original extreme positions, at least enough of them to count as a genuine
compromise.
It is worth trying to lay out some differences between the strong and the weak
forms of internalism.
The first is that the weak form can accept the relevance of brute facts (i.e., the
fact that q where not BSq) to justification when they act as enablers. Such facts do
not contribute to justification in the way in which reasons do, but they are not
because of that to be excluded from the story altogether. The strong form must
claim that all such facts are irrelevant, since it only countenances one sort of way
in which a consideration can be relevant to justification.
The second difference is that the strong version has a tendency to pull every-
thing relevant to justification into the set of justifiers. Since there is only one way
   89
in which a consideration can contribute to justification, everything that does not

contribute in that way is irrelevant. The weak version is more flexible, because it
offers two ways in which something can be relevant to justification. I venture here
the observation that the availability of the weak version can be hidden from one if
one writes one’s theory as a list of necessary and sufficient conditions for justifi-
cation, in the way that I expressed Alston’s position earlier. This format imposes
on one the idea that all relevant considerations are relevant in much the same sort
of way.
The third and final difference is in a way the most telling. The strong version
encourages us to suppose that we can divide the set of justifiers up into two halves,
those which begin BS . . . and those which do not. We are then tempted to hold that
where all the BS . . . members of the set are true, the relevant belief is in some sense
already justified, though its justification can increase. (Note however that the fact
that we are tempted to hold this does not mean that we can make very good
theoretical sense of it.) The move we are considering here is the one which gives us
the distinction between justification of act and of agent. The weak version does not
encourage this sort of move. On the weak version, when all the relevant BS . . .
clauses are true, justification is now possible, but there is no temptation to suppose
that the relevant belief is already to some extent justified, or that the believer is
justified though the belief is not.
This is the other reason why I am not happy with the compromise built round
the act/agent distinction. I find the story about enablers plausible in its own right,
and it fits what I want to say in the theory of motivation and elsewhere. I want to
find in other places the same tripartite structure that I find here between justifiers,
enablers and pure background. In the theory of motivation, I want similarly to
distinguish between motivating reasons, states which enable those motivators to
motivate, and states which play no such role. Equally, in the theory of moral
properties, I want to distinguish between properties that make the action good,
properties whose presence or absence enables the first properties to make the
action good, and properties that are pure background (here). So I have a general
approach of which this is merely one application. Given this, I find it hard to
accept as well the distinction between justification of act and justification of agent.
That distinction leads us to suppose that we can sort out the agent’s justification
before we turn our attention to the justification of the act (belief), in a way that
means that act-justification neither helps nor harms agent-justification. We find
ourselves supposing that in cases of well-intentioned failure we do not blame the
agent, so that agent-assessment is entirely independent of act-assessment. But that
is not at all the way my attempt to find a compromise between internalism and
externalism is supposed to work. For me, act-assessment depends upon agent-
related facts, and those agent-related facts (that S believes this or that) are largely
not in the set of justifiers at all. They are not thought of as being in the set of
justifiers, except in a special part of it labelled ‘justification of agent’. So it is hard
90     
for me to add the act/agent distinction in any serious form to what I already have
in place.
Of course this leaves me with the question what to say about a case where the
agent is blameless though misguided. The general view must be that one can say
everything that needs to be said here without any radical distinction between
assessment of act and assessment of agent. But I cannot hope to establish this here.
I end then by trying to lay out the picture as I see it. S’s belief that p is justified if
S’s reasons for believing that p are good reasons (are among the reasons). S’s
reasons will mainly consist of facts; that S believes those facts is probably not
among S’s reasons for believing that p, though the fact that one believes that q can
sometimes be a reason for believing that p. This is not ruled out, but it is not the
normal situation. The facts that are S’s reasons can only play that role if S believes
them, but that is not to mean that S’s reasons are beliefs of S’s. That S believes
them enables them to justify and is not itself a justifier. Often the facts that justify a
belief will not be very extensive. For instance, what justifies my belief that it is
raining outside? Answer: I can see and hear the rain. This might be the whole
answer. Don’t I need also to believe that I can hear and see the rain? Yes, but this
belief is an enabler and not itself among my reasons. Mustn’t it be true that human
perceptual mechanisms are in general reliable? Yes, but again this functions as an
enabler. Though it needs to be true, it does not need to be among my reasons. It
makes it possible for what are my reasons to be the reasons they are. Of course
I can believe that human mechanisms are reliable, and this fact (the fact that they
are reliable, not the fact that I believe them to be so) could then be a justifier; but it
doesn’t need to be. Mustn’t it be true that I am not asleep, dreaming that I hear
and see the rain? Yes, but this fact is another enabler. I do not need to believe it for
it to play this role. Only facts that are reasons need to be believed, and the fact that
I am not asleep dreaming is not among my reasons. My reason was simply that
I can hear and see the rain.⁹
⁹ I am very grateful to Richard Gale for discussion of an earlier and very different version of this
paper.
7
The Particularist’s Progress
Many people seem willing to call themselves feminists with little idea of what they
are committing themselves to. The same, in my experience, is true of particularism
in the theory of moral reasons. There is a common suggestion that to be a
particularist is, at the outset, only to admit that circumstances can make a
difference. But if that were all that particularism amounted to, it would be
uncontentious. In this paper I lay out what I think one commits oneself to if
one accepts the general claim that reasons are sensitive to context—a claim
sometimes called holism in the theory of reasons, and of which moral particular-
ism is merely one expression.
Of course holism here, as elsewhere, does come in degrees. The strongest form
of context-sensitivity would be the claim that every reason is somehow altered
with every change of context. The weakest form is the claim that some reasons are
on occasions capable of being altered by a change in context. The form of holism
that I recommend is pretty weak on this scale, so far as the modality goes, but
strong on the extent of the domain. I maintain that all reasons are capable of being
altered by changes in context—that there are none whose nature as reasons is
necessarily immune to changes elsewhere.
When I talk of altering a reason, I mean to suggest not that the consideration
which is a reason is altered, but that its nature as a reason changes. Instead of
being a reason in favour of some course of action, it ceases to be a reason for action
at all, or even becomes a reason against. One could express this by saying that the
practical relevance of the consideration at issue is sensitive to changes in context,
and the practical relevance of the consideration includes its polarity.
A consideration reverses its polarity when, having been a reason in favour of
action, it becomes a reason against, or vice versa. My holism holds that every
consideration is capable of having its practical polarity reversed by changes in
context.
It is hard to be sure quite how extreme a claim this is, partly because of the
awkward modality in its characterization. But I shall not be discussing that matter
much in this paper. Perhaps I will have to admit that not all reasons are sensitive
to context in this way—that there are a privileged few, including probably the
intentional inflicting of undeserved pain, which necessarily constitute the same
sort of reason wherever they occur. If so, I will have lost a battle but won the war.
For the main aim of my particularist position is to break the stranglehold of a
certain conception of how moral reasons function—the generalist conception
92     
under which what is a moral reason in one situation is necessarily the same reason
wherever it occurs. Generalism need not be false of every moral reason in order to
be largely false, and hence false as a general account of moral reasons and the way
they work. And if it is false as a general account of such reasons, rational
constraints on moral thought and action—in particular, accounts of what con-
sistency requires in these areas—must not themselves be based on generalist
assumptions.
It may be that my train of thought here is vitiated by being overconcerned with
one specific opposing account of how reasons function—Ross’s theory of prima
facie reasons. But I might as well admit that I do have this theory constantly in
mind, since it seems to capture so well the outlines of the position I am trying to
dislodge. Omitting Ross’s epistemology for the moment, the theory of prima facie
reasons holds:
1. What is a reason in one case is the same reason in all.¹

2. Judgement is the attempt to determine the balance of reasons, so conceived.
Holism in the theory of reasons
In this section I argue in favour of particularism in ethics. In the past I tended to

argue largely from example (e.g., Dancy 1993: ch. 4). This persuades some people
but not others. Here my argument will be more theoretically grounded—though
there will still be considerable use of examples as well.
As I said in the preamble, I see ethical particularism as merely one expression of
an overall holism in the theory of normative reasons—i.e., in the theory that
discusses the reasons that favour one thing (action, belief) over another. Such an
overall holism can be expressed as follows:
1. What is a reason in one situation may alter or lose its polarity in another.
2. The way in which the reasons here present combine with each other is not
necessarily determinable in any simply additive way.
There are theoretical reasons and practical reasons, reasons for belief and
reasons for action. My holism is intended to hold on both sides of that distinction.
I start by trying to establish that theoretical reasons are holistic. We will quickly
find that theoretical reasons are perfectly capable of changing their polarity
according to context, without anyone making the slightest fuss about the matter.
For instance, suppose that it currently seems to me that something before me is
¹ This is not fair to Ross: see McNaughton (1996).

 ’  93
red. Normally, one might say, that is a reason (some reason, that is, not necessarily
sufficient reason) for me to believe that there is something red before me. But in a
case where I also believe that I have recently taken a drug that makes blue things
look red and red things look blue, the appearance of a red-looking thing before me
is reason for me to believe that there is a blue, not a red, thing before me. It is not
as if it is some reason for me to believe that there is something red before me, but
that as such a reason it is overwhelmed by contrary reasons. It is no longer any
reason at all to believe that there is something red before me; indeed it is a reason
for believing the opposite.
As I say, it seems to me that nobody ever thought of denying what I am
claiming here. I know of nobody who has nailed themselves to an atomistic (i.e.
non-holistic) conception of how theoretical reasons function. If generalism is
taken to be the view that all reasons are general reasons, i.e. that if a feature is a
reason in one case, it is the same reason in any other case, generalism is uncon-
tentiously false of theoretical reasons.
Let us now turn to ordinary practical reasons. We will find just the same thing
there. There are plenty of examples to persuade us that such reasons are holistic (or
non-generalist, if you like). For instance, that there will be nobody much else there is
sometimes a good reason for going there, and sometimes a very good reason for
staying away. That one of the candidates wants the job very much indeed is
sometimes a reason for giving it to her and sometimes a reason for doing the
opposite. And so on. Now examples would be of little use if there were some
theoretical obstacle to taking them at face value. But again we should remind
ourselves that nobody has ever really debated the question whether ordinary prac-
tical reasons are holistic or not. There should be no parti pris on this issue; so the
examples, which are legion, should be allowed to carry the day without resistance.
Perhaps this is too quick. There is a theory-based reason for doubting my claim
that practical reasons are holistic, one that derives from the common thought that
practical reasons are grounded in desires of the agent in a way that theoretical
reasons are not. What one wants should not affect what one judges to be the case,
on pain of charges of bias or prejudice. But what one wants can perfectly well
affect what one has reason to do. Indeed, many find it hard to conceive of our
having any practical reasons at all if we had no desires. My own view on this
matter, however, is that desires do not give us, or ground, our reasons. Reasons
stem from the prospect of some good. If we have no other reason to do a certain
action, wanting to do it will give us no reason at all; nor can wanting to do a silly
action make it marginally less silly. (These are only the first moves in a long
debate.² I mention them here only to show the sort of way in which I find myself
denying the possibility of grounding practical reasons in desires of the agent.)
² For the remainder of the debate, see my Practical Reality (2000a: ch. 2).
94     
This view of mine is, of course, an independent input in the present debate.
I mention it only to show that a certain motive for doubting the analogy I have
been drawing between theoretical and practical reason is itself contentious.
It may be that here we come across the real motivation for generalism in the
theory of practical reason—an adherence to the view that reasons for action are
partly grounded in desires. For if we accept that view, and if we then think of
desires as giving the desirer the same reason wherever the desire occurs, we will at
least get the sort of generalism I discussed above. The right response to this,
however, is to claim that even if all practical reasons are grounded in desires, the
same desire need not always function as the same reason. Consider first the third-
person case. That he wants power and she does not may be a reason to give the
power to her rather than to him, as I have already said. (It may at the same time be
a reason to give it to him, since according to me one feature can be a reason on
both sides at once; but remember that here it is a reason not to give it to him, and
that it need not always be such a reason.) Now consider the first-person case.
Suppose that I am trying to train myself into indifference towards a girl. I want
very much to spend time with her. But I also want not to have this want, since she
is permanently indifferent to me. It is better for me not to think of her at all. If
I spend time with her, this will make things worse for me rather than better—so
long as I have not yet succeeded in training myself into indifference towards her.
Once I am indifferent towards her, I can spend time with her without loss. In this
situation, it seems, my desire to spend time with her may be a reason for me not to
do so.
Before carrying on to consider moral reasons, which have been claimed to be
non-holistic, I want to step aside for a moment to ask whether I have not already
made a mistake. There is a distinction between epistemic and what one might call
constitutive reasons. An epistemic reason is a reason for believing something or
other; a constitutive reason is a reason why something or other is the case. That
the butler’s fingerprints are on the murder weapon is a reason for believing that he
did the deed, but no part of what makes it the case that he did it or of why it is true
that he did it. That the hedgehogs are hibernating early is a reason for believing
that we will have a severe winter, but not any part of what makes it the case that
the winter will be severe. And so on. Now holism in the theory of reasons should
concern itself with constitutive reasons rather than with epistemic ones. But
I appear to have argued only that epistemic reasons are holistic, for my first
example, or fulcrum, concerned reasons to believe that there was a red thing
before me. It is, therefore, technically irrelevant.³
This is true, and I apologize for it. But matters can be redeemed. We should not
suppose that all that I have shown is that epistemic reasons are holistic, it being left
³ Thanks to Nick Zangwill for pointing this out to me.

 ’  95
entirely open whether constitutive reasons are or are not. For many, possibly most
epistemic reasons are also constitutive. For instance, that an action involves the
gratuitous inflicting of pain is held by many to make it wrong, but equally clearly
functions as a reason to believe that the action is wrong. It is both an epistemic and
a constitutive reason. Some epistemic reasons are not constitutive, and perhaps
some constitutive ones are not epistemic; this is all that can be said. Now could it
be the case that the epistemic ones are holistic but the constitutive ones are not?
I think that this is inconceivable. The mere fact of the overlap between reasons of
the two sorts should give us pause. But more importantly, can we suppose that the
very logic of epistemic reasons is capable of differing at a very deep level from that
of constitutive reasons? This supposition entirely undermines the sort of connec-
tion there needs to be between reasons why things are so and reasons for taking
them to be so.
I return, therefore, to the onward or outward spread of holism. So far we have it
that theoretical reasons (constitutive ones) are holistic, and so are ordinary
practical ones. Now could it be the case that moral reasons are quite different
from others in this respect, being the only atomistic ones? This is what many
have supposed, in supposing that moral rationality is based on the existence of a
range of moral principles. Moral reasons, they have held, necessarily behave in
regular (or rule-bound) ways, though other reasons see no need to behave in that
way at all. About this I want to say that straight off it just seems incredible
that the very logic of moral reasons should be so different from that of others in
this sort of way. Consider here the sad fact that nobody knows how to distin-
guish moral from other reasons; every attempt has failed. How does that fit the
suggestion that there is this deep difference between them? Not very well at all.
Then of course there are examples to be considered, examples of apparently
moral reasons functioning in a holistic way. I forbear to bore you with these. It
just seems inevitable that moral reasons should function holistically in the way
that other reasons do.
This certainly makes it hard to hold, as many do, that the very possibility of
moral distinctions, of moral thought and judgement, is predicated on the existence
of a range of moral principles. Moral principles, however we conceive of them,
seem all to be in the business of specifying features as general reasons. The
principle that it is wrong to lie, for instance, presumably claims that mendacity
is always a wrong-making feature wherever it occurs (pro tanto, of course, not
necessarily absolutely). It cannot be merely a generalization, a claim that lies are
mostly the worse for being lies, for if all moral principles were of this sort, the
argument that moral thought and judgement depend on the possibility of moral
principles would simply be the argument that such thought is impossible unless
there is a considerable preponderance of normal cases over abnormal ones. I have
never seen this argument made, and I doubt, what is more, whether it would be
persuasive if restricted to ethics.
96     
If moral reasons, like others, function holistically, it cannot be the case that the
possibility of such reasons rests on the existence of principles that specify morally
relevant features as functioning atomistically. A principle-based approach to
ethics is inconsistent with the holism of reasons.
All the same, it might be argued, we have to admit that there are some invariant
reasons—some features whose practical relevance is invariant. And surely I should
allow this, because holism, as I expressed it, concerns only what may happen, not
what must. It could be true that every reason may alter or lose its polarity from
case to case, even though there are some reasons that do not do this. If they don’t
do it, this will be because of the particular reasons they are. Invariant reasons,
should there be any, will be invariant not because they are reasons but because of
their specific content. And this is something that the particularist, it seems, should
admit. It is like the claim that a man can run a mile in four minutes, that Sam
Smith is a man, and that Sam Smith cannot run a mile in four minutes. These
claims are compatible, and so are the claims that reasons are variable qua reasons
though some reasons are (necessarily, given their content) invariant. The invari-
ance, where it occurs, derives not from the fact that we are dealing here with a
reason but from the particular content of that reason.
So can the particularist admit the existence of some invariant reasons? The
obvious examples are things like the causing of gratuitous pain on unwilling
victims. Surely, it is commonly urged, this is always for the worse, even if overall
we might in some case be morally forced to do it. Well, the first thing to say is that
admitting the possibility of some invariant reasons is a far cry from admitting that
the very possibility of moral thought and judgement is dependent on our being
able to find some such reasons. To support any such suggestion, we would
somehow need to be able to locate a sufficient range of invariant reasons, ones
that together somehow covered the moral ground entirely and themselves
explained the nature and role of the variant reasons. This is quite a different
matter from simply trying to refute particularism (which is merely an application
of holism in the general theory of reasons to the moral case) by producing one
counterexample of an invariant reason, which is normally what is going on.
Further, we should remember that the question whether reasons are atomistic
or holistic is a very basic question about the nature of rationality, of how reasons
function from case to case. It is, I suppose, conceivable that though the vast bulk of
reasons function holistically, there are a few that function atomistically. But if this
were true we would have a hybrid conception of rationality. There would just be
two sorts of reasons, each with their own logic, and moral thought would be the
uncomfortable attempt to rub such reasons together. It is much more attractive, if
at all possible, to think of our reasons as sharing a basic logic, so that all are
atomistic, or all holistic.
Let us consider, then, how the supposed invariant reasons function as reasons
in the particular case. Take the well-known example of the fat man stuck in the
 ’  97
only outlet from a cave that is rapidly filling with water from below. We and our
families are caught in between the fat man and the rising water. But we have some
dynamite. We could blow the fat man up and get out to safety. But the fat man is
unwilling to be blown up (he, at least, is safe from drowning); and, let us
immediately admit, he is blameless in being where he is, and in being fatter
than the rest of us. So what we propose to do involves the destruction of an
unwilling and blameless victim. As such, we might say, this is some reason against
lighting the fuse and standing back. The question I want to raise is whether this
feature (that we are causing the death of an unwilling and blameless victim) is
functioning as the reason it here is, in any way that is to be explained by appeal to
the (supposed) fact that it functions in the same way in every case in which it
occurs at all. It seems to me that this feature is the reason it is here quite
independently of how it functions elsewhere.
Of course, if the feature is genuinely an invariant reason, this fact, should we
discern it, will be of use to us in any case where we might be in doubt as to the
contribution it is making. We can say: ‘This is an invariant reason, it makes such-
and-such a difference there, and so it must be making that difference here.’ But
suppose that we were to treat one of these supposedly invariant reasons as
potentially variant, so as to deny ourselves the use of that inference. What sort
of mistake would we have made? Would it be a failure of rationality to treat an
invariant reason as potentially variant, or just a mistake of fact? I suggest that the
invariance of the reason is an epistemic matter rather than a constitutive one. That
the reason functions invariantly is a clue to how it is functioning here, but in no
way constitutes the sort of contribution it makes to the store of reasons here
present. In that sense, the invariance of its contributions is not a matter of the logic
of such a reason, and failure to treat the reason as functioning invariantly is not a
failure to understand how it functions as a reason. It is a perfectly good reason case
by case without our worrying about how it operates elsewhere.
I conclude, then, that particularism should accept the possibility of invariant
reasons, so long as the invariance is not a matter of the logic of such reasons, but
more the rather peculiar fact that some reasons happen to contribute in ways that
are not affected by other features. We can admit this without adopting a hybrid
theory of rationality, so long as we treat the invariance of any invariant reason as
an epistemic matter rather than as a constitutive one.
Holism in the theory of value
The next question concerns whether our holism in the theory of reasons spills over
to generate a holism about value. This new holism, value holism, can come in
various forms, just like the holism of reasons. In broadest outline, in my hands it
will amount to the claim that for any x that has value in some context, x may have
98     
a different value or none at all in other contexts, and if there is disvalue as well as
value, x may have value in some contexts and disvalue in others.
Is there any prospect of accepting a holism of reasons and denying value
holism? This is not a matter on which there is a long history of debate.
Presumably we should approach the issue by thinking about the relation between
reasons and values. Thomas Scanlon has revived the view that value is to be
understood in terms of reasons (1998: 95–7).⁴ He defines a valuable object as one
that has features that give us reasons to protect, promote, admire, respect, approve
of it (etc.: this list is open-ended). On this view, it seems inevitable that a holism of
reasons will generate a holism of value. There are, however, other views to be
considered. One might allow that wherever there is value, there will be reasons,
but leave it open whether there can be reasons that are not directly connected with
value.⁵ In this way, one would make room for certain forms of deontology,
without moving very far from the Scanlon position. Michael Slote (1983: ch. 5)
and Roger Crisp (1996) suggest that there are some reasons that do not derive
from values. One might also ask whether there are agent-relative values as well as
agent-relative reasons; perhaps agent-relative values do not ‘stem from’ values at
all. There is, then, a spectrum of views to be considered. On the most trenchant
views, a holism of reasons must be matched by a holism of values. But even on the
less trenchant views, it still remains possible that the holism of reasons must
match that of values. Suppose we agree for a moment that most reasons are linked
to values. (‘Linked to’ is the vaguest phrase I can think of.) And suppose that
values are atomistic, i.e. insensitive to context. Immediately we have the problem
of explaining how it can be that the reasons linked to those atomistic values are
able to be holistic at all. Why are they not atomistic too? Of course, if there are
some reasons that are not linked to values, and these too are holistic, we might
think this some justification for supposing that the explanation of their holism will
not make reference to any holism of values. And we could then argue that, by
parity of reasoning, the holism of other reasons is to be explained in a similar non-
value-related way. One trouble with this procedure, however, is that the propor-
tion of value-linked reasons seems to be enormous, at the least, and we are
therefore in danger of letting the tail wag the dog. Another trouble with it is
that the connection between reasons-holism and value-holism seems so plausible.
It is far easier to explain the holism of the few non-value-linked reasons in terms
of their relation to the holism of the many value-linked ones than to cast off the
only obvious prop we have.
I take it, then, that reasons-holism does not entail value-holism, since it is at
least possible that reasons-holism is to be explained in other ways. By far the most
⁴ Similar views were considered by Ross (1939: 278–83), and adopted by Ewing (1947: 148–9).
⁵ I consider this possibility in ‘Should We Pass the Buck?’ (2000b; repr. as Chapter 14 in this
volume).
 ’  99
plausible picture is that, just as most reasons are linked to values, so their holism is
linked to a holism of values. This leaves us with a strong incentive to be par-
ticularists in the theory of value. Of course this incentive needs to be supported by
examples of values varying according to context. But these are not too hard to
produce. The disvalue of pain may be affected by the question whether it is part of
a merited punishment. This does not mean that the pain hurts less, but that the
punishment is not as much the worse for involving the inflicting of the pain as we
might have been led to think by considering the disvalue of other pains that hurt
as much.
If value-holism shadows reasons-holism, the two views have to be structurally
similar. Now as far as reasons go, the holism that appeals to me holds:
1. What is a reason in one situation may not be the same reason in another; it
may even change its polarity.
2. The way in which the reasons here present combine with each other is not
necessarily determinable in any simply additive way.
By analogy, then, our value-holism should look like this:
1. A feature or part may have one value in one context and a different or
opposite value in another.
2. The value of a complex or whole is not necessarily identical with the sum of
the values of its elements or parts.
And this is therefore the form of value-holism that I adopt. In this I differ from
G. E. Moore, despite that fact that we could both be called ‘organicists’ in the
theory of value. Moore believed (2) above but not (1). He held that any feature or
element necessarily retained the same value as it moved from context to context,
but that it could contribute to a complex of which it was a part a value other than
the one that it had there. The whole, that is, could be more valuable because of the
presence of a certain part than could be explained by the value of that part; a part
can contribute more, or less, value than it actually has.⁶ I don’t believe any of this.
What explains the difference between Moore and me? This difference needs
explaining, since all the examples that impressed Moore are just the sort of
examples that impress me. How then have we come to such different conclusions?
The answer lies in the fact that Moore accepted without question a certain
doctrine of supervenience. He believed that the intrinsic value of something
supervenes upon its other intrinsic qualities, so that where an element does not
change its intrinsic qualities on moving from one complex or whole to another, it
⁶ See Moore (1903: 30 [1993: 81]).

100     
must retain the same intrinsic value. The examples of organic wholes that
impressed him required him therefore to say that elements can contribute to a
complex more, or less, value than they have got themselves there. I, however, do
not accept Moore’s doctrine of supervenience. I accept (or used to accept, and as
far as the present debate goes continue to accept) a slightly but crucially different
supervenience doctrine, that intrinsic value supervenes upon other qualities, but
not that the intrinsic value of one object supervenes upon its other qualities. For
me, it supervenes upon other qualities, including those of other objects.⁷
This may seem perverse. But there is a ready explanation of it, once one
remembers my commitment to particularism. First, I distinguish between those
features from which some value results (the good-making features, as we might
put it), and other features whose presence or absence would have made a differ-
ence. The latter features are obviously relevant to the value, but they are not
playing the same role as that played by the good-making features; they are not
themselves part of what we might call the ‘resultance base’. Given this distinction
between roles, I can announce that intrinsic value is value that results from
intrinsic properties of the object concerned, but also allow that that value can
vary because of changes elsewhere, that is, in those properties whose presence or
absence can make a difference to the ability of the intrinsic properties to generate
the value that they do. The notion of supervenience draws less fine distinctions
than that of resultance, and all we can say is that intrinsic value supervenes upon
other properties, not particularly upon intrinsic properties. And once we have said
that, there is no reason to stop short of allowing that the ‘other properties’ can be
properties of other objects, or relations between them and the first one, or
whatever.
My doctrine, then, is a sort of global supervenience, since the supervenience
base is cast so wide, while Moore’s is a sort of local supervenience. Who is right?
I do not see that there is any way of determining the answer to this question before
we get down to arguing about particular cases. It is not, that is, going to be a logical
question which of us is right. Nor is it going to be decided quickly by appeal to the
notion of a reason, or to naturalism in metaphysics, or anything like that.
So the overall situation, as far as value-holism goes, is that if values are to track
reasons, and if the structure of value-holism is therefore to be the same as that of
reasons-holism, we have to abandon one traditional formulation of the doctrine of
supervenience in favour of something less familiar. But to appeal to the traditional
doctrine to defeat my form of value-holism would be to beg the question.⁸
⁷ For an early proponent of this view of supervenience, see Ewing (1929: 166): ‘It does follow from
the conception of goodness or value that the value of something cannot be different except as the result
of some other difference, but this difference need not necessarily lie in the thing itself, it may lie in
something else. We cannot therefore say that the intrinsic value of any quality will always be the same
under all circumstances.’
⁸ It seems to me that Susan Hurley does make this appeal (1989: 235–7).
 ’  101
Holism in the theory of choice
Let us suppose now that I am right on both counts: my reasons-holism is the truth,
and my value-holism is the truth as well. What does this tell us about the
possibility of a full ordering, in which everything has its place, and where for
each A and each B, A is either better than B, worse than B, or roughly as good as B?
Well, so far as what we have so far seen will take us, there might still be a full
ordering of that sort, in which everything has its place in the table of values, from
best to worst. For all the considerations we have so far adduced concern the way in
which the value of a complex is determined from the values of its parts, and the
way in which parts may change their values as they move from one complex
(context) to another. Once the value of the whole is determined, however, it is not
going to vary, and we can enter the whole in its proper place in the great ordering.
This would mean, for instance, that transitivity is not threatened by particularism.
For with everything in its own place, we are never going to get a situation in which
A is above B, B above C, and C above A in the Great Order.
This pleasing picture is however not as secure as it initially appears, sad to say.
Let us remember that some of the things that have value are actions, and actions
are chosen out of a set of alternatives available to the agent at the time. Now to
adopt the picture I have just described is to suppose that each alternative action
has its place in the great ordering of values, a place that is not affected by the place
or nature of other alternatives. And this seems to require that the value of an
action is never affected by the question what the available alternatives are. Now
this is a very attractive doctrine indeed, partly because it enables us to retain a
plausible principle of rational choice which we might call the principle of the
indifference of independent alternatives (IIA):
IIA: If in one situation I prefer action A to action B, it can never be rational

for me to prefer B to A in other situations which differ from the first merely in
the fact that further alternatives are available.
In simpler terms, if I choose A where my choice is between A and B, I cannot

rationally choose B where my choice is between A, B and C. The availability of
C may indeed alter my overall choice, but it cannot affect the relative ranking
order of A and B that has already been established.
Though this principle is very attractive, I am not convinced that stubborn
adherence to it is fully compatible with the broadening particularist perspective
that I have been developing. For it is not obvious to me how one can prevent
available alternatives from counting as part of the context within which an action
is placed. And if one cannot do this, then the general particularist claim that
context can make a difference to value as well as to reasons seems likely to take us
102     
to the view that the value of an action or choice can be affected by the alternatives
that are available at the time.
There is a reply to this, however. The argument of the previous paragraph
might have been merely that every alternative is an object, though not all objects
are alternatives (to each other). Since every object may have its value affected by
others, every alternative may have its value affected by other objects, including
some that are alternatives to it. There could be no bar against this happening—no
bar that ensured that only those objects that are not alternatives to this one are
capable of affecting its value. But this, though true, does nothing to establish the
controversial doctrine that is really what we are after. That doctrine is that when
one object becomes an alternative to another, that change may make a further
difference to the value of the second—a difference beyond that made by the
existence and nature of the first object. And this doctrine does seem very peculiar.⁹
But not as peculiar as all that, perhaps. There may be examples of this
phenomenon—of the arrival of a new alternative making a difference to the
relative values of two existing alternatives, in ways that are not explained merely
by the existence or possibility of the thing that becomes an alternative, but rather
by its new status as an alternative. Suppose that we have two alternatives A and B,
and that we prefer B to A. Our original question was whether we might be
rationally led to prefer A to B if there appears a further alternative, C.¹⁰ Any
successful example of this must meet certain criteria. It must not be one in which
we simply change our minds about our initial ranking of A and B, perhaps for the
reason that the appearance of C as an alternative draws our attention to something
that we had previously missed. If nothing changes, however, it is hard to see how
the ranking of A and B can be reversed by the arrival of C. The question to bear in
mind is whether the examples offered contain the right or the wrong sort of
change.
Suppose that I have to buy a house in Reading, and have a choice between a
smaller house within walking distance of the University, and a larger and more
expensive one that requires a bus ride. I prefer the larger one despite the bus ride.
Then a third house, even larger but also further away than the second, comes onto
the market. I realize that if I buy the second house, I will always regret not having
⁹ I have expressed this doctrine in terms of a change. But that need not be the point. The question
could equally well be phrased in terms of the difference between the case where C is not an alternative
and the case where it is; can the difference between C’s being an alternative and its not being one make a
difference to the relative values of A and B? Here there is no talk of change. There is, of course, nothing
wrong with examples that do involve change. It is just that I should be careful to avoid supposing that
change is essential to the point.
¹⁰ It is not, of course, strictly necessary for us to find an example in which the order of the initial
choice is reversed. It would be enough if we found a case in which the relative values are altered, so that
the one that we originally preferred we still prefer, but not by so much—or by more, perhaps. Then we
could argue that this sort of change in relative preference is bound to lead, on occasion, to a change in
ordering. But it is more striking to produce an example in which the ordering is reversed.
 ’  103
bought the third. With this in mind, I buy the nearer house. Is this rational?
I suggest that it can be.
There is of course a change of information here, but it is not the sort of change
that I tried to rule out earlier—a change that leads to a change in my initial
ranking. If there were no further house available, I would still have chosen the
second. I have not changed my mind about that. So we are not dealing here with a
simple revision of the initial choice in the light of new information.
Have I cheated in the description of the example? The obvious mistake would
be a slide in the nature of the ‘objects’ of choice. Let us be sure that these ‘objects’
are my buying house A, my buying house B and my buying house C. House C was
there all the time, but it was not on the market. The mere possibility of a further
house C that is even larger than B, though still affordable, is not enough to cause
me to prefer the nearer house. There has actually to be such a house C available to
me before my continual regret at not having bought it can turn into a reason for
me to choose the nearer one.
The simplest example I know, and perhaps for that reason the best, is found in
Wilfred Thesiger’s Arabian Sands. He wrote, about his travels across the desert on
camel: ‘I would not myself have wished to cross the Empty Quarter in a car.
Luckily this was impossible when I did my journeys, for to have done the journey
on a camel when I could have done it in a car would have turned the venture into a
stunt’ (1959: 260).
So it does look as if there may be examples in which independent alternatives
are not indifferent.¹¹ Maybe, then, even if particularism does commit us to the
existence of such examples, this is not a disaster. But I raised a question earlier that
I have not yet answered. Does particularism itself constitute a reason for rejecting
IIA? My first attempt to show that it does was a failure. I argued that particularists
should not be surprised to see the nature of one thing making a difference to the
value of another. But this was irrelevant. The real question was whether the rather
special feature of ‘being an available alternative’ can make a difference to the
relative values of two things. Particularism does not show that this must happen.
Particularism is everywhere permissive rather than prescriptive; or perhaps we
should say that it forbids some things and prescribes nothing but suspicion. In the
present case it says that that we should be open to the possibility of such a thing
and not make a fuss if some crop up. The only reason for supposing that there
¹¹ I am grateful to Lars Bergstrom for very helpful discussion of this and other potential examples.
He referred me to Resnik (1993: 40), for another example with the same structure. Here is a further one,
which I owe to Eve Garrard. I have to choose between two men, Joe and Sam. Joe is dull but reliable.
Sam in unreliable but exciting. I prefer Sam, because if I chose Joe I would always be missing the
excitement that Sam would have given me. But then Sebastian comes along, who is even more exciting
but yet more unreliable. I realize that if I chose Joe I would always be missing Sebastian’s excitements,
even though Sebastian’s unreliability is so terminal that he is ineligible. And this tells me that my reason
for preferring Sam to Joe is no reason, in the new situation, since choosing Sam will not lead to my
having no lost excitement to regret. So I choose Joe.
104     
cannot be any examples would be the generalist claim that since the feature of
‘being an available alternative’ often makes no difference, it makes no difference
anywhere.¹²
There remains a difficulty. There appears to be an argument that there could be
no counterexample. If so, the situation is unstable. We would have an example,
our holistic position that there could be examples, and an argument that there
can’t be. Here is the argument. Sadly, it is one to which there can be no
particularist objection as such; it is not a switching argument, for instance.
Let us start with a supposed overall ranking, the Great Ordering. Everything has
its place in the order. We can compare the values of different objects, and compare
different objects in order to establish their relative placing, and these activities
cannot themselves alter the values of the objects compared or their relative places;
otherwise the very notion of relative value would be incoherent. One can compare
the values of merely possible objects, e.g. possible courses of action or possible
states of affairs. Could there be any difference between the values of merely
possible objects and the values of those objects should they become real? No: for
otherwise the activity of establishing the relative values of different possibilities
would be incoherent. And this would make deliberation before action incoherent,
if deliberation is the establishing of relative values of possibilities so as to decide
which to make actual. Suppose then that I ask you to rank ownership of each of
ten paintings. What you are ranking is a set of possibilities. Suppose then that
I give you all the paintings, and ask you to rank the ten actual ownings. There can
be no conceivable reason for a change in your ranking order (unless of course you
have changed your mind). Now: could there be a difference between an order of
preference and a ranking order for choice, where what one is dealing with is
alternatives? No: there is no possible relevant difference between a preference
order and a choice order. Suppose that instead of giving you all the paintings,
I give you the money to buy one. You should buy the one that came top in your
preference ranking. So the feature of ‘being an alternative’ cannot make a
difference.
Matters are more complicated than this. The complications do not make a
difference, but they are relevant to what happens later. There is an obvious
¹² An interestingly different avenue of approach, which I will not pursue here, starts from something
that Derek Parfit is apparently happy to admit, namely an analogous claim concerning not value but
‘ought’. Parfit’s view seems to be that it is possible that one ought to do A if B is the only alternative, but
that if C is also available one ought to do B. He denies, however, the claim that the values of A, B and
C can be related in a structurally similar way. See Parfit (1984: 429): ‘Whether I ought to act in one of
two ways may depend on whether it would be possible for me to act in some thind way . . . I then ask
whether, compared with A+, A would have been better. The relative goodness of these two outcomes
cannot depend on whether a third, outcome, that will never happen, might have happened.’ So in the
evaluative realm Parfit is what one might call a choice-atomist but in the deontic realm he is a choice-
holist. My own view is that this position is unstable. And my reason is fairly predictable: that if one
admits that there are examples of choice-holism in the deontic realm, structurally similar examples will
emerge in the evaluative one.
 ’  105
difference between buying just one painting and ranking them all. The difference
comes out when we consider a case where I give you the money to buy one
painting, then enough to buy another, and so on until you have all ten (though you
never knew in advance that I would give you the money for the next). The order in
which you buy the paintings need not be the same as your original order of
preference. To see this point, it is important to distinguish between two quite
different preference orders. The first has ten slots, in each of which one is asked to
put one item of the form ‘I own picture n’. The second has ten slots, the first of the
form ‘I own picture n’, the second of the form ‘I own pictures n and m’, the third
of the form ‘I own pictures n, m and p’, and so on. There is no reason whatever
why either of these two orders should be extractable from the other. The point is
that if you already have, say, six of the paintings, you might rationally choose to
add to those six a painting other than the one that came seventh on the list. To get
a true analogue of the original ordering, when it comes to choice, we have to
suppose that I give you enough money to buy one, but that just as you try to buy it
someone else gets in first; so you should go for the second on your original list—
but the same happens again, and so on down to the tenth. The order of choice
should be the order of preference.
What we have, then, is an explanation of why a certain feature, ‘being an
alternative’, cannot make a difference, and therefore of why particularism is
compatible with a full ordering. In one sense (epistemically, perhaps) it is possible
that ‘being an alternative’ can make a difference. But there is an argument (which
is not a switching argument) that no instance of this could be found. So the
situation seems to be that we have on the one side an example in which the feature
does make a difference, and a weak general reason derived from our holism to
expect this sort of thing to crop up, and an argument on the other side to show
that it is impossible. Now this is not one of those situations in which there can be
reasons on one side and reasons on the other, and we can just decide where the
balance of probability lies, leaving the defeated reasons in place. If we go with the
example, we have to show what is wrong with the argument on the other side.
Luckily this can be done. The property of being an alternative is incapable of
making a difference to a ranking order already established because there is no
relevant difference between overall preference and overall choice. And just as the
ranking preference order may be affected by the list of things to be ranked, so that
if we take something off the list, the rankings of the rest may change, so with the
ranking of alternatives. But all that this shows is that preference is like choice, and
like choice in the crucial respect that it deals with alternatives. Being alternatives is
the same as being mutually exclusive. Not all preference rankings are rankings of
objects conceived as mutually exclusive, as we have seen. But some are. And the
same is true whether we are ranking existing objects or possible ones. So the
explanation of why the feature of being an alternative cannot be the cause of a
difference between a preference ranking and a choice ranking is that this feature is
106     
present on both sides.¹³ Our conclusion should be that being an alternative can
make a difference to all three rankings: of possible objects, of actual ones, and of
objects of choice.
The existence of persuasive examples should then move us without further
resistance from value-holism to a sort of choice-holism, which holds that:
1. The value of one alternative can be affected by the nature of other available
alternatives.
2. Assessing the relative merits of different alternatives is not the same as
assessing the various alternatives one by one and then comparing the
results.
I want to end this section by comparing what I have said here with something
I wrote in Moral Reasons:
My daughter trod on a sea-urchin on holiday a few years ago, and we caused her
considerable pain (not entirely with her consent) in extracting the spines from
her heel. Was the pain we caused her something which made our actions worse
than they would otherwise have been? Here is a switching argument which says
that it was. Had there been available a painless method of getting the spines out,
we would and should have adopted it. We would have been wrong to continue
digging in her heel with a needle, because of the pain. Surely this shows that as
things were our actions were the worse for the pain they caused?
I don’t think it does show this. What we should say about cases like these is
that a feature which would have made this sort of difference had there been any
alternative choice need not necessarily make it if there is no alternative. It seems
to me quite consistent to say that as things stood our action was not the worse for
the pain it caused, though that pain should have led us to choose another method
had one been available. (1993: 65–6)
The idea, expressed in terms of reasons rather than, as above, in terms of values,
was that the pain is not a reason against the action if there was no alternative,
pain-free course of action available. It is not just that it is not sufficient reason; it is
not any reason at all. I presented this thought as an application of a style of
switching argument, whose general form is: if this action were less F, it would be
better; so its being F must detract from its overall value. But it can be seen
immediately that the example I gave goes further than is required for that purpose.
¹³ Thanks to Eve Garrard and David McNaughton here. It would be wrong to say, in reply to this
argument, that the feature we were originally discussing was that of being an available alterative, not
that of being an alternative. The notion of availability merely takes us from possible choices (prefer-
ences) between mutually exclusive options to actual choices.
 ’  107
My use of an example that hangs on a point about alternatives was more of a

distraction than a help, since the general point I was trying to make was nothing to
do with alternatives. Talk about available alternatives was intended more as an
explanation of the supposed fact that, in the example given, the action was not the
worse for the pain caused, even if it would have been better with less pain. Since it
was not possible to do it with less pain, the pain caused does not make the action
worse than it would otherwise have been. In possible-world terms: even though, in
the nearest world in which there was an available pain-free alternative, the action
we did was wrong, wrong because of the pain it caused, and the worse for that
pain, the actual action is not the worse for that pain. This is just an application of
the holistic thought that a feature can make a difference in one situation that it does
not make in another. Where this occurs, particularists admit that there must be
an explanation of it; the explanation is that in the actual world, there was no
alternative.
Perhaps, then, the situation is like this. Holism takes away from us one of our
two main reasons for sticking to principles like IIA. If one is a holist, it is going to
be hard to think that the question whether something was a real alternative cannot
make a difference. If IIA expresses a form of generalism, holists don’t have that
reason to believe it. They might have the other reason, which is that if we lose
principles like IIA (and all the rest), we lose what is really the only detailed account
of the ‘logic’ of choice. The loss of IIA seems to be another nail in the coffin of the
idea that there is such a logic.
One might suggest, however, that the loss of IIA is far worse for generalists than
it is for particularists. For generalists, the pillars of practical rationality would
really be tottering.
Choice-holism and the possibility of a full ordering of values
Suppose then that we accept a general choice-holism to go with our reasons-

holism and our value-holism. Have we totally lost any possibility of a full
ordering? There are two ways in which we could hope to retain anything like a
full ordering.
The first is to say that what we have established is only that the context of real
choice (i.e., the actually available alternatives) can affect the value of an option.
This result, we might say, is clearly disturbing. But it does not altogether disturb
our full ordering. Choice-holism concerns itself with real situations, in which the
question what alternatives are available is a serious practical one. As such, it is to
be distinguished from any thoughts about the effects of merely comparing one
option with another. The value of an option will not vary according to what we
imagine as the possible alternatives to it; it will only be affected by what actually
are the alternatives. And mere comparison is relevantly similar to imaginary
108     
choosing, we might say, so long as the purpose of the comparison is to establish

relative value. So an object’s value is not affected by the mere act of mental
comparison with another; it is only able to be affected if the two objects become
real alternatives for some agent.
Even this, of course, will do something to upset our full ordering, if we suppose
that the very same thing can occur in more than one actual choice situation. But
we might deny that possibility, supposing instead that objects of choice are
incapable of recurrence. We cannot have the same action again, that is, and we
cannot have any other choosable object again either. For the objects of choice are
not repeatables. If I offer you a chocolate bar today and you refuse it in favour of a
pint of beer, and I offer the very same bar to you tomorrow, the fact that it is the
same bar does not show that you have the same choice again. You only have a
similar one, and holists allow that objects that are intrinsically similar may yet
differ in value because of their context.
Be that as it may, the position that this move is trying to defend is surely
another unstable one. It holds that mere comparison of two objects A and B,
which we can do at any time at will, is incapable of affecting the values of A and of
B; but should they become actual alternatives, their values may be affected. So
I may compare A and B and prefer B, and yet when I have to choose between the
two I choose A without irrationality. This distinction between actual and imagined
choice, or between the effects of choice and those of comparison, is surely
unsustainable. And this means that the dream of a full ordering collapses entirely.
For if I cannot compare two objects without being in danger of affecting their
relative values by doing so, there is surely no sense left in which objects have their
own place in the ranking order. The ranking order must mean that objects have
their place on it whether one actually compares them or not; indeed, to compare
them is just to establish their relative placings in the order. If one could affect
those placings by the act of comparison, the notion of an order would be
destroyed.
So much for the first way of defending the possibility of a full ordering. The
second way involves us in re-describing each option in terms of the available
alternatives to it.¹⁴ Instead of thinking of ourselves as having the three options of
buying house A, buying house B and buying house C, our three options are:
1. Buying house A when we could have bought houses B or C.

2. Buying house B when we could have bought houses A or C.
3. Buying house C when we could have bought houses A or B.
¹⁴ I have borrowed this manoeuvre from Broome (1991: ch. 5; 1993).

 ’  109
Now, we might say, these three options have an unvarying value, and occupy a
fixed place in the full ordering. For if one of the three houses is taken off the
market, or a fourth house enters the equation, we no longer have any of these three
options, but either two or four new ones.
The difficulty that I see in this approach is that if it is to avoid the difficulty we
have already exposed, that comparison of the value of two objects is relevantly
similar to choice, we will have to relativize every item on the full ordering to all
other items, first severally and then in pairs and so on up until each is relativized to
all others at once. And there will be no way of predicting, from the value of an
option that is relativized to degree n, what its value will be relativized to other
degrees. Given this, the use of the full ordering will be limited indeed. Transitivity,
for instance, must fail. For if we rank ‘A when we could have had B’ above ‘B when
we could have had A’, and ‘B when we could have had C’ above ‘C when we could
have had B’, it in no way follows that we should rank ‘A when we could have had
C’ above ‘C when we could have had A’.¹⁵ It will be perfectly true, that is, that
every relativized option occupies one and only one place in the ordering, without
this doing anything to preserve the conception of rational choice that the ideal of a
full ordering was designed to promote.
There is a further problem. Suppose that we have a full ordering of all
relativized options. This locates each option with respect to every other.
Suppose now that I ask of item 32 in the list how it compares in value with item
33. It need not be the case that my answer is that item 32 is more valuable than
item 33. The option ‘33 when I could have had 32’ is a different option from the
simple option ‘33’, no matter how internally complex option 33 may be—and the
same goes for option 32. But if my ranking order does not even commit me to
claims about the relative values of the items ranked, it is pointless.
Holism and explanation
There remains one further matter that I think it worth bringing out. It seems to me
that particularism commits one to a highly debatable doctrine in the theory of
explanation. In Moral Reasons I was not so clear about this.
I start by considering the relations between two doctrines. One is the now
familiar holism in the theory of reasons. The other is a doctrine in the theory of
explanation, which has no agreed name that I know of. Here they are:
¹⁵ Of course, relative to one and the same three-way comparison, transitivity must be preserved—
or at least nothing that I have said gives us any reason to dispute that. If we do dispute it, we will
probably do so for quite different reasons, i.e. those stemming from comparisons in which many
different criteria are operating at once. For a recent rehearsal of such considerations, see Temkin
(1997).
110     
Holism: the ability of a consideration to stand as a reason for action can be

affected by the context in which it occurs.
Non-guaranteeing explanations: an explanation can be perfectly good without
being ‘complete’, where a complete explanation is one that is inconsistent with
the non-occurrence of the explanandum. Where E and O both occur, the
occurrence of E can explain that of O without guaranteeing it. Perfect explan-
ations can be ‘non-complete’. The idea that non-complete explanations are
enthymematic is a mistake.
The question that I want to start with is whether I was right or wrong to think, as
I once did, that my doctrine of reasons-holism was effectively the same as this
doctrine in the theory of explanation. If so, we have uncovered another awkward
consequence of particularism.
One preliminary first. Some philosophers maintain that though guaranteeing
explanations are not required when what we are explaining is an action, they are
required when we are explaining events. A scientific explanation, then, will have to
be a guaranteeing one.¹⁶ My own view, however, is that we would need a lot of
persuasion to say any such thing. The only reason for adopting it, I think, is a
sense that we would do well not to demand something of action-explanations that
we have no prospect of achieving; in science, by contrast, where prospects are
better, we should not be satisfied with anything less than perfection. The proper
riposte to this is that action-explanations are as good as any explanation needs to
be, and that, so far as the purposes of explanation are concerned, there is no reason
to think there is either need or room for anything better. We should therefore
accept the possibility of non-guaranteeing explanations on both sides or on
neither.¹⁷
The concept of a non-guaranteeing explanation requires that of an enabling
condition. If there is a non-guaranteeing explanation F of an event E, there must
be an event O such that:
1. The occurrence of O is not part of the explanation of E, and

2. If O were to fail, we would have a situation (F, not-E).
In such a situation, O would be an enabling condition for the explanation that

F gives us of E. The reason why there must be such things as enabling conditions if
there are to be non-guaranteeing explanations is that if every candidate for the role
of enabling conditions were to turn out to be part of the explanation of the event-
type E, all explanations would be guaranteeing ones (when ‘complete’). If we think
¹⁶ For a recent example of this view, see Foot (1995).

¹⁷ My views on the nature of explanation in general bear interesting similarities to those of Nancy
Cartwright (1983).
 ’  111
that there are any non-guaranteeing explanations, then, irrespective of whether we

think that no explanations are guaranteeers, we must be able to make sense of the
notion of an enabling condition—a condition whose satisfaction is required for
the explanation, but which is not itself a part of that explanation. And this sets us
something of a challenge.
But we face the same challenge in the theory of reasons for action, once we adopt
holism there. This is not perhaps too surprising antecedently, since presumably the
reasons favouring an action are reasons that explain that action’s rightness (or
whatever moral status it has). If we start from one case where there are reasons
R1–Rn making the action right, but allow that changes elsewhere may affect the
ability of those reasons, not merely to make the action right, but even to be reasons at
all, we will again say that features over and above the reasons must be functioning as
enabling conditions. They enable the features that are reasons to be the reasons they
are in this case, without themselves being among the reasons why the action is right.
To give a very simple example: suppose that ought implies can. Then if I cannot
do the action, the features which, were I capable of doing it, would be reasons why
I should do it, are incapable of playing that role. But we should not conclude that my
ability to do it is one of the reasons why I should do it. It is a condition that enables
the reasons why I should to be the reasons they are, but is not itself among those
reasons. Allow this, and you will probably allow the next: that I have the opportunity
to help is something without which the reasons why I should help would not be those
reasons. Her need is not a reason for me to help her if I have no possible opportunity
of doing so. It is only a reason for me to seek an opportunity, which is different. But
that I have an opportunity to help is not itself among the reasons for doing so.
Another similar example: if I were not alive, the reasons that there are for me to help
the needy would not be able to be the reasons they are. But this does little to show
that among the reasons why I should help the needy is the fact that I am alive.
So we see the same structure both times. Both holism and the claim that
explanations need not guarantee are committed to making sense of the notion
of an enabling condition. One might think¹⁸ that there are forms of holism that
don’t have this effect. We might try to adopt a sort of weak holism without
adopting any distinctive doctrine in the theory of explanation:
Weak holism: the ability of a consideration to stand as a reason can be affected

by what other features are present as reasons (but not by other things).
With this in hand, we might hope to avoid the need to talk about enabling
conditions at all, and avoid the need to allow non-guaranteeing explanations.
But we can only do this if we can add to our weak holism a sort of:
¹⁸ And I did think so, until Eve Garrard showed me that I shouldn’t.
112     
Holistic generalism: taking all the reasons here present together, they guarantee
the rightness of the action; i.e., where present together elsewhere, they will
always have the same effect (non-causally, of course).
This holistic doctrine is certainly compatible with the denial of a different form of
generalism:
Atomistic generalism: each reason has the same tendency as a reason, no matter
what the context in which it is found.
But in fact weak holism cannot be coherently combined with holistic generalism.
If the ability of a consideration to stand as a reason can be affected by what other
features are present as reasons, why is it that the presence of further reasons in a
second case, in addition to all those present in the first, is incapable of making any
difference to the original reasons? Surely the official statement of weak holism says
that a new reason can upset others whether we take them one by one or all
together. So the combination of weak holism and holistic generalism is not a
sustainable position.
Moving now to the theory of explanation, we find the same thing. There might
seem to be a form of holistic generalism available there, which is compatible with
the demand that all explanations be guaranteeing explanations. We might say,
that is, that though no individual feature has its own explanatory potency, which it
carries with it from case to case, regardless of changes in other explanatorily
potent features, still the entire complex of features that together explain the event
constitutes a guaranteeing explanation. It is a guaranteeing explanation because
that complex of features could not have occurred unless the event to be explained
had occurred (or been going to occur).
But this attempt to make room for guaranteeing explanations within a holistic
picture suffers from the same incoherence that we found in the theory of reasons.
No explanation is given of why the arrival on the scene of a new element is capable
of affecting the explanatory contributions of individual features, but not of a set of
such features. Why is it that a whole set of explanatory features is necessarily
invulnerable to the sort of difference that a new feature can make to individual
elements in that set? There is just no answer to this question. So there is no way for
a holist to avoid making room for the concept of enabling conditions, and denying
the possibility of guaranteeing explanations.¹⁹
One useful consequence of these thoughts is that the incoherence of the
combination of weak holism and holistic generalism gives us an answer to one
¹⁹ The same may be said about causal statements: a sufficient cause need not be a guarantee. But to
try to argue this would take us too far away from present concerns.
 ’  113
leading reply to particularism.²⁰ This reply is that my examples only concern what
we might call contributing reasons. It is true that contributing reasons do vary in
their polarity from one context to another. But it is not true of complete reasons.
These remain the same, as reasons, regardless of changes elsewhere.
There are two ways in which I can respond to this. The first is, as above, to point
out that there is no obvious explanation of why a new reason in a new case is
supposedly able to change the behaviour of one reason but not of all of them at
once. Why is it that the whole pack of them is immune to change, when no
individual one is? Note that this question is asked within the constraints of weak
holism; it would be a different matter to maintain that considerations that are not
reasons in the new case are able to make changes in this way.
The second way in which I can respond to this attack is to challenge its notion
of a complete reason. This cannot be identical with all the reasons present in the
particular case, since we have already seen that there is no justification for the view
that the pack of all such reasons is invulnerable to changes brought about by the
presence of a new reason in a new case. It must therefore be something greater
than that; it will presumably contain all the enabling conditions, as well as the
absence of disabling conditions (specified one by one). But, first, this collection is
ceasing to look like a reason at all, and, second, there is beginning to be a prospect
that a complete reason will expand indefinitely.²¹
What I really want to say, of course, is just that my interests are in the ordinary
notion of a contributing reason, not in the concocted notion of a complete reason,
which in my view is designed merely to save a dubious philosophical theory from
refutation. But this may seem to be a definitional sulk; so I support it by attacking
the notion of a complete reason that is designed to save generalism from
counterexample.
²⁰ I mentioned this reply earlier, in discussing holism in theoretical reason. A recent example is to be
found in Bennett (1995: 80).
²¹ Jonathan Lowe pointed out to me that a valid deductive argument is something we could
reasonably call a complete reason, since it is monotonic, i.e. no addition of premises will affect the
validity of the argument. In this, deductive reasoning differs from probabilistic reasoning. So there is
one model of a complete reason that I cannot undermine. This model is, however, not applicable to
moral reasoning, unless there are absolute moral principles; reasoning that runs in terms of prima facie
principles is non-monotonic. There is certainly available a notion of deductive reasoning in ethics. The
only question is whether the reasoning that takes us from premises specifying the features that make an
action right to the conclusion that it is right is ever deductively valid. (I mean by this form of words to
exclude such premises as ‘all actions of this sort are right’, ‘this feature is the only relevant one’ and ‘this
feature is a pro tanto reason’.) I think not.
8
Necessity, Universality, and the A Priori
in Ethics
Kant held that moral laws are necessary truths, that they are universal in form and
that they are known a priori. These three claims are interconnected. He believed
that only universal truths could be necessary, and that if they are necessary they
must be known a priori, if known at all. He also held that moral laws are synthetic
practical propositions (Groundwork 4.420). In what follows I consider these
various views. I argue that what I think of as basic moral truths are indeed
known a priori, despite being synthetic. But I do not accept that their being
known a priori requires them to be necessary and universal in form. And I do
not only reject the inference. I maintain that its conclusion is false: basic moral
truths are contingent and particular.
Some of my arguments will be brief. This is because I have already laid out the
relevant points in greater detail elsewhere.¹ The position from which I view all
these matters is what has come to be called particularism in ethics. The particu-
larist holds that the possibility of moral thought and judgement in no way depends
upon a suitable provision of moral principles. Kant, by contrast, held that if there
is no true Supreme Principle of morality, moral distinctions are null and void.
There could not be a greater distance between his view and mine.
One important point of contrast is that the particularist starts by thinking about
moral reasons, not about moral duties, moral obligations or more generally about
what overall we ought morally to do. But Kant does not seem to operate with an
idea of a moral reason. If one does think in terms of moral reasons, one will
suppose that there can be reasons on both sides of the question. But Kant’s whole
approach to ethics is in terms of maxims and their ability to serve as universal
laws. The maxim on which one acts, whatever its content—and the exact way to
understand the sort of content that Kant supposed his maxims to have is hotly
debated—does not specify a reason for doing the action one proposes to do. Or
rather, if it does, there seems to be no ‘opposing’ maxim specifying the reasons for
not doing that action. One might imagine that a maxim might include some such
phrase as ‘despite the fact that she does not want me to do this, I will do it in order
to gain an advantage’. A maxim with such a ‘despite clause’ would have specified a
¹ Most recently in Ethics Without Principles (2004).
, ,       115
consideration as a reason against. But how it is that this feature stands as a reason
against is left completely mysterious in Kant’s philosophy. Contrary to the claims
of some eminent Kantians, therefore, I am of the opinion that Kant was not trying
to capture the idea of a moral reason. He went straight for the overall judgement
that this is what I ought to do, understanding that in terms of the universalizability
of a maxim of the form ‘I will do an action of this sort, whenever I find myself in
circumstances such as these’.
This contrast between Kant’s top-down approach (though I have here sug-
gested that he didn’t get very far down) and the particularist’s bottom-up
approach is significant for our general topic. For Kant, it seems, facts about
one’s overall duty in this or that sort of situation are the basic facts of
morality—and such facts are universal and necessary. For the particularist, basic
moral facts are facts about what is a reason for what, case by case. What I mean by
‘case by case’ is that the particularist supposes that what is a reason in one case
may well not be so in another; it may even be a reason against. I call this view
‘holism in the theory of reasons’. Particularism is built on this form of holism,
which can be re-expressed as the claim that reasons do not need to be invariant.
A reason is variable if the feature concerned is a reason in some circumstances and
not in others. If moral reasons are like this, then even though moral thought and
judgement is firmly based on reasons, it does not need to be based on the sort of
invariant reasons that would be captured in universal moral principles. Moral
principles, after all, can be thought of as specifying a feature and giving its moral
relevance. A principle is nothing if not universal. So if a principle is true the
feature it specifies must be of invariant relevance; for the principle is trying to
capture something invariant. The particularist, supposing that there may be no
invariantly relevant features, does not deny that there are moral reasons but does
deny that there must therefore be moral principles to hold those reasons in place.
For the particularist, the basic moral facts are facts about what is a reason in a
particular case. This is what we all start from, and it is more or less all that we have
to rely on when trying to come to an overall judgement about what course of
action is most suited to the present situation. The particularist’s moral epistem-
ology has to fit that perspective. Are those basic moral facts necessary or contin-
gent? A basic moral fact is such a fact as this; that you are dealing with someone in
distress is a reason for you to go gently (here). There is no suggestion that distress
(even the distress of this person, say) is always a reason to go gently, only that it is
so here. On another occasion, things might be different. It follows from this that
the basic moral fact is contingent, not necessary. It is a fact that might have been
otherwise, and would be otherwise in situations that are relevantly different from
this one. Basic moral facts are contingent, therefore, and they are not universal in
form. Are we also to suppose that they are not knowable a priori? I don’t think so.
After all, even when all the empirical information—that is, the information that
we think of as ordinarily available to the senses—is in, we still have to determine
116     
which aspects of the present situation, as revealed to us in experience, count which

way as reasons. That decision seems to take us beyond anything that the senses
can inform us about. And this is so even if we allow that some a posteriori
knowledge is the product of inference. Certain sorts of inference from sensory
evidence are generally allowed to yield a posteriori knowledge—such as my
knowledge, here in New Mexico where I am writing, that the weather will
continue fine for at least the next minute. But the forms of inference that are
involved in these cases, which take us from one piece of a posteriori knowledge to
another, do not seem to be involved in the process by which we pass from
recognition of empirical matter of fact to knowledge that (for example) her
distress is here a reason for us to go gently. Even if we have the benefit of previous
experience of such cases, the most we can get out of that is the recognition that her
distress is probably a reason. Now this sort of knowledge would only be a poster-
iori if the evidence on which it is based is a posteriori. And that just returns us to
the primary question here: in a case where we don’t have the benefit of previous
experience, is the ‘process’ by which we come to recognize that her distress is a
reason one that can be thought of as taking us from a posteriori knowledge (that
she is in distress) to new a posteriori knowledge (that her distress is a reason to go
gently)? And the answer seems to me to be no. If our decision that it is a reason is
capable of counting as knowledge, then, it seems that it will have to be a priori
knowledge that is at issue, since we only have two choices—a priori and a
posteriori—and the latter is ruled out. So the particularist is left in the uncom-
fortable position of holding that some contingent and particular truths can be
known a priori.
Now though this position is indeed uncomfortable, I don’t think it is eventually
untenable. But the present point is, of course, that it is flatly at odds with Kant’s
claim that only necessary universal truths can be known a priori. Something has to
give, therefore, and my aim here is to show that it is Kant that has to give, not me.
If that turns out to be right, we will have to find something to say about the
discomfort I referred to just above. The problem is the old problem of the
synthetic a priori, but intensified by the denial that what is a priori is universal
and necessary. How could we know particular contingent truths a priori?
Particularism is at a dialectical disadvantage, because it sets its face not only
against a long and influential tradition as well as against the supposedly untutored
intuitions of ordinary folk. To topple that dominant perspective, it needs to
provide impeccable argumentation. But in addition to that, it faces a tough
explanatory task. Why is it that so many people are unshakeably attached to
what particularism sees as an error? Some of the blame can be laid at the door of
the Christian churches, at least so far as we are trying to explain habits of thought
in Europe. But that cannot be the whole story. There are other ways in which one
might come to think that basic moral facts must be universal in form.
, ,       117
Consider W. D. Ross. For him, basic moral facts are facts about our prima facie
duties. Ross maintained that we know these facts with certainty, and we know
them a priori; but he also maintained that their truth is revealed to us by what we
know about particular cases. What we know about particular cases is epistemo-
logically prior, then, but not morally basic. (For him, as for me, what is basic is a
matter of moral metaphysics, not of epistemology.) The question for him, then,
should have been what we can know about the particular case that would enable us
to extract from it the kind of general moral knowledge of prima facie duties that he
supposed we all possess. What could the particular case reveal? All that we can
find there is a feature counting in favour of, or against, some response. How could
such a thing manage to reveal to us the truth of a general moral principle to the
effect that this feature must always play that role? Ross’s answer was that the form
of inference involved is ‘intuitive induction’. This is induction because it takes us
beyond anything present in the premises, but intuitive because the inductive
process leads us to recognize the truth of the conclusion directly. (‘Directly’ here
means that if the process is successful one’s knowledge of the conclusion ceases to
be inferential, if it ever was.) But what about our knowledge of the epistemically
basic facts about features counting in favour or against in particular cases? Is our
knowledge of that sort of thing a priori or a posteriori? Ross did not address this
question. He is clear that one’s knowledge of the principle of prima facie duty is a
priori. The question I am after is whether this requires him to suppose that one’s
knowledge of the premise (that this feature counts in favour here) is a priori too,
or whether intuitive induction is a process that can take one from a posteriori
knowledge to a priori knowledge. The paradigm instance of intuitive induction
may be misleading in this respect. It is the way in which we come to know that a
form of inference, modus ponens, say, is valid by being shown instances. How are
we to conceive of the validity of an instance? It is presumably a necessary truth,
and known a priori; but is it universal in form? It is possible to say that it is,
because the validity is not potentially dependent on any other features of the
context; the idea will be that these premises take one necessarily to that conclu-
sion, and we move by intuitive induction to the realization that this is a matter of
form. Now compare that paradigm case with the moral case that is our real
concern. Is what is revealed by the particular case, when we discern that this
feature is here a reason, covertly universal in form? I suspect that Ross would say
yes, because he is already supposing that these truths are context-independent—
the very thought that the particularist denies. But why is he supposing that they
are context-independent? The answer must be that the general principles of prima
facie duty are known a priori, and this requires that what we see in particular cases
that reveals the truth of those principles must somehow be the general truth itself,
only clothed in particular form. My suggestion then is that Ross rightly took it that
moral knowledge is a priori, and wrongly supposed that if it is a priori, it must be
118     
knowledge of universal truths, that is, of true propositions that have universal
form. This supposition he may be said to have inherited from Kant.
There is a fundamental Kantian argument to be addressed, then, which goes
like this:
1. Whatever is knowable a priori must be necessary.

2. Whatever is necessary must be universal in form.
3. Moral truths are known a priori.
4. So moral truths are universal in form.
There is a simpler argument in similar style which does not mention the a priori:
1. Whatever is necessary must be universal in form.

2. Moral truths are necessary truths.
3. So moral truths are universal in form.
I will restrict myself for the moment to this simpler version, which does not
mention the a priori. The main difference between the two arguments, as I see it, is
that the first, though it entails that moral truths are necessary truths, does not
actually use that as a premise. Now, as I have said, particularism denies that basic
moral facts are necessary truths. Why did Kant hold the opposite for his basic
moral facts? Four things need to be mentioned in answer to this question.
The first point, the one most stressed by Kant, is that moral obligations bind all
rational creatures, not just humans. Moral obligations, therefore, are not
grounded in the contingencies of the human situation, but require a ground of a
different sort so that they can apply to all possible rational beings. Kant wrote
(Groundwork 4.389):
Everyone must grant that a law, if it is to hold morally, that is, as a ground of an
obligation, must carry with it absolute necessity; that, for example, the command
‘thou shalt not lie’ does not hold only for human beings, as if other rational
beings did not have to heed it . . . that, therefore, the ground of obligation here
must not be sought in the nature of the human being or in the circumstances of
the world in which he is placed, but a priori simply in concepts of pure reason;
and that any other precept, which is based on principles of mere experience—
even if it is universal in a certain respect— . . . can indeed be called a practical rule
but never a moral law.
Again (Groundwork 4.408):
[If the moral law] must hold not only for human beings but for all rational beings
as such, not merely under contingent conditions and with exceptions but with
, ,       119
absolute necessity, then it is clear that no experience could give occasion to infer
even the possibility of such apodictic laws . . . And how should laws of the
determination of our will be taken as laws of the determination of rational beings
as such, and for ours only as rational beings, if they were merely empirical and
did not have their origin completely a priori in pure but practical reason?
These remarks are not perfectly clear. In considering them, I want to grant Kant,
for purposes of argument, what seems to be his main premise here, which is a
claim about who is subject to the moral law—the domain of application, as one
might call it. The moral law governs all those who are capable of responding to it
(or, as Kant would put it, capable of acting in accordance with a representation of
it—Groundwork 4.412). But this does not entail that there cannot be laws that only
really apply to humans, that is, to those subject to our conditions, liable to our
failings, emotions, etc.—and I don’t really think that Kant supposed otherwise.
There could be a law saying that if you have just been humiliatingly defeated at
squash, and experience a strong desire to smash your opponent over the head with
your racquet, you should walk off the court quickly and try to calm down. (I think
this is Michael Smith’s example.) This law governs the behaviour of all rational
beings, but is only able actually to affect the behaviour of those liable to strong
emotions of a certain sort.
Kant might however be thinking that since the moral law governs the behaviour
of all rational beings, it governs our behaviour only as rational beings. But if so, his
point can only be that it is as rational beings that we are capable of responding to
those rules, of controlling our emotions in accordance with them. There can
clearly be rules about the appropriateness of emotions, and there must be some
sense in which they only apply to those capable of responding to them. Further
Kant seems to say that the ground of the obligation cannot lie in the nature of the
human being, but must lie in the concepts of pure reason. But this does not show
that rationality, or the things that pure rationality can recognize, is the ground of
the law. That ground might lie in contingent truths about the emotions.
Another possibility is that Kant is thinking that if the moral law governs the
behaviour of all rational beings, and of ourselves only as rational beings, it must be
possible for a purely rational being to come to know that law. Since a purely
rational creature is restricted to a priori knowledge, the grounds for the law cannot
be empirical. But even if we allow that those grounds cannot be empirical, this
would do nothing to determine the possible contents of such laws; it rather
addresses our ability to know that they are laws. The reason why the law is law
(if there is such a reason) cannot be an empirical reason, but must be some feature
that a purely rational creature could come to know.
Kant also makes a point about exceptions. There cannot be exceptions to a
moral law; whatever the law says is true in all conditions whatever, that is, as we
would now say, in all possible situations or worlds. It is therefore a necessary truth.
120     
My question is whether there is anything here—anything true, that is—that a

particularist could not accept. My general view is that so long as we do not assume
in advance what sort of thing the moral law is to say, or the form of basic moral
truths—and Kant is in no position to make any such assumption right at the start
of the Groundwork—it remains perfectly possible that though all are subject to the
law, the law itself is particularist, by which I mean that its basic pronouncements
are truths about what is a reason for what, case by case. Differences between
situations may affect what is a reason for what in them.
The last point above, the one about exceptions, is the sort of thing I mean to
exclude by saying that we should not assume in advance what sort of thing the
moral law is to say. If the moral law specifies laws, that is to say universal
principles, those principles must be true, and for them to be true is for them to
be exceptionless. Our basic moral principles could not, for instance, say that lying
is normally wrong, if only because that would be no guide to someone trying to
work out whether this case was one of the exceptions. The most that such a
principle could tell him would be that it probably was not an exception, which,
though it is of some help, is not really what was required. But this is just to specify
a consequence of holding that the moral law specifies universal principles; it is not
to give any reason to suppose that this is what the moral law is in the business of
doing, which is what the particularist is wanting to deny. So it seems that we
cannot extract the conclusion that the moral law specifies necessary truths from
the claim that it specifies universal principles, because that claim is also in dispute.
Kant seems sometimes to give a different sense to the idea that moral laws hold
with necessity. In this new sense, even if they do not express necessary truths, they
necessitate us. They require us to act, and accept no excuse. Of course this sort of
necessitation is different from causal necessitation, since one may well not do what
one is necessitated to do in this sense, what one is required to do, while what is
causally necessitated must happen, cannot fail to occur. The moral law, as Kant
often says, concerns what ought to be done but very often is not done. Now there
is here the special point that purely rational beings cannot fail to do what they are
rationally required to do, because they have no non-rational motivation. So they
will always do what they ought (and of course moral requirements are just one
species of rational requirement, for Kant). But nonetheless the laws which they
observe do not impugn their autonomy; they remain free to do the other thing—it
is just that they won’t do it, since their nature prevents them from having any
incentive to act contrary to the requirements of morality, or of rationality more
generally. So the crucial point is that Kant’s second conception of necessitation
allows that one is free not to do what one is necessitated to do. All this, however, is
simply irrelevant to the question whether moral truths are necessary truths or
universal in form. The particularist would happily say that we are necessitated by
the constraints present in the case before us, seeing no need to think that anything
capable of necessitating one in this way must be of a certain form, or a necessary
, ,       121
truth. An obvious analogy here is with the idea that an event that causally
necessitates another event may itself be stubbornly contingent; and the causal
laws that govern the process may be contingent too. Causal necessitation takes
place in a contingent world and according to contingent rules, and moral neces-
sitation might be the same.
What I take myself to have argued so far is that moral truths, though they
necessitate, need not for that reason be necessary truths. My next point is that they
do not need to be universal in form in order to govern the behaviour of all rational
beings. Kant’s point about the domain of application, which I am allowing for the
moment, really tells us nothing about the form of the things that apply in that
domain. If the basic pronouncements of the moral law were claims about what is a
reason for what in particular circumstances, those claims would apply to all
rational beings without exception. Even if the claims contained some restriction
on the sort of person, or even just the person,² for whom the reason is said to exist,
this would not mean that their domain of application is thereby restricted. In
logic, the domain of quantification is not itself restricted by the antecedent of a
conditional. If the domain of quantification is all rational beings, the proposition
that if one is a parent, one should care for one’s child is a proposition to be
assessed as true or false within that wider domain. One could say that there is a
narrower domain, that of parents, and what the law says within that domain is
that all should care for their children. But one need not say this, and there is a
reason not to, which is that otherwise the notion of the domain of quantification
would cease to have much of its point. Analogously with the moral law: the
domain (domain of application, now) is that of all rational beings, and the law
is that parents should care for their children. The particularist would see this
rather differently: the domain is all rational beings, and the moral law pronounces
that this person has a reason to care for that child, since he is the child’s father.
(It may be that another person has a child for whom she has no reason to care; but
if so there will be some explanation of this fact.) The main point here is that the
universality of a domain of application does not show that every truth within that
domain is universally quantified.
So far I have argued that the pronouncements of the moral law need not be
necessary truths, even if they apply to all equally. But now there is the a priori to be
considered. Is it true, as both Kant and Ross seem to think, that since moral truths
are knowable a priori, they must be necessary and universal in form? If this were
true, then though an a posteriori morality might be as the particularists suppose all
morality to be, a morality knowable a priori would have to consist of necessarily
true moral principles.
² I am thinking here of the possibility of agent-relative reasons.

122     
The connection between the epistemological distinction between a priori and a

posteriori knowledge and the modal distinction between necessary and contingent
has been loosened by Saul Kripke’s ‘discovery’ of a posteriori necessities. This
certainly helps us with respect to one premise that occurs in both of my Kantian
arguments above. But it does little to help us in the present case. The question is
the reverse one about a priori contingencies: how something that is not necessary
could be knowable a priori. The normal routes to a priori knowledge are the
conceptual route and the semantic route, and both of these seem to be excluded in
the moral case. Moral truths are not true merely in virtue of the concepts involved,
and they are not true merely in virtue of meanings either. Conceptual and
semantic competence, then, cannot be what explains our ability to know what is
a reason for what in the moral sphere. Kant would, I think, phrase this question in
terms of what synthesizes such judgements. If I understand this way of putting
things at all, it amounts to asking how we know that the two terms of our
judgement (subject and predicate) are connected, or, in more objective vein,
what it is that connects the two terms. In the case of ordinary empirical judge-
ments, the answer is experience. This does not mean that experience puts the
terms together, but that it is experience that puts us in a position to see a
connection between them. If I judge that Italian ice cream is nice, the connection
between being Italian ice cream and being nice is entirely provided by experience,
and therefore can only be contingent. So what is it that enables us to put the
two terms together when we judge that her having asked me not to do it is a
reason for me not to do it? It cannot be experience, because experience gives out
before we get to this sort of judgement. Must it then be reason? Kant would
say yes, and then insist that the primary object of reason cannot be something so
particular, but must be something universal and necessary. I think that this is
too swift.
A truth is synthetic a priori if it is not analytic, and we can know it in a way that
is not derived from our experience. What counts as being derived from our
experience is somewhat vague, but I suppose that what is being ruled out here is
either being provided by experience (presented as part of the surrounding scene)
or being inferable from what is so provided. It is not that if something is a priori it
cannot be inferable from experience, but only that it can be known in a way that
does not involve such inference, even if it can be known a posteriori as well. Now it
seems to me that there are quite a lot of truths that fit this recipe and which are
neither necessary nor universal. An example I have given before concerns our
judgements of similarity. Suppose that I am asked to judge of four things A, B, C,
and D whether A is more similar to B than C is to D. Is Mahler’s music more like
that of Richard Strauss than Mozart’s music is like Haydn’s? I take it that I make
such judgments, at least some of the time. And some of them will count as
knowledge. Mozart’s music is more like Haydn’s than Beethoven’s is like Bach’s.
Now such judgements are mediated by experience, in the sense that without the
, ,       123
relevant experience I could not even begin. But they are not provided by experi-
ence, nor are they straightforwardly inferable from what experience does provide.
Such judgements are much too nuanced for that; to make them, one has to have a
sense of which similarities really count for something (for present purposes only,
of course) and which are not very telling for those purposes. It seems to me,
therefore, that we are dealing here with a priori knowledge, and also that it is
particular, not universal in form.
It also seems to me that it is also contingent rather than necessary; but this is a
further matter, and needs further discussion. Mozart might have written different
music, and if he had, his music might have been no more similar to Haydn’s than
Beethoven’s is to Bach’s. But still, one might say, it is a necessary truth that
Mozart’s music, given its actual empirical nature, is very similar to Haydn’s,
given its empirical nature. And this is what one is determining a priori, if
anything. Perhaps, then, the example I gave in the previous paragraph is inad-
equate. It is an example of the a priori determination of a particular necessity, and
not, as I suggested, of a particular contingency. So it does not serve to defend the
particularist’s suggestion that basic moral claims are particular and contingent but
known a priori.
I don’t think this challenge is sound. It is a necessary truth that Mark, given his
actual height, is taller than Jonathan, given his actual height. But the necessity of
that truth is compatible with its being a stubbornly contingent truth that Mark is
taller than Jonathan. The necessary truth is a consequence of the contingent one,
and, I would say, known only by knowing the contingent one. I maintain that the
contingency of the fact that Mozart’s music is so similar to Haydn’s is not
impugned by the fact that in all worlds in which Mozart wrote what he wrote in
this world, and Haydn did too, Mozart’s music will be just as similar to Haydn’s as
it is here. The matter is analogous to the situation with reasons. It is a contingent
truth that this feature is a reason to go gently here, since in other circumstances
this same feature might not have been a reason. It is a necessary truth that this
feature is, in this context, a reason to go gently, since wherever that context recurs,
this feature will be the same reason that it is here. But this necessary truth (which
requires, of course, a suitable conception of ‘this context’), is a consequence of the
particular truth, and, I would say, known only by knowing the contingent one.
It may seem to be a mystery that a contingent truth can entail a necessary one.
But that is not quite what is at issue. What is really the case is that if we start from
the contingently true proposition and ask how things would be in other situations
in which everything possibly relevant to the question whether that proposition is
true or not is held constant, the answer will be that in all such situations we will get
the same contingent truth. This is not the same as saying that the original
contingent truth was not contingent.
So for a second example, consider judgements of relevance. Our judgement of
what is relevant to what seems to me to be inescapably particular, though
124     
I understand that some people are working on principles of relevance.³ Further,

I would say that judgments of relevance sometimes amount to knowledge. But
none of them seem to be respectably a posteriori.
Of course that leaves the question what synthesizes these judgements. At this
point I want to say that we just have the capacity to judge these things. The
relation of similarity is one that we would not be human if we could not track to
some extent. Some similarities are perceivable, experienceable, and others are not.
(Similarities between the views of Kant and of Aristotle would be of the latter
type.) Is it reason that enables us to track such things? Not in any very restricted
sense. (If reason is the capacity to track reasons, as some suggest, then knowledge
of what is a reason for what will be rational knowledge of the bluntly particular
and contingent, in my view. But this only shows that it does not help in this
context to appeal to such a conception of reason.) We have the capacity to assess
similarities, and it is a rational capacity if anything, I suppose; but calling it so does
not advance the argument in any direction. So if the question is what synthesizes
the two terms in our similarity judgements, I am tempted to suggest that the
answer ‘nothing’ is as good an answer as we are going to get. (Another answer is
‘we do’.) And the same applies to judgements about relevance.
So far I have been suggesting that Kant fails to establish that morality must be
based on necessary universal principles known a priori. I have not tried to argue
that since particularism is true, Kant must be mistaken. My argument was rather
that Kant’s inferences, or associations, are mistaken, and that we can see that they
are mistaken by considering an alternative conception of morality, the particu-
larist one, and asking what Kant would have to say against it. If, as I suggest, the
answer is nothing, or nothing effective, that is enough for me. The crucial question
is whether Kant has given us reason to suppose that there must be moral
principles if there is to be morality at all. I maintain that no such reason has
been provided, and none can be extracted from the fact that basic moral know-
ledge is a priori.
There remains the great Kantian connection between morality and rationality.
Does the fact that moral truths are not known empirically show that they are
known by pure reason? Not if we restrict pure reason to the discovery of necessary
universal truth. What then happens to the notion of pure reason if we extend its
capacities to the discovery of the sorts of contingent particular truths that I have
been talking about?
Suppose we think of pure reason as the capacity to recognize and respond to
reasons. This is not impossible; there are reasons for moral judgements, however
we conceive of those judgements. There are also reasons for judgements of
³ I am referring here to the work of D. Wilson and D. Sperber (e.g., 2002). In my view the notion of
relevance that interests them is one whose opposite is not ‘irrelevant’, and hence is a different notion of
relevance from the one that interests me here.
, ,       125
similarity, and for judgements of relevance. Now the rational being will have to be
aware of more than just what is a reason for what. She will have to be aware of
whatever considerations can affect that sort of thing, and in my view that is a
considerable expansion. As I take it,⁴ there are features that enable other things to
stand as reasons, without being reasons themselves. If so, our rational being will
have to be able to be aware of these enablers—as of anything else that plays an
analogous role. (There are disablers as well as enablers, and there are also
intensifiers, which strengthen the reason given us by something else, and attenu-
ators, which do the opposite.) All these things must come within the recognitional
capacity of our rational being.
In a way there is nothing wrong with thinking in these terms. The only
reservation I have derives from Bernard Williams’s characteristic question what
we are to say about someone who fails to recognize a reason, or even just an
enabler. Are we to think of such a person as irrational, at least to this extent? Have
we now retained the central Kantian idea that moral failures are rational failures?
Of course there are two sorts of moral failures: failures to discern what is a reason
for what, or how one ought (that is, morally ought) overall to respond to the
present situation, and failures actually to respond in that way once one has
recognized that one should. The latter are failures of rationality only to the extent
that what one might call rational motivation should defeat whatever other forms
of motivation there might be. Since I am not myself convinced that there is any
such general rule, I leave this aside. What about the other sort of failure? Suppose,
to adapt a well-known example whose author I cannot discover, I fail to discern
that a child is alone in a crowd. Is this worth calling a defect of rationality? My
own view is that to think in those terms would be to stretch the notion of
rationality, and of failures of rationality, too far, and for no reason. For instance,
often we fail to notice a reason because we are tired, or otherwise engaged. I do not
think it is worth labelling such failures as marks of irrationality. We should not
forget, after all, that what has been driving us in this direction is the supposedly
exhaustive distinction between experience and reason. Whatever drives the
thought that this distinction is exhaustive does little to unsettle one’s sense that
there are failures of many different types, and that it serves little purpose to
maintain that all are signs either of blindness (as a general name for a defect of
sense) or irrationality.⁵
⁴ I have argued for this view in many places, most recently in chapter 3 of Ethics Without Principles
(2004).
⁵ I am grateful to Mark Sainsbury and to Paul Woodruff for their comments on this paper.
SECTION 2
MORAL METAPHYSICS
9
Two Conceptions of Moral Realism
In this paper I distinguish two conceptions of moral realism, strong and weak, and
argue the merits of the former. In fact, I argue that the weak conception is not a
form of moral realism at all. I take moral realism to be a view about the nature of
moral properties; the realist holds that moral properties are real properties of
objects. The two conceptions differ on the sense they give to the notion of a real
property. According to the weaker conception, a property is a real property of an
object if it is a property which is there anyway, independent of any particular
experience of it. A popular (but not the only nor necessarily the best) way of
capturing this notion of independence is to require that real properties persist
when unperceived. As Hume held, continuity entails independence. On this
approach, real properties are real (i.e., are there to be experienced) because they
persist (i.e., are there waiting to be experienced). Another way of capturing this
notion of independence is to turn our attention away from continuity in our
world, and consider instead characterizations which turn on the nature of other
worlds. Thus we might require the real properties to be ones which objects can
have in certain counterfactual circumstances. On the first approach, then, sub-
junctive conditionals about what would be experienced if one were to look are
sustained by statements about the actual continuity of the objects of experience.
On the second, the subjunctive conditionals are grounded in statements about the
properties that objects can bear in worlds very different from ours, typically
worlds which lack perceiving minds. Such worlds are unfortunately not much
use in ethical theory, not because it is so obvious that moral properties have no
place in worlds devoid of minds to perceive them, but because the only possible
objects of moral approval and disapproval are necessarily absent in such worlds.
There are no actions and no agents there. Whether or not a suitable version of the
relevant notion of independence can be constructed by some similar appeal to
possible worlds, I do not here presume to judge. For present purposes, it will be
sufficient to work with the simpler notion of continuity. Real properties, then, on
the weaker conception, are those that are there waiting to be experienced.
There is, however, a stronger account of what it is for a property to be a real
property of objects, which offers a stronger sense in which the moral properties of
actions exist independent of our awareness of them and so generates a stronger
130  
conception of moral realism. On this account, real properties are those which are
not constituted by the availability or possibility of a characteristic human
response. This differs from its predecessor because properties which are real on
the weaker conception may still count as unreal on the stronger. Neither sense of
‘real’ is in itself better than the other, of course. But in deciding what the moral
realist can and should claim, we need to hold the two apart. The question is
whether the realist should and can restrict himself to holding that moral proper-
ties are real in the weaker sense.
We should distinguish questions about the real from questions about the
objective and subjective. In my terms this last is a distinction between two types
of concepts, the question in ethics being ‘Are moral concepts subjective?’.
A concept of ours is subjective if it could not be grasped by a being that did not
share in characteristic human concerns. Our moral concepts are subjective if
beings that do not share in those concerns could not make any sense of the way
of carving up the world that our moral vocabulary represents. I take John
McDowell to have provided an effective argument that moral concepts are
subjective (1981: esp. 144–5). This is what is sometimes called the disentangling
argument, which holds that there is only one way in which beings who do not
share our concerns could hope to come to understand the distinctions we draw
between right and wrong, good and bad (and all the ‘thick’ distinctions between
the generous, tactless, rude, kind, etc.). This would be for them to substitute for
our way of drawing these distinctions, which naturally depends upon their
relation to our concerns, a different way which latches onto natural similarities
between objects, similarities which fit and act as ground for our regular responses
to them. But we have no reason to suppose there to be any such natural similar-
ities. Why should there be only one way of getting to be rude, a natural similarity
between rude actions (or people) to which our perception of them as rude can be
seen as a response? And if there are many ways of being rude, our aliens will have
no way of catching onto the shape we give to our concept of rudeness. Our moral
concepts will for them be shapeless because their natural grounds are shapeless.
The natural properties from which rudeness results in particular cases will differ
in ways whose irrelevance the aliens will have no hope of understanding, and if
they cannot understand this they will be unable to predict our uses of the concept,
and thus unable to find any substitute way of joining in with (or at least appearing
to join in with) our moral thought.
Moral concepts, then, are subjective, and this claim is one, but only one, sense
of the idea that moral properties are anthropocentric. They are anthropocentric
because the concepts of those properties could not be grasped by beings who did
not share in characteristic human concerns.
There is however another sense we could give to the idea that moral properties
are anthropocentric, which has already emerged in discussion of the two accounts
of what it is for a property to be a real property of objects. We might say that moral
     131
properties are anthropocentric because they are constituted by the possibility or

availability of a characteristic human response. And this would give anyone who,
though tempted by moral realism, was also convinced by the disentangling
argument that moral concepts are subjective a strong incentive to stick to the
weaker conception of real properties. For such a person ends up in the apparently
comfortable position of maintaining that moral properties are anthropocentric in
both senses, but still real properties of objects.
I see no reason in advance, however, to suppose that the two notions of the
anthropocentric necessarily go together. There still remains the possibility, for all
that has been said so far, of maintaining that moral properties are anthropocentric
in the first sense (because moral concepts are subjective), but not anthropocentric
in the second sense (because they are not constituted by the possibility of or
availability of a characteristic human response). Someone attracted by the stron-
ger form of moral realism would find himself in this position; for such a person,
moral properties, qua real, are not anthropocentric in the second sense.
I do not see any argument which is going to show it straightforwardly incon-
sistent to hold both that moral concepts are subjective and that moral properties
are, in the strong sense, real properties of objects. There seem to be plenty of
examples where we want to adopt this combination of strong realism and sub-
jectivity. For instance, I take the pain of another to be a real property which she
has unfortunately got, but I want to hold that the concept of such a pain is
subjective. But there is no doubt that this is an uncomfortable position in ethics;
though there seem to me to be good reasons for getting into it, it is difficult to be
entirely happy when one is there. This sense of tension in the position, which
I share, is probably what has led some to accept the weaker sense of realism, one
which claims less and suffers from correspondingly fewer threats of lurking
inconsistency. It is this search which has led some of those sympathetic with the
claims of moral realism to take seriously an analogy between moral properties and
secondary qualities.
Secondary qualities of objects, conceived of in Locke’s way, are unreal proper-
ties of objects on the strong conception of what it is for a property to be a real
property. As dispositions to cause certain effects, they are permanent properties of
the objects that have them, but their nature is constituted by the nature of the
characteristic effect which they are dispositions to cause. The effect is a charac-
teristic human response, and so this leaves the secondary properties as unreal
properties. But if we adopted instead the weaker conception of what it is for a
property to be real, then colour, viewed in Locke’s way as a disposition to cause a
characteristic effect in us, is a real because persisting property of objects. It is real
because the disposition persists even when not being triggered by events, as
commentators on Locke have been so rightly fond of emphasizing.
The point now is, of course, that we can easily conceive of moral properties as
real in the weaker sense, while agreeing that moral concepts are subjective,
132  
without even the appearance of inconsistency. All we need to do is to conceive of

moral properties in a way analogous to colours as Locke conceives them, i.e. as
dispositions in the object to cause or elicit certain characteristic human responses
in us; and then the analogy with secondary qualities does the work for us. This is
the approach of John McDowell’s ‘Values and Secondary Qualities’ (1985)¹ to
which the present paper is intended as a reply. And for his purposes there we do
not need to drive the analogy very far. All that we need it to do is to show that it is
possible for moral concepts to be subjective although moral properties are real
properties of objects. The analogy relies on our agreement that colour is a
distinctively phenomenal property and a real property at the same time, i.e. that
phenomenal aspects of objects can be real aspects of the world. And we then
simply say that this is how moral properties are too. But the analogy pretends to
achieve no more than this, and respects in which colour is disanalogous with
moral properties, which are admitted to be numerous, are irrelevant; they do not
impugn its limited use for this one purpose, as McDowell insists.
Despite the attractions of this weaker conception of moral realism, I want to
argue that it should be viewed as satisfactory by nobody, and that realists have got
to be brave. The only possible form of moral realism lies in the stronger concep-
tion. This is because the weaker conception is inconsistent with the main argu-
ment for moral realism, and hence has no possible provenance. It is not a form of
moral realism at all. I shall also argue that the weaker conception collapses into the
stronger one at the end of the day.
It is not necessary for this purpose that I say more in defence of the stronger
conception, for which I have offered no defence beyond a non-ethical example,
that of another’s pain. But I shall need to deal at some point with one direct
argument in favour of the weaker conception, given by McDowell (1985: 110).
This is that only by adopting some such view can we achieve a conception of an
objective circumstance which is internally related to the will. Now it may be that
not all realists are committed to or even attracted to such a conception; not all may
or need share his idea of a cognitive state all the possessors of which see exactly the
same reasons to act. Indeed, it is not clear exactly how McDowell pictures the
content of such a state. But even for those who do share this idea, I do not believe
that the weaker conception succeeds in making good sense of objective circum-
stances internally related to the will. Reasons for this will emerge later, after
suitable preparation.
¹ I have also been very much helped by Michael Smith’s paper ‘An Argument for Moral Realism’.
[Added in 2021: that paper was never published; the main ideas were repurposed in Smith (1993).]
     133
II
The first step must be an expression of the main argument for moral realism. This
is the argument from moral phenomenology; it is beginning to acquire its own
history and bibliography.² I take there to be two forms of the argument, one
simple and one more complex.
The simple form can be found in Mackie’s view that we take moral value to be
part of the fabric of the world; taking our experience at face value, we judge it to be
experience of the moral properties of actions and agents in the world. And if we
are to work with the presumption that the world is the way our experience
represents it to us as being, we should take it in the absence of contrary consid-
erations that actions and agents do have the sorts of moral properties we experi-
ence in them. This is an argument about the nature of moral experience, which
moves from that nature to the probable nature of the world.
We shall return to the question whether the weaker conception is consistent
with this simple form of the phenomenology argument. But before doing so it is
worth spelling out the more complex form. This starts from Wiggins’s suggestion
that non-cognitivism distorts the phenomenology of moral choice. In moral
choice, we struggle to find not any answer that we can bring ourselves to accept,
nor any answer that we can accept in consistency with previous answers, but the
right answer. We present our search to ourselves as one governed by a criterion
which does not lie in ourselves; our fear is that we may make the wrong choice, a
fear not allayed by the thought that we might make that choice quite happily and/
or consistently with previous choices. Why otherwise should moral choice be so
dodgy? Why otherwise should I agonize about what is the right thing to do in the
circumstances? Non-cognitivism leaves moral life too easy, and is inconsistent
with the phenomenology. It distorts our sense of the authority that the action
which we choose as right has over us. That authority is one to which our choice
must conform, and cannot itself depend on our choice; our choice is a recognition
of an authority which it could not itself create. We can express this in another way
by saying that in this area at least desire is impossible without desirability. Moral
desire and commitment do not appear as free inclinations of the will, but as
responses demanded by the desirability of the object/action. We recognize that
desirability, and our choice begins to be constrained by that recognition. The
criterion for choice then is the desirability of the action, rather than the criterion
for desirability being our choice (desire).
It is the simple form of the argument that will concern us for the remainder of
this paper, for only the simple form is concerned with our experience of the moral
properties of actions we perform or see performed in front of us. The complex
² See Mackie (1977: ch. 1); Wiggins (1976); Lovibond (1983); Nagel (1980: esp. 100–1).
134  
form is more to do with the phenomenology of moral choice than that of moral
experience, and this means that the sense in which it combines with the simple to
make a single case is not perfectly clear. However they do combine, I do not want
to say that the two arguments are conclusive, though they do seem to me in
combination to be very strong; at the limit, they aim to show that we abandon
moral realism at the cost of making our moral experience unintelligible. The point
here is that anything that presents itself as a form of realism needs to be consistent
with their general thrust. I hold that the weaker conception fails just at this point.
There are two ways to make this out. The first is to raise general doubts about
the compatibility of the sort of primary/secondary distinction to which the weaker
conception appeals with any serious use of the phenomenology argument.
Perhaps the best way to do this is to move away from ethics for a moment and
look at a distinction between two sorts of direct realism in the theory of percep-
tion. The direct realist holds that we are not aware of physical objects in virtue of a
more direct awareness of some intermediary objects, sensa perhaps. One of the
arguments for this position is a feeling that we cannot make good enough sense of
the intermediary objects, in particular that they do not seem to have sufficient
properties of the right sort to sustain the role here assigned to them. But another
argument starts from the phenomenology of perception. The sort of double
awareness of which indirect realists speak so warmly is not one of which we are
aware; for us there only seems to be a single awareness, which when we are lucky is
an awareness of the world around us. This phenomenology argument is far from
conclusive, but it represents perhaps the first reason for taking direct realism
seriously (just as in ethics the phenomenology of moral experience is the first and
perhaps the only argument for realism, remaining thoughts being used for
defence/offence). And the point of all this is that it is not clear that the primary/
secondary quality distinction is compatible with the realist’s attitude to the
phenomenology. For that distinction amounts to taking the phenomenology
seriously only some of the time; the more we draw the secondary qualities away
from the objects and nearer to ourselves, the more we step away from the direct
realist’s feeling that in perception we are open to the world as it is before us. For
some of the way the world presents itself to us is abstracted from it by the
distinction between primary and secondary qualities, becoming instead more an
aspect of our response than an aspect of that to which perception is a response.
It might still be objected to this that the weaker conception does not draw the
secondary qualities away from the objects and nearer to ourselves, since it places
those qualities firmly in the objects as dispositions to cause characteristic
responses in us. This leads me to the second and more powerful argument that
the weaker position is not consistent with the phenomenology argument(s): it is
hard to make good sense, within that position, of the idea that moral properties
can be experienced. For the experiences we have, as effects of a disposition in the
objects to cause those experiences in us, do not seem to be experiences of that
     135
disposition. More is required for an experience that is caused by a dispositional

property to be an experience of that property, and that more is lacking in the case
we are now considering. Our moral experiences, on this approach, seem to be
more like manifestations of the disposition than experiences of it.
So far, this complaint has only the status of a suspicion. But if it is sound, it will
be fatal to the weaker conception’s claim to be a form of moral realism, for in
undermining the idea of moral experience it removes all possibility of reliance on
the phenomenology argument. In the next section, I attempt to make the best
sense of this thought. One difficulty will be that it threatens to prove too much, for
if sound it bids fair to have the same consequences for colour perception, on the
Lockean primary/secondary model. Can we really accept that Locke was unable to
make good sense of the idea that we experience the colour of objects? There is
some reason to believe that he was, but we will return to this question later and our
rejection of the weaker conception will not initially depend upon our answer to it.
I shall however end up suggesting that current versions of the primary/secondary
quality distinction leave the secondary qualities beyond the reach of experience.
III
What more is required for an experience caused by a dispositional property to be

an experience of that property? One suggestion, firmly within the Lockean
tradition, requires our experience of the property to resemble that property ‘as it
is in the object’. But this fails to provide what we want, for two reasons. First, our
experience of moral properties on the weaker view does not resemble them as they
are in the object, for there they are only dispositions which our experience can
hardly be said to resemble at all. Second, the classic Lockean account of the
primary/secondary quality distinction in terms of ideas of properties that do
and do not resemble those properties as they are in the object is completely
flawed, and must be rejected if that distinction is to survive. The reason why it
is flawed is that Locke requires us to make sense of an instruction to distinguish
between primary qualities which resemble purely phenomenal objects (ideas) and
secondary qualities which don’t. But this is a hopeless muddle, worse in its
account of primary qualities than of secondary ones. Primary qualities, viewed
in this light, will be impossible because we can make no sense of properties which
are objective and able to exist independent of perception but yet resemble our
experience in purely phenomenal respects. (This is the main thrust of Berkeley’s
objections to Locke, based on his claim that an idea can be like nothing but
another idea.³) So the primary/secondary distinction would collapse in favour of
³ The objection is used also by McDowell (1985: 113–17).

136  
the secondary. We need, therefore, a new account of the distinction before we can
really judge the strength of the analogy between moral properties and secondary
qualities.
The only way in which a distinctively different account can be constructed is to
abandon the hope of running the distinction in terms of an intrinsic resemblance,
or the lack of one, between external object and internal subjective state. Instead we
should drop talk of resemblance in favour of talking of an identity between how
the objects are and how our experience represents them as being.⁴ Here there is no
suggestion that experiences act as intermediaries between ourselves and the world,
having representational properties only in virtue of their intrinsic ones. Instead we
should think of experience as intrinsically representational (cf. McDowell 1985:
115–16). This makes room for us to think of the content of experience as being in
favourable cases identical with the way the world is; the world can be the way it
seems to be. And with this we can recover the primary/secondary distinction as
the difference between those properties that are represented to us in experience as
distinctively phenomenal (the secondary ones) and those that are not.
One of the purposes of the shift from talk of resemblance between perceptual
state and object to talk of identity between representation and property repre-
sented is to avoid the insidious attractions of a question which looks very like the
one I asked at the beginning of this section. Representative (or indirect) realists
tend to ask ‘What more is required for a perceptual state that resembles an object
to count as a perception of that object?’, and then to try causal answers. If the
question is well conceived, and in particular if the concept of a perceptual state
which it requires is sound, some causal answer must be right; though this faith has
hardly been borne out by any conspicuous success in the answers offered.⁵ But the
attempt to buttress resemblance with causation in order to arrive at an account of
perception is rendered void by the shift from talk of resemblance to talk of
identity, and with it falls the ‘What more?’ question mentioned above (as well as
a traditional ground for the argument from illusion). But this does not mean that
our ‘What more?’ question is no longer applicable, for ours can be asked within
the new approach. In asking what more is required for an experience caused by a
dispositional property to be an experience of that property, we are asking under
what conditions a dispositional property can be the content of an experience
which it causes, or under what conditions an experience caused by a dispositional
property can represent that property to us and be an experience of that property.
My strategy will be as follows: first I shall pretend that we can make good sense
of colour experience as the experience of something which is, in the object, only a
disposition, and argue that even then we cannot make similar sense of moral
experience. This will mean that colour experience is secure and moral experience
⁴ Here I follow McDowell and also Evans (1980: esp. 94–9). ⁵ See Snowdon (1981).
     137
is not. But I will then turn to suggest more strongly that the experiences caused by
the disposition which is what colour is in the object are not experiences of that
disposition, on our present understanding. So if you find the first part of the
argument unconvincing, you should remember that it is succeeded by a stronger
and more general one.
So let us suppose for a moment that there is some answer to this question,
which persuades us that we can make good sense of the experience caused by the
colour of an object being an experience of that colour, even though colour is to be
conceived dispositionally. Can our moral experience similarly represent objects
(actions) to us as having the dispositions which are moral properties in the
objects?
Now there seem to be two ways in which our experience can represent
dispositions to us. The first is what one would think is the normal one, that of
actually having the sort of perceptual experience which the disposition is a
disposition to elicit. (We are supposing for the moment that this works at least
in the case of colours.) The second is perhaps rather rarer, but none the less
interesting for that; it is when we experience the disposition unactualized. A good
example of this is our ability to experience the sharpness of a knife by running a
thumb across the blade, so that we experience its ability to cut without experien-
cing any actual cutting. Which of these models is most suitable for our present
purpose? The second model looks initially unpromising. If we are trying to
conceive of a moral property as a disposition to elicit certain responses from us,
are we not to take it that the standard way to experience that disposition is when
we find ourselves approving or disapproving? If so, the first model has first claim
on our attention.
To discern the difficulties presented by the first model we need to notice one
respect in which moral properties are disanalogous with the secondary quality of
colour (and other secondary qualities too). We can consider the experience of
colour either determinately or determinably, either as the experience of some
particular colour or as the experience simply of some colour or other. Considering
the matter determinably, a coloured object is one which has a disposition to cause
a distinctive sort of perceptual experience, colour experience. What is it for an
object to be red? Analogously, it is for the object to have a disposition to cause a
distinctive sort of colour experience. But all objects have a disposition to cause that
sort of experience in some conditions; a red object is one which has a disposition to
cause that sort of experience in standard or normal conditions. Turning now to
ethics, we might hope for a reasonable analogy for the determinable; an object that
has some moral property is one with a disposition to elicit a certain sort of
response, let us say approval and disapproval (stones can’t do this, perhaps).
But when we turn to the determinate, we cannot suppose that a determinate
moral property is a disposition to cause a certain moral response in normal
circumstances. Normal situations are precisely not very good ones for discerning
138  
the moral properties of the choices we face. We have to deal with distractions,
contrary influences, prejudices, lack of time and similar problems. More promis-
ing, it seems, is to conceive of the moral property of rightness, for example, as a
disposition to cause a certain response (approval) in ideal circumstances. (I allow
this notion of ideal circumstances to package together what are really two require-
ments, that the perceiver be ‘ideal’ and that the situation be ‘ideal’ too.)
Our question now is whether it is possible for our moral experience, which
never occurs in ideal circumstances, to represent to us moral properties conceived
now as dispositions to cause certain responses in us in ideal situations, which we
never experience. In such a case, though our experience may be indistinguishable
from the experience which the relevant disposition is a disposition to elicit, it is
not that experience itself. And this means that our first model turns out to be
inapplicable. For the model is written around the notion of an identity; we
experience a disposition if we have the experience which the disposition is a
disposition to cause. Our normal moral experience may be indistinguishable
from the experience which the moral property is a disposition to cause, but it
will not be identical with it, for the phrase ‘in ideal circumstances’ is an inelimin-
able part of the specification of that disposition.⁶
But this is not yet the end of the matter. After all, we are able to tell from an
experience of colour in an abnormal situation what colour the object would look
in a normal situation; at least we can do this with a bit of gathered knowledge
about the relevant relations between normal and abnormal situations. Can’t we
hope to do the same in ethics? I don’t see how we can. The problem is not that the
sort of response we are here talking about is one of which we cannot conceive; as
far as the response goes we can conceive of it because we have experienced it, or
responses indistinguishable from it, in more normal, less ideal circumstances. Nor
do I think that the right way to express the problem is by claiming that we are in
such cases involved in inference from what we perceive rather than in perception
itself; this would matter because of the constant need to remain consistent with the
general thrust of the phenomenology argument, but I am not confident enough
about the distinction between perception and inference to rely on it for the present
purpose. The real point is that though we might be said, in the case of colour, to
see what colour the object really is in seeing what colour it appears in this light,
this is because we have sufficient independent experience of the normal and of the
abnormal in this case. But in the ethical case, we lack the independent experience
which would, if it existed, enable our present non-ideal moral experience to count
as experience of the real moral properties of the object. We cannot be said to
experience here the object’s disposition to elicit a certain response in ideal
circumstances.
⁶ I find a slightly different use of the ideal/normal distinction in McGinn (1983: 150–3).
     139
Nor will it help to return to the second model, which we initially ruled out as
unpromising. To do so would now amount in this case to saying, by analogy,
that our only way of telling that a knife is sharp is by running our thumb along
its edge. But surely, though running one’s thumb along in that way may perfectly
well be a way of perceiving the sharpness (rather than inferring that the knife is
sharp from something else that we perceive), still this can only be so because we
have established a way of correlating the way the knife now feels with an ability
to cut. But in the ethical case no such correlation is available, because of the
unavailability (and indeed the redundancy) of experiences of those moral
properties which are dispositions to elicit characteristic responses in ideal
circumstances.
This argument, and the distinction which it uses between the normal and ideal,
is what undermines McDowell’s main independent argument in favour of the
weaker conception of moral realism, referred to at the end of Section I. The crucial
consideration is that a moral property, conceived as a disposition to elicit a certain
response in ideal circumstances, is not for that reason internally related to the will
of a perceiver who is not himself in those circumstances.⁷ If, for instance, I am
drunk or overcome by greed, or even surrounded by and distracted by ordinary
concerns, why should I care about the disposition an action may have to cause
certain responses in me in situations in which I do not at present find myself? We
may for these purposes admit that if I were in such a situation, the objective
circumstances with which I take myself to be faced would be internally related to
my will. But I am never in such circumstances, and this means that the sort of
internal relation that McDowell is seeking to establish is never (or at least not
normally) present. So the weaker conception succumbs itself to his objection to
the stronger, if that objection is valid at all.
It is beginning to look, then, as if there is a real difficulty in the idea that moral
properties, conceived as analogous to secondary qualities, are genuinely experi-
enceable. And the argument to this conclusion does not have the unfortunate
consequence of ruling out any experience of secondary qualities. For it traded,
first, on a distinction between moral properties and ideal circumstances, on the
one hand, and secondary qualities and normal circumstances, on the other; and
second, on the resulting availability of independent background knowledge to
extend the bounds of the perceivable in the case of colour but not in the ethical
case. We have not yet argued that on the revised account of secondary qualities
colour cannot be experienced. Of course, on the faulty account derived from
Locke, which we rejected, there is such an argument, for Locke cannot make
good sense of there being an idea in our mind resembling colour as it is in the
object, as he admits (indeed insists). And this seems to entail that though we can
⁷ I owe this thought to David McNaughton.

140  
have experiences caused by the disposition which is colour as it is in the object,

and in this extensional sense have experiences of colour, still our experience in this
case is not of colour any more than an experience caused by a dangerous object,
and caused by the disposition which is its dangerousness, is for that reason an
experience of danger. But the revised account of secondary qualities, as distinct-
ively phenomenal properties which our experience represents objects as having,
does not admit this sort of attack. Our objects are, on this account, represented to
us as having certain dispositions to elicit responses from us in certain circum-
stances; this relation is irreducibly intensional.
This does not mean that no doubts can be raised about the account of
experience of secondary qualities given here. We may well wonder whether our
experience really does represent objects to us as having these dispositions, and
suppose instead that however dispositional our understanding of them may be,
still our experience of them is as non-dispositional. Can we take seriously the way
in which colour appears to us, which seems to me at least to be stubbornly non-
dispositional, if we insist that the right account of the ‘distinctively phenomenal’
nature which colour is represented as having is to be understood as its being (or
existing in virtue of) a disposition to present a certain sort of appearance? It may
turn out that real realists will eventually be able to think of moral properties as
analogous to secondary qualities, but only at the cost of abandoning the disposi-
tional account of both which is the core of the weaker conception of moral
realism, for on it hangs the argument that moral properties are real properties
of objects. The central questions here are ‘Can dispositions appear, and how do
they do it?’. Two important matters depend on our answers. First is the question
whether the revised account of the primary/secondary quality distinction is in the
end any improvement; it may seem that whereas Locke’s account fails to make
good sense of the primary qualities, as McDowell argues, McDowell’s own account
fails to make sense of the secondary ones as possible objects of experience. Second
is the question whether the sense, if any, in which moral properties are distinct-
ively phenomenal bears any relation to the sense (or senses) in which colour is
distinctively phenomenal.
So far we have been discussing the question whether the weaker conception of
moral realism is really a form of moral realism at all, and arguing that it is not
because it does not make satisfactory sense of the experience of moral properties
of objects, and therefore is eventually inconsistent with the phenomenology
argument, which is the main incentive to moral realism in the first place. But
we have now reached a point at which our considerations take a rather different
tack, tending to show that the weaker conception collapses eventually into the
stronger. The difference between the two conceptions lies in a dispute about
whether to introduce a richer conception of moral properties as they are in objects
than that of a disposition to cause or elicit a certain response, or to present a certain
perceptual appearance. McDowell’s formulation of the position (1985: 111), which
     141
speaks of ascriptions of secondary qualities as only able to be understood as

true, if true, in virtue of the object’s disposition to present a certain sort of
perceptual appearance pretends to leave it open whether the moral property is
to be seen as the disposition itself, or whether the moral property is something
which results from the disposition. But perhaps the main competing view is that
the disposition itself results from the moral property: this not merely causally, of
course, just as it is not possible that red objects should not have a disposition to
look red. Speaking in general of secondary qualities, a possible position is that
though we continue to see them as distinctively phenomenal, we understand the
disposition to present certain sorts of perceptual appearance as lying between the
secondary quality as it is in the object and ourselves as perceivers. This position
involves giving an account of what it is to be distinctively phenomenal other than
the McDowell one in terms of a disposition or of that which exists in virtue of a
disposition. But with a new account, the question will be whether colour is
distinctively phenomenal in any sense in which moral properties are. There
must be some such sense for McDowell’s purposes, for otherwise the reality of
the phenomenal, established in the case of colour, will not spread over to establish
the reality of the moral.
The present point, however, is that an account of moral properties as disposi-
tions to elicit certain responses in ideal circumstances seems itself to introduce a
more substantial conception of those properties than the dispositional one. This
emerges when we ask, what is ideal about the ideal circumstances? What are the
ideal circumstances ideal for? Presumably for the elicitation of moral attitudes.
But the question is what is better about the attitudes elicited in ideal circum-
stances. They, like other attitudes formed in more normal situations, are responses
to a disposition—a disposition which shows no need to be understood in terms
of its effect in certain very special circumstances. It is only if we smuggle in the
idea that the disposition cannot reveal itself as it really is except in those special
circumstances, that those circumstances come to play an important role. But this
idea is uncongenial to the weaker conception of moral realism (whether it is
consistent with the phenomenology argument or not), for it comes naturally
only to someone who says that as well as the disposition to elicit attitudes of
certain kinds, there is something further in the objects/actions which the atti-
tudes caused by the disposition only fit, or can only be guaranteed to fit
(whatever sort of fit is in play here), in ideal conditions. The conditions are
ideal because they are the ones in which our response fits the moral property,
which cannot therefore be conceived of as a disposition. So the conclusion
seems to be that the weaker conception collapses into the stronger one, because
the attempt to see moral properties as dispositions to elicit certain responses in
ideal circumstances depends covertly on the richer conception to which it was
supposed to be an alternative, if it is to be able to explain why the ideal
circumstances are ideal.
142  
IV
This concludes my argument against the weaker conception and in favour of a

stronger alternative. But it may seem that even if my arguments succeed in
demolishing the target at which they are aimed, there remains a further and better
version of the attempt to use an analogy between secondary qualities and moral
properties, which escapes what has been said so far. For towards the end of his
paper, McDowell (1985) himself begins to stress a feature of disanalogy between
the two. This is that while a secondary quality like colour is adequately conceived
as a disposition to cause certain responses in us, a moral property should be seen
rather as a way in which an object merits some distinctive responses.
McDowell has his reasons for this shift, of course; they concern the way in
which he thinks we should seek to avoid Mackie’s charge (1977: 19–20) that we
have empirical reason to avoid taking the deliverances of experience seriously in
certain cases, because the properties which our experience represents objects as
having are explanatorily redundant. But this shift seems to make the use of a
secondary quality model harder to sustain, at least at first sight. Admittedly the
model was never intended as more than a model. There was no intention of
holding that moral properties are secondary qualities in just the sort of way that
colour is. As I have already suggested, moral properties may not be perceptual in
just the way that colour is; they are not in the same way ‘distinctively phenom-
enal’. This may mean that McDowell is wrong to insist on the notion of a
perceptual appearance as central to the distinction between primary and second-
ary qualities; there seem to be some properties such as those of being dangerous
(to which he refers—1985: 120) and interesting (to which he doesn’t) for which a
secondary quality model looks promising but which do not seem to fit a centrally
perceptual version of that model. (Even if dangerousness is a quality of Locke’s
third sort, interestingness should, as a power in objects to cause ideas in us, count
as secondary.) But talk of meriting a response rather than causing it seems to make
the perceptual model harder to cash for ethics, and thus to distance it from what
we are still maintaining to be analogous. It strengthens the sense in which the
distinctive phenomenality of colours differs from that of moral properties. So the
purpose of the analogy begins to become mysterious.
I said that the use of the notion of meriting seems at first sight to have the effect
of undermining the secondary quality analogy. But in fact my opinion is that is it
not in the end distinct from a notion we have already considered and found
wanting, that of a disposition to elicit a certain sort of response in ideal circum-
stances. It is not as if we have here two differences between moral properties and
colour, the first lying in the distinction between normal and ideal circumstances
and the second in that between eliciting and meriting a response. The only way to
understand the notion of meriting a response is to see a merited response as the
one which would be elicited in ideal conditions. We can give no good sense to the
     143
thought that an object should merit a response which it would never receive, even
in ideal circumstances; that there should be something about a good or a right
action which lies beyond any possibility of recognition; that an action might
be right even though there are no circumstances in which we could hope or be
expected to notice the fact; or that the rightness of an action might only be
revealed to us in less than ideal circumstances. So it seems to me that the notion
of a response which an action merits is identical with the notion of the response
which it would elicit (has a disposition to elicit) in ideal circumstances. And this
means that we are left with our two original thoughts, that there is something odd
about the experienceability of dispositions, and that the distinction between the
normal and the ideal undermines the weaker conception of moral realism. The
detour through merit is in the end no improvement.
What then remains of the primary/secondary distinction? Abandoning the

attempt to run this distinction in terms of those properties which in the object
are merely dispositions to cause certain perceptual states in us, what importance
can we attach to the idea that the distinction can or must survive as part of an
attempt to separate properties that are distinctively phenomenal from those that
are not? This separation must rest on a difference which is not merely a matter of
degree.
What we are looking for as direct realists is a distinction between two types of
property which is not mirrored in the epistemology. (This is because any attempt
to mirror it in the epistemology is likely to open the door to indirect realism.)
McDowell’s idea is that secondary qualities are something distinctively, even
essentially, to do with the way things look. There is a contrast in this idea, but it
must not be that between appearance and reality; we are not separating properties
which concern the way things look from those which concern the way they are—
not at least if we intend to wind up with some form of realism about the secondary
qualities. Nor can we be here contrasting ways things can look with ways they
can’t look. There is no future in denying that things can look square. Perhaps
instead the distinction is between the secondary qualities, which are essentially to
do with how things look, and the primary qualities as other ways things can be but
which they don’t need to look. Primary qualities, on this account, can get by
without the looks.
But don’t square things need to look square? On the Lockean resemblance
theory, now rejected, they perhaps don’t. It might be a contingent question
whether our ideas of square objects really resemble those objects in respect of
squareness. (Actually, we have already decided on Berkeleian grounds that they
can’t; this feature of Lockean theory is not a strength.) But on the identity-based
144  
theory which we have substituted for the resemblance theory, it is surely not a
contingent question whether our perceptual states do or don’t represent objects to
us as square, or whether objects that are square are or are not as our perceptual
states represent them to be. This gives one sense in which both primary and
secondary qualities are essentially phenomenal. Is there another sense in which
the essential phenomenality of the secondary is distinctive?
We earlier considered the suggestion that to be red is to look red in normal
conditions; to be coloured is to (be able to) present a certain kind of perceptual
appearance. And we suggested that it is not possible for red objects not to have a
disposition to look red. But this takes us no further forward; on the identity-based
theory it is not possible for square objects not to have a disposition to look square,
although the conditions in which they will look the shape they are (i.e., those that
are to count as normal for shape perception) will be of a different style from those
in which they will look the colour they are. In a way, the identity theory makes it
impossible for us to ask which leads here—whether our conception of what it is to
look F is derived from our conception of what it is to be F, or vice versa. It
collapses this contrast, and with it should fall our trust in any recognizable version
of the primary/secondary distinction. With this collapse moral realists are in no
danger of suggesting ludicrously that moral properties are or are like primary
qualities. But neither will they derive much sustenance from an analogy with
secondary qualities. Real realists don’t need the distinction at all.⁸
⁸ I acknowledge with gratitude the helpful comments I received on earlier drafts from David
Bakhurst, Robert Gay, Harry Lewis, Richard Swinburne and particularly David McNaughton and
Michael Smith.
10
Contemplating One’s Nagel
Thomas Nagel exhibits such a sense of philosophical puzzlement, such engage-

ment with philosophical issues, and such attractive diffidence in the answers he
proposes, that I wish I found it easier to accept his perspective on the problems
with which he works.¹ By this, I do not mean the high level of generality, though
this book is certainly written with a very broad brush; except perhaps in the
discussion of ethics towards the end, there is little in the way of detailed analysis
and argument. Nagel may feel that he has done all that before—and indeed much
of the material in this book has been presented to us before in different forms. But
the main claim of The View from Nowhere, and its particular contribution if
successful, is not so much in the discussions of the different topics that occur as we
go along, but rather the framework within which those discussion take place. It is
the links between the different topics that create the central message, and it is
Nagel’s account of those links that I find it hard to accept.
Nagel’s book starts thus:
This book is about a single problem: how to combine the perspective of a

particular person inside the world with an objective view of that same world,
the person and his viewpoint included. It is a problem that faces every creature
with the impulse and the capacity to transcend its particular point of view and to
conceive of the world as a whole.
The perspective of a particular person inside the world is called a subjective view,
and so the ‘single problem’ is how to combine a subjective and an objective view.
Nagel wants to persuade us, first that we cannot avoid the attempt to combine
subjective and objective views, second that the attempt to achieve an objective
view without totally abandoning the subjective one leads at best to tension and at
worse to incoherence, third that these tensions and incoherences recur in very
much the same form in many of the main problems of philosophy, and so, fourth,
that our best hope is to understand the need to combine subjective and objective
views and to go as far in that direction as is possible, even if perfect objectivity is
unattainable. He then makes different suggestions as to how this last might be
done in different areas, but these suggestions are not supposed to offer ways in
¹ This paper was a critical notice of Nagel (1986). I owe its irresistible title to Paul Snowdon, and am
indebted even in my first sentence to Jim McLaverty.
146  
which we might hope to avoid the tensions between objective and subjective,
which are in some way fundamental to the human condition.
There are occasions in The View from Nowhere when the strength of Nagel’s
vision seems overpowering. I have in mind here particularly his discussion of how
it is possible that one of the people in the world should be me, and the final chapter
on birth and death. Here Nagel is at his impressive best, and it is passages like
these that persuade me that this is an important book, even if not a new one.
Elsewhere however, the rhetoric of objective and subjective is less convincing.
It is perhaps natural, when told that all these different problems are really
symptoms of the same underlying tension, to doubt whether this is true. And
I find that it is not true, and that much of the recurring tension which Nagel
discerns between objective and subjective points of view is in fact a tension
which derives from the attempt to run two versions of the objective/subjective
distinction simultaneously.
Nagel announces that for him objective facts are those which are accessible to a
special sort of understanding; objectivity is that method of understanding. One
point of view is more objective than another if the world as seen from that point of
view is more able to be understood in the objective way. There are degrees of
objectivity; a point of view may be more objective than some, and less objective
than others. The attempt to leave our subjective point of view and strive for a more
objective one is called objectification. And it seems to me that if one looks carefully
at Nagel’s descriptions of objectification, one can see that there are in fact two such
processes. The first appears on page 4: ‘To acquire a more objective understanding
of some aspect of life or the world, we step back from our initial view of it and
form a new conception which has that view and its relation to the world as its
object.’ Let us call this the first or Hegelian model of objectification. Notable
features of it include:
1. It is linear, and apparently infinite. There is no prospect of reaching a view

of the world which we cannot move beyond, by a further step of the
same sort.
2. In this process of objectification, nothing is left behind. Every aspect of each
succeeding view is retained, though maybe somewhat altered, in each of its
successors.
3. There is therefore some reason to say that on our first step away from our
initial, most subjective point of view we change our view of the world.
4. Despite the Hegelian overtones of this sort of objectification, we are not
forced into the rise towards objectivity by any sort of contradiction, but
nonetheless we cannot avoid the desire for an objective view, since that
desire is a feature of our humanity. We crave objectification.
5. So objectification is not a luxury, but a necessity for us. But though we have
no choice in the matter and so don’t need reasons for seeking an objective
 ’  147
view, there are still good reasons to be found because of the benefits such a
view can offer. The obvious benefit is an increase in understanding.
The second sense of objectification is expressed on the next page:
The distinction between more subjective and more objective views is really a
matter of degree, and it covers a wide spectrum. A view or form of thought is
more objective than another if it relies less on the specifics of the individual’s
makeup and position in the world, or on the character of the particular type of
creature he is. The wider the range of subjective types to which a form of
understanding is accessible—the less it depends on specific subjective
capacities—the more objective it is.
The suggestion here is that the move towards a more objective view involves
abstracting from one’s own view of the world any element which can be seen as
the product of one’s own perspective. The hope presumably is to arrive eventu-
ally at a view which has no element of that sort. This makes it appropriate, in the
light of the popularity of the notion of the absolute conception, to call this
second sort of objectification absolute objectification. Prominent aspects of
it are:
1. It is linear and infinite unless there is a most objective view no element of

which can be attributed to the perspective from which that view is reached.
2. A great deal is left behind in the process, since at every stage we separate off
into a category called ‘appearance’ something which earlier was conceived as
part of ‘reality’.
3. Since the whole aim of the process is to achieve a view of the world as it is in
itself, uncoloured by the contributions of the perspective from which we
view it, the nature of the world changes (reduces) as our view becomes more
objective. At each stage, further qualities are eliminated as mere appearance.
(A few, such as electric charge and charm, are added.)
4. We are not forced into the rise towards objectivity either by our humanity or
by some Hegelian contradiction, but we can expect a more objective view to
be nearer the truth about the world, so that we can correct earlier views from
the perspective of later ones. The truth about the world as it is in itself is
contained in what can be discerned by a form of understanding which does
not depend at all on any subjective capacities in respect of which another
view might be different.
5. The drive to abstract from one’s own perspective, in the attempt to reach a
view to which one’s individual perspective makes no contribution, makes us
think of this form of objectification as essentially a search for generalization;
we search for the general in this and only this sense, that we attempt to
148  
reduce ourselves to that which can be affirmed from any perspective what-
ever, whatever its individual peculiarities. This is the view from nowhere in
particular, or the view from nowhere.
Are the two forms of objectification compatible?
Let us start by noticing some differences between Hegelian and absolute

objectification.
1. Nagel says that ‘The standpoint of morality is more objective than that of
private life, but less objective than the standpoint of physics’ (p. 5). This is
true for the second sense of objectification, but false for the first. No one
could say that physics is an attempt to form a conception of the world which
has the relation between our moral view and the world so viewed as its
object, nor that it is likely to occur as a later stage in the process of Hegelian
objectification. Physics is not concerned with the relation between our view
of the world and the world so viewed at all. It is odd that Nagel continues:
‘We may think of reality as a set of concentric spheres, progressively
revealed as we detach gradually from the contingencies of the self.’ For
this notion of concentric spheres fits only Hegelian objectification. Absolute
objectification, by its very nature, does not tend to build a broader concep-
tion of reality around a narrower one, but rather to banish to the realm of
appearance parts of what were previously taken to be reality.
2. Nagel is prone to talk about detachment, as when he says that physics is the
science in which we have achieved our greatest detachment from a specif-
ically human perspective on the world. But the notion of detachment varies
according to which sense of objectification is at issue. In the first or Hegelian
sense, the sense in which we leave anything behind is dubious. There is
certainly no antecedent guarantee that the results of the process will be our
abandoning as unreal many aspects of the world as we originally conceived
it. We may of course come to change our minds about something; but this is
to take something to be false which we previously took to be true, not to take
something to be mere appearance which previously we took as reality.
Detachment for the Hegelian amounts to this, that we stand back from
our ordinary engagement with the world and view that engagement as an
object. The difficulty in taking simultaneously a more and a less objective
view, if any, derives from the need to be more and less engaged at the same
time. This is certainly difficult, but there is no reason to suppose it to be
impossible. Hegel of course had special reasons for saying that it is impos-
sible to be at two levels of consciousness at the same time. This is because for
him the higher level ‘sublates’ the lower, since we have only risen to the
 ’  149
higher because of a contradiction at the lower level. There is something of

this in Nagel. Having achieved a more objective view, one cannot hope to
return to one’s earlier subjective view with pristine innocence. In this sense
one cannot combine an earlier and a later view, but there is no reason why
one should wish to do so. Having achieved the greater understanding of the
objective point of view, we should not wish to abandon that advantage. But
this should not mean that we cannot return to the subjective view at all: only
that on our return we will find it altered. On the second sense of objectifi-
cation, by contrast, the whole purpose of the enterprise is to hive off certain
aspects of the world as we originally take it to be in the attempt to improve
our conception of independent reality. The difficulty of taking simultan-
eously a more and a less objective view will now be that we are committed to
thinking of some aspect of the world both as real and as mere appearance.
This is surely impossible on any account. But I would say that the need to
achieve this impossibility is less pressing than the need to stand back from
our engagement with the world for the sake of a gain in understanding. It is
the latter which is required of us by our humanity; the former seems to offer
much more technical advantages.
3. For Nagel, objective reality (that part of reality which can be understood
from no perspective in particular) is incomplete in two ways. First, there
may be aspects of the world which we can never understand at all; this is an
expression of extreme metaphysical realism. (He does not take this position
for ethical truth.) Second, there are aspects of reality which an objective view
leaves behind. The nature of any sensory state (even one of ours) is
something of which our objective understanding is at best partial. This
sort of thought seems to fit the second sense of objectification much better
than it fits the first. There seems to be no reason why our own objective
understanding of our sensory states need be incomplete, for Hegelian
objectification is not an attempt to conceive of those states from a perspec-
tive with no distinguishing features at all, and hence the end result is not
likely to be a perspective from which our subjective states are likely to be
incomprehensible. Hegelian objectification would be a failure if that were its
result, while absolute objectification would not.
4. Nagel suggests that a subjective view is ‘correctable or confirmable’ by
reference to a more objective view. This is true on both accounts but in
different ways and for different reasons. On the Hegelian sense, our exam-
ination of the relation between our view of the world and the world may lead
us to make a change in that view. (An example of this might be our response
to Hume’s doubts about induction.) But there is no sense in which the
results of objectification are intrinsically more real than the contents of a
more subjective view. By contrast, there is a definite feeling that absolute
objectification is a move towards the truth. ‘The fundamental idea behind
150  
the objective conception is that the world is not our world’ (p. 18). In the
objective view what we reach is the result of a process of separating out as
much as possible as a facet of our response to the world, and what has been
separated out cannot be put back again. For the sense of reality at issue is
one generated by the process of separation, and the continuing contrast
between the real (independent) and our reaction/response to it.
5. For Nagel, objectification is generalization. ‘The search for generality is one
of the main impulses in the construction of an objective view—in normative
as in theoretical matters. One takes the particular case as an example, and
forms hypotheses about what general truth it is an example of ’ (p. 152). The
appeal to the example of the theoretical is revealing; we are clearly intended
to think of physics. But the generalizing tendency of physics, and its
dismissive attitude towards the particular case, is not a facet of objectifica-
tion in either sense. There is admittedly a (different) sort of generalization
involved in absolute objectification, for we are expected to abstract from our
own perspective and reduce our account of the world to those truths which
can be reached and affirmed no matter what the perspective from which we
start. In ethics, we are told that ‘the detached objective view takes in
everything and provides a standpoint from which all choosers can agree
about what should happen’ (p. 183). Elsewhere, the very nature of the search
for objectivity guarantees that its results will be shareable, understandable
by all regardless of their particular perspective. But such shareable, mutually
comprehensible thoughts need not be general in content; they are not
general in the way that physics is general. They may have entirely particular
contents, so long as they admit of general access. And even this much is
restricted to absolute objectification. For Hegelian objectification, there is no
reason to suppose that the results of my attempts to understand the relation
between my subjective point of view and the world I so view need be subject
to the demand that creatures relevantly different from myself should
understand them.
6. The difference between objective and subjective points of view is a matter of
degree. Is there a most objective view? This depends. Hegelian objectifica-
tion seems to me likely to continue as a possibility indefinitely, unless we
meet a stage where there is (as Hegel would put it) no distinction between
our consciousness and its object. But on Nagel’s approach there is no reason
to expect such a stage. The first sort of objectification is Hegelian but not
Hegel’s. With absolute objectification, however, there is a reason to suppose
that there is a final stage of objectification, though there is no reason to
suppose that we might reach it. The reason is given by the very nature of the
process we are engaged in. For surely there must be such a thing as the
independent nature of the world, and the most objective view will be that
which has this and only this as its content.
 ’  151
7. This does not mean that the drive towards objectivity is a drive towards
truth and away from error. There is no stage at which error is impossible;
the distinction between truth and error is not really related to the object-
ive/subjective distinction at all, despite Nagel’s attempt to argue in one
place that a view is subjective if it does not admit of the distinction
between truth and error (p. 32). Objectification of the second sort is
concerned, on Nagel’s showing, with the distinction between appearance
and reality. But there is no suggestion that appearance and error are
somehow connected.
8. Or rather there should be no such suggestion. But sometimes there is a
distinct hint that what merely appears and what is false are connected, so
that the move from a more subjective view to a more objective one is a move
towards truth. Nagel’s official view is that the objectively understandable
world is not the whole world. ‘The aim of such understanding . . . is to go
beyond the distinction between appearance and reality by including the
existence of appearances in an elaborated reality. Nothing will then be left
outside’ (p. 18). But even this does not go so far as to say that the
appearances themselves will remain; what remains is the fact that there
are appearances, which is rather different. And elsewhere there is a distinctly
dismissive tinge to the notion of an appearance. ‘The old view then comes to
be regarded as an appearance, more subjective than the new view, and
correctable or confirmable by reference to it’ (p. 4). Elsewhere: ‘Why are
we not content to regard the internal perspective of agency as a form of
clouded subjective appearance, based as it inevitably must be on an incom-
plete view of the circumstances?’ (p. 114). This sort of downgrading of the
subjective view is unjustified; the objective view is equally incomplete. And
it is also dangerous, for if we are not careful we will find ourselves saying
that an objective view is one from which all (mere) appearances have been
expunged, so that the objective world is one which does not appear at all.
This would be an error, but I suspect that some thought of this sort underlies
some of Nagel’s sense that the view from nowhere is incoherent, despite
being required of us.
With these comments on the differences between the two forms of objectifica-
tion, we can try to decide whether they are compatible. In one sense, of course,
they are. There is no reason why someone should not interest himself in both at
once, so long as he is clear about the differences between them. But is it probable
or even possible that the same conception of the world should result from both
processes? These are the crucial questions. The only reason I can find for thinking
it probable that the same conception will result from both forms of objectification
is an unwitting acceptance of an Enlightenment viewpoint. Given that viewpoint,
there may seem to be a reason why the results of one form of objectification should
152  
be identical with those of the other. But, quite apart from the fact that we should
not in this context assume any viewpoint of that sort, an Enlightenment viewpoint
amounts here to a determination to separate out the elements of our view of the
world which are due to the peculiarities of our own perspective; and this prevents
one from seeing Hegelian objectification in any other light at all, for one has
decided in advance that the only acceptable result of an enquiry into the relation
between our view of the world and the world we view is one that embodies the
Enlightenment position. All this, however, assumes that it is at least possible that
one conception should answer to both processes. And, given the differences
mentioned above, I doubt whether this is so.
First, absolute objectification gives a sense to the notion of detachment which is
incompatible with that of Hegelian objectification. Second, it is admitted that
absolute objectification leaves subjective states incomprehensible; we know that
there are such states, but we cannot hope to grasp their content. But this, which is
the anticipated result of the success of absolute objectification, would amount to
the failure of Hegelian objectification. Third, the absolute conception of the world
(the Enlightenment view) marks the final point of absolute objectification. But
there is no reason even to suppose the possibility of such a final point for Hegelian
objectification.
It seems to me, then, that we cannot hope to admit the difference between the
two forms of objectification and argue that the same conception might result from
both of them. If Nagel wishes to continue to talk of the objective view, then, he has
only one option: to abandon one of them, and work entirely in terms of the other.
But to do this would be to abandon some of the oppositions central to his book,
which only arise if one conflates aspects of both sorts of objectification. In the
space that remains to me I comment on two examples of this, in the discussion of
knowledge and of freedom and autonomy, and end by suggesting that part of
Nagel’s conception of moral realism also results from mistakes about his own
notion of objectification.
Knowledge
Is objective knowledge possible? Nagel expresses the problem thus: ‘But if initial
appearances are not in themselves reliable guides to reality, why should the
products of detached reflection be different? Why aren’t they either equally
doubtful or else valid only as higher-order impressions?’ (p. 67). Two things
should be noticed about this way of getting into the problem of scepticism. The
first is that there is a suggestion that appearances are intrinsically unreliable or
dubious; that what we designate as appearance is something which naturally
deserves suspicion. But this epistemological loading, to the prospect of which
I have already referred, is no part of either sense of objectification. Even on
 ’  153
absolute objectification, the intention to discover the nature of the independent,

objectively understandable world by separating off those aspects of our subjective
conceptions which can be attributed to peculiarities of our perspective, need only
leave us supposing that appearances are not part of objective reality. It is still
possible that an appearance is a reliable guide to reality, in one sense, as colour is.
And it is also still true that reality is not just objective reality, so that the
unreliability of appearances, if any, might be simply our fault for mistaking their
relevance.
The second thing to say is that if the process of absolute objectification is
successful, the products of later stages will not be merely more ‘mere appearances’.
For there is no reason in advance to suppose that the attempt to abstract from the
peculiarities of our own perspective cannot succeed, and if it does succeed the
absolute conception we eventually reach will neither count as intrinsically dubious
in whatever sense appearances are intrinsically dubious, nor count merely as
higher-order impressions. It is true that at no stage can we be sure that we have
reached the most objective view. But that gives us no reason to doubt that the
process of objectification necessarily takes us nearer to the truth about the
objective world.
The reason for fearing that, no matter how far we travel along the path towards
objective truth, we have made no progress at all emerges on the next page, where
we read:
If the initial appearances cannot be relied upon because they depend on our
constitution in ways that we do not fully understand, this more complex idea
should be open to the same doubts, for whatever we use to understand certain
interactions between ourselves and the world is not itself the object of that
understanding. However often we may try to step outside of ourselves, something
will have to stay behind the lens, something in us will determine the resulting
picture, and this will give grounds for doubt that we are really getting any closer
to reality. (p. 68)
The place where Nagel has conflated Hegelian with absolute objectification is in
the words ‘something will have to stay behind the lens, something in us will
determine the resulting picture’. The first of these is true on the Hegelian con-
ception, the second refers more to the absolute conception. Crucially, one cannot
argue from the first to the second. The fact that something will have to stay behind
the lens does not show that any individual feature or features of that something
will determine the resulting picture. On the Hegelian conception, as I said earlier,
there is no general supposition that the process of objectification has a natural
terminus; and if there were a terminus of a Hegelian sort it would be one where the
whole of the resulting picture was determined by the lens or what lies behind it.
Either way, the availability of a natural terminus has nothing to say to sceptical
154  
doubts deriving from the thought that we can never reach a perspective which has
no distinguishing features.
It seems to me, then, that Nagel only comes to suppose that ‘the very same ideas
that make the pursuit of objectivity seem necessary make both objectivity and
knowledge seem, on reflection, unattainable’ because he confuses the proper aims
and likely products of Hegelian objectification with those of absolute objectifica-
tion. If he were to speak only in terms of one or the other, the danger of scepticism
would be much less endemic.
Autonomy
Autonomy is concerned with the status of the agent as a free chooser. According
to Nagel, ‘to be really free we would have to act from a standpoint completely
outside ourselves, choosing everything about ourselves, including all our prin-
ciples of choice—creating ourselves from nothing, so to speak’. And he adds:
This is self-contradictory: in order to do anything we must already be something.

However much we incorporate from the external view into the grounds of action
and choice, this same external view assures us that we remain parts of the world
and products, determined or not, of its history. Here as elsewhere the objective
standpoint creates an appetite which it shows to be insatiable. (p. 118)
The desire for autonomy is the desire to form one’s beliefs on the basis of
principles of choice which one can oneself judge to be correct, rather than on the
basis of influences that one cannot know, understand, assess or endorse. Why is
this desire incoherent? Because it leads to the attempt to assess everything about
ourselves, including all our principles of choice, at once. And this attempt is
incoherent because assessment is only possible if one retains some principles of
choice.
One can consider the desire for autonomy in either of two areas, though
these should not in the end be taken to be unconnected. The first is that of
cognitive autonomy, where we desire to understand and approve the principles
by which we form and assess beliefs; the second is that of the autonomy of an
agent, where we desire to understand and assess the principles by which we
form practical projects and so choose actions. The suggestion that the object-
ive view is one from which we can understand and assess all our principles at
once is as well motivated and as incoherent in the first area as in the second.
Nagel’s discussion is targeted more on the second; but I shall speak more of
the first in what follows, since I think this is the easiest way to express my
doubts about his conclusion that the desire for true objective autonomy is both
compelling and incoherent.
 ’  155
In what sense is the desire for autonomy related to the search for objectivity?
There seem to be two possibilities. The first is that the search for autonomy
necessarily involves an attempt to assess the relation between the world and our
view of it. For we want to discover the extent to which our principles of choice suit
the world within which we are choosing, and this whether we are choosing actions
or cognitive stances. We accept that our enquiry might leave us with the view that
our principles don’t suit that world at all; Hume’s doubts about induction are the
classic example, again. I am sure that we are all interested in this sort of attempt to
establish our own autonomy, and our interest is one of the reasons for engaging in
Hegelian objectification. But what about the two further supposed aspects of this
enquiry? They are, first, that we must (hopelessly) attempt to assess all our
principles at once and, second, that a principle is only acceptable if it is affirmable
by all rational beings, irrespective of their perspective.
This second demand derives from the generalizing aspect of objectification. If
objectification is the attempt to abstract from our own perspective (as in absolute
objectification), and if in objectification we aim to come to an assessment of our
own principles (this is Hegelian objectification), the conclusion is that the prin-
ciples that survive that assessment must be those which are acceptable from a
perspective other than our parochial one. But this shows that this demand derives
from a conflation of the two distinct sorts of objectification.
What lies behind the first demand? Why should we not suppose instead that
our method should be the one of examining each principle in turn and rejecting
those that fail? One traditional way of drawing this contrast, in the cognitive
sphere, is to talk about Neurath; another is to appeal with Descartes to two ways of
getting the rotten ones out of a barrel of apples. What is it about objectification
that insists on one way rather than the other? The answer lies in Nagel’s expres-
sion of why the preferred route is impossible. He says that ‘we cannot assess and
revise or confirm our entire system of thought and judgement from outside, for we
would have nothing to do it with’ (p. 118). But the attempt to revise and assess our
entire system from outside is an attempt fostered by the aims of absolute object-
ification. The desire to work from outside is a desire to have a view from nowhere,
a view with no special characteristics but which is nonetheless sufficiently rich to
enable us to assess our own perspective from it. It is true that this desire is
incoherent, since it could never be satisfied. But it is no part of the aims of
Hegelian objectification. Hegelian objectification is a process with no evident
end. There is no reason in advance to suppose that there should be an end to it,
as there is with absolute objectification. And, that being so, there is no reason
other than a conflation of the two sorts of objectification to insist that Hegelian
aims are incapable of fulfilment. Hegelian objectification aims at an increased
understanding, derived from an investigation into the relation between our
subjective point of view and the world so viewed. There is nothing here which
leads us to expect and aim for a final understanding which is not itself susceptible
156  
to the same questions as before and which therefore cannot be superseded. That
sort of desire and expectation seems to me to derive from thoughts about absolute
objectification.
In his discussions of knowledge and of autonomy, therefore, Nagel shows that
he needs both conceptions of objectification to get his problems going. But the
intractability of the problems, which so much impresses him, seems to me to
derive not from one and the same tension between subjective and objective points
of view, but more from one and the same tension between two distinct concep-
tions of objectification, once one fails to see the differences between them.
Generalization in ethics
I now turn to the second half of Nagel’s book, which is concerned with ethics.
Many of the views expressed here are familiar, though there is one important
change on the topic of agent-relative reasons. However, the discussion is
everywhere informed by what has preceded it. Nagel, as a moral realist, is
concerned to stress the sense in which our moral reasons claim objective
validity. But the objective self, that part of us which strives to make its
judgements from the most objective point of view, finds it hard to live with
our subjective self. He says that ‘the pursuit of objectivity requires the culti-
vation of a rather, austere universal objective self ’ (p. 63). What does it mean
to say that the objective self is universal, and to what degree of austerity must
such a self aspire?
If we cannot achieve the ideal of true autonomy, how far can we go in the way of
objectifying our motives, that is of achieving a motivation which can be under-
stood from a perspective which relies less on the specifics of our nature and
position in the world? One might think that a view from nowhere would perhaps
recognize the fact of subjective motivation, but would not itself be subject to
motivation at all. Nagel certainly thinks that many of our actual motives are like
this; they can be recognized by the objective will, but cannot expect the endorse-
ment of that will. The most they can hope for is objective tolerance. But other
motives can hope for more than this. Motives of prudence are motives which the
objective will can not only tolerate but also accept. One’s susceptibility to motives
of prudence is not something that depends on any particular facet of one’s point of
view. So the objective will is motivated, but the range of its motivation is severely
reduced. In this sense it is austere.
But there are not only motives of prudence. There are also other values. Some of
these values are subjective values, values comprehensible only from the point of
view of the person whose values they are. But others are objective values. Most
objective values are impersonal ones, values which make the same claim on each
objective will. But some objective values are agent-relative or personal; these create
 ’  157
reasons for action only for those persons to whom the reasons make essential
reference. For example, there is no objective reason why anyone should stand on
the top of K2. But I might have an objective reason for me to stand on the top of
K2. This is an objective reason because it can be understood (and affirmed? Yes,
p. 153; no, p. 170.) from a point of view which is not mine. But it is agent-relative
because it creates no reason for anyone other than me to do anything.
Actually I disagree with Nagel on this last point. Indeed, he is not consistent on
the point. Sometimes he says only that agent-relative reasons do not give rise to
comparable or corresponding neutral values; at other times the view is that ‘the
achievement of a typical personal project or ambition has no value except from the
perspective of its subject’—though he then adds ‘at least none in any way
comparable’ (p. 169). My own view would be that if you have a strong desire to
stand on K2 I have at least some reason to further your plans so long as this does
not require too much of my time or resources. Of course my reason is in no way
comparable to yours. But I do not see room for a sort of reason which can be
understood and affirmed from a perspective other than that of the person whose
reason it is, without that affirmation having some consequences for the actions of
the objective will that makes it.
Nagel’s reason for wanting to say that agent-relative reasons, though objective,
do not generate neutral reasons is that he can use this to defeat a preference-based
ethics. But I think that this purpose would be equally well served by the admission
that agent-relative reasons do not create comparable neutral reasons. All we need
for the purpose is to establish that agent-relative reasons are not copied wholesale
as neutral reasons.
What now is the sense in which the objective will is not only austere but
universal? Let us start by reminding ourselves of several distinct ways in which
a value might be said to be universal.
1. It can be understood and affirmed from any point of view.

2. It concerns not just how some one individual should behave but how all
individuals should behave.
3. It concerns not just how some one individual should behave here, but how
that individual should behave in all relevantly similar cases.
4. It is a value with respect to which all individuals have equal claims.
Of course, a value may be universal in all these senses at once. But it need not be;
the different senses are distinct, and none follows from any of the others. Now
Nagel, in establishing the nature of the objective will, is intending to rely upon
very restricted resources: namely, the concept of objectification itself. The ques-
tion I want to raise is how much of the claim that values accepted by the objective
will are universal is a simple consequence of objectivity. To focus attention, here
are some claims of Nagel’s:
158  
From the objective standpoint, the fundamental thing leading to the recognition
of agent-neutral reasons is a sense that no-one is more important than anyone
else. (p. 171)
Because [objective values] are accepted from an impersonal standpoint, they
apply not only to the point of view of the particular person I happen to be, but
generally. They tell me how I should live because they tell me how anyone should
live. (p. 135)
The search for generality is one of the main impulses in the construction of an
objective view—in normative as in theoretical matters. One takes the particular
case as an example, and forms hypotheses about what general truth it is an
example of. (p. 152)
This position is reached by a fairly minimal generalisation from one’s own case . . .
One translates one’s own reasons into a form that can be accepted by people with
different preferences, so that it can be used by anyone to account generally for his
own reasons and those of others. Nothing but the need for an account that can be
understood from no particular perspective would require the adoption of such
general principles. (p. 150)
The detached, objective view takes in everything and provides a standpoint of
choice from which all choosers can agree about what should happen. (p. 183)
What I want to suggest here is that Nagel has conflated the various different senses
in which values can be universal in order to reach the conclusion that the urge to
objectivity (an urge whose reasons vary between a desire for understanding and a
desire for the truth) itself drives us towards the formation of an ethical view of a
very specific sort. It will be one which consists (a) of a set of moral principles
which (b) treat all persons as of equal importance, and on which (c) all choosers
can agree. Now I am nearly sure that the search for such an ethical view is entirely
admirable, but I cannot see that the need for it derives purely from the urge to
objectification.
Objectification (of the absolute sort) is the search for an account of the world
which abstracts from one’s subjective view those elements which derive from its
peculiarities, leaving only elements comprehensible from a point of view which
has no peculiarities at all—the view from nowhere. This seems to validate the
presence of condition (c) above. But what of (a) and (b)? Nagel presents the first as
if it is obvious that objectification and generalization go hand in hand. But the
appeal here is to the example of physics, and in my view, as I have said before, the
objectifying tendency of physics is quite distinct from its generalizing tendency.
The fact that physics seeks to tell us about the independent nature of the world
and the fact that physics is not interested in the particular case seem to me more or
less unconnected. Even if this were not so, the sort of induction used to take us
from particular cases to general truths in physics is quite different from that which
 ’  159
we might hope to use to reach moral principles. This is because moral principles
are not made more probable by an accumulation of instances, and seems to me to
constitute a serious weakness in any attempt to argue by analogy from physics to
ethics. The view that theories essentially generalize in such a way that the only
proper task of a moral theory is the production of a set of moral principles seems
to me to derive really from the fact that Nagel is what one might call a generalist in
the theory of reasons. Every reason is for him a general reason, of such an extreme
sort that he seems to hold that specific features of actions cannot be reasons at all.
‘If I have a reason to take aspirin for a headache or to avoid hot stoves, it is not
because of anything specific about those pains but because they are examples of
pain, suffering or discomfort’ (p. 158). It is this position which takes him from
ordinary moral reasons to moral principles, not anything to do with the process of
objectification. I reject this form of generalism about reasons; but whether it is true
or false, the point is that Nagel’s realism about values and his generalism about
reasons are distinct positions, neither providing any support for the other. The
search for objective value does not itself validate the conception of a successful
moral theory as a set of moral principles.
So much for moral principles. The second element (b) in Nagel’s conception of
a moral view seems to me to be simply a mistake. A position which thought of
some persons as more important than others, or even one which held simply that
only two of the people in the world are important, might be held to express a
general value in any but the last of the four senses above. In particular, I see no
reason in advance to suppose that such a position might not be one on which all
objective wills could agree, unless we have an independent proof that the position
is incoherent. (Remember that the truth of a view is not a reason for supposing it
to have any special claim on the objective will.) So I conclude that most of Nagel’s
conception of an objective moral view derives from mistakes about the generality
or universality that is associated with the products of objectification.
There remain many elements of Nagel’s book which merit discussion. That
I have concentrated on the notion(s) of objectification is due to the fact that it is
here that the central message of The View from Nowhere lies. My criticisms of that
message do not diminish my sense that this is an important book from which
I have learnt much, though perhaps not in the way that Nagel intended. My main
conclusion is that the need to achieve the absolute conception is by no means as
pressing as is sometimes claimed.²
² I am grateful to David McNaughton and Christopher Hookway for discussion, and to David
Bakhurst for very helpful criticism of an earlier draft.
11
In Defence of Thick Concepts
As standardly conceived, thick concepts somehow hold together a property and an

attitude.¹ The attitude is either pro or con; the Aristotelian virtues, for instance, are
thick pro-concepts, the vices thick con-concepts, as it were. The property is a more
complex matter. An honest man may be conceived as one who respects the truth,
and who is to be valued for that. Here we have property and attitude—or, as it is
also sometimes put, description and evaluation: not mere evaluation, of course,
but evaluation in the light of the relevant description (the attitude is adopted to the
property-bearer for having the property). An honest man is not a truth-respecter
and good, but a truth-respecter and good for being so. The evaluation, therefore,
cannot be peeled off from the description so as to stand as independently
comprehensible. But neither can the description be peeled off from the evaluation,
if we are dealing with a genuinely thick concept. The leading feature of the thick
was supposed to be that the descriptive ‘side’ is not independently comprehen-
sible. A person from another culture who failed to see the evaluative point of a
thick concept would not be able to predict local use of it on the basis of descriptive
similarities alone. The descriptive qualities of the instances of the concept would
indeed bear resemblances to each other case by case, so that repeated uses of the
concept are visibly instances of ‘going on in the same way’, but that way is a way
invisible to the outsider. In this sense, the concept would be ‘naturally shapeless’;
the concept does have a shape, of course, but its shape is partly an evaluative one.
We can sum this up by saying that, with the thick, evaluation partly determines
description.²
Non-cognitivist analyses of thick concepts, however, such as those produced by
C. L. Stevenson (e.g., 1944: 207), took them as having a descriptive element which
is capable of standing alone, to which is added an evaluative one which is not. It
seemed to follow from this that even if the point of being interested in the
¹ It is standard to attribute the thick/thin terminology to Bernard Williams (1985). Tim Williamson
has pointed out to me, however, that the anthropologist Clifford Geertz popularized a notion of ‘thick
description’; see, e.g., his ‘Thick Description: Toward an Interpretative Theory of Culture’ (1973).
Geertz attributes the idea to Gilbert Ryle.
² In this opening paragraph I write as if all thick concepts function in exactly the same way. But there
may be many different types; for a good range, and worries derived from that range, see Scheffler
(1987).
     161
collection of features that constitute the descriptive element may be obscure to

outsiders, it would still be possible in principle for outsiders to come to mimic the
practice of insiders with respect to the thick concept. Outsiders, that is to say,
would be able accurately to predict which objects the insiders would see as falling
under the thick concept, even if they could not really see why objects of this sort
were found so interesting. From the cognitivist perspective, this was an implaus-
ible consequence. But non-cognitivists could not escape it, because it was essential
to their whole approach to conceive of description as independent of evaluation.
Here is a summary of what has been said so far, in slightly different terms.
Thick concepts involve ‘property-attitude pairings’.³ The attitude is fairly simple,
and somehow lexically coded into the concept, i.e. you would expect to find it
signalled in a dictionary. The property is a much more complex matter, but the
only thing we have been told about it is that it is not independently discernible.
There need be nothing flat about the description, then; it is full of evaluative life.
(Evaluation partly determines description.) This makes it impossible for non-
cognitivism, with its dead conception of description, to capture the nature of these
concepts. It is reduced, futilely, to denying their existence.
II
In a recent article, Simon Blackburn has argued that this whole debate has been
misconceived.⁴ If there were thick concepts as conceived above, there would have
to be a lexically signalled, and therefore uncancellable, attitude present in each
one. But Blackburn suggests that when one looks for such things, very few present
themselves, and those largely derogatory, e.g. ‘nigger’, ‘ponce’, ‘tart’—not at all the
sorts of concept that the proponents of the thick took themselves to be talking
about. Concepts like this can perfectly well be handled within the constraints of
traditional non-cognitivism, because they do not show the sort of internal com-
plexity that would make trouble. But in the supposed examples of the thick that we
find offered in the writings of cognitivists, we are not going to find any hint of a
single lexically signalled attitude. Even the Aristotelian virtues and vices are not
always good and bad, respectively; courage may be for the worse, in familiar ways.
Blackburn takes from Allan Gibbard, his co-symposiast, the nice example of the
lewd. One would have thought that this is a thick concept, but it shows no trace of
a lexically signalled single attitude. A lewd action is not one of which we auto-
matically disapprove; some lewd actions are called for in their circumstances, and
others are just plain fun. It is not even just that we approve of some and
³ See Wiggins (1987).

⁴ Blackburn (1992). This paper was a reply to Allan Gibbard (1992), in a symposium on ‘Morality
and Thick Concepts’.
162  
disapprove of others. There is just a welter of competing attitudes, more than one
of which will be in place on any one occasion. An excellent example of this,
suggested to me by David Bakhurst, is that of the lewd in Restoration comedy.⁵
What sort of attitude, or what attitudes are called for, by this sort of lewdness on
the stage? The answer is, I suppose, a combination of pleasurable shock, of mild
titillation, and of a sort of self-directed embarrassment (and no doubt there is
more to be said than just this; I am not very good at this sort of thing).
There were two ways, then, in which the original examples of the thick fail to fit
the model of a single-attitude concept. First, they are ‘variable-attitude’ concepts,
since they call for different attitudes on different occasions. Second, they are
‘multi-attitude’ concepts. They can ground, or call for, more than one attitude at
the same time.
A third feature of the standard examples of thick concepts is that they are much
more fluid than the original picture suggested; indeed, their fluidity is essentially at
odds with any idea that one, or even many, attitudes, are signalled lexically. Lexical
signalling, according to Blackburn, involves a sort of concretization of something
which is in a process of perpetual change and development, causing us to doubt
the possibility that there can be a change in the range of attitudes associated with a
term without a change of meaning.
Blackburn sums all this up by saying that the standard examples of thick
concepts are in fact ‘complex, fluid and mobile’ (1992: 296)—quite different
from the way that they were originally conceived. The sort of complexity that is
at issue here involves change from case to case. The sorts of terms we are talking
about come to us with a rich accretion of attitudes, but even if each has its own
limited range, different cases in which the same term is applicable will call for
quite different mixes of attitudes from that range. Blackburn argues therefore that
there are really no thick concepts at all. To be a thick concept one would have to be
single-attitude, with lexical signalling; but all the interesting cases are quite
unlike this.
This now gives Blackburn an opening in the continuing battle between non-
cognitivism (expressivism, in his case) and cognitivism. For he can now say that
his expressivism is far better equipped to handle terms of this fluid sort, where
though there is a crust of attitude it is not part of the concept, than is any
cognitivist alternative. In trying to squash the flexibilities of these terms into the
straitjacket of their notion of a thick concept, the cognitivists really choke the life
out of them.
The final issue concerns the relation between evaluation and description.
Cognitivists claimed, of their examples, that the interconnection between evalu-
ation and description was beyond non-cognitivist capturing. Non-cognitivists had
⁵ See also Blackburn’s remarks (1992: 296) about the Dionysiac.

     163
to see the description as essentially independent of the evaluation, in a way that

made it always accessible to the outsider. Cognitivists, by contrast, could under-
stand the description as partly determined by the evaluation. Blackburn now
claims, however, that though it is true that in these cases evaluation does partly
determine description, this is something that expressivists can happily counten-
ance. There is nothing to choose between expressivists and cognitivists on the
point. All can agree that these concepts are naturally shapeless, since evaluation
part-determines the extension of the description. Gibbard suggests that there is
‘not enough descriptive meaning’ to stand alone; Blackburn does not endorse this,
but argues, in a case that I will discuss in more detail below, where attitude is
carried by contemptuous tone, that ‘it is quite compatible with the attitude being
carried by the tone, that it then play a role in determining the extension and in
ruling out of the extension things which, for quite different reasons, escape the
attitude’ (1992: 291).
When Blackburn has finished, then, the tables have been turned. Instead of the
thick constituting a problem for non-cognitivism, it turns out to expose weak-
nesses in the position of those who were trying to use it as a weapon—a genuinely
two-edged sword.
III
What are we to make of all this? I want to start by suggesting that cognitivists
should just accept with gratitude everything that Blackburn has pointed out about
the behaviour of the terms that have been used as standard examples of the thick,
and put aside the mistaken model of the single-attitude concept. (Let us take the
case of the lewd as our chosen example.)
Do the cognitivists in fact know all this already? The best cognitivist work in
this area is by Wiggins. He has tried to show that a term like ‘funny’ may be
essentially contestable in a way that is not at all at odds with recognition of the
‘mutual inextricability . . . of the anthropocentric properties of an object and the
attitudes towards it that are made for these properties’ (1987: 212). These thoughts
mean that it may be a matter for debate whether an object falls under the thick
concept or not, and that nothing in the concept need determine the answer to this
question. Wiggins says, in the passage of which the remark quoted above is
intended as a partial defence, that ‘without serious detriment to the univocity of
the predicate, it can now become essentially contestable what a thing has to be like
for there to be any reasons to accord that particular appellation to it and corres-
pondingly contestable what the extension is of the predicate’ (1987: 198). But this
concerns only our decision whether the action was lewd or not. Wiggins in fact
takes it for granted, at least in the cases that he is discussing (the funny et al.), that
there is a single attitude associated with the relevant term. He writes:
164  
If a property and an attitude are made for one another, it will be strange for one
to use the term for the property if he is in no way party to the attitude and there is
simply no chance of his finding that the item in question has the property. But if
he is no stranger to the attitude and the attitude is favourable, it will be the most
natural thing in the world if he regards it as a matter of keen argument what it
takes for a thing to count as having the property that the attitude is paired with.
(1987: 199)
Three comments on this. First, Wiggins seems to take it for granted that there is
a property and an attitude in what he calls a ‘property-attitude pairing’; the
former may not be contentious but the latter certainly is. Second, he only argues
for the essential contestability of the property, not for that of the attitude. This
may be because he thinks that the contestability of the attitude is identical
with that of the property. But, even though this might fall out from his single-
attitude picture, with our new picture it is surely worth distinguishing two
essential contestabilities, that of the property and that of the attitudes. Third,
the notion of essential contestability is just a different notion from that of the
variable and fluid. These last were the ones that Blackburn was really pressing,
and they are located in a different place. We can see this because Wiggins makes
his points about essential contestability in the context of talk of a property-
attitude pairing, which he expresses in terms strongly reminiscent of the old
single-attitude model.
IV
Blackburn’s strategy could be expressed as a dilemma: either thick concepts are as

originally conceived, in which case there are hardly any instances, or they have the
sort of fluidity and variability that we have seen, in which case we are not dealing
with thick concepts at all. The second horn of this dilemma is grounded in two
independent claims, either of which would be sufficient for Blackburn’s point. The
first concerns fluidity, i.e. change over time. Cognitivists have to rebut the charge
that their account of the thick introduces an uncalled-for concretization, turning
too many things into changes of meaning, and increasing the danger of two
disputants being taken to be ‘talking past each other’, i.e. to be failing to talk on
the same topic. The second concerns variability; variability shows that no attitude
or mix of attitudes is lexically signalled as required by the use of the term, and
according to Blackburn this shows that the associated attitudes are not themselves
part of the concept.
I take the first point first. Blackburn argues that if there were thick concepts,
disagreement that involves them would need to be seen as a difference in concepts
rather than as a difference in judgement. He writes:
     165
[The appeal to thick concepts] implies too ready a tendency to diagnose the
discussions and re-evaluations that are the essence of ethical activity as exercises
in talking past each other. For if you don’t respond to lewdness as I do, then on
the thickie picture your amalgamated concept of lewdness is not mine, and we
are left in incommunicable solitude. (1992: 299)
The question is why this result is supposed to follow. The clue lies in the word
‘amalgamated’. Blackburn, reasonably supposing that defenders of thick concepts
(‘thickies’, as above) must offer some account of how descriptive and evaluative
‘elements’ are combined in a thick concept, offers on their behalf the only account
he thinks available to them. He claims that:
the idea must be that the two [elements of attitude and description] form a
compound or amalgam, rather than a mixture; the attitude and the description
infuse each other, so that at the end, in the repertoire of the mature speaker, the
two elements are no longer distinguishable. . . . The thickie must claim that a
person loses the concept if a particular amalgam dissolves back into its elements.
A form of life has gone, a way of seeing the world has vanished. (1992: 298–9)
Why then is it that if you don’t respond to lewdness as I do, your concept of
lewdness is not mine? If we are to get this result, we will not be talking about two
people who come to different conclusions about the practical relevance of a
particular lewd act, or even disagree more generally about how to respond to
lewdness case by case, but rather about two people who think of each other as
differing systematically, not on what is lewd and what is not, but on the appro-
priate response to it case by case. If the difference is not systematic, it can be
presumably catered for in the sort of way discussed in the previous section. We
have, after all, moved on from the single-attitude picture of thick concepts that
Blackburn rightly challenged. These multi-attitude concepts are doubly essentially
contestable, and understood as such. But if we systematically agree on the descrip-
tion (‘yes, this is lewd’) and not on the evaluation, the concept has been dissolved
for one of us, and hence our disagreement is to be explained as due to a difference
in concept rather than in judgement.
Should defenders of thick concepts accept the picture that Blackburn is offering
them and try to mount a reply within the constraints it creates? One obviously
dubious aspect of that picture is hinted at by Blackburn’s continual talk of
‘elements’. Though language of this sort seems to be almost unavoidable in the
present context, it remains unfortunate. At least, it remains unfortunate if we do
not want to beg the question against one leading cognitivist idea about the thick,
that it does not have elements, there not being enough description to make a
descriptive element at all. A good example of this is Blackburn’s own account of
the descriptive side of lewdness: ‘something to do with sexual display, something
166  
to do with mockery of normal proprieties’ (1992: 297). If this sort of thing is all
that is available, there is no possibility that an amalgam of which it is ‘part’ could
be dissolved back into its parts, since nothing like this is capable of standing alone.
The correct picture, I think, as I will try to explain later, is not that there are two
‘really’ distinct elements which by a pseudo-chemical reaction somehow become
indistinguishable from each other. There are no elements at all, in any normal
sense. There is indeed a property and an attitude (pretending for the sake of
simplicity that we are dealing with a single-attitude case), but these things are not
elements of a concept. They are incapable of being so, because the property is best
characterized as being that of meriting the attitude, and the attitude is best
characterized as the appropriate one given the presence of the property.⁶ So
there are not two things to amalgamate—there is not even one which, with the
addition of something incapable of independent existence, is somehow trans-
formed into something other than itself (description turning into something
indistinguishable from evaluation). The so-called amalgam cannot be disassem-
bled or dissolved, since there is not enough to dissolve it into.
The differences between this picture and the one offered by Blackburn may
seem, however, only to make things worse rather than better. For surely, if the
concept is indissoluble in this sort of way, it will be true that there can be no debate
between those who reject wholesale the very suggestion that whether something is
lewd or not makes a difference (by its links to attitudes) to how one should behave,
and those who use the concept of the lewd without any sense of strain. With that
suggestion (is it an element?) removed, there is on the present account nothing in
common between the full-blooded user of the concept and the one who hopes to
be able to describe things as lewd without buying in to any of the standard
(socially authorized) practical consequences. So how could they begin to debate
the issue between them? Are they not condemned to ‘talking past each other’?
My attitude at this point is that the supposed difficulty depends on an exag-
geration.⁷ The difficulty is supposed to be that the protester wants to say,
‘Lewdness just isn’t important at all, in the sort of way it is generally taken to
be’, but that someone who thinks this cannot even have the concept of lewdness,
though he needs it in order to make his point. The danger that is held to follow
from this is that the thing which the protester claims to be unimportant, whatever
it is, cannot be the thing that everyone else thinks important. So they are talking
past each other; neither is denying what the other is saying. But this supposed
danger is spurious. Someone who rejects the common understanding of (i.e., the
common use of the concept of) the lewd can still be perfectly well aware of how
⁶ This is how Wiggins describes the situation: ‘There will be then no saying, very often, what
properties these names stand for independently of the reactions they provoke. . . . But . . . there will often
be no saying exactly what reaction a thing with the associated property will provoke without direct or
indirect allusion to the property itself ’ (1987: 195).
⁷ I think that at this point I am in agreement with Wiggins (1987: 212), but it is hard to be sure.
     167
that concept functions. In saying that we should not think of the lewd like that, he
is not saying that he wishes per impossibile to retain part of the concept of the
lewd. For the way in which acts are collected together by that concept is at least
partly dependent on the social practice of responding to lewd acts in a certain (no
doubt complex) way, and it is to that social practice that he is objecting. So such a
person is really abandoning the concept of the lewd altogether, in a way that
creates no danger of talking past one another. For everyone knows perfectly well
what is going on.
The only person who is prevented from joining in the discussion is the person
who has never been able to see the point of the concept of lewdness in the first
place. But such a person is no threat to the thick. To reject a concept, one has to do
more than fail to see its point; one has to know what its point is and reject it for
that reason. But this last is perfectly well available to those who wish to reject the
mix of social attitudes on which rests the comprehensibility of the thick concept.
The person who has never been able to see the point is prevented, in a way in
which the person who wishes to withdraw from the practice is not, from con-
structing a sort of successor concept—one which lacks the ‘life’ of the original
thick concept but which we can reasonably see as a development from it. The
extension of this successor concept will no doubt gradually begin to vary from that
of the original one, but this is only to be expected, since there is nothing to anchor
it or to keep it in step with the concept from which it sprang. In particular, there is
no requirement that the successor concept should retain an ‘element’ from the
original. There are forms of inheritance available that are more complex than that.
Indeed, there may be no specific requirements which one concept must satisfy in
order to be the successor of another.
It is easier to make this point, and easier in general to see how there can be a
range of differences over concepts, without any of them leading to a total
breakdown in communication, if we remember how far we have moved from
the old single-attitude picture of the thick. The multi-attitude approach opens up
the possibility of a wide range of complex differences over the use of a single
concept, so that the distinction between those who see the point that stay in line
and those who don’t is clearly too simple. Among those who see the point, there
will be too many possible differences compatible with common possession of a
single concept for the sort of danger that Blackburn is talking about to be a real
threat.
Now for the second idea, that the variability of the relevant expressive response
shows that no attitude can be lexically signalled, and therefore that there are no
genuine thick concepts. This is nothing to do with essential contestability; in
168  
however many forms or places that rears its head, it is not a feature whose
presence demands to be lexically signalled. I take Wiggins to have shown that
contestability is not at odds with the univocity of the relevant predicate, and hence
with the idea that we are dealing with a genuine concept. The variability is the
crucial point.
The variability of the attitudes associated with a thick concept (or term) shows
that no attitude or mix of attitudes can be lexically signalled. If it could, that mix
would be uncancellably required of anyone who uses a thick concept. One could
never say, of some thick concept C and an attitude A that is in the signalled mix,
‘This object is C but I do not take attitude A to it’. But we find no signs of anything
like this sort of uncancellability. Because of variability, each attitude within the
relevant mix will be occasionally inappropriate.
This argument would clearly have been effective against someone committed to
the rejected single-attitude model; but we have to evaluate it in the context of the
new, flexible, conception of a thick concept. We now see competence with a thick
concept as requiring a general understanding of the range of attitudes associated
with the concept, as well as the ability to agree with others on which possible mixes
from that range are plausibly called for in the present case, and the ability to make
and defend one’s own choice. (There will be more than one plausible mix, because
of essential contestability.) Now why should we suppose that, in order to count as
conceptual competence, this complex competence must be capable of being
captured in a lexicon? I can think of no reason why. What is capable of appearing
in the lexicon will admittedly fall far short of expressing intelligibly and accessibly
what it is that competent users of the concept know, on the present conception,
without that fact constituting a criticism either of the lexicon or of our conception
of thick concepts. It would be bizarre to insist on uncancellability as a test of
conceptual content, when the picture we are offering contains an explicit explan-
ation of why there is no such thing available.
VI
So far, I have argued that Blackburn’s more flexible picture of the thick does not
itself make difficulties for cognitivism. I now turn to the way in which he tries to
steal the cognitivists’ favourite clothes, by holding his expressivism compatible
with the view that evaluation can partly determine the extension of the
description.
We find this claim made in discussion of the use of intonation to convey much
that at present is (or at least may be) conveyed lexically. The example is ‘fat#’,
which is meant to represent ‘fat’ said with a sneer, by a fattist. Consider Pavarotti.
‘Pavarotti is unquestionably fat, but many fattists would recoil from calling him
fat#. They want to overlook the fact that he is fat to an extent that would normally
     169
repel them, since he is so transcendentally uncontemptible in other ways’

(Blackburn 1992: 290). Here we have a concocted case where there is no sugges-
tion of a subtle connection of any sort between description and evaluation, and
where what is going on can be perfectly well characterized by an expressivist. But
even here, Blackburn claims, attitude can contribute to determining extension. So
expressivists are in a good general position to allow the cognitivists’ main point
without damage.
But this seems to me quite a different thing from the one that cognitivists
have made so much of. There is nobody in this debate who ever denied that
the extension of a thick term, considered as a whole, is determined partly by the
evaluative side. With the old single-attitude concepts, one is only able to apply
a term of approval if one approves. The crucial question was supposed not
to be that, but whether non-cognitivists must think of the extension of the
descriptive side as unaffectable by what happens on the evaluative side. In the
present case, then, the question should be not whether fattists will recoil from
calling Pavarotti fat#, but whether they can avoid calling him fat. They may,
as Blackburn says, wish to overlook the fact that he is fat, but they are in no
position to deny it.
This is, of course, because there is a fairly clear descriptive content to the term
‘fat#’, whose operation cannot rationally be affected by attitude, as the expressi-
vists understand it. Attitudes are adopted to things and people, in the light of
certain properties that they have got. It is no good saying that whether they have
those properties is partly determined by whether we adopt this rather than that
attitude. This must, within the expressivist picture, be regarded as irrational. And
the situation would be no better for a cognitivist who thought that the thick
concept was some sort of mixture of descriptive and evaluative meaning. That is
why it seems so central to the cognitivist story, properly told, that it deny the
presence of descriptive meaning, so understood, altogether.
So far, then, it seems to me that Blackburn has failed to establish an expressi-
vist’s right to what is close to the cognitivist’s best point. And there is a principled
reason why one should expect him to fail—a reason to do with the fact that the
nature of the contrast between attitude and description is not something on which
expressivists and cognitivists are agreed. Now it is true, of course, that the term
‘fat#’ is a rather special type of example. But the general point surely remains
sound. Everyone, even the traditional non-cognitivist talking in terms of single-
attitude concepts, can agree that evaluation partly determines the extension of the
whole concept. One needs resources of which expressivists intentionally deprive
themselves, however, to make sense of the extension of the ‘description’ being
partly determined by the evaluation. Cognitivists earn the right to this result by
holding that the property is best characterized as being such as to merit the
attitude, and the attitudes best characterized as the ones appropriate to the
presence of the property. Expressivists have no such resource.
170  
VII
Now we must turn to the vexed question of rationality in conceptual practice.

Does the existence of thick concepts show that the expressivist is committed to
something disputable? I suggested in my Moral Reasons (1993: ch. 5.5) that their
account of thick concepts commits non-cognitivists to a suspect account of
practical rationality. For where the extension of the ‘description’ is insensitive to
evaluative considerations, what can constitute rationality in the use of a thick
concept? Rationality is consistency, understood in a special way. How can it be
other than this, that where one has previously valued an object highly in respect of
certain features, one must value equally highly a new one with the same features?
This is all it can be to ‘go on in the same way’. Rational constraints must be driven
by similarity in respect of the (independent) descriptive element. To be rational in
this respect is to evaluate similarly (pro tanto, of course) those things that are
descriptively similar in the relevant respect. This concept of rationality depends
crucially upon the idea that a thick concept has a natural shape, and this, of course,
was the point that the cognitivists fastened on.
It would be wrong to reject this picture on the grounds that everyone allows
that evaluative considerations partly determine the extension of a thick concept.
This does not mean that everyone can claim that thick concepts have no natural
shape. The sense in which old-style non-cognitivists can admit that evaluative
considerations partly determine the extension of a thick concept is merely that if
you do not accept the relevant evaluation, you cannot assent to the use of a thick
concept, even for the first time. In another sense, however, the satisfaction of the
relevant description is enough to require assent to the use of the concept,
including the relevant evaluation, once one has assented in a previous and similar
case. The move from one case to another is driven by the descriptive motor, and
old evaluations require new ones wherever the descriptive side repeats, on pain of
inconsistency.
That, then, is the status quo. Before entering the debate, we should distinguish
two questions which are in danger of being confused, and which are run together
in the argument I gave above. The first is whether the expressivist is committed to
a generalist account of practical rationality, i.e. one that accepts that if the very
same features that were our reasons for an evaluation in one case recur in a
second, we are committed to the same evaluation as before. The second is whether
expressivists have to suppose that thick concepts, if there are such, have a natural
shape.
The question that I want to raise about these two questions is whether our
answer to them is affected by the picture of thick concepts that has been devel-
oping. I take the second question first. It seems to me that we can now see three
distinct ways in which the expressivist can avoid commitment to a natural shape.
     171
First, the expressivist can appeal to the variability of the associated attitudes.
A natural shape for a thick concept would be a set of natural properties conceived
of as individually necessary and jointly sufficient for an object to come under the
concept. But if there is this common natural base for every instance, how could
any variability of attitude be called for? Variability is at odds with natural shape.
Second, there is essential contestability. There is strong pressure to admit that
the natural and the essentially contestable are incompatible. This pressure derives
from the idea that essential contestability stems from some intrinsic relation to
human concerns and purposes, and the natural is conceived as only extrinsically
related to those concerns and purposes. If we yield to this pressure, we have to
allow that if it is essentially contestable whether this new object comes under a
particular thick concept, the question cannot be resolved by appeal to natural
shapes.
Given this, what is to prevent expressivists from holding with cognitivists that
whether a new object is properly to be subsumed under a thick concept will be an
essentially contestable matter? It might seem that there is an answer to this
question, for I argued in the previous section that expressivists are in no position
to allow that evaluation can partly determine the extension of the descriptive side
of the concept. But this is just a different point. To say that a concept which is
partly descriptive is essentially contestable is not itself to say that evaluation can
partly determine the extension of the description case by case. The notion of
essential contestability is concerned with a far wider sensitivity to human aims
and purposes than one restricted to the question whether one approves or
disapproves of the relevant object for satisfying the relevant description.
So expressivists can also appeal to essential contestability as a reason for
denying the possibility of a natural shape for a thick concept. Natural shapes
cannot be essentially contestable; perhaps, then, an expressivist should talk rather
of the descriptive shape of a thick concept. So the argument I gave at the beginning
of this section fails; it illegitimately assumes that descriptive shapes are
natural ones.
What is more, there is a third argument available to expressivists. The inter-
esting feature of thick concepts, if they exist at all, is that they are capable of
grouping together objects that do not show enough natural similarity to be rightly
grouped together on natural grounds alone. As I would put it, the resultance base
of these concepts varies dramatically from case to case. It follows from this that
the family similarities between different instances need not be discernible to
the outsider. So even if in each case we adopt our attitude as a response to the
property, and even if the present instance of the property is taken as a certain
grouping of natural properties, it still does not follow that consistency in applica-
tion from case to case can be driven by natural similarities alone. There may be a
natural base in each case, without those bases sharing a common natural shape.
172  
And the important thing here is that, if this is true, there is no reason why the
expressivist cannot say so.
This gives us three reasons why the expressivist need not maintain that
thick concepts have a natural shape; they only have a descriptive one. But this
does not tell us how the expressivist can avoid the generalist claim that where
the same description recurs we must adopt the same attitude, on pain of
inconsistency. There will be more to practical rationality than this, no doubt,
because of the variability of the resultance base. But won’t this still have to
be among the rational constraints on the use of a thick concept? Essential
contestability is irrelevant here. It may be contestable whether the description
recurs. The constraint only tells us that if it recurs, consistency requires
the same attitude in the new case. And variability in the resultance base is
similarly irrelevant.
The idea of a descriptive shape that recurs from case to case is one that leads us
to distinguish, in a given case, between those properties that are elements in that
shape here and those that are not. Now our question at the moment is whether an
expressivist can be a particularist rather than a generalist. A particularist is likely
to say that changes in respect of a property that is not an element in the descriptive
shape in one case may justify withholding use of the concept in another, even
though there has been no change in respect of those that were such elements in the
first. Can the expressivist allow this? Is the notion of descriptive shape flexible
enough? Can a group of properties that constitute the entirety of the descriptive
shape of a concept in one case be less than the entirety in another? If so, why
cannot an expressivist say so? The reason would have to stem from the difference
between expressivist and cognitivist accounts of the relation between description
and evaluation in a thick concept.
Both sides have allowed the idea of a descriptive shape; but they surely mean
very different things by this notion. Expressivists are distinguished by the sharp-
ness of the distinction they draw between description and evaluation. (Compare
the cognitivist idea that there can be a property of being ‘such as to deserve these
attitudes’.) For cognitivists, though the shape is descriptive, it is also evaluative.
This is something that expressivists cannot say. At the end of the day they must
think of evaluation as a response to independent description, the two being worlds
apart (two directions of fit). Are they then committed, in a way that the cognitivist
is not, to the idea that where the same description recurs, independent of other
changes, the attitudes merited by that description in the one case must be merited
in the other? The fact that cognitivists can use, that differences elsewhere can
affect one’s evaluation in respect of these features, and hence alter the relevant
description, is not available to the expressivist. And, though I am not sure about
this, it seems to me to mean that the way in which attitude is focused on essentially
independent description in the expressivist picture does eventually commit
expressivism to generalism. For them, where the same description recurs,
     173
regardless of changes elsewhere, the attitude they adopt in the light of that
description must be the same.
We should note that the first of the three reasons why expressivists can avoid
the notion of a natural shape is a generalist reason. The idea was that if there was a
natural shape that recurred from case to case, differences elsewhere should not be
able to justify differences in the attitudes we take to the various instances case
by case.
If so, we have the result that, unlike cognitivists, expressivists have to be
generalists. Combining the two conclusions of this section, then, expressivists
can think of thick concepts as having a descriptive rather than a natural shape, but
they have to be generalists with respect to that shape.
VIII
I now turn to the difficult question of exactly how cognitivist defenders of the
thick should conceive of the relation between description and evaluation in a thick
concept. Although something has been said on this topic in what has gone before,
more is required. We know this because there is more than one possible model of
the matter still available to us. In this section, I consider the relative merits of
different models.
The rejected non-cognitivist model understands the thick as a combination of
description and thin evaluation, and we have already heard plenty of objections to
this as a universally applicable model. The first alternative suggestion that I want
to consider comes from a non-cognitivist, Alan Gibbard, who offers the thought
that a thick concept does not have enough descriptive meaning for that meaning
to be the whole meaning of a purely descriptive term. I want to reject this, perhaps
not surprisingly given its source, because it seems to me incomprehensible. (Timeo
Danaos et dona ferentes, again.) I just don’t understand the notion of ‘not enough
descriptive meaning’. Any amount of descriptive meaning is enough descriptive
meaning to be the whole meaning of some term, even if it does not make for a very
interesting or useful one. If a term has descriptive meaning, its descriptive ‘aspect’
is capable of standing alone as a neutrally descriptive concept.
A second suggestion would try to link the description to the evaluation in the
same sort of way as the evaluation is linked to the description. Remember that the
word ‘therefore’ was used to prevent the charge that the non-cognitive analysis
was unsuitably atomistic. We could introduce a similar link going in the other
direction. We could say that an object which comes under the thick concept has
features 1, 2, 3, . . . in such a way as to merit approval. Here we admit that the
concept is genuinely thick, since one can learn from its use something about the
nature of the action held to fall under it. (It has features 1, 2, 3, . . . in some form.)
And it presents a kind of interconnection between description and evaluation.
174  
This second proposal is still not acceptable. First, it admits, so far as I can see,
that there is something pretty substantial present in the way of a description, and
it is hard to see why that should be unable to stand as an independent descriptive
meaning. That the object has features 1, 2, 3, . . . is surely just a not very specific
description, and a not very specific description is still a description. Second, the
whole picture is a single-attitude picture, and we have decided that this need not
be the rule. Third, there is a question how to combine this link with the link that
goes in the other direction. To have both links, we would need to combine two
thoughts so as to say that an object which comes under the thick concept has
features 1, 2, 3, . . . in such a way as to merit approval for having those features.
This is difficult because it seems that it is not for having those features, exactly, that
the approval is bestowed. On the present story, it seems that the approval is
bestowed for having those features in the right way. Let us take these points in
reverse order.
There is something right about the last idea, that the approval is bestowed for
having those features in the right way, for it has it that the ‘description’ and the
‘evaluation’ are truly intertwined. And we should recognize that the move from
approval ‘for having those features’ to approval ‘for having them in the right way’
is a significant move away from the picture that motivated the original non-
cognitive introduction of a link between description and evaluation. For that
which is approved, or, rather, that for which the object is approved, is no longer
a purely descriptive feature. The ‘evaluation’ partially determines the ‘description’,
since the only way to specify the nature of the object of which we approve was to
say that it has these or those features in the (or a) right way.
This reveals three constraints on a satisfactory account:
1. It will have less description, somehow.

2. It will make room for multi-attitude thick concepts as well as potential
single-attitude ones.
3. It will capture the idea that that for which the object is approved is itself
partly evaluatively determined.
And we are moving towards meeting the third of these, anyway. We should be
careful, however, not to misread the phrase ‘in the/a right way’. It does not mean
‘in the/a (morally) correct way’ or ‘in the/a way that we approve’; for if it did mean
this, it would indeed enshrine the single-attitude approach, breaching the second
constraint. A ‘right way’ must be merely the sort of way that merits a certain sort
of complex response—the sort of response that can only be captured by saying
that it is a response merited by the lewdness. When we call something lewd, we
take it to merit the term ‘lewd’ because it has features 1, 2, 3, . . . in such a way that
this sort of response is right for it; whatever attitude we adopt to it on that count
will be our response to that same fact, namely that it has features 1, 2, 3, . . . in the
     175
way that by making it lewd, grounds the complex of attitudes that the lewdness
here calls for.
Here we see the interconnectedness of ‘evaluation’ and ‘description’. For,
understood in this way, the similarity between different lewd actions will not
be entirely natural. This is why there is no chance that outsiders will be able,
even by lucky accident, to light on the purely descriptive element of the thick
concept. They will not be able to do this because there isn’t any such thing. That
which was set up to play the role of a descriptive meaning is partly evaluatively
determined.
This way of capturing the interconnectedness of evaluation and description in
a thick concept shows no signs of being restricted to a single-attitude approach.
If it should turn out that the concept of lewdness is a multi-attitude concept,
this would do nothing to disturb the structure of the account we are
constructing.
We have now reworked the previous proposal in such a way as to meet
the second and third constraints, but what about the first? We still face the
difficulty that there is a distinct descriptive meaning contained in the idea
of features 1, 2, 3, . . . . But how can we have description without descriptive
meaning? This seems an almost insoluble puzzle. One possibility which appeals
to me is to take a leaf from Aristotle’s book, in a way unwittingly exemplified by
Blackburn. When Aristotle discusses the different virtues, he characterizes them
by their domain. Courage, for example, is a mean concerning feelings of fear
and confidence.⁸ Liberality is a mean ‘with regard to giving and taking money’.
Megalopsychia is a mean ‘with regard to honour and dishonour’. Sophrosune is a
mean concerning pleasure. Gentleness concerns rage, and so on. In all these
remarks, Aristotle is pinpointing the different virtues of character by specifying
the domain of their application—the range within which they operate. In this way
he locates each virtue-concept on the conceptual map. He can say more; each is a
mean with respect to that domain. And there are further remarks to be made of a
similar nature. But they do not amount to anything like a descriptive meaning of the
word ‘courage’, or a descriptive content for the concept of courage. Such a thing is
not to be available. Of course, when we praise someone’s courage, their action will
have features that make that description appropriate, and some of those features
will surely be properly called ‘neutrally descriptive’. But those features are no part of
the concept. They are restricted in relevance to the particular case.
What Aristotle says about the virtues can be borrowed for thick concepts of
other types. Lewdness is not a virtue nor a vice, but the concept has a range or
domain of operation in a way that tells us the sort of thing we are dealing with
⁸ Nicomachean Ethics, 2.7, 1107a28ff. The translations I have offered are taken from Barnes (1984: ii,
1748–9).
176  
without exactly describing it. In this way, it seems to me that Blackburn’s

characterization of the ‘semantic anchors’ of lewdness (something to do with
sexual display, something to do with mockery) is exactly the right sort of charac-
terization to give.
Locating a thick concept by specifying the domain of its operation is different
from giving even a vague specification of its content. Telling us where to look for a
concept is different from telling us what the concept is a concept of. To say that the
concept of courage concerns fear and danger is not to say anything at all about its
content. Being about fear and danger is not what an action has to be like to count
as courageous. We are, of course characterizing the concept in saying that it
concerns fear and danger, but we are not doing so in a way that offers to generate
even part of a descriptive meaning for the predicate.
This suggestion therefore seems to me to fit the three constraints we have
unearthed. There is description at issue, but not in a way that bids fair to construct
anything like an independent descriptive meaning, however slight. There is no
reason why courage, conceived in this Aristotelian way, should not be taken as a
multi-attitude concept. And the need to recognize that evaluation partly deter-
mines description is met by the idea that a courageous act, say, is one that
concerns fears and good heart in the right way.
Questions will arise about whether it is possible for two distinct thick
concepts to have the same domain of operation. If they can, the only way of
distinguishing them will seem to be to appeal to descriptions within that
domain, and that will cause us once more to fail to satisfy the first constraint.
This difficulty did not worry Aristotle. For each domain, as he saw things, there
are at least three different concepts available: that of the mean, that of excess
and that of defect. Such structural ways of distinguishing thick concepts do
nothing to reconstruct a descriptive meaning or independent descriptive elem-
ent in a thick concept. They might, however, seem to return us to a single-
attitude account. Surely excess and defect are exactly to be understood as where
we ought not to be, and the mean is defined as the proper place in the relevant
domain. This only shows, however, that Aristotle’s concerns are not ours. In fact
the difficulty is only apparent. If there were to be two distinct thick concepts
with the same domain of operation, this does nothing to threaten their distinct-
ness. It is not as if the domain constitutes a descriptive meaning, nor is it as if
there is nothing to be said about the thick concept beyond the characterization
of its domain of operation. There is more, but that more, being partly evalua-
tively determined, lies beyond the grasp of those who have no understanding of
the social role of the concept. It may well be that we have more than one type of
interest in the domain of sexual expression and display, each with an attendant
concept, and that all an outsider can determine is that we have two distinct
concepts that operate in broadly similar types of situation. This failure would do
nothing to undermine the distinctness.
     177
IX
The notion of a thick concept has undergone some changes during the course of
this paper. I end with a short comment on what I take to be one significant
consequence of those changes. This is that it is no longer possible that the
evaluative in general should supervene on the descriptive.
I have to be careful here in using the terms ‘evaluative’ and ‘descriptive’.
Though I have used these terms fairly freely, that has been because the debate
that I have entered here has been phrased in those terms. My own view is that the
notions of descriptive and evaluative meaning are not really coherent. Evaluation
and description are indeed distinct as speech acts, but that distinctness does
nothing to establish a distinction between evaluative and descriptive properties,
meanings or anything else much. Still, the supervenience thesis with which I am
concerned here does not need to be expressed in those terms. It could be expressed
as the supervenience of the moral on the natural. Nothing that I have said does
anything to destabilize the notion of a natural property.
Let us ask, then, whether a property like lewdness supervenes on natural
properties. The answer, if what I have been arguing is anything near right, is no.
If the relation between thick concepts and what I earlier called descriptive
meaning is as I have suggested, we cannot suppose ourselves driven from indis-
tinguishability with respect to the natural to similarity or identity of evaluation.
For two objects that are indistinguishable from each other in natural terms—it is
easiest of course to think of two actions rather than two objects—need not for that
reason be indistinguishable evaluatively. Whether they are so or not will depend
on whether they are indistinguishable at the intermediate level of the thick. But
this matter is not determined by whether they are naturally distinguishable, since
the applicability of a thick concept is not fully determined on natural grounds.
This means that there need be no ‘description’ such that we know in advance that
two objects sharing that description must be evaluated similarly.⁹
It is often held, and I have myself suggested in the past, that the thesis of the
supervenience of the moral on the natural is held in place by the concept of a
reason; it is effectively an expression of the thought that where exactly the same
reasons are in place, one must make the same moral judgement. But since there
now seems to be nothing to persuade us that the idea of a reason eventually takes
us all the way down to the natural, since we will often more or less come to a halt
at the thick, this claim about reasons does nothing to establish the supervenience
claim. The latter now seems to me to be false.¹⁰
⁹ In Moral Reasons, I failed to see this (1993, ch. 5.1–5), and in general was confused about the
importance of the notion of a thick concept.
¹⁰ I am very grateful to David Bakhurst and Simon Blackburn, and to members of my department at
Keele, especially David McNaughton, for helpful criticism of earlier versions of this paper.
12
McDowell, Williams, and Intuitionism
In his ‘What Does Intuitionism Imply?’,¹ Bernard Williams considers the relation
between what he calls ‘methodological intuitionism’ (MI) and ‘epistemological
intuitionism’ (EI). MI admits a plurality of principles that may conflict, and has no
explicit method or priority rules for resolving such conflicts. Williams accepts MI,
but he does not accept EI. EI, however, comes in two forms. The first form
supposes there to be ‘some instructive analogy between ethical propositions and
the self-evident truths of mathematics’ (p. 183). Williams finds it hard to take this
position seriously. He supposes the central idea to be that one can discern
particular ethical truths one by one and directly, rather as one discerns very
simple truths of mathematics, and comments that this leaves us with ‘virtually
no account of how such convictions might be corrected or rejected’ (p. 184). He
also notes that EI of this sort tends to deal with very abstract, general moral
propositions using terms such as ‘good’ and ‘right’, and is thus a form of what
Susan Hurley called ‘centralism’.
Williams then turns to a second form of EI, one that relies instead on an
analogy with sense perception, and this he attributes to John McDowell. This view
presents a range of particular ethical convictions that are immediate, that involve
more specific or thick ethical concepts such as treachery or cowardice or promise,
and that claim to represent something objective. It is this last claim that Williams
challenges. He thinks that in ethics ‘we can do nothing with the claim of object-
ivity’ (p. 186).
In this paper I start by considering the nature of McDowell’s position. I then
turn to examine the merits of Williams’s criticisms. In the final section I try to
decide what justice there is in the common claim that McDowell is an intuitionist,
which he himself seems forcibly to deny.
In his paper ‘Values and Secondary Qualities’, McDowell offers a general account of
the nature of value on the model of the secondary qualities, in particular colour.²
¹ Williams (1988), repr. in Williams (1995: 182–91). Page references below are to the latter version.
² References to various papers by McDowell will be to the versions printed in his Mind, Value, and
Reality (1998).
  , ,   179
He takes from Gareth Evans a broadly Lockean understanding of the sensory

qualities as dispositions in the object to cause certain experiences in us. Colour, so
conceived, is not to be identified with the relevant primary quality base, but with a
power or disposition possessed by physical things, a disposition to affect appropriate
perceivers in certain ways. This enables McDowell to maintain that value, like
colour, is both objective and anthropocentric. It is objective because, like experi-
ences of colour, experiences of value are experiences of a property that is there
anyway, waiting to be experienced, and independent of those experiences, though
obviously not independent of experience in general. It is anthropocentric because it
is understood, like colour, as a disposition to affect creatures like us in certain
specific ways.
Colour, then, is understood as a disposition to cause experiences of colour in
appropriate perceivers. What is it that value is a disposition to cause? Before we
ask this question, we must be sure that value is a disposition to cause something.
But, though the discussion in ‘Values and Secondary Qualities’ is not entirely clear
on the point, McDowell seems to accept a causal understanding of the relevant
disposition, even though he wants to add that the thing caused, the relevant
response, is ‘merited’ by the valuable object.³ This is of course in line with his
general claim that the space of reasons is not a non-causal space: it is to be
understood rather as that space where explanations do not operate by subsuming
particular cases under laws. What then is this (merited) response? There are three
potential candidates. The first is an experience of value; McDowell often talks in
such terms in this paper. The second is a recognition or apprehension⁴ of value,
which I take it is not to be conceived as an experience of the value, but more as a
judgement that the value is present, or perhaps as simple approval. McDowell
talks in these terms too; in the first paragraph he speaks of evaluative thought, and
early in section 2 he criticizes the perceptual model thus:
The perceptual model is no more than a model: perception, strictly so called, does
not mirror the role of reason in evaluative thinking, which seems to require us to
regard the apprehension of value as an intellectual rather than a merely sensory
matter. (1998: 132)
This remark about the role of reason will return to haunt us. But for the moment it
looks as if we should be more interested in the recognition than in the experience
of value. However, there is a third possibility, which emerges at the end of the
paper, when it appears that the response is not so much epistemic as volitional—
an inclination of the will. Now this seems entirely appropriate given one of
³ See for instance his fn. 42: ‘I do not mean that satisfying explanations will not be causal. But they
will not be merely causal.’
⁴ The notion of apprehension has a long and distinguished history in intuitionist thought.
180  
McDowell’s aims in the paper, which is to show the possibility of a conception of

the world that is ‘not motivationally inert’. Value, conceived as the disposition to
cause an inclination of the will, or a merited inclination of the will, seems just the
ticket for this purpose.⁵
But if we supposed that McDowell intends us to conceive of the relevant
response as volitional rather than as recognitional, we would probably be making
a mistake. As he says in his paper on aesthetic value, plenty of our responses to
value are not volitional at all; take, for instance, our response to a great painting.
So it seems better to go for a variable view: value is a disposition to elicit a merited
response, recognitional or volitional.
This variable view has the advantage that it applies in cases where the will is not
involved. This is an advantage because much ‘recognition’ of value is made in the
absence of the relevant object. Where that object is an action, it may be one done
by someone else in the past, one of which we know only by having read about it.
And it may also be one that we are about to decide not to do, because of the
disvalue that doing it would have or bring about.
It has another great advantage, but one that is less commonly noticed. It gives
us a way of getting in between what are supposed to be the only two options in the
theory of moral motivation, internalism and externalism. The question normally
asked is what relation obtains between moral judgement and moral motivation.
(And exactly the same question can be asked about the relation between recogni-
tion of value and motivation.) One possibility is that moral judgement necessarily
involves motivation: this is internalism—Moral Judgement Internalism, to give it
its full name. The other possibility is that there is no necessary connection between
moral judgement and motivation, so that any relations there are between the two
are entirely contingent. Now what I want to say about this list of options is that it
is plainly too short, and that both the options on the list are manifestly absurd.⁶
There is plainly some definite link between moral judgement and motivation, so
we are not dealing here with pure contingency, and equally plainly it is possible to
recognize a wrong and not to care two hoots about it. Somehow we have to get in
between the necessity/contingency distinction, and the dispositional conception of
value enables us to do this. For one can experience something as meriting a
⁵ Though it is worth pausing here to note something he says in his paper on aesthetic value: ‘The
phenomenology of value experience in general suggests a visual model for our dealings with value. In
the moral case we are prone to be tempted away from that model by the distracting influence of the
concept of choice or decision; whereas in the aesthetic case—so long as we are not corrupted by any
easy philosophical assimilation—that temptation is not operative. (I think the temptation should be
resisted in the moral case too, but that is another story.)’ (1998: 117).
⁶ It is worth pointing out that attempts to finesse the issue by requiring that moral judgement
motivates the rational agent are not a third alternative here, since they amount to a form of the view
that the connection between judgement and motivation is in fact contingent. Perhaps one ought
rationally to be motivated, or ought by one’s own lights to be motivated, but this does nothing to
show that any actual moral judge is motivated, nor even, so far as I can see, that they can be expected
to be.
  , ,   181
response, or as disposed to elicit a merited response, without actually responding

in that merited way.
It is important to see what is going on here. Colour is a disposition, and the
normal way of experiencing that disposition is by having the experience which
that disposition is a disposition to elicit. In such a case we might say that the
experience realizes the disposition. And a similar structure can be found on the
value side of the analogy. Talking in terms of experience, we might say that we
experience the value by having the inclination of the will that the valuable object is
as such disposed to elicit. Our inclination realizes the disposition, in such a case.
And here we have the sort of thing that internalists are keen to concentrate on,
namely a nice tight link between experience and motivation. But of course it does
not always work that way. It is possible to experience a disposition unrealized,
unactualized. One can, for instance, experience the sharpness of a knife by
running one’s thumb along the blade without being cut. One experiences the
ability to cut, but no cutting occurs. Perhaps this is also possible in the case of
value. We can experience, or recognize, the disposition to elicit appropriate
motivation without being appropriately motivated thereby. This seems to be the
sort of thing that externalists are keen to concentrate on. And when we do this, we
seem to be having the recognitional response to the disposition; our response
realizes the disposition, but is a recognition of the unrealized aspect of it, the
motivational aspect. So in a way the dispositional conception of value, in the more
complex, variable form, offers us a way of combining the salvageable elements of
internalism and externalism. I take this to be a great advantage.⁷
Sadly, however, I don’t think that the analogy with colour experience is in the
end defensible.⁸ This is partly because there is a worry whether either the
recognitional or the volitional response to value is sufficiently similar to an
experience. But perhaps that doesn’t matter too much. More worrying for me is
the fact, as I take it to be, that colour as we experience it does not present itself as a
disposition at all. The colour that we see ‘on the objects’ (Berkeley’s phrase) seems
to be a brute feature of those objects, rather than the sort of disposition that Locke
was talking about. Of course it remains true that a coloured object is disposed to
cause a certain response in suitable perceivers. But an experience of colour, even if
it is the realizer of that disposition, is not an experience of that disposition. More is
required for an experience that is the realizer of a disposition to be an experience
of that disposition, and that more is missing in this case.
One might ask, in response to this, why, if colour is a disposition, experience of
colour should represent it as such. An experience of some feature does not have to
⁷ I don’t say that the dispositional conception is the only approach that achieves this desideratum;
I suggest another in Ethics Without Principles (2004: 112).
⁸ Here I repeat points I first made in 1986 (‘Two Conceptions of Moral Realism’, repr. as Chapter 9
in this volume); I now think that I was, if anything, too charitable on that occasion.
182  
represent every aspect of that feature. Why is it so significant that our colour
experience does not represent the colour experienced as that disposition which it
must be? The answer to this is that it is not merely that our experience of colour
fails to mention (as it were) its dispositional nature. The difficulty is that it is
impossible to make any sense of the idea that the colour of an object in plain
view before one is a disposition, even though we all admit that a blue object is
necessarily one that has a disposition to look blue. And remember that McDowell
is hoping to produce an account of colour, and thereby of value, that is true to the
phenomenology of the relevant experience. But it is about that phenomenology
that he is wrong. This leaves us with a choice. Either colour is a disposition,
and colour experience misrepresents it as non-dispositional, or colour is not a
disposition, with all the awkward metaphysical consequences that may flow
from that.⁹
Two consequences flow from this. The first is that we should probably abandon
the dispositional understanding of sensory qualities that McDowell took from
Gareth Evans. Sensory qualities are somehow essentially to do with how things
look, smell, feel, taste or sound (to perceivers like us). But it seems to me
impossible to maintain that the right way to capture this essential relation to
experience is by the dispositional route, because this falsifies the nature of our
experience of the sensory. We need a different understanding of the sensory—
though I fear that I do not have one to offer.
The second consequence is that we should abandon all dependence on the
analogy with secondary qualities.¹⁰ Nice though it was to appeal to them to
establish the compossibility of weak objectivity (as characterized above) and
anthropocentricity, that which we were appealing to is not as it needed to be for
the analogy to hold. Now this does nothing to show that McDowell’s conception
of value is mistaken. For all I have said so far, it may be that value is properly
conceived as a disposition in the valuable object to elicit a merited response in
⁹ We ought to ask ourselves what sort of identity statement is at issue when we consider the
question whether colour is a disposition. This is not just one more example of attempts to identify
aspects of the manifest image with aspects of the scientific image. The identification here proposed is
between something that is sensory and (as I would contentiously say) something that is not, though it is
defined by reference to the sensory. (My own view is that the dispositional conception of colour renders
colour invisible.) The model of heat, identified with mean kinetic energy, is not a good model here,
because heat is not a sensory quality—even though it is of course sensible. The phenomenology of the
sense of heat is, I would say, compatible with the claim that heat is a disposition, while that of colour is
not. Thanks to Bob Adams and Kryster Bykvist here.
¹⁰ Michael Smith suggested, in the discussion of this paper at Leeds, that the analogy could be to
some extent restored by rewriting the account of colour in terms of meriting, so as to make it more
similar to the eventual accounts of fearfulness and value. And Chris Megone offered the idea that
colour experience be understood in terms of the proper functioning of human perceptual systems, so
that a blue object is such as to cause a certain experience in a properly functioning system of our sort.
(There is something of this sort already present in the idea of ‘appropriate perceivers’.) But there would
still remain the difficulty of making such conceptions of colour compatible with the phenomenology of
colour experience.
  , ,   183
appropriate apprehenders. We will have to check, of course, to be sure that the

value we apprehend is suitably dispositional as we apprehend it. Perhaps we can
do that; let’s hope so. But the model that we used to show how it is all supposed to
work has to be dropped, and this is a loss because that model seemed to have some
explanatory force. The result is that the dispositional conception of value has to
stand or fall on its own merits, without help from any thoughts about the sensory
qualities.¹¹
McDowell does however offer another model, the model of fearfulness. The
purpose of that example lies elsewhere (and is to be treated below), but I start
by taking it in its own right. According to him, the fearfulness of the fearful object
is to be understood dispositionally, as a disposition to elicit (merited) fear. I want
to suggest, however, that there are two dispositions to be thought about, not just
one. There is the disposition to harm or damage, which is one thing, and then
there is the disposition to cause merited fear, which is another. I maintain, against
McDowell, that the fear is merited, not by the second disposition, but by the first.
Further, the fear is a response to the first disposition, but not a realizer of that
disposition; the realizer of the first disposition would be the damage caused, or the
causing of that damage. Our fear is a realizer of the second disposition, but it is not
a response to it, nor an experience of it. So it seems to me that the fearfulness
model has complications that tend to disable it, or at least to prevent it from being
much help.
On these fronts a yet further model might be preferable, which is that of the
humorous or the funny; both McDowell and Wiggins appeal to this. And it will
indeed be preferable if we can see amusement as the realizer for the disposition to
which it is a response, and of which it is an apprehension or recognition. Our
amusement, then, will have to be a response to the disposition to amuse, which is
the disposition to elicit merited amusement. What worries me here is whether this
disposition is what we laugh at when we laugh at the funny. It doesn’t seem to me
at all clear that this is so. But I don’t mean by this to challenge the view that the
funny is that which merits laughter, and that laughter is the/an appropriate
response to the funny. This thought, which lies at the basis of what McDowell
later calls the ‘no-priority’ view, remains unchallenged. It is the dispositional way
of casting it that seems challengeable.
Perhaps the best thing, then, is to abandon these various analogies and to tackle
value independently and see how we get on. So what form should our dispositional
¹¹ Others have drawn this same conclusion, but for different reasons; see Darwall et al. (1992:
115–89).
184  
account of value take? McDowell’s form, as I said, sees the valuable as that which is
disposed to cause (or elicit) a merited response; and I suggested that the relevant
response might be either recognitional or volitional. So my first question is
whether we should see any difference between meriting a response and being
disposed to elicit a merited response. The main point of the dispositional model,
of the insertion of the disposition, as it were, was to generate a situation in which
the realizer of the disposition was also, and thereby, an experience, or at least an
apprehension, of that disposition. If we just talk in a far more natural way of the
object as meriting a response and of the response being made to the object as so
meriting, we lose that central feature. So if we are trying to run a dispositional
account, we have to distinguish between meriting a response and being disposed
to elicit that merited response. We cannot thin out the disposition so much that it
effectively vanishes—any more than we managed to thin out the disposition in the
case of colour so as to keep in play the thought that our experience of colour is an
experience of the object as disposed to cause that experience.
Still, what are the phenomena here? What is it actually like to discern a value?
There is no simple answer to this question. Sometimes that discernment is
volitional; the object or situation is presented to us as calling for such and such
a response. The badness of what is about to happen, or of what is going on in front
of me, presents itself to me as demanding a certain response from me, an
inclination of the will; in inclining my will accordingly, I respond to the value
(or disvalue, in this case) in the way that it merits. As we might say, I recognize
that value in the way I respond. Not all volitional cases will be simple, however; we
can imagine cases where I feel the need to respond somehow but am not sure how,
so that volition is, as one might put it, active but stymied. And if we are not dealing
with a volitional response but a (merely) recognitional one, the situation seems
effectively similar. Approval is such a response. I read a really good essay by my
star undergraduate, and it seems to me to call for enthusiastic approval. Again,
I recognize the value in the way that I respond to it. But all this can be said without
any appeal to dispositions at all. And better said, too. For it seems to me to be a
distortion of the phenomenology to insist that what I am responding to is not the
meriting of the response that I give, but the disposition to elicit that merited
response it. So as I see it, talk of meriting and responding to merit in the way
merited is just fine, but the insertion of a disposition spoils everything. If you thin
out the disposition completely, it effectively vanishes and if you don’t, it gets in
the way.
There was however one supposed merit of the dispositional account that I have
mentioned but not paid enough attention to. This is the way in which it expresses
the so-called no-priority view. This view is presented as a reason for dissatisfaction
with projectivism. The projectivist supposedly maintains that we can achieve a
satisfying conception of a certain attitude which we then project onto the world by
constructing a concept that applies to the things that we respond to in that way. So
  , ,   185
we are amused, which is a response; and we construct on the basis of that response
the concept of the funny; but there is no property of funniness out there to
respond to; there is just what amuses and why it amuses. In a similar way, there
is no objective quality of ticklishness. There is just what tickles us and that about it
that tickles us. The difficulty for this approach, according to the no-priority view,
is that there is no independent conception of the relevant attitude to start from in
the way that the projectivist wants.¹² The attitude can only be understood as that
of finding funny; and the funny can only be understood as that to which amuse-
ment is an appropriate response. Property and attitude are in this way made for
each other. There is no priority. We are not dealing with a parent–child relation-
ship, where the attitude is parent and the property the child; we are dealing rather
with two siblings.
The point now is that the dispositional account seems perfectly set up to
capture the relevant idea of siblings. The object has the feature of being disposed
to elicit the relevant response, a feature that cannot be understood independently
of the response merited; and the response is a response to the object as disposed to
elicit just that response. The response is the realizer of the disposition and so
cannot be understood independently of the disposition to which it is a response.
The two sides are intertwined.
What I want to say about this, however, is that the dispositional way of
capturing the no-priority view, nice though it is, is not the only way to do it. It
is equally good to talk of meriting or calling for a response and responding to that
merit, or to that call, in the way that is merited or called for. The no-priority view
and the dispositional conception are not a package deal.
I now turn to the way in which McDowell uses the dispositional account to
respond to one of Mackie’s charges against objective value. This is that the
supposed objective value-qualities, even if not conceptually incoherent, are
explanatorily redundant; they do not earn their place in the objective order.
MacDowell’s response to this is that the same could be said about colour. All we
need for the relevant explanations is the primary quality ground, in the object, and
the nature of the sensory systems in us. But this does not persuade him that real
colours are explanatorily redundant in the relevant sense. The question, he says, is
not whether they pull their weight in causal explanations, but whether it is possible
to run those explanations while denying that there are real colours. And he claims
that this is not possible; nobody could explain a colour experience by appeal to the
¹² For similar reasons I would say that McDowell’s conception of value cannot be presented as an
example of ‘fitting attitude’ analysis.
186  
primary qualities of the coloured object without thereby allowing that the object is
disposed to cause the relevant experience in a suitable perceiver. If we deny the
existence of such a disposition we undercut the causal explanation we were trying
to appeal to. Real colours, then, have a secure place in the objective order.
What interests me here, of course, is the extent to which such things can
properly be said of objective values. The obvious first point is that McDowell’s
remarks about real colours are predicated upon the dispositional conception of
objective colour that I have been challenging. If we abandon the dispositional
conception of colour (which I maintain makes colour invisible), and think in
terms of the sort of non-dispositional colour that, on my understanding, is what
objects display to us, we can no longer say that without objective colour of that
sort the supposedly pared-down ‘scientific’ explanation of colour experience
would collapse.
It is not my purpose to defend, or to attack, the non-dispositional conception of
colour as we experience it. Maybe we are going to have to admit that there is no
such objective quality to be experienced. But it is my purpose to wonder whether
we can construct an independent argument to the effect that, though objective
value-qualities do not pull their own weight in causal explanations of certain
responses, still we cannot deny the existence of those qualities if we want to run
the causal explanations. What, then, about the relevant responses—approvings,
say, or inclinations of the will—makes it undeniable that the relevant value-
qualities exist? I will tackle this question in two steps. First I will ask the question
in terms of the dispositional conception. Then I will ask it in terms of the
alternative conception, which I will be calling ‘pure meriting’.
So there is the object, all set up with its natural qualities, and here am I, warmly
approving or moved in some way or other. And we assert that my warm approval
or movedness is the effect that the object has on me. This is the causal explanation.
By giving this explanation, are we thereby forced to allow that the object is
disposed to cause me so to approve? This is what, on the dispositional account,
it is for the object to have objective value. (There is of course the thought that the
approval is merited, but I don’t think that affects the present point.) Now
I certainly allow that the object can cause me to approve. But surely the disposi-
tional account is supposed to say more than that. I can also allow that the object is
such as to cause, or so set up as to be able to cause me to approve; these things,
certainly, cannot be denied either. But this addition of little words doesn’t seem to
me to change anything at all. The fact that what does happen can happen does not
turn a non-dispositional account of what is going on into a dispositional one.
What someone who is trying to run a dispositional account wants is somehow to
ensure that the object is doing the work, but it seems to me perfectly deniable that
this is so in the case of value. Again, if we thin out the disposition so that it
effectively vanishes, there is no objection, but once we put the dispositional flesh
back on it, it gets in the way.
  , ,   187
But although this point seems sound to me as far as it goes, I think it misses
McDowell’s real thrust. There are two possible reasons for saying that when we
explain valuings, we have to allow that, as he puts it, reality contains something in
the way of value. The first is the dispositional conception of value, and the idea
(which I have just rejected) that we cannot run the relevant explanation while
denying that the object is disposed to cause the response in us. This is the way
things ran in the case of colour. The second, however, seems to make no use of the
dispositional conception at all. It appeals rather to the distinctive nature of the sort
of explanation we are offering when we explain fear—and when we explain
valuations. McDowell writes:
But if what we are engaged in is an ‘attempt to understand ourselves’,¹³ then

merely causal explanations of responses like fear will not be satisfying anyway.
What we want here is a style of explanation that makes sense of what is explained
(in so far as sense can be made of it). This means that a technique for giving
satisfying explanations of cases of fear . . . must allow for the possibility of criti-
cism: we make sense of fear by seeing it as a response to objects that merit such a
response, or as the intelligibly defective product of a propensity towards
responses that would be intelligible in that way. (1998: 144)
The idea here seems to be that we cannot makes sense of fear as a human response
unless we suppose that some instances are merited; and we will explain some
instances just by showing that they are merited. At this stage, we seem to have
abandoned any resort to the dispositional analogy, and the whole thing is being
done in terms of pure meriting. So McDowell writes:
For an object to merit fear just is for it to be fearful. So explanations of fear that
manifest our ability to understand ourselves in this region of our lives will simply
not cohere with the claim that reality contains nothing in the way of fearfulness.
Any such claim would undermine the intelligibility that the explanations confer
on our responses. (1998: 144)
Note how dispositions have vanished from the scene here. What he is claiming is
that we cannot run explanations of fearings in ways that deny that there are things
to be feared, even if we allow that this feared object is not among them.
Projectivists could try to run a causal explanation of fearing the fearful while
denying that there are things that merit fear, or they might allow that some things
merit fear and others don’t, but see themselves as merely expressing in this their
own endorsement of people’s being afraid in some cases and not in others. But in
¹³ This phrase is taken from Blackburn, who is McDowell’s target at this point.
188  
doing so, they suppose that the appropriateness of the response is not itself any
part of its explanation. And this is what McDowell is challenging. I take it that his
challenge is an expression of the distinction he later draws between two general
styles of explanation:
To recognize the ideal status of the constitutive concept is to appreciate that the
concepts of the propositional attitudes have their proper home in explanations of
a special sort: explanations in which things are made intelligible by being
revealed to be, or to approximate to being, as they rationally ought to be. This
is to be contrasted with a style of explanation in which one makes things
intelligible by representing their coming into being as a particular instance of
how things generally tend to happen. (1998: 328)
McDowell is supposing that explanations of fear, in virtue of their critical dimen-

sion, are to be placed in the group of ‘explanations of a special sort’. It is this, not
something to do with dispositions, that explains why we cannot run the relevant
explanation without admitting features that the pared down ‘scientific’ explan-
ations do not even mention. And once we have seen this, we can see that the same
would hold of explanations of valuing.
I now turn to Bernard Williams’s discussion of McDowell’s position. In the

present section, I devote myself to Williams’s general charge against intuition-
ism that it suffers from various explanatory incapacities. One aspect of this
is that it needs, but cannot provide, a theory of error, by which is meant
something we can say that might justify sticking to our own view when
others around us are equally convinced of the opposite. But that is not quite
Williams’s focus in ‘What Does Intuitionism Imply?’. He focuses more on
McDowell’s claim that value is objective, in the sense of being there to be
recognized or otherwise responded to, independent of that particular response.
Williams offers a sort of test for such objectivity claims. In the case of colour, to
maintain our sense of the objectivity of the colours we perceive we need a
picture of the world that can contain other perceived colours (those perceived
by birds, say) together with those that we perceive. Williams allows that this
is possible in the case of colours, but not in the case of fearfulness, nor in the
case of value.
Unfortunately, the way in which Williams sees us as constructing a conception
of one world that contains the colours perceived by humans as well as those
perceived by birds is the dispositional way. The world can be easily conceived as
disposed to present certain sorts of visual appearances to humans, and other sorts
  , ,   189
of visual appearances to birds. These different dispositions do not conflict. If, by

contrast, we were to accept the non-dispositional conception of colour as we
experience it, matters would probably be different. Could we conceive of the object
as coloured all over in our way, so to speak, and in a different, avian way at the
same time? Somehow I doubt it. If not, these non-dispositionally conceived
colours are going to fail Williams’s test for objectivity. Perhaps that is right;
perhaps it isn’t. It might even show that Williams’s test is a good test. But this
tells us nothing yet about values.
Williams then addresses the case of fearfulness, and here I find it very hard to
understand what he says. He writes:
Consider a people who are filled with terror, perhaps of a rather special, numin-
ous kind, by certain features of their environment. They have a word to pick out
things to which they react in this way. It is not a blankly causal, still less a merely
individual, reaction, and children are instructed in what does and what does not
merit it. We—the ethnographers—come to understand these reactions, and the
word that picks out things in terms of that reaction. We do not share the reaction,
except to the extent that we imaginatively enter into their view of things; for
instance, we do not share beliefs and attitudes that make this reaction intelligible.
Is the quality for which they have the term ‘there to be perceived’? It is part of
their world; it is not part of our world. Is it part of the world? (1995: 185)
For this to work at all, we have to allow that we are dealing with genuine terror,
even if it is of a special, numinous kind (though I have no real idea what that
means). But now the issue is whether the things these people are frightened of are
at all dangerous. If not, those things lack the first disposition, namely the dispos-
ition to damage or hurt. They might have the second disposition, namely the
disposition to elicit fear. And it might even be that these people’s fear is rational, in
the sense that they have a coherent explanation of the supposed danger. But if in
fact there is no danger there, there is nothing to fear, and the quality for which
they have the term is not part of the world. They take these objects to be fearful in
some way, and respond to them appropriately as meriting the fear, but they don’t
merit it. So here I just don’t see the difficulty. But even if I am right about this, the
prospects of objective value are hardly thereby improved. The analogies have all
failed, and we have to take the case of value on its own merits.
So I now turn to value. Williams’s test requires us to ask whether we can
construct a picture of the world that can contain other values besides those that we
recognize ourselves. And he writes:
[With the secondary qualities] we can form a picture of one world, differently
perceived, and that picture can contain these qualities. With the example of fear,
and equally with what are clearly values, this is not so. We again have the
190  
phenomenon of a shared practice that requires psychological explanation.¹⁴ But

here we can do nothing with the claim of objectivity, because here the explan-
ation of why things are like that for them and not so for us does not bring in that
quality [sc. the supposed objective value-quality, there to be perceived]. We could
not, as in the case of secondary qualities, fit together into one world the qualities
that they are supposedly perceiving and those that we are supposedly perceiving,
let alone all the qualities demanded by the various value systems that there are or
might be. . . . There is no room for all those qualities, and, even if there were, there
would be no coherent explanation of why we are able to perceive some of them,
and they others. (1995: 186–7)
The first thing to say here is that the objectivity test that Williams is running seems
to be different from his standard test, which appeals to the absolute conception of
the world. This is not because the new test admits the objectivity of colour when
the standard test does not. I take it that colour, conceived as a disposition, does not
earn its own place in the world absolutely conceived, because it characterizes the
world in ways that relate it to potential perceivers. But one might nonetheless
allow colour a derivative place there, since colours are partly grounded in and
partly explained by aspects of the world absolutely conceived. So colours pass both
tests, in a way, while values pass neither test.
The difference between the tests seems to be that the appeal to the absolute
conception is an attempt to understand what is ‘there anyway’, while the new test
appeals rather to the impossibility of making room for all supposed value-
properties. Williams is willing to allow that objects have all the colours they
appear to have to creatures whose colour-perception is functioning properly for
them. But he is not willing to allow that all the values supposedly discerned by
people in different cultural traditions are there to be discerned. This would
overpopulate the world.¹⁵
What world would it overpopulate? Williams is suggesting here that there are at
least two ways in which a practice can be ‘world-guided’, the second of which
appeals to socialization in a way that the first does not; the world that guides can
sometimes be the product of a practice. And that is how it is with values. He
writes:
Nevertheless, the nature of the shared practice shows that it is world guided, and
explanation will hope to show how that can be. What the explanation exactly
may be, is to be seen: but we know now that a vital part of it will lie in the desires,
attitudes, and needs that we and they have differently acquired from our different
¹⁴ The psychological explanation tells us ‘how they can go from one situation to another in using this
term, and why one thing rather than another elicits from them the related reaction’ (1995: 186).
¹⁵ Thanks to Miranda Fricker for help at this point.
  , ,   191
ways of being brought into a social world. The explanation will show how, in
relation to those differences, the world can indeed guide our and their reactions.
‘The world’ in that explanation will assuredly not be characterized merely in
terms of primary qualities; the account of it will need to mention, no doubt, both
secondary qualities and straightforwardly psychological items. (1995: 186)
In terms of the new test, Williams thinks that adopting a dispositional concep-
tion does not achieve the same results in the case of value as it does in the case
of colour. We cannot, apparently, make progress with the idea that, just as one
and the same object is disposed to elicit one sort of visual experience in us and
another in the birds, so one and the same action is disposed to elicit approval
from us and disapproval from the Erewhonians. But as far as I can see, a
dispositional conception will achieve this much perfectly well. There will be a
common world, disposed to present different appearances according to the
different receptivities of those responding to it, whether those responses are
perceptual, recognitional or volitional. But I would agree with Williams that this
is not a consoling resolution. It is not consoling because those dispositions are
not what value is in the object. In my view, the dispositional conception,
expressed as I expressed it above, simply dispenses with objective value. We
need to remember that the response is not supposed to be just elicited; it is
supposed also to be merited. But where their responses differ from ours,
are both responses to be thought of as merited? If so, are we thinking of them
both as merited by how things are? It would be nice to think that the explan-
ation of the meriting appeals to the supposed objective value-quality out there
in the object; that response is merited that best fits the facts. But Williams is
right to say that this is unconvincing. If both responses are merited, there
must be two such value-qualities out there, and if only one is merited, we
need some explanation of what the other one is lacking. But there need be
nothing lacking in either response; the only thing wrong with their response is
that it is not ours, and vice versa. This is why Williams maintains that the
relevant explanation is psychological, not objective. Neither response is to be
explained by showing that it fits the facts; each is explained as the correct
product of a social practice.
But suppose that there are objective values, and that we are blind to some of
them. As far as I can see, there will be no insurmountable difficulty here unless we
think of our openness to value as rather like our openness to colour. If we do think
in that way, we are supposing that we have just to open our eyes, and either we will
see the value or we won’t. If we don’t, and it is there, there is something wrong
with our ‘value-perceptual’ apparatus; perhaps it needs better cultural tuning. But
suppose I said to myself, ‘They clearly see something here that I don’t; I wonder if
they can explain it to me. Perhaps with a little good will on both sides I can be
helped to cotton on.’ Why should it not be that my conception of value can be
192  
enlarged, or if not enlarged at least changed, in this sort of way?¹⁶ The main
answer to this question is that the new values may be incompatible with the old
ones, in such a way that they cannot be added. But our present value system has
plenty of apparently incompatible values which we struggle to find some way of
holding in place all together. One example (suggested to me by David Frank) is the
combination of communitarian and individualistic values; another might be
tolerance and commitment. The idea here is not just that we may be faced with
situations in which we cannot achieve everything that we value, and so we have to
choose which values to promote on this occasion, leaving the others for later. It is
rather that we can have values that are intrinsically in conflict; to promote one is
necessarily to infringe the other. These values do not just get in each other’s way
accidentally, as it were; the tension between them is essential. Of course we seek an
accommodation in the face of such competing values, and such accommodations
are possible, though perhaps always with a sense of strain. There is nothing in this
to challenge the values so accommodated.
More generally, a very significant difference between values and colours is that
values are connected to reasons, and with that connection comes the possibility of
manageable disagreement, where we understand what someone is saying though
we think they are wrong and that there is some account of why. Perhaps they are
over-impressed by one aspect, and insufficiently sensitive to another. If so, it is not
as if nothing whatever can be done about this, and maybe in the discussion our
own picture will have to be adapted somewhat. The right attitude here seems to be
a sort of optimism rather than the pessimism that Williams evinces. Further, we
should make room for the possibility of a certain amount of indeterminacy or
vagueness in ethics, and one can certainly learn from the different responses of
others that matters are far less clear-cut than one thought. Irreconcilable differ-
ences where both sides think the other to be clearly in the wrong remain a
possibility, but as far as I can see they are not a theoretically damaging possibility.
My general conclusion here is that Williams’s explanatory demands can be met.
Finally I turn to consider whether Williams is right to call McDowell an intu-

itionist. His reason for doing so is that McDowell’s view ‘presents a range of
ethical convictions that are immediate and claim to represent something objective’
(1998: 184). McDowell himself, however, shows some dislike of intuitionism. So
things need to be sorted out here.
As I see it, the best way to characterize intuitionism is not by offering a list of
necessary and sufficient conditions (which is, after all, the wrong way of doing
¹⁶ McDowell asks this question with respect to aesthetic values. As he points out, confronting other
cultures often enlarges one’s aesthetic horizon.
  , ,   193
most things), but by offering a series of ‘marks’ of intuitionism. The idea here is
that not all intuitionists will meet all the marks, but that all meet a fair number of
them. Of course there will be some people, such as Tim Scanlon, who are
especially hard to map; this seems entirely predictable.
Mark 1: intuitionists are pluralists in the theory of the right; they are suspicious
of artificially architectonic theories, and especially of the idea of a Supreme Ethical
Principle; they assert several equipollent principles if they deal in principles at all,
and if they don’t they assert that there are many distinct ways of getting to be right
or wrong.
Mark 2: intuitionists are realists, asserting that there are facts of the matter in
ethics as elsewhere.
Mark 3: intuitionists are cognitivists, asserting that moral judgement is a
cognitive state. (If challenged, they would probably say that this is a straight
consequence of Mark 2, since belief is the appropriate attitude to a fact.)
Mark 4: intuitionists are non-naturalists, asserting that the facts at issue are
non-natural facts, not merely that they can be properly characterized using non-
natural concepts.
Mark 5: intuitionists are metaphysical quietists, showing little interest in the
metaphysical issues that are so often pressed against them, in virtue of Mark 4.
Those who press those issues can be naturalists, on one side, and Kantians on the
other; the Kantians are driven by the thought that the normative needs a very
special explanation—something for which the intuitionists see no need. (A lot
more could be said about this.)
Mark 6: intuitionists think that at any rate a wide range of these normative facts
are self-evident, and known a priori.
There is a possible Mark 7: intuitionists think that the right is independent of
the good.
We are now in a position to ask a sensible question. How many of these marks
fit McDowell? The answer, I would say, is pretty well all of them. There is room for
debate on some points. One might assert that he is not really a non-naturalist but
someone who presses a distinction between first and second nature, or between
the space of reasons and the space of law. One might say that he has more to say
about the operations of a moral sensibility than would really be appropriate for
someone who supposes that our moral knowledge is a priori, or even that moral
judgement is a cognitive state. One might say that the whole idea of a disposition
to elicit a merited response is evidence of far greater sensitivity to metaphysical
issues than we see in the British intuitionist tradition in general. All these things
could be said. But still, I would say, the general picture is clear.¹⁷
¹⁷ It is quite interesting to use my marks in assessing the work of such other prominent writers as
Joseph Raz.
194  
However, we should also recognize the distinctive nature of what has become
known as sensibility theory, always remembering that a sensibility theory is not a
moral sense theory. It has the following notable features:
• the appeal to dispositions in the object;

• the combination of anthropocentricity and objectivity;
• the ‘no-priority’ thesis;
• non-centralism;
• a primary focus on value.
The last two of these come together. Although McDowell’s early paper on
hypothetical imperatives focuses on those imperatives that tell us what we should
do, later work seems much more interested in the sort of judgement that involves
the application of thick evaluative terms, without the constant ‘centralist’ interest
in the way the applicability of such terms contributes to the dominant ethical
question of where one’s duty lies—that is, to the question which action is right and
which is wrong—or to the attempt to get what some think of as contributory
considerations to add up to an overall determination of what is best in the
situation. These signs of centralism seem to be largely absent in McDowell.
After all this, it is disconcerting to read what he says himself about ethical
intuitionism in ‘Values and Secondary Qualities’:
The perceptual model is no more than a model: perception, strictly so called, does
not mirror the role of reason in evaluative thinking, which seems to require us to
regard the apprehension of value as an intellectual rather than a merely sensory
matter. But if we are to take account of this, while preserving the model’s picture
of values as brutely and objectively there, it seems that we need to postulate a
faculty—‘intuition’—about which all that can be said is that it makes us aware of
objective rational connections: the model itself ensures that there is nothing
helpful to say about how such a faculty might work, or why its deliverances
might deserve to count as knowledge. (pp. 132–3)
The suggestion must be that this bad sort of intuitionism can offer nothing by way
of a serious epistemology. But then we want to know how a better intuitionism is
supposed to work. Apparently the bad sort is capable of revealing objective rational
connections—which is not bad so far as it goes, one might have thought. What can’t
it do? Does it give too thin a conception of what there is out there to respond to?
McDowell repeats some of these thoughts several times in ‘Projectivism and
Truth in Ethics’, for instance when he writes:
The assimilation to the senses gives this intuitionistic position the superficial
appearance of offering an epistemology of our access to evaluative truth, but
there is no substance behind this appearance. (p. 154)
  , ,   195
Now this particular remark comes early on, in an exposition of Blackburn’s views,
which are his target. But the idea is endorsed in propria voce at various places later
on, for instance:
This appearance reflects an assumption that, at the metaphysical level, there are
just two options: projectivism and the unattractive intuitionistic realism that
populates reality with mysterious extra features and merely goes through the
motions of supplying an epistemology for our supposed access to them. (p. 157)
So obviously his own position is supposed (as is common in McDowell) to be one

whose possibility is revealed by the diagnosis of a misguided assumption that there
are only two views to choose from. His position is to be adopted because it shares
none of the flaws of the two positions we were originally being expected to choose
between. But how is this supposed to work? We might have thought that the
crucial difference is between cognition of an ‘extra fact’ and something that
involves the operation of reasons. Here we could appeal to the resultance of the
relevant concepts, and try to argue that our cognition of instances of those
concepts must somehow be grounded in cognition of the features of the case
from which the present instantiation of the concept results, and which could be
seen as the reasons why it obtains in this case. But I don’t think this is McDowell’s
view. I think he follows Evans in holding that the epistemology here need not
follow the metaphysics. We can, that is, have a bottom-up metaphysics (if the
action is unkind, that is because of certain other features that it has, features that
make it unkind) married to a top-down epistemology, according to which what we
first cognize might be the unkindness, it being left till later to work out the exact
respect in which it was unkind.
Obviously populating reality with mysterious extra features is not an option.
One might have thought that what is needed here, by way of an alternative, is the
ability to recognize and respond to reasons—so long as we don’t conceive of
reasons as extra features, which we need not. We can instead conceive of reasons
as rational connections between considerations and responses. But apparently that
is no better—at least so long as we stick to the picture of values as ‘brutely and
objectively there’. If we do that, ‘the model itself ensures that there is nothing
helpful to say about how such a faculty might work, or why its deliverances might
count as knowledge’. Now clearly we are not supposed to abandon the idea that
values are objectively there. So it must be the bruteness that is at issue. We must
abandon the view that objective rational connections are brutely there for the
intellect in the sort of way that sensible features are brutely there for perception.
But this is still obscure, and it is obscure partly because it is not so obvious that we
are all committed to the bruteness of perceptual presentation. (Perhaps the
problem is that if we contrast the merely sensory with the intellectual, we leave
the sensory too brute.)
196  
So obscure is it that I wrote to McDowell and asked him to cast his mind back to
those early times and see if he could think what he might have meant. His reply
was this: ‘It is the immediacy of the supposed cognition, not its object, that makes
the epistemology a mere pretence. There is nothing wrong with objective, and so
knowable, rational connections so long as you don’t suppose they simply impress
themselves on some supposed quasi-perceptual faculty. The idea of being on to
objective rational connections is all right if you put it in a context in which you
talk about acquired capacities to reason.’ This reply reinforces my feeling that
Williams’s distinction between two forms of EI (one using an analogy with simple
mathematics, the other with sense perception) has turned out to be inadequate to
the case. It looks, by contrast, as if it might be correct to categorize McDowell as an
intuitionist, but that in order to do that we need to distinguish between the bad old
intuitionism (that did indeed think in terms of the mind’s eye in one way or
another) and a good new McDowellian intuitionism that focuses on acquired
capacities to reason.
But then we still need to work out what this amounts to. There are acquired
capacities, and there are capacities to reason. As far as acquired capacities go, we
should allow that some of these are perceptual; the jazz aficionado acquires a
capacity to hear what originally sounded as a welter of disorganized noise as the
gradual development of a theme. And then there is the point that McDowell tends
not to contrast sense and reason, but to suppose rather that human perception is
rational in a way that the perception of the lower animals is not. So an acquired
perceptual capacity might itself count as an increase in one’s capacities to reason.
Not only this, however, but there is also the motivational side of things to be
considered. The notion of a capacity to reason must itself not be conceived too
intellectually, unless we are willing to allow the possibility of practical intellect.
(We should be as wary of the notion of the merely intellectual as we are of the
merely sensory.) Here one of the themes of Section 1 above returns: the distinction
between recognitional and motivational responses to value. The good new intu-
itionism will want to make sense of the idea that a response can be both recogni-
tional and motivational at once, which after all is just another way of expressing
the idea that the world is not motivationally inert. But this is not the place to
elaborate on such possibilities.
13
Practical Concepts
Introduction
In this paper I revisit themes that I first treated in my ‘In Defence of Thick
Concepts’ (1995b; repr. as Chapter 11 in this volume). That paper grew out of
the comments I had prepared to give as Chairman at the 1994 Joint Session
Symposium on thick concepts, where the speakers were Allan Gibbard and
Simon Blackburn. Those comments were suppressed on the occasion, for lack of
time. But I had been impressed by some of the things Blackburn said about the
thick, and thought that they required not total abandonment of, but still signifi-
cant adjustment to the standard conception of thick concepts.
The standard conception held that thick concepts somehow combine descrip-
tion and evaluation. The evaluative element would be either pro or con; some
thick concepts are pro-concepts and the others are con-concepts. This picture
seemed to me to be undermined by the example introduced by Gibbard, that of
the lewd. The concept of lewdness looks like a perfectly standard thick concept, if
there are any such, but it fails to fit the standard conception in two ways. First,
some actions are the better for being lewd, though many lewd actions are the
worse for it. So lewdness has a variable rather than a constant practical relevance.
Further, when one calls something lewd one may be expressing a complex of
attitudes rather than just one of the two super-attitudes of approval and disap-
proval. Another way of putting this point is that our response to the lewd is mixed,
at least on occasions. There is a certain amount of lewdness on display at Mardi
Gras, and my response to this is certainly neither simple approval, nor simple
disapproval. In fact the choice between approval and disapproval, pro and con,
doesn’t seem to be exactly the point. I’ll return to this later.
To sum up: the standard conception understood thick concepts as used to
express just one attitude, and to express the same attitude on each occasion. They
are all single-attitude concepts, and they are invariable in the attitude they express.
(Sometimes, as here, I speak of the concept as expressing an attitude, but of course
no concept is capable of such a thing; it is the speaker, or the use, that really does
the expressing.) My suggestion was that we abandon both of these elements of our
conception of the thick. In their place, we should allow that thick concepts are, or
at least can be, both multi-attitude concepts (expressing a mix of attitudes at once,
most of which will be neither pro nor con—think of the embarrassment associated
198  
with even acceptable instances of lewdness) and variable in their practical

relevance. Sometimes, as I said, an act can be the better for being lewd, for
lewdness is occasionally exactly what is called for; isn’t that exactly what carnivals
are about (among other things, of course)? A display doesn’t cease to be lewd for
being acceptable and appropriate in its context.¹
It is an interesting question, which I will address briefly here, whether all those
thick concepts that are variable in this sort of way do enjoy nonetheless a sort of
default valence. Lewdness, we might say, is generally not desirable, and when it is
desirable, this is always because the circumstances are unusual in one way or
another. Very many thick concepts are like this. Think of the courageous, the
honest, the amusing, the light-hearted—and so on. I might be able to find a sort of
case in which something, or someone, is the worse for being amusing, but this will
only be because something in the situation interferes to prevent things from being
the way one would expect.² I don’t have any trouble with this suggestion.
Technically it allows the variability I want, and that is enough for me. But it
would be good to find at least some examples of thick concepts that don’t have a
default valence in this sort of way. David Sosa suggested the provocative; another
such would perhaps be the seductive.
There was a second theme in that early paper of mine. The interesting thing
about thick concepts is not really whether they are single-attitude or multi-
attitude concepts, variable or invariable in their practical relevance. It is the fact
that they somehow hold together elements, or aspects, or evaluation and descrip-
tion. When I say that an action was lewd, I am telling you quite a lot about what
sort of action it was, even if by doing so I also express a certain mix of attitudes.
(Note the way that attitude and evaluation are being run together here.) Now
some people, known as the disentanglers, think that these two aspects can be
disentangled from each other; though co-present here, one at least of them could
be instantiated separately. I say ‘one at least’ because it is common ground
between entanglers and disentanglers that the evaluation is not capable of separate
instantiation, since the evaluation at issue is linked to the description.
A regrettably lewd action is not of such and such a sort, and bad; it is bad for
being of that sort. All accept that the evaluation is linked to the description in this
way. The interesting question is whether the descriptive element is capable of
separate instantiation. The disentanglers say yes, the entanglers no. The disen-
tanglers offer a reductive picture: use of a thick concept informs us that the
relevant act (or whatever) has certain specifiable, and specified, features, and is
the better or the worse for being so. If we take things in this way, it is evident that
¹ I am told that North Americans are more likely to think of the bawdy than of the lewd as
functioning in this variable way.
² I am intending these remarks to be reminiscent of the discussion of default reasons, especially the
work of Lance and Little (2004; 2007).
  199
the first bit of information can stand alone even though the second cannot; the
action had those features, and this can be said, or thought, without any associated
evaluation, any expression of attitude, at all.
So now the question arises for the entanglers: how can it be that the supposed
descriptive ‘element’ is incapable of standing alone? Is it that there is not enough
description here to make the content of a separable assertion? Hardly: any amount
of description, no matter how slight, seems to be enough to make a content. But if
not that, then what? The answer to this question is that just as the evaluation is
linked to, penetrated by, the description, so the description is linked to, penetrated
by, the evaluation. There is ‘interpenetration’. But this is just a name for the thing
we are trying to capture. Use of a thick concept characterizes the object in some
way, describes it, ascribes some property or properties to it, and then evaluates it,
expresses some attitude or response towards it. The interpenetration of descrip-
tion and evaluation is an interpenetration of property ascribed and attitude
expressed. And that sort of interpenetration can be characterized in a way that
is owed to David Wiggins: the property is to be understood as that of being such as
to merit the response, and the response is to the object as meriting that response.
Taken in this way, the response can only be understood in terms of the property to
which it is a response, and the property is to be understood as the property of
meriting that response. This overall picture was later called the ‘no-priority’ view
(McDowell 1987), because it maintained that neither property nor attitude
enjoyed any sort of priority over the other. We cannot construct the property
out of the attitude, nor construct the attitude out of the property, because as soon
as we have one of the pair we already have the other.
Another way of making what is effectively the same point is to say that use of a
thick concept does not express an evaluation of the relevant object for having
certain specific features. The object is evaluated not just for having those features
but for having them in the right sort of way. What is that way? It has those features
in such a way as to make the evaluation appropriate.
The no-priority view, as I have expressed it so far, is merely formal. We have
been shown a possible arrangement where the interpenetration is there for all to
see, but we have not been shown how it works in any particular instance. No
matter: our task at the moment was merely to make some sense of the entanglers’
claim that neither description nor evaluation was capable of standing alone, and
I was persuaded that the no-priority view did achieve that end. One difficulty that
immediately arises for it will be how to distinguish between different thick
concepts. These are going to differ on the descriptive side, and it may be hard to
see how to characterize descriptive differences without returning to a list of
features that will, once more, be capable of standing alone. The answer that
I gave to this difficulty was to appeal to an Aristotelian conception of a domain.
A courageous act is one that is to be commended in the domain of response to fear
and danger. This specification of a domain seemed to me distinct from a list of
200  
features for which the action is to be commended. The act is not commendable for
being in that domain, but for other features not specified.
There is a further attractive aspect to the no-priority view. Although property
and response cannot be understood independently of each other, still there is an
asymmetry of explanation involved. As Wiggins points out, the property both
explains and justifies the response; he might have added that the response does
neither of these things to the property. McDowell later gives the example of
fearfulness (1985: §4). We can explain fear by showing that the feared object
was indeed fearful, to be feared; that is, the fear is explained by being merited,
appropriate, justified.
In this section, I bring out some assumptions that have pervaded, or underpinned,
recent discussion of the distinction between thick and thin concepts. I am not
suggesting (yet) that these assumptions are sound.
First, the distinction between thick and thin is not just a matter of degree.
Indeed, it is not a matter of degree at all. There may be more or less thick concepts;
some are thicker than others, that is, some offer richer description than others. But
thinness is not a matter of degree; all thin concepts are equally thin.
Second, there is therefore something that thick concepts have got and thin
concepts don’t have, and that something is the thing that those that are more thick
have more of than do those that are less thick. We can (warily) call this something
‘descriptive meaning’.
Third, the thick therefore combines something which it shares with the thin
(call this ‘evaluative meaning’) with the descriptive meaning that it does not. Some
thick concepts have more descriptive meaning than others, but there is no such
thing as more evaluative meaning; each thin concept has as much evaluative
meaning as any other, and the thick concepts have that much too.
Having got so far, we arrive at a position in which we can ask how the thick
‘combines’ these two sorts of meaning, and we have two possible answers, the
disentanglers’ answer and the entanglers’ answer.
Finally, we notice the way in which the thick concepts are understood as, as one
might put it, tending towards the thin. The paradigm thin concepts are good, bad,
right and wrong. Thick concepts have that sort of meaning in them added to some
descriptive meaning. One wields a thick concept en route to a final, overall
evaluation of the for/against sort. This sense in which the thin dominates was
called ‘centralism’ by Susan Hurley, and the name has stuck. She wrote: ‘What
makes an account centralist is the conceptual independence and priority it allows
to the general evaluative component in relation to specific reason-giving concepts’
(1989: 13). And this is just what we see in the standard picture of the relation
  201
between the thick and the thin. The thin is conceptually independent of the thick
and the thick is not so independent of the thin. We can picture the situation as
I have characterized it so far in Figure 1. The triangle in the middle is the area of
the (decreasingly) thick. The amount of descriptive meaning diminishes to zero,
and at that point we encounter the thin. It has to be confessed, however, that it is
not obvious why the line representing the thin has any significant length at all. For
all thin concepts are, according to the assumptions, equally thin.
Descriptive meaning
Evaluative meaning
Figure 1
Descriptive meaning
Evaluative meaning
Figure 2
Perhaps the diagram would look better as in Figure 2. Here the length of the line
representing the thin has been reduced almost to nothing; I wanted it to be almost
invisible, but it still needs to be there. One could have similar worries about the
long line representing those concepts that are neither thick nor thin. It is not easy
to find an appropriate (that is, non-contentious) term for these concepts. We seem
to have to choose between ‘natural’ and ‘descriptive’. Both have their difficulties.
First, it is not obvious that all these concepts can usefully be called ‘natural’
concepts. Clearly this depends on how one characterizes the natural, but if one
took the standard approach and said that the natural is the domain of the natural
sciences, there will be plenty of concepts that are neither thin nor thick nor
natural. This does not mean that ‘descriptive’ is much better. The term ‘descrip-
tive’ must get whatever sense it has from what it is to describe—which is
something that people do, rather than concepts or terms. And it is not at all
obvious that we misuse the notion of describing when we say that he described his
star pupil as outstanding, excellent in every way, though even stronger in criticism
than in invention. My solution to this terminological nightmare is to contrast the
evaluative simply with the non-evaluative. This choice leaves it open to me to
divide the left-hand line into two parts, the natural on the left and the non-natural
on the right, should that turn out to be worthwhile; to do this would of course be
to adopt a conception of the natural that is narrower than is usual.
202  
Non-evaluative meaning
Evaluative meaning
Figure 3
Non-evaluative meaning
Evaluative meaning
Figure 4
So my diagram ends up looking like Figure 3. When I presented this diagram at

the conference,³ it was suggested that some thin concepts have some non-
evaluative meaning. The example offered (by Pekka Väyrynen) was that ‘ought’
implies ‘can’, but not all thin concepts imply ‘can’; perhaps ‘is fitting to’ does not,
and maybe ‘is a reason for’ does not either. Now plainly, if some thin concepts
have non-evaluative meaning, the diagram needs to be rethought entirely. And the
general idea need not depend upon this contentious example. Michael Smith, in
his paper at the conference,⁴ argued that the distinction between thick and thin is
indeed a matter of degree, a sliding scale of evaluative concepts taking us from the
least specific at the far right to the most specific where we hit the vertical line as we
move towards the left. So on his view there is no such thing as the sort of purely
evaluative meaning standardly ascribed to thin concept, and the diagram should
look like Figure 4.
In the remainder of this section, I respond to these suggestions. First, the
argument that some thin concepts have non-evaluative meaning relies on the
view that something non-evaluative can be inferred from the applicability of a thin
concept. But this would not be enough to show that the thing inferred is itself part
of the concept. It is in fact certain that many things that can be inferred in this way
are not themselves part of the concept they are inferred from. For if they were (and
this gives some sense to the dubious notion of ‘part of a concept’), competence
with that concept would require competence with the supposed part. But the
concept ‘is a cow’ entails an enormous number of other concepts, such as ‘is not a
horse, ‘is not a mountain’, ‘is not the Aga Khan’, and so on; and these things
entailed cannot sensibly be supposed to be part of the concept ‘is a cow’, on pain of
rendering that concept unlearnable.
The question then is whether this implication is part of the concept of ‘ought’
and not part of the concept of ‘reason’. I would say that the contentiousness of the
implication is evidence that accepting the inference is no part of competence with
³ ‘Thick Concepts’ conference held in July 2009 at the University of Kent. ⁴ See Smith (2013).
  203
the concept, since neither side in the debate accuses the other of conceptual
incompetence. For what it is worth, then, I am not convinced by the examples
that are supposed to show that some thin concepts have non-evaluative content.
What is more, even if some thin concepts are a little thick, there will be some that
are not (such as the concept of the fitting, which is allowed not to entail ‘can’); so
the idea of a concept that is purely thin will survive. But Michael Smith’s points are
quite general, and do not depend on particular examples, and if correct would
undermine the very idea of the ‘purely’ thin. For him, there is only the distinction
between more or less specific.
There are two parts to Smith’s paper, and I am going to take them in reverse
order. The second part concerns Susan Hurley’s anti-centralism. As Hurley
characterizes this position, it amounts to the denial that ‘the general concepts
right and ought are logically prior to and independent of specific reason-giving
concepts such as just and unkind’ (1985: 56). Of course this does not reveal what
logical priority is intended to mean. But it is clear that this denial is compatible
with two positions. The first asserts the reverse priority, and the second simply
denies any priority. I take Hurley to have intended the second of these. She
contrasts a centralist account, not with the sort of view we associate with
Bernard Williams, who definitely ascribed priority to the thick, but with what
she called a ‘coherence account’. The passage from her Natural Reasons which
I quoted earlier continues: ‘A coherence account, by contrast, would hold the
general and the specific concepts to be conceptually interdependent’ (1989: 13).
But Smith supposes her to be committed to the reverse priority, because of the
following remark: ‘According to [non-centralism], to say that a certain act ought
to be done is to say that it is favoured by the best account of the relationships
among the specific values that apply to the alternatives in question’ (Hurley 1985:
56). Smith understands this sentence as saying that the concept of what ought to
be done is a theoretical construct out of the values that we take to be relevant to
whether something ought or ought not to be done. It must be a theoretical
construct because on the non-centralist picture there is no prior concept of
what ought to be done. But if so, he says, two people who operate with different
sets of specific values could not even be said to disagree about what ought to be
done, since they do not have the same concept. One’s concept of ‘ought’ is built
out of the values that one takes to be relevant to determining ought-ness; so people
with different values must have different concepts.
This is, however, a mistake. The last sentence of Hurley’s quoted above can be
understood as offering an account of the concept of ‘ought’ that is entirely formal,
as we see if we insert a caveat in it thus: ‘According to [non-centralism], to say that
a certain act ought to be done is to say that it is favoured by the best account of the
relationships among the specific values that apply to the alternatives in question
(whatever they may be).’ With this reading we can still respect the distinction
between conception and concept. People who disagree in the values that they take
204  
to determine what ought to be done can still be said to have the same concept, and
hence to be disagreeing about something, even though they have different con-
ceptions. The difference in conception is not in danger of making a difference in
concept, as it would be if one’s concept was a theoretical construct out of one’s
conception.
Turning now to the body of Smith’s paper, I am going to apply the distinction
between concept and conception to the way in which Smith lays out a certain
debate between Hare and his opponents. Hare starts by saying that calling an
action right commits one to accepting a self-directed imperative instructing one to
do any action that shares with the first action the features that led one to call it
right. It was complained against him that not all such self-directed imperatives
could be thought of as moral imperatives; morality has a certain restricted subject
matter, and moral imperatives must relate to that. Foot, for instance, urged that
moral requirements must have something to do with benefit and harm. Suppose
we accept this amendment to Hare’s picture. We now have a more specific picture
of what moral judgement amounts to. And we could continue in the same vein,
saying that considerations of justice and fairness are also relevant, or that consid-
erations of simple identity are not to be relevant, ruling things out or in as we go
along; but all we are doing is rendering more specific the thing we started with. But
also, as we go along, we will be introducing the sorts of considerations that are
standardly located within the thick: justice, fairness, impartiality, tolerance, and so
on. So the difference between thin and thick is really nothing but a simple
distinction between the more and the less specific. There is no difference that is
not a matter of degree.
My response to all this is that we start with a concept of the right, namely of those
responses that we have most moral reason to perform. This concept is thin, as is the
concept of a reason that figures in it, and the broader concept of relevance that is
lurking somewhere there too. What we then do is to develop a conception of the
things that are morally relevant, which I take to be an elaboration of the moral
domain as we conceive it. We thereby develop a conception of what counts morally
in favour of what, and of the sorts of actions that are right or wrong. The increasing
thickness of our conception of the morally relevant does nothing to undermine the
pure thinness of our concept of the right, of reasons and of relevance.
My conclusion at the moment is that the distinction between thin and thick is
not a matter of degree, but of type.
I ‘solved’ the interpenetration problem, on behalf of the entanglers, by appeal to

the no-priority view, which I still think gives a good sense to the idea that property
and attitude cannot be disentangled. But I confess to serious doubts about whether
  205
that view can be used to give a general account of thick concepts. Once we have a
suitable attitude in place, and a suitable property, we get a nice picture. But it may
not be right to think of all thick concepts as explicable in these terms. As I said
above, the no-priority view as I expressed it was only formal. And it seems to be
an open question whether all the concepts we want to think of as thick are capable
of being explicated in its terms. So the question is: does the no-priority view
generalize?
The only way to answer this question is to throw examples at it. The best case
for the no-priority view, I think, is unsurprisingly the one that Wiggins and
McDowell most often appeal to. This is the concept of the amusing, or funny.
An amusing joke is one that merits a certain response, that of amusement or
perhaps laughter; this is the property of being amusing. And when one is amused,
one is amused by the object as meriting that amused response. The amusing joke
calls for amusement; it says, as it were, ‘laugh at me’. Of course one can be amused
by a joke that is not funny. But this is a defective instance of the response.
This central example will only work if we can sustain the idea that our amuse-
ment is a response to the object as meriting just that response. And I have to confess
that it is not absolutely clear to me that this account is true to the phenomenology of
amusedness.⁵ But I leave that aside, because my main focus here is on the question
whether an account of this style can be given across the board.
Wiggins, in his presentation of these themes (1987), restricts his attention to
two classes of concept. The first class is what one might call the ‘-ings’: the
amusing, exciting, enticing, frightening, engaging, interesting, disturbing, alarm-
ing, challenging, and so on. We might include in this class the attractive, suggest-
ive, provocative, impressive, festive, and other such; also the respectable,
admirable, acceptable, understandable, desirable and lovable. In all these cases it
seems that things fit the pattern required. This is because in each case the
distinction between property and response is already given; there is a distinct
response-type served up by the concept. We are enticed by the enticing, disturbed
by the disturbing, amused by the amusing, attracted by the attractive, provoked by
the provocative and alarmed by the alarming. We desire the desirable and fear the
terrifying. Here, then, things work well.
Wiggins also considers thin concepts: the good and the right, just for a start.
And here again things work well, because with the thin we are dealing with overall
approval and disapproval. These are responses, and they may come in different
forms; but at the level of the thin those different forms will need to be amalgam-
ated into general-pro and general-con.
But there seem to be many thick concepts that do not work in this kind of way,
since no clear response is there to be appealed to. In saying this, I am of course
⁵ For a then recent expression of these doubts, see Dancy (2012).

206  
vulnerable to the charge that some of the examples I am about to give are not
really examples of the thick at all. My response can only be that there are just too
many of them for this style of rebuttal to be convincing overall. Consider then the
list of thick concepts that Samuel Scheffler gave in his 1987 review of Williams’s
Ethics and the Limits of Philosophy:
(1) justice, fairness, impartiality

(2) privacy, self-respect, envy
(3) needs, well-being, interests
(4) autonomy, consent
And here is another list:
(1) relevant, rational

(2) turn, share, help, promise
(3) subtle, distinguished, sensible, responsive, fine, gross, coarse, balanced,
lop-sided, harmonious, flabby
(4) imaginative, resourceful, inventive, dull
In very few of these cases is it easy to fix on a response that is served up by the

concept in the sort of way that we have seen. Take the concept of a turn, as in ‘It is
not your turn’. There is the institution of waiting one’s turn, which comes with
respect for the fact that it is not one’s turn but someone else’s, not my turn but
yours. It also comes with the practice of not pushing oneself forward, not going to
the front of the queue, and so on; these are practical responses, but they are still
responses. Similar things could be said of privacy, for instance. But the trouble is
that the only way of making sure that there is a suitable response in each case is to
read the response off the property. Take subtlety. What is the appropriate
response to that? The only one I can think of is recognition of the subtlety, but
neither that nor the sort of admiration that appropriate subtlety attracts seems to
play the sort of role that the no-priority view adduces. It seems that in these cases,
and in many others in the lists above, the property is prior to the response. Our
response to subtlety is not a response to the subtle object as meriting our response,
though if it is subtle it does merit that response. Its subtlety is not its meriting the
appropriate response, but a feature in virtue of which it merits that response.
These are the reasons for thinking that the no-priority view does not generalize
in the way required if it is to stand as the basis for a general account of the thick.
Have I somehow cheated here by imposing invariability or the single-attitude
view on the discussion? Would things look any different if we adopt the positions
I suggested above, understanding thick concepts as multi-attitude concepts and as
variable in their practical relevance. I don’t see that they would. One’s response to
subtlety can be as mixed as you like: it is still a response to a property in the object
  207
that is prior to that response. Maybe, again, some things are ruined by being subtle
(which is not the same as being too subtle—sometimes subtlety is exactly what is
not wanted); but this does not seem to change the situation either.
I conclude that the no-priority view cannot do everything that it is being asked
to do.
Another aspect of the interpenetration problem for entanglers was the need to say
something about the different non-evaluative meanings of different thick terms.
‘Elegant’ and ‘clumsy’ are both thick, and if we were invariabilists we would say
that one is a pro-concept and the other a con-concept. But they clearly carry
different ‘non-evaluative meaning’, no matter how wary one is of that notion.
I tried to deal with this difficulty by appeal to Aristotelian domains: the courage-
ous is the commendable (roughly) in the domain of responses to fear and danger.
And that manoeuvre works well in general for the virtues, so long as we accept the
Aristotelian way of distinguishing them, which I am not concerned here to
dispute. But again it is not obvious how to make similar remarks across the
board. One needs to be able to offer an independent specification of the relevant
domain; otherwise nothing is gained. I would say that we can do this for one of
those on Scheffler’s list, impartiality, and for some on mine, such as resourceful-
ness, which is something like skill at finding solutions to problems; but then these
two look very much like virtues. (It helps that they are predicated of people.) But
what about the harmonious? This looks very much like that which is to be
commended so far as harmony is concerned. And what is the domain for turns?
Perhaps it is the domain of deciding who is to go next when there is competition
for precedence. But this doesn’t look sufficiently independent.
The interpenetration problem is proving very recalcitrant for entanglers. Their

attempts to give a general characterization of the glue that holds what we are
calling non-evaluative and evaluative meaning together in a thick concept are
not meeting with much success. And I know of no further resource beyond
those we have already tried and found wanting. A new start is needed. And
what one tends to do in such cases is to examine as carefully as possible the
contrasts in terms of which the debate has been posed, to see if one can unearth
some challengeable presupposition in them. What are those contrasts? Here
are some: descriptive vs. evaluative, property vs. attitude, attitude vs. response,
concept vs. property.
208  
Now my general feeling is that the debate about entangling and disentangling
has been conducted in ways that reflect the main preoccupation of early writers in
the field, namely the battle between cognitivism and non-cognitivism in value
theory and ethics. It was that preoccupation that led to matters being presented in
terms of property and attitude or response. The no-priority view was itself first
presented as a weapon against non-cognitivism. Non-cognitivists, primarily
Blackburn, supposedly maintained that the attitude of amusement, say, had the
priority, and the so-called property of funniness was a construct out of that.
Considered solely as a response to non-cognitivism, the no-priority view had no
need to generalize. All it needed was a few good cases where the non-cognitivist
account could not be got to work. If it could be run for the virtues and for the
various concepts that I listed in Section 2 above, it had no further pretensions and
could be deemed successful in its own terms. It was not primarily intended as a
contribution to an independent debate about thick concepts in general.
It now seems to me that we would do better to go even further back, to Bernard
Williams’s presentation of the thick/thin distinction in his Ethics and the Limits of
Philosophy. The terms that Williams appealed to are quite different from the ones
that we have come to use in our thinking about the thick. For Williams, thick
concepts are world-guided and action-guiding. Non-evaluative concepts are only
world-guided, and thin concepts are not world-guided at all.
I propose to try to make sense of this distinction. Note, first, that if it can be
done it avoids entirely any worries about the glue that holds non-evaluative and
evaluative meaning together in one concept. In place of such worries, we get
instead a worry about the relation between being world-guided and being action-
guiding in those concepts that are both. And it remains to be seen whether this
new (or rather old) worry is any more tractable than the one that it is intended to
replace (or that replaced it). The general idea, then, is to abandon the attempt to
knit property and attitude together, and to try to make sense of the view that some
concepts are intrinsically action-guiding, and that some of those are world-
guided too.
Nobody could maintain that the contrast between world-guided and action-
guiding is terribly happy as it stands. The obvious questions are whether the sort
of guiding concerned is the same both times, and what is meant by ‘the world’
here. It looks as if some prejudicial conception of the world may be at issue. But in
fact, as Williams uses it, this is not so. He writes, quite simply: ‘At the same time,
their application is guided by the world. A concept of this sort may be rightly or
wrongly applied, and people who have acquired it can agree that it applies or fails
to apply to some new situation’ (1985: 141). This sort of ‘guided by’ is not very
different from ‘answerable to’. But when we ask what is meant by ‘the world’ we
get a richer picture. It sounds initially as if ‘the world’ is supposed to mean ‘what
there is anyway’—a sort of nod towards the absolute conception which figures so
largely in Williams’ thought. But in fact he has two conceptions of ‘the world’. The
  209
first is that of the world absolutely conceived, the world of primary qualities
(roughly). The second is the human world, the world of desires, attitudes and
needs. Williams writes elsewhere:
Nevertheless, the nature of the shared practice shows that it is world guided, and
explanation will hope to show how that can be. What the explanation exactly
may be, is to be seen: but we know now that a vital part of it will lie in the desires,
attitudes, and needs that we and they have differently acquired from our different
ways of being brought into a social world. The explanation will show how, in
relation to those differences, the world can indeed guide our and their reactions.
‘The world’ in that explanation will assuredly not be characterized merely in
terms of primary qualities; the account of it will need to mention, no doubt, both
secondary qualities and straightforwardly psychological items. (1995: 186)
So the idea that our use of thick concepts is world-guided is not based on some
appeal to supervenience or anything like that. Perhaps we should not be spending
time on the notion of world-guidedness anyway, because Williams’s real use of
that notion lies in his denial that application of thin concepts is world-guided, and
I do not want to go down that path at all. The idea that where non-evaluative
meaning lapses we lose our sense that concepts of this sort can be rightly or
wrongly applied, and our sense that those competent with it can agree on its
application to new cases, is not one that commends itself to me. In my view, both
thick and thin concepts are world-guided and action-guiding. The difference
between thick and thin lies elsewhere.
So for me the real issue is how to characterize the action-guiding nature of thick
concepts. Somehow we have to make sense of the idea that these concepts have an
intrinsic practical relevance. Other, non-evaluative concepts may be such that
their applicability makes a practical difference on occasion, but that relevance is
extrinsic to them. It may make a difference that this car goes faster or uses less
petrol than that one, but that difference is not intrinsic to those features.
It would be a mistake to read these initial remarks about having intrinsic
practical relevance as remarks about being necessarily relevant. Only the thin
concepts are necessarily practically relevant. The practical relevance of the thick
must be consistent with the variability that I am persuaded is a feature of many
thick concepts, and variability includes not only the idea that the applicability of
the concept can make things now better and now worse, but also the idea that it
may on occasion make no difference at all. The point has to be that in such a case
there is something to explain, which there is not in the case of the non-evaluative
concepts. By contrast, when the instantiation of a thick concept does make a
practical difference, there is as yet nothing to be explained.
We explain the relevance of a non-evaluative concept by appeal to the co-
instantiation of some other concept. (In less clumsy terms, the relevance of a
210  
non-evaluative property will be explained by the co-presence of some other

property.) This sort of explanation is not necessary when a thick concept has
practical relevance, for such a thing is to be expected. This is what makes it
possible to think of these concepts as practical concepts. And competence with
such concepts will be practical competence, since it consists in knowledge of the
sort of difference it can make that the concept is here instantiated. This sort of
knowledge brings with it the ability to tell one case from another in this respect;
the competence is not just an ability to determine whether the concept is instan-
tiated or not, but also the ability to determine what difference this makes on the
present occasion. By contrast, competence with a non-evaluative concept requires
only the former ability.
I said, however, that I intend this picture to be compatible with the variabilist
picture that I adumbrated at the start of this paper. And here there is a difficulty
for me. For suppose that a thick concept is here instantiated, and that it makes a
difference. Still, as variabilism has it, the instantiation of that concept might have
made a different difference, since it is capable of making any of a range of
differences. Which difference it makes here will surely need to be explained by
appeal to other features of the case. But haven’t I said that if that is so, we are
dealing with a non-evaluative concept?
Matters would be manageable if every thick concept had a default relevance.
For in that case, those instances where the actual relevance is the default relevance
are ones where no further explanation is needed, and that is all that is required to
distinguish the thick concepts from non-evaluative ones. But I doubt that it is true
that all thick concepts have a default relevance in the way required.
Luckily there is another way to distinguish the thick from the purely non-
evaluative. For the instantiation of a thick concept may be practically relevant in a
way that does not depend upon the co-instantiation of other concepts, even if the
actual relevance that it has is so dependent. That is to say, that it is relevant at all
needs no further explanation of that sort, even though the actual relevance it has
may do so. And this is enough for the purpose of distinguishing the intrinsic
practical relevance of a thick concept from the extrinsic practical relevance of the
purely non-evaluative.
If we think in these terms, we make room for those thick concepts, such as a turn
and a share, where the identification of a suitable attitude is rather a challenge, and
undermines the general programme of characterizing the thick in terms of some
combination of property and attitude, held together by some suitable sort of glue.
But we can also allow a special place for the sort of thick concept that has dominated
recent discussion, the sort that does indeed hold together property and attitude, or
property and response. The no-priority view still seems to get it right about these
concepts, but they are special in a way that cannot be extended to the thin (for
example, the fitting), and that does little to sustain the sort of sentimentalism that
Wiggins wanted to sustain for the evaluative quite generally.
  211
I suggest, then, that thick concepts are those with an intrinsic practical rele-
vance. But perhaps not all relevant relevance is practical relevance. Simon Kirchin
asked me about aesthetic concepts; these are often used in the appreciation of art
rather than in the making of it. Do they have an intrinsic practical relevance? Yes,
they do, because they are relevant to the choices of the artist. So there is a way out
here, but still one might think that placing all the weight on the practical is a
distortion, and will lead to a lop-sided conception of the thick. A different
suggestion is that we should return again to Williams, and resuscitate his idea
that thick concepts are more generally those that have an evaluative point.
Miranda Fricker suggested this change of emphasis to me. Is there any reason to
resist it? The notion of an evaluative point is perhaps less clear, less well focused
than that of practical relevance. But this may be an advantage. We could say that
competence with a thick concept requires an understanding of its evaluative point,
and that this includes, but is not restricted to, a grasp of the sorts of practical
difference it can make that the concept is here instantiated.⁶
In this final section, I return to the topic of variability. Why have theorists been so
tempted to suppose that thick concepts have an invariant relevance? I can think of
three explanations for this. The first is simple. It is that, as commonly happens in
philosophy, we work with a very limited range of examples. I am sure that this is
part of the story, but it is not the only part, nor the most respectable. Another
factor is a tendency to concentrate on the totalizing distinction between pro and
con. I think of this tendency as a sign of a sort of centralism. Evaluative concepts
are in the business of making cases for or against responses. That being so, they
must carry elements of pro or con with them, to contribute to the mix. But I think
this is a mistaken picture. Evaluative concepts are concepts with an evaluative
point, and this is perfectly compatible with their not having elements of pro or of
con in them. There is the appropriately lewd and there is the inappropriately
lewd, the deliciously lewd and the unattractively lewd, and the fact that lewdness
comes in all these forms does nothing to show that it is not a concept with an
evaluative point.
Still one might think that there is some a priori consideration operating here in
favour of invariabilism, and I end by considering one pressed on me by Michael
Smith in conversation. He asked how it is possible for variabilists to explain the
evaluative nature of thick concepts. There are two standard ways of doing that.
The first says that thick concepts are evaluative because they entail a pro tanto
⁶ Both Kirchin’s and Fricker’s comments were made at the 2009 ‘Thick Concepts’ conference.
212  
evaluation. But this is what we have seen already, and it is not available to
variabilists. The second says that thick concepts are evaluative because an ability
to determine, or notice, instances of them requires what one might call an
evaluative eye, an understanding of human practical purposes. The problem
with this suggestion is that it seems possible for a non-evaluative concept to
require the evaluative eye. I say that this might be so; I don’t have a fully persuasive
example. But one that has always appealed to me is Margaret Little’s example of
the child alone in a crowd. Loneliness is, I would say, a thick concept, but
aloneness might not be, and one can imagine saying that it is a non-evaluative
matter whether the child is accompanied or not, but that the ability to notice such
a thing requires an understanding of human practical purposes. I’m not sure of
this; one might say instead that what one notices is that the child is not appro-
priately accompanied, or accompanied by an appropriate person. Uncertainty
here leads me to think that the second option is not reliable either, and variabilists
need to look elsewhere.
My positive suggestion really appeals to points made earlier in this paper. It is
that thick concepts are evaluative because competence with them requires a
general understanding of their evaluative point, including the range of their
practical relevance, the sorts of difference it can make that the concept is here
instantiated. Thick concepts differ from non-evaluative concepts and from thin
concepts in this respect.
14
Should We Pass the Buck?
My topic is the relation between reasons and values, and that between the right
and the good. I introduce these matters by relating some aspects of the debate
between various British intuitionists in the first half of the twentieth century.
In Principia Ethica (1903), G. E. Moore claimed that to be right is to be productive

of the greatest good. He wrote ‘This use of “right”, as denoting what is good as a
means, whether or not it be also good as an end, is indeed the use to which I shall
confine the word’ (p. 18). By the time he wrote his Ethics (1912: e.g., p. 6) he seems
to have weakened his position, and offers conduciveness to the good not as a
definition of ‘right’ but as an account of the one and only property that makes acts
right. Even if it be the only right-making property, conduciveness to the good will
not be identical with the rightness that it makes.
One might ask why Moore changed his view, and an obvious answer is that he
came to see that the notorious Open Question Argument, by which he strove to
establish that goodness is not identical with any good-making feature, can be used
equally well to show the same thing of rightness. If it is an open question whether
goodness is conduciveness to happiness, it is equally an open question whether
rightness is conduciveness to goodness. And if, as Moore claimed in the first case,
its being an open question shows that the answer to it is no, the same applies in the
second case.
W. D. Ross details all these matters with further references, and with his
customary clarity, in the early pages of his The Right and the Good (1930: 8–11).
Though he argues that Moore’s second view is a vast improvement on his first,
Ross’s own position is quite different. Rightness and goodness are utterly distinct;
indeed, no one thing can be both right and good. Goodness, for Ross, is a property
of motives and outcomes, and rightness is a property of acts. An act can be
intrinsically right or wrong, but never intrinsically good or bad. Acts can be
instrumentally good, or conducive to good; but Ross announces, surely correctly,
that instrumental value is not a form of value at all. So it turns out that acts can
have no value at all. Motives, by contrast, can be intrinsically good or bad but
never right or wrong. On this picture the very idea that one might define the right
in terms of the good is quite peculiar. Also peculiar is the idea that the only way
214  
that an action could get to be right is by having the best consequences, since that
would prevent actions from ever being intrinsically right.
H. W. B. Joseph’s response to this in his Some Problems in Ethics (1931) was
that Ross’s position was absurd. ‘Why’, he asked, ‘ought I to do that, the doing [of]
which has no value (though my being moved to do it by the consciousness that
I ought, has), and which being done causes nothing to be which has value? Is not
duty in such a case irrational?’ (p. 26). Rightness, for Joseph, must be in some way
dependent on goodness, whereas on Ross’s account the two properties float more
or less free of each other. Joseph pursued this idea by claiming that the word ‘right’
is ambiguous. In one sense it means ‘obligatory’, and Joseph writes (1931: 61) that
‘obligatoriness is not a character of actions. There is no ought-to-be-done-ness, or
ought-to-be-forborne-ness. To say that an act is obligatory means that the doing it
is obligatory on me.’ In the other sense, ‘rightness is a form of goodness, to the
realizing of which the actions belong; and it is the thought of this goodness which
moves us when we do an action from a sense of obligation’ (1931: 104). In this way
Joseph rejects Ross’s claim that rightness and goodness are utterly distinct.
In his second book, The Foundations of Ethics (1939), Ross’s position becomes
more complex. He has already argued that there are two uses of ‘good’, attributive
and predicative. The attributive use is at issue when we speak of a good liar or a
good knife. The predicative use is the one that is of importance for ethics, and it is
found when we speak of a good man, or claim that virtue, knowledge and pleasure
are good. Ross claims that when we say that the pleasure of others is good, we
mean that it is a proper object of satisfaction. This is a ‘definition’ of this use of
‘good’. When we speak of a man, or of a motive, as being morally good, however,
we mean something else, something that cannot be defined but only paraphrased
(1939: 283). The paraphrase is that the good, in this use of ‘good’, is a proper
object of approval, worthy of approval or admiration.
Why is this not a definition? Because Ross is still sticking to his original view
that goodness, in this sense, is an intrinsic property of objects, not a relation. If
being good in this sense were being worthy of approval or admiration, it would be
a relation. But it is not; this sort of goodness is the property that in approving or
admiring we take the object approved or admired to have. For to approve is to
think good, and ‘admiration is not a mere emotion; it is an emotion accompanied
by the thought that that which is admired is good’ (1939: 278–9).
It is worth pausing to note what Ross means by a relation here. It is not what we
would ordinarily mean, because we would ordinarily think that for a relation to
obtain, there must be at least two relata and both must, in the relevant sense, exist.
But in suggesting (even if only to reject the idea) that goodness might be a relation,
Ross is clearly not thinking of relations in this way. For something can be worthy
     ? 215
of approval even if no approval and no approver is forthcoming.¹ Further,

although Ross thinks that goodness in this sense is not a relation, that is not his
reason for rejecting the claim that goodness is identical with being worthy of
admiration and approval. Ross’s real point is that the goodness that we take the
object to have cannot be identical with its being worthy of our so taking it, because
it must be that in the object that makes our so taking it an appropriate or fitting
response.
It would have been possible to avoid this result if we had been more catholic in
our choice of the attitudes or responses that the good action is worthy to elicit. In
his The Definition of Good (1947), A. C. Ewing defines the good as that which
ought to be the object of a pro-attitude (pp. 148–9). He attributes the term ‘pro-
attitude’ to Ross, and continues:
‘[P]ro-attitude’ is intended to cover any favourable attitude to something. It

covers, for instance, choice, desire, liking, pursuit, approval, admiration . . .
When something is intrinsically good, it is (other things being equal) something
that on its own account we ought to welcome, rejoice in if it exists, seek to
produce if it does not exist. We ought to approve its attainment, count its loss a
deprivation, hope for and not dread its coming if this is likely, avoid what hinders
its production, etc. (1947: 149)
By stressing the broad variety of attitudes that may count as pro-attitudes,

Ewing seems to avoid Ross’s only argument that we cannot define the good (in the
relevant predicative sense) as what is worthy of a certain response. Ewing seems,
that is, to be in a position to say that goodness is not a distinct evaluative and
intrinsic property in objects, one whose presence we can discern and to which we
do, or at least should, respond with approval and admiration. The goodness of the
object just is the relational fact that we should respond to it with approval,
admiration or other pro-attitude. The evaluative ‘good’ has been defined in
terms of the deontic ‘should’. And with this result, the intuitionists reversed
Moore’s position, the position with which I started this brief history. Moore
defined the right, that which we ought to do or should do, in terms of the good.
Ewing defined the good in terms of how we should respond.
This is the end of my brief historical introduction. At the point we have reached
(the late 1940s), the intuitionists’ broadly cognitivist approach to ethics was
eclipsed by the onrush of non-cognitivism. Fifty years passed before these sorts
of issues could again be debated with any sense of seriousness.
Fifty years on, Thomas Scanlon writes as follows in his What We Owe to Each
Other:
¹ It is just possible that Ross thinks of the relevant relation as ‘being worthy of our approval’, which
would bring him back into line on this point.
216  
To value something is to take oneself to have reasons for holding certain positive
attitudes toward it and for acting in certain ways in regard to it. . . . To say that
something is valuable is to say that others also have reason to value it, as you
do. . . . [T]his account . . . takes goodness and value to be . . . the purely formal,
higher-order properties of having some lower-order properties that provide
reasons of the relevant kind. [It holds] that it is not goodness or value itself
that provides reasons but rather other properties that do so. For this reason I call
it a buck-passing account. (1998: 95–7)
My question is whether we should in this way seek to pass the buck. Should we
take it that to be valuable is to have features that give us reasons, in the way that
Scanlon suggests?
II
I start with some comments on the buck-passing view as expressed above by

Scanlon. First, I think there must be a slip in the way that Scanlon tries to capture
the buck-passing account of taking something to be valuable. It cannot be right to
say that to take something to be valuable is to take it that others also have reason
to value it, as you do. For in valuing it we do not take ourselves to have reasons to
value it; at least, not on the buck-passing view. On that view, to value it is to take
oneself to have reasons of certain other sorts. The quotation I gave above does not
reveal this, but Scanlon does detail these sorts of reason. He talks of reasons for
admiring, respecting, preserving and protecting; of reasons to be guided by the
goals or standards that the value involves; of reasons for promoting; and of
reasons to act in certain ways. To value something is to take oneself to have
reasons of these sorts, reasons that are given one by features of the object (or
perhaps more broadly of the situation) which act as ground for the reasons and, in
doing so, as ground for the value. But when we value the object, we are not taking
ourselves to have reason to value it, exactly. So when we come to think of the
valuable, rather than just of the valued, we should surely be thinking of others as
having the same reasons as we do, reasons to admire, protect, promote, etc.
With that slip corrected, Scanlon’s view is very similar to Ewing’s. Scanlon’s is
expressed in terms of reasons, while Ewing spoke of how we ought to respond. But
a definition of value in terms of reasons is very similar to one in terms of oughts,
especially when we notice that Ewing was speaking of prima facie oughts, and
Scanlon is speaking of what one might call contributory reasons. Crucially, the
notion of what one ought to do is a deontic notion, as I take that of a reason to be,
and the notion of a value is evaluative; Ewing and Scanlon combine to give a
deontic definition of the evaluative, and this is what is at once attractive and
worrying about their position.
     ? 217
What I mean by this is that we should not forget the position they are trying to
undermine. We might think that there are just two families of normative concepts.
The first family is the evaluative family. Here we are dealing with the good and the
bad, the noble, the fine and the evil. Second is the deontic family. Here we are
dealing with the right and the wrong, shoulds, oughts and musts, obligations,
requirements and prohibitions. Ross took the view that these two families are
utterly different. His picture of the normative realm is therefore far more complex
than is that of Ewing or of Scanlon—or of Moore, of course, at the other end.
Complexity has its advantages and its disadvantages, as we will see. But we may be
permitted to hope that the Ewing/Scanlon view, the buck-passing view, has more
than the advantages of simplicity. If we are to accept it, there must be more to be
said in its favour than just that.
And of course there is. At least, there is if Scanlon is to be believed. He argues
for the buck-passing view by arguing against two alternative positions, supposing
that, if they fail, the buck-passing view will emerge as the sole remaining con-
tender. Unfortunately, I will now suggest, he does not consider every available
alternative. So his argument by remainder ends up looking a bit weak.
The opposing views, according to Scanlon, are (1) the ‘teleological’ view, and
(2) the view that value is a property the presence of which grounds or explains
reasons. Interestingly, these two views look very much like the earlier and the later
views of Moore. In Moore’s terms the ‘teleological’ view is the view that rightness
is to be defined in terms of value rather than the other way round. The second
view is the view that value is the right-making property, though not identical with
rightness. Expressed in terms of the relation between values and reasons, rather
than between value and rightness, the teleological view will be that reasons are to
be defined in terms of some relation to value, and the second view will be that
goodness is a (or the) reason-giving property. Now I accept that both of these
views are wrong. We abandon the teleological view because we suppose that we
cannon define reasons in terms of values, whatever the proper account of the
relation between reasons and values turns out to be. We abandon the second view
because we recognize immediately that the badness of a toothache, for example,
does not add a further reason to the reasons for going to the dentist that are
already given by the nature of the toothache, its painfulness. As we might put it,
the badness of the toothache exists in virtue of certain features, features that give
us reason to act in certain ways; the badness of the ache adds nothing to the
reasons given us by those lower-level features. In short, value adds no reasons to
those generated by the ground for that value.
Does it follow from this that to be of value is just to have reason-giving features?
I don’t see that it does. Two further views remain unrefuted. The first is Ross’s
view, expressed in terms of the relation between reasons and value rather than in
terms of the relation between rightness and goodness. Values and reasons are just
utterly different, and neither can be defined in terms of the other. This was clear,
218  
for Ross, because the things that are duties have no value at all, and the things that
have values cannot have duties. There is indeed a link between values and reasons,
since where there is intrinsic value we always have a reason. Ross, after all, thought
that ‘one of our main duties is to produce as much that is good as possible’ (1939:
257). The prospect of value always gives us a reason, then, and this is something
that needs to be explained if values are as different from reasons as Ross thought
they were from duties. But Ross incurs this explanatory debt by adopting a picture
for which he thought there were perfectly good independent reasons.
The second view that remains is one that, without committing itself to any vast
difference between values and reasons, supposes nonetheless that the evaluative is
distinct from the deontic, even if both sorts of properties result from the same
ground. If they result from the same ground, one and the same object can have
both evaluative and deontic properties; it can be good, and can be something that
we have sufficient reason to do. This final view incurs the same explanatory debt
as Ross’s did, if it is combined with the claim that a ground for value is always (or
even therefore) a ground for reasons as well.
Here then is a map of the five possibilities I have just mentioned, where
r = reasons, v = value and g = ground, and the relation ‘!’ is the grounding
relation:
1. Early Moore (teleological view): g ! v = r (reasons defined in terms of value).

2. Later Moore: g ! v ! r (reasons grounded in but not defined in terms of value;
value adds to the reasons given by the ground).
3. Ross: g¹ ! v, g² ! r (i.e. the ground for reasons differs from the ground for
value).
4. Buck-passing View: g ! r
↑
v (i.e. being of value is having features that ground
reasons).
5. Last View: g ! v, g ! r (reasons and values are distinct but may have the same
ground).
My only point here is that one does not establish the buck-passing view by
refuting the two views of Moore. If that is right, and fair criticism of Scanlon,
the main argument in favour of the buck-passing view will be its theoretical
neatness. This neatness largely consists in the fact that it does not require us to
say why it is that wherever there is value in the offing, there are at least some
reasons. Ross’s view is vulnerable to this requirement, as is the Last View.
At this stage, I need to introduce a different topic, attitudes to which will help us
to decide between the buck-passing view and its alternatives. This topic is the
relation between ground, rightness and reasons.
     ? 219
Remember that we agreed with Scanlon that the badness of the pain cannot add
to the reasons generated by the features that make the pain bad (i.e. effectively, by
the way it feels). Now are we to say the same sort of thing about rightness and
wrongness? Are we to suppose, that is, that though goodness and badness cannot
add to the reasons coming up from below, rightness and wrongness can? Should
we suppose that the thin properties (or relations) are in general incapable of
adding to the reasons thrown up from below? It is perhaps a bit surprising that
Scanlon’s answer to this question is no. According to Scanlon, the wrongness of an
action can add to the reasons not to do it, though the badness of the action cannot.
There are the reasons given us by the features that make it wrong, and then there
are further reasons given us by the fact that it is wrong, or by its wrongness.
Even if, considering the matter simply in structural terms, we would probably
expect all the thin properties to behave as reason-givers in much the same sort of
way, that expectation could be overturned by a closer investigation of what
actually happens. Derek Parfit considers the relation between two claims:
1. This act violates standards of conduct that we all have very strong reasons to
regard as important and to follow.
2. This act gives the agent reasons to feel guilty and gives others reasons for
indignation and resentment.
Suppose now that (1) specifies what it is for an act to be wrong. Parfit continues:
‘[I]f our violating such standards would give us a reason to feel guilty, and would
give others reasons to be indignant or resentful towards us, these facts do seem to
give us further reasons not to act in this way.’²
If Parfit is right, the picture of the relation between ground, wrongness and
reasons is like this:
g!w!r
While if the alternative view is right, it is more like:
g!r=w
This is because those who think that the wrongness of an action cannot add to the
reasons against doing it that are thrown up from below, are motivated by a certain
conception of what wrongness and rightness are. We might call this the ‘verdictive
conception’.³ In deciding whether an action is right, we are trying to determine
² This quotation was taken with permission from a draft of Parfit’s then forthcoming Rediscovering
Reasons, which never appeared.
³ I take this name from Philip Stratton-Lake (2000: ch. 3).
220  
how the balance of reasons lies. Our conclusion may be that there is more reason
(or more reason of a certain sort, perhaps) to do it than not to do it, and we
express this by saying that it is therefore the right thing to do. The rightness-
judgement is verdictive; it expresses our verdict on the question how the reasons
lie. It is incoherent, in this light, to suppose that the rightness can add to the
reasons on which judgement is passed, thus, as one might say, increasing the sense
in which, or the degree to which, it is true. And the same is true of wrongness. So
when I expressed the ‘alternative view’ above as I did, I meant by this to show how
that view understands wrongness-judgements as expressing, or capturing, how
things are at the level of reasons. And I supposed that it this is correct, the
wrongness of the action cannot itself add to the reasons that it captures.
We might think of this as a buck-passing view about rightness and wrongness.
Scanlon wrote that his view ‘is called the “buck-passing view” because it takes the
normative force of a claim that something is good or valuable to be inherited from
that of the reasons which it asserts to be present’. To claim that something is good
is to claim, according to the buck-passing view, that it has features that give us (pro
tanto) reasons to take certain attitudes to it. Similarly, we might say, to claim that
an action is right is to claim that it has features that give us overall reason to do it.
This is a buck-passing view because it holds that the normative force of a claim
that something is right or wrong is inherited from that of the reasons which it
asserts to be present. So for Scanlon to adopt what I called the ‘alternative view’
about the relations between rightness and reasons would, one would have thought,
have been for him to be a consistent buck-passer.
Think now about the arguments that Parfit produced in support of the non-
buck-passing conception of rightness and wrongness. They amounted to the claim
that if the wrongness of our action would give us a reason to feel guilty, and would
give others reasons to be indignant or resentful towards us, these facts give us
further reasons not to act in this way.
Should these arguments persuade us not to be consistent buck-passers? I doubt
it. The same arguments seem to apply to badness. Suppose that my acts are bad as
well as wrong. I would say that the badness of my behaviour gives me reason to
feel ashamed, and gives others reasons to take certain attitudes towards me. Won’t
those facts then give me further reasons not to behave in those ways, if this is what
happens in the case of wrongness as Parfit suggests?
Alternatively, we could look again at the suggestion that if the wrongness of our
action would give us a reason to feel guilty, and would give others reasons to be
indignant or resentful towards us, these facts give us further reasons not to act in
this way. It would be possible to hold that the wrongness of our action does not
give us reason to feel guilty. What it does is to show that if our action is wrong, we
have reasons to feel guilty, and others have reason to feel indignant or resentful
towards us. It does not show that the wrongness of our action adds to those
reasons.
     ? 221
Be this as it may, we have now come across three possible overall views. The
first is Scanlon’s and Parfit’s. They are buck-passers about the good, but not about
the right. I have been suggesting the possibility of a more consistent buck-passing
view, which passes both bucks. But I have also aired the possibility of a view that
passes the rightness buck but not the goodness buck. This is because I have yet to
be convinced of the buck-passing view for goodness. In the final section of this
paper I try to say why.
III
The points I raise here are not supposed to constitute a refutation of the buck-passing
view for goodness. (I will call this just ‘the buck-passing view’ from now on,
forgetting any problems there may be about passing the rightness buck.). They are
supposed only to raise doubts. I try to raise doubts by pointing to significant areas of
debate in meta-ethics which the buck-passing view would automatically resolve. My
suggestion is that this is probably not the right way to resolve these issues, and that
they would be better resolved by argument than by a sort of peremptory definition.
Of course, if the buck-passing view were true, the issues would indeed be resolved,
and that would be the end of the matter. But remember that the only effective
argument we have seen for the buck-passing view is that it offers a pleasing
theoretical neatness. I like neatness, but I don’t like to see apparently significant
issues resolved by a definition whose only recommendation is that it tidies things up.
The first point here involves the contrast between consequentialism and
deontology. The buck-passing view threatens to resolve this debate in favour of
consequentialism. Deontologists have suggested in one way or another that there
are duties, and so reasons, that are not value-involving. An action can be one’s
duty even though doing it has no value and its being done generates nothing of
value. Standard examples here were of trivial duties. Suppose that I promise my
children that I will tie my right shoe lace before my left shoe lace on alternate days
of the week if they will do their homework without fuss. One can imagine arguing
that though I ought to tie my right shoe lace before my left shoe lace today, since
I did the opposite yesterday, my doing so has no value of any form. The buck-
passing view rules this out in advance. To have value is to have reason-giving
features, we are told, and since this is an identity statement it goes both ways. So to
have reason-giving features is to be of value. So the deontological view expressed
above is ruled out in advance of any significant debate.⁴
⁴ Derek Parfit suggested to me that one could avoid the thought that the buck-passing view, being an
identity claim, goes both ways by distinguishing two sorts of reasons—or, probably better, by distin-
guishing two sorts of things that reasons can be reasons to ‘do’. I suggested above that the buck-passing
view is pretty catholic about the sorts of reasons it is talking about, talking of reasons for admiring,
respecting, preserving and protecting; of reasons to be guided by the goals or standards that the value
222  
We might try to recover the issue we thought we had, by recasting the debate
between consequentialism and deontology as a dispute about whether actions are
made right solely by the value of their consequences, or whether they cannot also
be made right by their intrinsic value.⁵ This debate is still alive, even if the buck-
passing view is true, for on both sides of the issue there is the sort of relation
between reasons and value that the buck-passing view can allow. But there would
still be one position on this issue that is ruled out in advance, namely that of Ross
(and also that of Prichard, one might say). For Ross would not accept that actions
that are right are made right by their intrinsic value, or by the features that give
them that value. As I briefly pointed out at the beginning of this paper, his view
was that no duty has value of any sort, and that nothing that has value is a duty.
And he had independent philosophical reasons for the two parts of this view. Can
it be right to say that we know his view to be false because of the attractions of the
buck-passing view, when those only amount to theoretical neatness?
There is a further point of the same sort, which concerns the notion of agent-
relative value. There were in fact two reasons for the initial deontological view that
moral reasons are not essentially related to any sort of value. The first I alluded to
above, namely the possibility of valueless but still right acts. The second is a worry
about maximizing. Suppose that we allow that every right action has some sort of
value, just in virtue of being right. It should then turn out that we have reason not
to do the action ourselves if we can thereby enable two other agents to do actions
of similar value. And this is at odds with the basic deontological picture of duties.
This picture is built on a distinction between two sorts of rule. To take an example
that I think I owe to Philippa Foot: there is the rule ‘do not shout’ and there is the
rule ‘see to it that as little shouting as possible takes place’. The first of these is a
deontological rule; it is addressed to the agent, saying, as it were ‘don’t shout—this
means you’ or ‘don’t be a shouter’; and we break that rule if we shout. The second
rule is quite different. We do not necessarily break this rule by shouting; indeed,
we would possibly only be able to keep the rule by shouting, as when we need to
shout in order to shut everyone else up. In this sense, deontological rules are not
maximizing rules. You should not shout even if, by shouting, you can minimize
the incidence of shouting. A maximizing rule is not the sort of rule that deontolo-
gists take themselves to be talking about. But if we introduce a link between
reasons and values, we undermine this deontological picture of rules (obviously,
and especially, of moral rules). For we reintroduce the possibility that the value at
issue in keeping the rule is best served by breaking it.
involves; of reasons for promoting; and of reasons to act in certain ways. If some of these sorts of reason
are involved in value, as we might put it, and others are not, the identity between having reason-giving
features and being valuable will fail. I agree that this is so, but my present view is that there is no
effective way of carving the reasons up appropriately.
⁵ Philip Stratton-Lake suggested this to me.
     ? 223
There is, however, a possible way out of this difficulty. It involves the intro-
duction of agent-relative value. If the sort of value that doing one’s duty has is
agent-relative value, it might be possible to shatter the maximizing picture even
though one has retained a general connection between reasons and value of the
buck-passing sort. Now I don’t want here to go into the details of how this might
be done, if indeed it can be done at all. My point at this juncture is going to be
merely that, if we are to try to prevent the adoption of the buck-passing view from
undermining a significant aspect of deontology by introducing a conception of
agent-relative value, this is a considerable theoretical cost, and it is a cost, once
again, that we have committed ourselves to paying just for the sake of theoretical
neatness. What is more, many people doubt the coherence of the notion of agent-
relative value in the first place. If the buck-passing view can only be sustained by
introducing a piece of dubious philosophy, it is looking much less attractive.
I turn now to a completely different source of disquiet about the buck-passing
view. There is a history to this too. For Ross held that goodness is an intrinsic
property, while rightness is a relation. Rightness is the relation of being fitting to
the situation (1939: 52), while goodness is a property of motives and that is not a
relation at all. Moral goodness, in particular, is a monadic property of ‘acts of will,
desires, and emotions, and finally relatively permanent modifications of character
even when these are not being exercised’ (1939: 292). Had such goodness been
able to be defined as whatever is worthy of admiration and approval, it might have
been seen as a relation (though, as discussed earlier, this might not be quite what
we would now mean by a ‘relation’). But we saw that Ross refused to allow this
‘definition’, though he did not reject it as a paraphrase.
I don’t mean here to appeal to the detailed argumentation by which Ross defends
his position. I don’t even mean to defend his view that goodness, at least moral
goodness, is not a relation at all. The real point, I think, is one about the polyadicity of
rightness and of goodness—and this is a point that translates into thoughts about the
polyadicity of reasons. Let us allow, without asking why for the moment, that
rightness is a many-place relation. The point will then be that even if goodness is
also a many-place relation, it has fewer places than rightness does and fewer than
reasons do. Now if this is true, it cannot be correct to define goodness as the presence
of reason-giving features. For the presence of reason-giving features will have more
places in it, so to speak, than the goodness has.
The reason for supposing that goodness is less polyadic than reasons is that
reasons belong to, are for individuals. There are no reasons hanging around
waiting for someone to have them. If the situation generates a reason for action,
it must allot that reason to someone. (I don’t mean to suggest that this is always or
often difficult.) Perhaps the same reason is allotted to everyone (all have the same
reason to avoid extreme pain); but there is still an empty place in the specification
of that reason, which is filled in this case by a universally quantified variable.
Goodness, however, is not like this. Something can be good or bad without
224  
specification of an agent. The desolation or destitution of someone is bad even if

there is nobody around to do anything about it, nobody who has any opportunity
to do anything about it, and so nobody who can be said to have a reason to do
something about it. Someone’s destitution, then, has features that would ground
reasons for any agent suitably situated, but it does not follow that those features
already ground reasons. And if it does not, we can be sure that to have value or
disvalue is not itself to have reason-giving features.
One might reply that there are surely reasons for the destitute person, created
by his destitution. In that case, a better example might be the lonely and sudden
death of someone without friends or relatives, far from any possible help. We
could say that this is bad, even if there is nobody who has reasons to grieve or
indeed reasons of any other sort.
Derek Parfit suggested to me in conversation that there may be a weaker form
of the buck-passing view that is consistent with the original motivation but which is
less vulnerable to this sort of worry about polyadicity. This weaker form understands
having value, not as having reason-giving features, but as having features that are
potentially reason-giving. To have potentially reason-giving features is a less polyadic
matter than to have actually reason-giving features, since something could be of the
former sort without our needing to specify any particular individual for whom the
reasons are reasons (since the reasons don’t yet exist). This manoeuvre seems sound,
so far as that goes, but it raises worries of another sort. For it seems far too easy to
have features that are potentially reason-giving—features that would, in certain
circumstances, give us reasons. Something that has no value at all might well have
features that would, in certain circumstances, ground reasons. So this weakening of
the buck-passing view seems to me to enfeeble it.
The failure of these replies seems to me to establish that goodness and reasons
do not have the same degree of polyadicity. Is that result consistent with the buck-
passing view? I doubt it. To have features that ground reasons, where reasons are
polyadic to degree n, is, I suggest, necessarily itself a property that is polyadic to
degree n. Just having those reason-grounding features is, of course, not necessarily
polyadic at all. But that is irrelevant. Being in pain may be the reason-grounding
property, without itself being other than monadic. But that this feature is reason-
giving must itself share the polyadicity of the reasons given, since it is a feature
that can only be instantiated when the various empty spaces in the specification of
the reason have all been filled up.
IV
My overall conclusion, then, is that the buck-passing view needs more defence
than it has so far received. The buck-passing view I am talking of is, of course, the
view that to be good, or valuable, is to have features that give us reasons of certain
     ? 225
sorts. The discussion above tended to favour the other buck-passing view, the view
that for an action to be right is for it to have features that give us overall reasons
(perhaps of a certain sort) in favour of doing it. If we yield to this pressure, neither
we nor Scanlon will be consistent buck-passers. We will pass only the rightness
buck, and he passes only the goodness buck. And a challenge might here be
mounted in favour of consistent buck-passing.
In reply to that challenge, the appropriate response is to point out that rightness
and reasons are concepts of the same normative family, the deontic family. It is
not, therefore, so surprising if there are the sorts of links between them that are
involved if we pass the rightness buck. There would be no worries about poly-
adicity, for instance, for all the deontic concepts seem to be polyadic to the same
degree. Equally, the deontologists may feel that goodness does not always generate
reasons, or, more probably, that reasons do not always pass via goodness. The
maximizing worries that support this sort of thought have nothing to say against
passing the rightness buck. So I think that, although it would be prettier if we
could pass both bucks, and that there is an initial attraction in the cry ‘Both bucks
or neither!’, the terrain is not flat enough to allow us to do this.
Refusing to pass the goodness buck does nothing to reinstate something that we
all think mistaken, the thought that the value of an object adds to the reasons that
come from below, the reasons given us by the features that generate the value. This
thought is a mistake, but that does nothing to support Scanlon’s buck-passing
view. Nor can we defend that view by asking ‘How can there be value without
reasons?’ (the opposite of Joseph’s question how there can be reasons without
value). This question is irrelevant. The question before us has been whether to be
valuable is to have reason-giving properties, not whether all valuable things are
ones that have features that give us reasons to treat them in one way rather than
another. The latter is not disputed. But to allow that anything with value will have
reason-giving features is not to accept the buck-passing view. These things may go
regularly, constantly or even invariably together without being identical.
Suppose, however, that they do go invariably together. Or suppose that to be
valuable is to have what we might call ‘potentially reason-giving features’, ones set
up to give reasons for any suitable agent, should there be one. These thoughts
create a link between values, between being valuable, that is, and reasons. And we
still need to find a way of explaining this link. My final remark is the admission
that on this score the buck-passing view has an advantage, though not, I think, a
final one. All reductive views, after all, give peremptory answers to questions about
interconnections between the ‘two’ features that they reduce to one, and those
peremptory answers are often, as here, unsatisfying.⁶
⁶ Many thanks to Derek Parfit and to Philip Stratton-Lake for many discussions of these topics and
helpful criticism and suggestions; also to Roger Crisp, Jurriaan de Haan, and Irwin Goldstein for their
comments.
SECTION 3
ACTION AND REASONS
15
Arguments from Illusion
We all know about the argument from illusion. In this paper, I suggest, with
several examples, that the argument from illusion is just one example of a style or
form of argument which recurs in many other places in philosophy, with effects
which, according to one’s point of view, are either very damaging or distinctly
enlightening.
I start with a version of the argument from illusion. A state which is one of
genuine awareness of a world may be indistinguishable from one which is a mere
appearance. One cannot tell from the phenomenology of one’s awareness which it
is—an illusion or real awareness of an external reality. The argument from illusion
urges us to conclude from this that real perceptual awareness is a combination of
two elements. The first element is that which genuine perception and illusion have
in common; we could call this appearance. (If what we are dealing with is an
illusion, it is a mere appearance.) The second element is the extra bit that
perception has over and above the appearance. A theory of the following sort is
tempting: an appearance is genuine perception when it is caused by an external
object which it sufficiently resembles. Causal theories like this are theories about
what one has to add to appearance to get perception.
It would be a mistake to suppose that causal theories are all forms of represen-
tative realism, that is, theories that analyse a perceptual state as a combination of
act of awareness and inner object of awareness. Adverbial theories of the nature of
a perceptual state are (or at least can be) equally causal. At issue between adverbial
and representative theories is the nature of the state caused in us by the object
when we are perceiving it. The adverbial theory describes this state adverbially, as,
e.g., ‘my being appeared to redly’; the representative theorist describes it as an
awareness of an inner object or percept of a certain sort. Both agree that there is a
basic state in common between cases of illusion and cases of genuine perceptual
awareness, which is the real conclusion of the argument from illusion.
In fact, however, many of those who accept causal theories of perception make
no use of the argument from illusion. The classic instance is Frank Jackson (1977),
who rests his case for a representative theory more or less entirely on quite
different considerations, those designed to support the act-object analysis of a
perceptual state. My own view is that this form of argument is far better, whatever
the merits of the theory which it supports. The point here, however, is that Jackson
agrees that the argument from illusion is of dubious force. It proves nothing.
All allow that genuine perceptual experiences are or may be phenomenally
230   
indistinguishable (to the perceiver at the relevant time) from illusory or hallucin-
atory experience. A non-causal theorist is forced to say, therefore, that two states
which are completely indistinguishable to their owner in their general nature are
in fact very different. The first, the successful case, is not to be understood as
consisting of what is present in the unsuccessful case plus something else (the
outer thing). There is no common element at all. And this certainly seems
awkward. The phenomenal similarity between the two states (i.e., the fact that
the person concerned cannot tell them apart) appears as a reason for avoiding any
analysis of them that makes them radically different in nature. So the causal
theorist concludes that the best analysis is to suppose that both states share a
common element, but only one has an external object. Since the existence and
nature of the external object are not intrinsic to the nature of the perceptual state
of the perceiver, the phenomenal similarity between the states does not prevent us
from saying that one is graced by an external object when the other is not.
So no party to the debate supposes that the argument from illusion is conclu-
sive. It is still possible to suppose that the two states do differ thus fundamentally,
despite their phenomenal similarities; the argument from illusion merely acts as a
reminder that there is at best something awkward about this, and that it would be
more attractive to avoid it. But the appeal to the deliverances of introspection is
not generally allowed to be conclusive elsewhere, and there seems to be nothing
special about the present case to warrant any more respectful attitude to intro-
spection here. So the awkwardness is to be admitted, without being allowed to be
decisive.
This weak conclusion is one with which I do not want yet to quarrel. Among
my suggestions will be, however, that in other areas of philosophy arguments
from illusion have been wrongly taken to be conclusive. I now offer five further
examples.
First example: acting and trying
I quote from a recent book on the philosophy of action:
The same event of trying occurs both when we fail and when we succeed in
bringing about the act-neutral event [of one’s body moving, say]. This presence
of trying is concealed by the fact that we generally succeed, so that we reserve the
word ‘trying’ only for those cases where we failed or where failure was foresee-
able. But since failure is always a possibility, we also try when we succeed.
(Moya 1990: 22)
The author here is reporting an argument of Brian O’Shaughnessy, who himself

announces it as a version of the argument from illusion (1980: ii, 93). To check
   231
this, we can read it through again, substituting ‘appearance’ for ‘trying’ and
‘perceiving an external object’ for ‘bringing about the act-neutral event’. One
might still doubt this self-description, however. For there is here no claim that
the event of trying and failing is always indistinguishable to the agent (at the
relevant time, etc.) from the event of trying and succeeding. How could there be?
Nothing is more different than trying to jump over the river and succeeding, and
trying and failing. So how can there be any argument from indistinguishability
here? Doubts continue when we read later in O’Shaughnessy’s book that he
actually uses two arguments, an argument from error and another from illusion.
What is the difference between an argument from error and an argument from
illusion? These names are not well defined in the philosophical profession.
Without distorting O’Shaughnessy’s position, we can say that both start from
cases where we make mistakes, but that arguments from illusion are distinctive in
that they focus on cases of indistinguishability. (I would prefer to reserve the name
‘argument from error’ for sceptical arguments that appeal to indistinguishability.)
O’Shaughnessy’s argument from error makes no use of considerations of
indistinguishability. It asks us to consider a case where a man contracts to perform
an easy but very important action such as throwing a switch to stop a bomb from
going off. He does perform this action, but immediately a team of conjurors,
psychological experimenters and master technicians, all in the employ of some
Philosopher King (improbable though such staffing levels may appear these days),
persuade him with unimpeachably rational arguments that he did not perform it.
What is he to say to those who upbraid him for failing to throw the switch? There
is one thing he can certainly say, according to O’Shaughnessy, and that is ‘At least
I tried to do it’. Since the conditions for his saying this are effectively perfect, he
knows that he tried to do it. This shows that he did try to do it. It is also true that
he did it. And since this scenario could be played for any action, we can conclude
that all cases of success are cases of trying. As Moya said in the quotation above,
‘since failure is always a possibility, we also try when we succeed’.
This argument is not an argument from illusion. It does not claim that some
case of trying or of having tried is indistinguishable from one of succeeding or
having succeeded. The poor dupe believes that he tried and did not succeed. But
O’Shaughnessy distinguishes between this argument from mistake and a different
argument which asks us to consider a case where a man’s arm is hidden behind a
screen so that he cannot tell whether he has succeeded in raising it. Suppose that
his arm is so paralysed that, when he thinks he is raising it, it in fact remains
immobile. What would he say on being confronted with this stark reality?
(Suppose again that it is very important to all that he should succeed, so as to
stop the bomb going off.) He could and should say ‘I tried to raise it’. But now we
have a genuine argument from indistinguishability. The action of trying to raise
one’s arm and failing is indistinguishable to the agent at the time from the quite
different action of trying and succeeding. O’Shaughnessy concludes:
232   
Then since all these . . . items [willing, awareness of trying, seeming to succeed]
can obtain independently of the occurrence of [the arm moving], and the internal
situation was indistinguishable from that which normally obtains, these
must be further psychological ingredients of the normal successful voluntary
act situation. (1980: ii, 266)
My own view here is that the former argument, from error, reduces to the latter
one from illusion. Suppose that just when you had, so far as you could tell,
succeeded in ringing the bell (for instance), a crowd of philosophers turned up
who convinced you that you did not ring the bell after all. What are you to think?
Would you think that you had anyway tried to ring it? My own view is that you
would reasonably doubt this as well, on the grounds that if you had tried you
would have succeeded. What is more, surely the proper excuse in this peculiar case
is not ‘I tried to do it’ but ‘I thought I had done it’. The latter seems to me a lot
nearer the truth than the former. If the former is in place at all, this must be
because there is a suggestion that if one thinks that one succeeded and discovers
that one did not, one must suppose that one at least tried. But this will itself
depend on the claim that, to the agent whose faith in his grasp on external reality
has been shaken, an action of trying and failing is indistinguishable from an action
of succeeding. I learn that the latter did not happen, and so I suppose that the
former, which is indistinguishable to me now, must have been what happened. But
why should this establish that when I succeed I also try, let alone that trying is a
distinct element in succeeding? Take some other case in which we cannot distin-
guish A from B and learn to our surprise that we did not do A. If we then
conclude that we must have done B, this does nothing to establish that doing B is
an element or ingredient in doing A, nor even just that when I do A I also do
B. Suppose, for example, that I have a toggle switch which, when pressed, either
locks or unlocks my briefcase; it reverses the existing state of the lock, whatever
that was. It is only by using extraneous information that I know whether by
pressing it I am locking or unlocking the briefcase. Suppose further that, having
pressed the switch to open the case, I find to my surprise that instead of
unlocking it, I have locked it. This should do nothing to persuade me that
locking it is an ingredient in unlocking it, nor even that when I unlock it I also
lock it. It seems then that the only use of indistinguishability that will serve
O’Shaughnessy’s purposes is the one in the argument from illusion. The argu-
ment from error adds nothing.
So it seems that there is an indistinguishability argument here after all, despite
initial appearances. It is not the argument that successful actions are commonly
indistinguishable to the agent from unsuccessful ones, but that if agents are
restricted entirely to their own ‘internal’ resources, they cannot draw the distinc-
tion in a given case, and that this ‘proves’ that trying is an ingredient in
succeeding.
   233
Before moving on, I want to pause to draw two distinctions that have gradually
emerged in the discussion of trying.
The first distinction is between the claim that trying is an ingredient in acting,
and the claim that when one acts one also tries.¹ Arguments from illusion, as I am
presenting the matter, promote the latter conclusion by promoting the former.
Hence rebuttals of the argument from illusion can aim either to deny the former
(leaving the latter undetermined for the while) or to deny the latter and thereby
the former. I will call the first sort of response anti-conjunctivism, and the second
the incompatibility thesis. Both responses can of course be discussed and evalu-
ated on grounds other than any thoughts of indistinguishability; but arguments
based on indistinguishability are my official concern here.
The second distinction is between those arguments from illusion that announce
that all A-cases are indistinguishable from B-cases and that therefore A is an
element in B, and those that only hold that some A-cases are indistinguishable
from B-cases. In my view it is only because of their association with sceptical
arguments (with which we are not here concerned at all) that arguments from
illusion have so often been presented as if they needed the stronger claim. In fact,
what I am calling ‘the argument from illusion’ in this paper is a structural
argument in the philosophy of mind, and it is sufficient for this argument to get
going that it finds some cases (a reasonable range, no doubt) of indistinguish-
ability. These distinctions will play a role in what follows, because they help us to
map various moves and responses against each other. I now turn to my second
example.
Second example: facts vs. beliefs as reasons/motives
Here I have no nice neat quotation, but I think that the sort of argument I am after
is perfectly familiar anyway. Can a fact be anyone’s reason for action? Here is an
argument that our reasons are always beliefs and never facts. Suppose that Owen
claims that his reason for investing in a company is that the stock is undervalued,
and let us suppose that he is right—the stock is undervalued. We know, however,
that he would have invested even if the stock had not been undervalued. What was
required for him to invest was not that the stock be undervalued, but that he
should have thought it was undervalued. So Owen’s real reason for investing lay
not in the (purported) fact that the stock is undervalued, but in his belief that it is
undervalued. And this argument can be repeated for every case. It follows that no
fact can be a reason, for an agent would have had the same reason for action even
¹ I owe my recognition of the importance of this distinction to Tim Williamson.

234   
if that fact had not obtained.² And the same argument will work for reasons for
belief rather than for action. Since one would have believed even if what one takes
to be fact had not been fact, one’s reasons must lie not in the facts themselves but
in one’s beliefs about the facts. The facts themselves cannot be reasons.
Though I think that this argument is familiar, I do not think that it is clear. To
make it clearer, let us compare a version which talks about reasons for belief with
one which talks about causes. The causal version says that facts are not causes of
action because one would have acted the same way in their absence, so long as one
still believes. Is this a version of an argument from illusion, or is it a simple appeal
to some principle of causal reasoning? The relevant principle in this case would be
that A is not a cause of B if B would have happened even if A had not happened.
But that principle is not at issue in the present case. We have three players: belief,
fact and action. The argument we are considering has it that the fact is not a cause
of the action because the action would have taken place in the absence of the fact.
But to get the action without the fact we would still need the belief, and sometimes
the belief would not have been present in the absence of the fact either. (If it had
not been true that I have two hands, I would not have believed that I have two
hands.) So the admission that with the belief but without the fact we would still
have had the action does nothing to show that in this case the fact does not cause
the action. To show that, we would have to show that without the fact there would
still have been belief, which will not always be true.
A better causal argument has the opposite conclusion, that the belief always
plays some role in the causal story, since without belief there is no action. In a
suitably weak sense we can allow this, because in discussing the argument from
illusion we are not trying to avoid the conclusion that when one acts in the light of
a fact, one also believes and acts because of that belief. We are officially only
disputing the conclusion that the latter is an ingredient in the former. This leaves it
open to us to select our own story about how fact and belief are actually related in
a successful case. We might for instance say that the belief is an enabling
condition—something whose presence makes it possible for the fact to cause the
action, but which is not itself a part of the causal story proper. This would give us
two possible causal routes to the action, the first where the cause is the fact
(always, of course, mediated by the belief) and the second where the cause is the
belief. Notice that though we of course hold that no one example is an instance of
both of these at once, we do allow that an action caused by a fact is one where the
fact is believed, and where the belief is relevant to the causal story. There is no
incompatibility claim here.
Even if this causal argument is a good one, I do not think it is a version of
the argument from illusion, because it says nothing about indistinguishability.
² For an expression of the argument I am trying to characterize, see Bernard Williams’s claim that
the difference between true and false cannot affect the form of the appropriate explanation (1981: 102).
   235
Nor can I see how to adapt it so as to include thoughts about indistinguishability

in any significant way. Such thoughts only appear in a version of the argument
which focuses on reasons rather than causes. Here we may say that believers
cannot distinguish between the situation in which they believe falsely and that in
which they believe truly. If their beliefs are false, they still have sufficient reason to
act. The truth of their belief should not therefore be conceived as changing
anything in that respect, and so can only be taken to be an added ingredient.
This approach is much more like the perceptual case, and it has the proper
conclusion for an argument from illusion, namely that there is a common
ingredient to cases of success and of failure. It attempts to drive us from what
we want to say in a case of error or defect (mere appearance, in the perceptual case,
or mistake about the facts in the other) to what we say about a case where there is
no error or defect. As such an argument, it can be resisted. The principle on which
it relies is that a successful case must be some sort of function of an unsuccessful
case, and there is no general reason in advance to believe anything of the sort. We
should be stubborn here as elsewhere and hold that a fact can be a reason for
action as for belief, even if we would still act or believe even if the facts were not as
we take them to be. What this would mean is that where we are mistaken we do
not have the reasons for our actions and beliefs that we take ourselves to have.
Note that it is common for both my first examples to be taken as conclusive.
O’Shaughnessy takes his example to provide a proof, and many suppose that facts
are unable to motivate or be reasons on the sort of grounds I have suggested, as if
there can be no two views on the matter.
Third example: knowledge and justified belief
The next two examples are from epistemology. The first concerns the view that
belief is an element in knowledge. We might argue that knowledge and mere
justified belief are indistinguishable to the cognizer. Therefore knowledge must
consist of justified belief plus something else, and the extra bit must be evidence-
transcendent. Truth, being an evidence-transcendent property, is a good candi-
date for this role; but whatever our choice for that role, we have already established
that belief is an element in knowledge.
This is another version of the argument from illusion. As such, it is not
conclusive. It only uncovers a difficulty. This difficulty is the familiar one of giving
entirely different analyses of two states which are experientially indistinguishable.
Perception is phenomenally indistinguishable from mere appearance, and know-
ledge is phenomenally indistinguishable from mere justified belief. This fact is
some reason for giving similar analyses of the two states, but hardly conclusive
reason. We are not forced into the position recommended, that of supposing
that there is a common element to mere justified belief and genuine knowledge.
236   
We can be stubborn. We can give an entirely distinct account of knowledge, as a

state in which a fact makes itself manifest to us, not by causing in us some
appearance or proxy for itself, but simply by being transparently present to the
mind. In genuine knowledge, the mind is open to the world, and when it grasps
the world, the world (or the relevant aspect of it) enters the mind. In this vein John
McDowell (1982) offers us a different conception of knowledge, as a sort of
transparent openness to the world. Any state of this sort is quite different from
an opaque state founded on reasons other than the fact believed. So knowledge is
not to be thought of as a state reached by adding conditions onto something which
in its own right is inherently unreliable. Indeed, if we are in the business of adding
conditions so as to make things better, we can tell that we are not talking about
knowledge at all. A state reached by piling on extra helpful thoughts, such as
that human cognitive systems are truth-effective, will only be a better form of the
opaque state, never the transparent state which is knowledge. Knowledge does not
need this sort of support. It is already perfect in its own right.
We should be careful to distinguish this position from the stronger claim that
knowledge is not a form of belief at all. This incompatibility claim goes further
than McDowell needs or wants to go. We may, for all that has yet been shown,
continue to think of the opaque state and the transparent state as distinct forms of
belief, forms which are, on occasion at least, indistinguishable to their owner. One
characteristic of knowledge as McDowell conceives it is that it is infallible, in
the sense that if one is in such a state there is no possibility that the world should
not be the way one conceives it as being. With the opaque state, no matter how
clever, there will always be this possibility; but any state which can exist without
the world being as that state presents it as being will not be knowledge of
McDowell’s sort.
Fourth example: the internalism/externalism distinction
This example follows on from my second example. It is not merely the case that if
the facts were not as we take them to be we would still believe and act in the same
way. It is also commonly held that in doing so our beliefs and actions would be
justified. Justification is a normative status, and no complaint can be levelled
against us for holding beliefs that are supported by the other beliefs we hold. For
by our lights we are doing as well as we possibly could. There is no constraint or
requirement which we are capable of recognizing when it applies and capable of
meeting when it does, and which we have here failed to meet. Our position is
indistinguishable to us from one in which we are in perfect epistemic shape. So no
complaint can be raised against us. Our side of the story (‘internally’, as it were) is
faultless. We deserve epistemic praise, of the sort normally handed out by calling
our epistemic behaviour rational or reasonable, and our belief justified. But if we
   237
are justified in a belief merely because it is well supported by other beliefs we hold,
it follows that what justifies is not the facts, but one’s beliefs about the facts. For
our belief is justified even if the necessary facts are absent; so the only possible
candidate justifiers are our other beliefs. It seems that the facts themselves make
no contribution to justification beyond that already made by the beliefs.
Suppose, however, that we feel (for whatever reason) that this form of intern-
alism is too restrictive. We want to say that even when internal justification of this
sort is fully present, something more is added if certain facts can be established, for
instance that the epistemic methods being used are in fact reliable rather than
merely (reasonably?) believed to be reliable. An internalist can allow this move in
the interests of flexibility, but is likely to say that what is added by these externalist
considerations is not more of the same thing as we had at first. Even if the facts do
add something, what they add is essentially independent of what has already been
established, namely that as cognitive agents we are irreproachable in a case
like this.
Let us call this the independence thesis. It is the result of an argument from
illusion. The independence thesis here has a perfect mirror in the theory of
perception. The perceiver’s (believer’s) side of the story is one that can be told
without reference to what lies beyond the perceiver’s (believer’s) ken, and it
follows from this that such things as success-conditions must be placed elsewhere
by the theory, in a separate story which is independent of but can be added to what
we have established so far. This is the place where we see specification of causal
conditions in many currently popular theories of perception—and of justification.
What this means is that we do best to start from a perceiver who is in error
(reasonable error will probably be the most helpful since it introduces the fewest
complications). This is the best starting point because it ensures that we do not
confuse elements which are properly part of the perceiver-story with elements
which are not. Then perceivers who are not in error are people of whom essen-
tially the same sort of story can be told, as far as their side of the whole goes. Surely
the pattern here is exactly the same as the one we have been seeing in the theory
of justification. And the reply to it will be the same both times. There is no
independent reason to start one’s analysis of a cognitive state (whether perceptual
or not) from an example of such a state which is in error and then to attempt to
build on that one’s account of the case where there is no mistake.
There are two aspects of this reply that are worth stressing. The first is that it
makes the distinctive move of holding that epistemology is best done by starting
from the best cases, those in which there is no mistake, and working back from
that. The second is that it rejects a basic Cartesian theme underlying the argument
from illusion, namely the distinction between what is internal to agents and hence
cognitively available to them, and what is external to agents and so must be placed
elsewhere in the analysis. The latter facts are logically independent of the former
ones, and this makes possible the view that the agent’s mental health (which
238   
includes the question whether the agent’s beliefs are justified) is to be established
in ways that make no appeal to the world external to the agent. This move in the
philosophy of mind is a straight expression of Cartesianism and its conception of
the self-contained and autonomous individual. It is what drives the sense that our
analysis must start from a case of mistake rather than a case of success. So the
second aspect underlies the first one. It is one thing, however, to say that the
independence thesis has a dubious provenance (i.e., that it is based on a form of
argument which I take to be dubious), and another to suggest that it is false. But
there are reasons for holding it false. We might agree that we make no complaint
about the sort of believers I have been discussing, those who perform perfectly by
their own lights, even where those lights lead them astray in ways which they are
unable to rectify. But does this necessarily mean that their position could not be
improved, so that any conceivable improvement must be attributed to some other
aspect of the story? I do not see why we should allow this. In my view there is at
least one respect in which their position could still have been better than it was,
namely that they could actually have had the reasons they thought they had. They
do not have those reasons because the facts that would have been their reasons do
not obtain. So the believers here are not in the best imaginable cognitive position.
Fifth example: the act/agent distinction
The very same moves, mutatis mutandis, are the ones which generate a significant
distinction between act and agent in moral theory. There is an independence
thesis here, driven by the thought that all that we can require of moral agents is
that they act as best they may by their own lights. This thesis can, I think, allow
that the way they have come by those lights makes a difference. An agent whose
position, moral or epistemic, has been formed in ways of which some epistemic or
moral complaint can be made is not someone whose conscientiousness from then
on is a full defence. An indifference to relevant and available information, an
insensitivity to counterevidence and the opinions of others, slovenliness in the
gathering and marshalling of reasons—these and other similar faults undermine
any attempt to justify agents as having ‘done as well as they could by their own
lights’. But where no such complaint can be made, it is said, no further complaint
is in place when an agent does an action which a perfectly informed and sensitive
agent would not have done. It is this thought that drives the act/agent distinction
in its traditional form in ethics. The act is something that could have been
improved upon; the agent could not. The agent’s side of the story (the ‘internal’
story) is independent of any features of the act in any case of discrepancy.
Against this view we can say, first, that it is another version of the argument
from illusion. Second, we can doubt the conclusion as before. Rewriting a previous
section: we might agree that we make no complaint about the sort of agent I have
   239
been discussing, those who perform perfectly by their own lights, even where led
astray by those lights in ways which they are unable to rectify. But does this
necessarily mean that their position could not be improved, so that any conceiv-
able improvement must be attributed to some other aspect of the story? I do not
see why we should allow this. In my view there is at least one respect in which their
position could still have been better than it was, namely that they could actually
have had the reasons they thought they had. They do not have those reasons
because the facts that would have been their reasons do not obtain. So the agents
here are not in the best imaginable moral position.
But is it not true that we make no complaint about such ‘blamelessly benighted’
agents? And doesn’t this force us to accept that, once the satisfactoriness of the
agent has been established, every further advance must be ascribed to something
else, the action? My response to this sort of pressure is to suggest that we are being
offered too few choices. It is as if our moral vocabulary contained only one word of
approval and one of disapproval, and we are asked what we think of a blamelessly
benighted agent. We clearly cannot apply the term of opprobrium, and so we
apply the term of approval. But we in fact have a wide variety of ways in which to
think of agents, and a wide variety of terms in which to do it. Surely the best
approach is just to say that we think of agents who do not have the reasons they
reasonably take themselves to have in precisely that way. People in this position do
not deserve the reproach they would have done had their conception of what they
were doing been different, and equally, had they been doing the action they took
themselves to be doing, they would have deserved the highest praise, perhaps. As
things are, however, they merit sympathetic understanding and encouragement,
rather than straight reproach or moral abuse, on the one hand, or blanket
approval on the other.
This suggestion, that we are being offered too few choices, has been made
before, but not in this area. J. L. Austin argued against A. J. Ayer’s use of the
argument from illusion that it only appeared sound if one worked within an
unnecessarily restricted vocabulary.³ In aesthetics he argued that we should
sometimes turn our attention to the dainty and the dumpy. I presume that this
was not only because these concepts are interesting in their own right. The
suggestion is rather that we will not understand the role of the thinnest predicates
unless we come to see how they relate to the thicker predicates that underpin
them.⁴ To try to work with the thinnest predicates alone is a cause not only of a
limited but of a distorted view.
³ I owe this thought to David McNaughton.

⁴ The thick/thin distinction, originally owed to David Wiggins, appears to some effect in Williams
(1985: chs. 8–9). Thick properties are so called because they have more empirical content than thin
ones do. The thinnest properties in ethics are those of goodness, rightness, badness and wrongness. One
knows less about the action if one knows that it has one of these properties than one knows if one
knows it to have some thicker property, e.g. courageousness.
240   
The five examples compared
In this section, I draw out some similarities between the six cases I have discussed.
Most of the material for this section has already appeared along the way.
The point I want to stress is that the argument from illusion, wherever it
appears, has what we might call a ‘constitutive’ conclusion; it encourages us to
discover a more complex constitution in something (state, process, object) than
we might have expected. This is more obvious in some examples than in others.
We see it operating clearly in the idea of the complex constitution of a veridical
perceptual state, of a successful action and of knowledge. In each of these
examples the idea of a common ingredient is an idea about the constitution of a
successful case, which sees the successful as the unsuccessful plus something
independent. But essentially the same point can be made about my other three
examples. It is the independence thesis that is the crucial link. In the motivation
case, the idea was that the truth of the relevant beliefs is incidental to the
motivation generated by those beliefs. The extra bit present in the successful
case is independent of what is imported from the unsuccessful one. In the act/
agent case, again, the idea was that we should start from the conception of the
unfortunate agent, see that as a common element which is entirely independent of
the features distinctive of the successful case, and so see the structure of the
situation assessed in terms of the structure of assessment. Assessment of the
agent is completed before we turn our attention to the question whether things
actually are as the agent takes them to be. Hence the action is constructed as the
bearer of any properties beyond those born by the agent. Since these matters are
structural, they are constitutive.
The same is true of the internalism/externalism case. It might seem that we
have turned with this example from questions of constitution to questions of
assessment, or of evaluation, which may rest on constitutive matters, without
being themselves constitutive. I don’t accept this. The point in the discussion of
internalism and externalism at which we reached the role of the argument from
illusion, supposedly, was the one where the independence thesis appeared. This
thesis asserted that what the facts may add to assessment is essentially independ-
ent of what has already been established, namely the agent’s own perspective on
events. In my view what this means is that the structure of assessment and the
structure of the situation assessed should match each other. In the situation
assessed, the question whether things are in fact as the agent takes them to be is,
according to the argument from illusion, quite independent of the prior question
how the agent takes them to be. The role in assessment played by answers to the
former question must therefore be independent of and secondary to that played by
answers to the latter question. These are structural matters, and this is enough for
me to think of them as constitutive.
   241
Responses to the argument from illusion
If the argument from illusion has a constitutive conclusion, responses to it will

reject that structure or constitution. Instead of seeing the relevant situation as a
conjunction of two independent elements, they will give a different account of the
matter. Two such non-conjunctive accounts have already been mentioned along
the way. The first was the incompatibility thesis, more plausible in some areas
than in others. We saw an instance of this in the claim that when one acts, one
does not try to act. Another instance would be the idea that knowledge is not a
form of belief at all. The second arose in McDowell’s claim that though knowledge
is a form of belief, belief is not an element in knowledge. We could express this in
the language of determinates and determinables: belief is a determinable of which
knowledge is one determinate. (There may be other models than the determinate/
determinable one, but I shall not consider any such here.)
As well as these two non-conjunctive accounts, however, there is also what has
come to be known as the disjunctive approach—normally, the disjunctive theory
of experience, since that is its home concern.⁵ I want to end by considering the
relation between these various forms of non-conjunctivism. Disjunctivists offers a
disjunctive account of, for instance, its looking to S as if there is an oasis before
him, as:
either (1) there is an oasis manifesting itself to S

or (2) it is for S as if there is an oasis manifesting itself to S
The second disjunct can be read simply as ‘(2) it is for S as if (1)’. Let us, therefore,
invent similar accounts of trying and of belief:
S tries to V iff:
either (1) S Vs/S succeeds in V-ing
or (2) it is for S as if (1).
But this is peculiar. Trying is not the same as its seeming to one that one is
succeeding. A similar attempt at a disjunctive theory of belief might be:
S believes that p iff:

either (1) S knows that p/the fact that p is manifest to S
or (2) it is for S as if (1).
⁵ See McDowell (1982); Snowdon (1981).

242   
This is clearly wrong, and for similar reasons. Believing in the sense of (2) is not
always indistinguishable from knowing. What is happening here?
One thing that has gone wrong is that we have restricted the disjunctive
conception to two disjuncts. This was wrong even in the home case, the perceptual
one. For after all, there are plenty of experiences that are easily distinguishable
from perceivings, and the provision of only two disjuncts denies this. The dis-
junctive conception of experience should really be:
S has an experience of O before him iff:

either (1) O manifests itself to S
or (2) it is for S as if (1)
or (3) S has a recognizable O-illusion.
What we are now in a position to see is that the standard formulation of the
disjunctive theory is rather peculiar. What it was really trying to say was that there
are two sorts of experience (we could call them an outer experience and an inner
experience) which are indistinguishable from each other. It was no part of
disjunctivism to say that all inner experiences are indistinguishable from outer
experiences—only that some are. So we should allow the existence of experiences
of a third sort. What is more, that the experiences of the second sort are
indistinguishable from those of the first, the thought captured by the phrase ‘(2)
it is for S as if (1)’, is a sort of comment on them rather than a positive description
of their nature. The indistinguishability of experiences of types (1) and (2) should
be a contingent matter, not a matter of definition.
So what does the new disjunctivism amount to, from this perspective? It
amounts to the assertion that there are broadly two types of experience, which
we can call manifestings and non-manifestings, and that some of the latter are
indistinguishable (to their owners, at the relevant time, etc.) from the former. We
are told virtually nothing about the natures of these different types of experiences.
What sort of an answer, then, does disjunctivism offer to the argument from
illusion? In what way does it go at all beyond what I have already offered, in my
description of what is required if one wants to reject the argument from illusion?
I don’t think it goes any further at all. It just repackages the idea that those who
reject the argument from illusion in the perceptual case are bound to say that there
can be two states of mind which are indistinguishable to their owner though
neither is an ingredient in the other. It doesn’t make any move beyond this at all.
The disjunctive theory, then, does not amount to more than a structured way of
expressing opposition to the argument from illusion.
But in that case, two questions arise. The first is how there can be other ways of
expressing that opposition. The second is whether we can actually write suitable
disjunctive conceptions for my other examples. After all, the ones we sketched for
   243
trying and believing were clearly unworkable. If disjunctivism merely expresses

the form of opposition to the argument from illusion, and if I am right in claiming
that there is an argument from illusion at work in each of my examples, there
should be disjunctive conceptions available in each case. Let us deal with this
second worry first.
What was wrong with our first attempts at disjunctive conceptions of trying
and acting was that we illegitimately took aspects of the perceptual case and
imported them into the general structure required of a disjunctive analysis.⁶ The
disjunctive account of perception really says there are two quite different sorts of
oasis-experience, which may nonetheless be indistinguishable to their owner. The
first is the genuine article, and the second, though it is indistinguishable, has
nothing in common with the first other than the fact that they are both oasis-
experiences. In the standard formulation of the account, misleadingly, this is
explicitly the way in which the second disjunct is characterized; we characterize
it solely by saying that it is like what it is not. Presumably, however, there may be
available a more direct characterization of the second disjunct, and in a totally
explicit version of the theory it would indeed be characterized in that better way.
The current characterization is just a sort of place-holder, showing what has to be
said about the relation between first and second disjunct.
Suppose, then, that we try disjunctive conceptions of the following quite
different forms (there being no reason why all disjunctive accounts should be of
exactly the same form), first for trying:
S tries to do V iff:
either S does V
or S does something with the intention of thereby doing V.
The account of belief will probably be more like the account of perception:
S believes that p iff:

either S knows that p
or S takes it with reason that p
or S has a blind faith that p.
These accounts, though they do allow that when we know, we also believe, and
when we act, we also try, do not allow that the latter is a component in the former
(that (2) is a component in (1)). So they are genuine alternatives to the approach
which the argument from illusion is trying to promote. Even though when one
⁶ I owe this thought to Bernard Williams; the previous, wrong, approach was the one
I originally took.
244   
acts one also tries, acting is not to be understood as trying plus anything. Similarly,
knowing is not to be understood as believing plus something else. There is no
conjunctive account of knowing or of acting derivable here. Knowing and acting
are determinates of the determinables believing and trying respectively.
A notable difference between these and earlier attempts is that we have aban-
doned any attempt to inscribe the primacy of the successful case in our formula-
tions, This might be thought a defect. Seeking to establish the primacy of the
successful case was a leading motivation for adopting the disjunctive conception.
So should not the primacy of the successful case be retained in any subsequent
formulation? The argument from illusion is an attempt to establish the opposite
primacy, that of the unsuccessful, after all, and one purpose of disjunctivism was
to block that attempt.
But those with this purpose were surely confused. They ran together two things
which should be kept separate: that each disjunct is, at least on occasion, indis-
tinguishable from the other; and that the successful case enjoys a certain priority.
They ran them together by giving the original account of the two disjuncts given in
my text, where the second disjunct can only be understood in terms of the first,
since it reads ‘(2) it is for S as if (1)’. (In different terms, they captured the
indistinguishability of seeing and seeming to see in an asymmetric way that
leans towards the first.) That this was tendentious is shown by two considerations.
First, we can see in the case of belief and knowledge that there need be no
suggestion that the worse case can only be understood in terms of the better
one. ‘Opinion’, whatever it may be, need not be defined in terms of knowledge,
and those who think otherwise will have to do more than show the possibility of a
disjunctive conception of belief. Second, even in the home, perceptual, case, the
contrast between seeing and seeming to see is tendentious: it encapsulates the
priority of seeing in a way for which there is as yet no justification. For although
we may be persuaded to read the disjunctive conception as holding that there are
two states, a better one and a worse one which is indistinguishable from the better
one, we could equally well have read things the other way around, saying that
there are two states, a worse one and a better one which is indistinguishable from
the worse one. Here the apparent primacy of the better state has vanished, leaving
only its preferability, which was never in doubt. Calling the worse state ‘seeming to
see’ expressed that primacy, but was inessential to the main purpose of responding
to the argument from illusion.
Let us take it, then, that these new disjunctive conceptions are satisfactory. Our
other question was whether disjunctivism is the form of opposition to the argu-
ment from illusion, or whether there are indeed other forms. Earlier I suggested
that there are in fact three forms: the disjunctive approach, a determinate/deter-
minable approach, and the incompatibility thesis. But now I think that, with our
improved understanding of the disjunctive approach, there is no further difference
between it and the determinate/determinable one. The incompatibility thesis is the
   245
odd one out; this does go beyond anything required to rebut the argument from
illusion. One would move from the disjunctive approach to the incompatibility
thesis if one took knowledge, say, to be of a completely different sort from belief.
To the extent that one sees knowledge, as McDowell does, as a sort of transparent
openness to the world, and mere belief as an opaque, closed state, to that extent
the more or less metaphysical difference between the two sorts of state may
prevent one from seeing knowledge as a special case of belief, with the concept
of belief ranging from knowledge at one end to blind faith or mere hunch at the
other. But this move is separate from anything required to dispel the argument
from illusion, and requires a separate motivation.
Conclusions
What conclusions do I wish to draw, then? There is a weaker one and a stronger
one. The weaker, which I have been stressing along the way, is that no use of an
argument from illusion should be taken to be conclusive. In the parent case, in the
theory of perception, claims of conclusiveness have been long abandoned. In the
other cases, however, the relevant arguments are largely held to be conclusive.
I think that this is a mistake, and that comparative investigation of different
versions of the argument can cure us of it.
Associated with this is the idea that arguments from illusion stem from
defective Cartesian presumptions about the independence of the mental. I have
certainly not shown this in the present paper. I have only suggested that it is a
ground for treating such arguments with suspicion.
There is, however, a stronger conclusion available. In the first section of the
paper I effectively accepted that because veridical and illusory appearance are
indistinguishable, it is awkward to deny a common-ingredient account of appear-
ance. But now we can see that the determinate/determinable version of the
disjunctive approach offers a competing explanation of the indistinguishability.
That these states are both oasis-experiences is the basis of an explanation of why
the owner cannot tell them apart. There are other obvious cases where the fact that
two states are determinates of the same determinable explains the respect in which
they are similar: two different colours, say red and green, are similar in respect of
being colours, and this is a real, not a merely nominal similarity.⁷ That they are
both colours captures the respect in which they are similar without explaining that
similarity in terms of any common ingredient. In the same way we can offer an
explanation of the similarities between different disjuncts in other cases—between
acting and ‘merely trying’, as it were—by pointing to their common determinable.
⁷ This point is Tim Williamson’s.

246   
We can thus offer to explain the indistinguishability without alluding to a com-

mon ingredient. If this is so, then the greater plausibility of the common-
ingredient model vanishes, and with it the idea that it is awkward even in the
parent case to admit the indistinguishability of perception and illusion without
allowing a common element.⁸
⁸ I am very grateful to David McNaughton, Jennifer Hornsby and Paul Snowdon for comments on
earlier versions of this paper, to Bernard Williams for helpful discussion, and especially to Timothy
Williamson for his reply to it, read at the Oxford Philosophical Society, which largely reshaped my
understanding of the area.
16
Why There Is Really No Such Thing
as the Theory of Motivation
One of the very many distinctions influential in moral theory is that between
motivating and justifying reasons. According to this distinction, there is a differ-
ence between the reasons why an action is or was right and the reasons why the
agent did the action. As a distinction, this is quite uncontentious. It is perfectly
obvious that the reasons why an agent did the action may not be reasons why it is
right, either because it was not right or because, though it was right, the reasons
why it was right were not the reasons why the agent did it. Contention comes in
when we try to give an account of the two sides of the distinction, to back up and
make theoretical sense of the difference. The theoretical account that I want to
reject is one that is becoming increasingly common in the literature of moral
theory. This account takes justifying reasons to be facts about the world, and
motivating reasons to be combinations of beliefs and desires. The most striking
feature of this account is that no justifying reason can be a motivating reason and
no motivating reason a justifying one. The two classes of reason come to be seen as
mutually exclusive.
I will eventually come to look at the arguments underlying this account. For the
moment I want to elaborate on the importance attributed to the distinction, so
understood. One use to which it has been put is in the attempt to undermine
internalism in the theory of moral judgement. William Frankena, who is respon-
sible for the account in its present form, argued that the internalist claim that
moral facts are capable of motivating in their own right succumbs to the motiv-
ating/justifying distinction (1958: 43–5). A moral fact is a justifying reason for
action, and thoughts about what is capable of motivating in its own right belong
not in the theory of justifying reasons but in that of motivating ones. It is a simple
mistake to suppose that the commitment to moral facts as justifying reasons
carries with it a commitment to any view about motivation. This mistake is
covered up by the use of phrases like ‘the moral fact is in and of itself a reason
for acting’, read as a remark about motivation.
More subtle, perhaps, is David Brink’s appeal to the distinction in his attack on
internalism (1989: ch. 3). He uses it to divide internalism into two halves—
internalism about motivation and internalism about reasons. Brink argues, not
as Frankena does that the motivating/justifying distinction is itself sufficient to
disprove internalist claims, but that once we recognize the distinction, we can
248   
disentangle matters well enough to see that internalism about motivation is false
for moral judgements, and internalism about reasons is false for moral facts.
Motive internalism is false because it is at least possible for an agent to be totally
indifferent to what she allows to be morally relevant considerations, since she
lacks the necessary desires. Internalism about reasons is false because it is not
incoherent to admit that one has a moral obligation to act but to ask whether this
gives one any reason to act.
A second use of the distinction, with which I will be more concerned, is to
defend some form of Humeanism in the theory of motivation.¹ The suggestion
here is that attempts to deny the truth of Humean belief/desire theory appeal to
considerations that properly belong to the theory of justifying reasons, and that
are improperly transferred to the theory of motivation. We make a mistake if,
having accepted that there are facts about what is right and wrong (thus becoming
realists in the theory of justifying reasons), we then suppose that this gives us
ground for a purely cognitive theory of motivation (i.e., a non-Humean one).
The common suggestion is that this is what protagonists of cognitive theories of
motivation have done, and that it is necessarily an error because the theory of
motivation is quite distinct from the theory of justifying reasons. So different are
these theories supposed to be that we even read that the word ‘reason’ is ambigu-
ous, and that this ambiguity is offered as the explanation of why people find it so
easy to slide from thoughts about one sort of reason to thoughts about the other.
The mistake that we cognitivists have made, supposedly, is to notice that Hume
was wrong about justifying reasons, and to suppose that in order to show this we
have to show that he was wrong about motivating ones, or that since he is wrong
in the one area he must also be wrong in the other. The distinction is thus used as
part of the defence of Humeanism in the theory of motivation. It is also used,
similarly, to reconcile a broadly Humean theory of motivation with some form of
realism about values. It has been common to see some form of subjectivism or
projectivism supported by remarks in the theory of motivation. But this is a
mistake. Nothing in the theory of motivation could have such consequences,
because of the sheer ambiguity in the notion of a reason.
At its most extreme, therefore, the role of the distinction is to create a limited
area for Hume to have been right about. To the extent, then, that we set our face
against admitting the truth of Humeanism in the theory of motivation, to that
extent we are probably going to feel that there is no such thing as the theory of
motivation, so conceived, at all. And that will be the position that this paper is
trying to defend, though not only for this reason.
It might seem miraculous that so much can be extracted from the little
distinction with which we started, between the reasons why an action was right
¹ See Smith (1984: esp. ch. 4.2). I am very grateful to Michael Smith for his repeated attempts to
explain his view to me.
      249
and the agent’s reasons for doing it. It is not so much the distinction itself that is
the culprit, however, as the account of it that sees motivating reasons as complexes
of beliefs and desires, i.e. as complexes of psychological states of whatever sort,
and sees justifying reasons as truths. It is this account, which puts into form the
attempt to combine value realism with Humean philosophical psychology, that
leads to the results I have outlined above.
Let us start by taking it as a maxim in the theory of practical reason in general

that reasons must be at least potentially explanatory of actions. A reason
must be something for which someone could have acted, and in any case where
someone does act for that reason, the reason contributes to the explanation
of her action.
This maxim is already in conflict with the account of the distinction between
motivating and justifying reasons that I outlined above. For the maxim appears to
say that justifying reasons must be capable of being motivating ones, and this is
directly denied by the claim that motivating reasons are beliefs and desires, while
justifying reasons are truths. The categorial difference between truth and psycho-
logical state means that no one thing could be a reason of both sorts.
One would suppose, therefore, that protagonists of the account would deny the
maxim—or, more probably, admit it of motivating reasons but deny it of justifying
ones. In fact, however, some do and some don’t. Michael Smith is one of those
who don’t. Almost the only thing that reasons of the two sorts have in common,
according to him, is that they are potentially capable of rendering actions intel-
ligible. But it is hard to see how this can be true within the constraints that he
accepts. According to his picture, it is complexes of beliefs and desires that explain
the occurrence of actions, and truths do not contribute to this at all. They do
contribute, of course, to the explanation of why the action was right; but the
maxim does not concern itself with that sort of explanation at all. So it seems that
justifying reasons fall foul of the maxim.
There is one move that could be made in reply to this complaint, which is that
not all talk of ‘rendering an action intelligible’ is talk of explaining the action. The
idea might be that the notion of rendering intelligible is itself ambiguous between
justifying an action and explaining it. But to make sense of this we have to ask
what the difference between rendering intelligible and explaining is supposed to
be. To this question we will not find any very satisfactory answer.
One suggestion that Smith has made to me in correspondence is as follows.
Suppose that someone returns to its owner a wallet that she found in the street.
Without having the slightest idea why she did this, we already think that it is an
intelligible thing to do. And the reason why we think of it as intelligible is that it is
250   
right. So justifying reasons necessarily render intelligible the actions they justify.
But they do not thereby explain them.
There are two things wrong with this attempt as it stands. The first is the idea
that if an action is right, it is at least intelligible, whatever the agent’s reasons for
doing it. This idea of a sort of intelligibility which prescinds from the agent’s
reasons leaves us with too little to work on. One is tempted to say that if we do not
take into account the reasons why the agent did it, we are not yet in a position to
ask the question whether the action is intelligible. There is no such thing as an
intelligibility that can be established in this way, irrespective of reasons.² What we
do get from the thought that the action was right is that it is at least intelligible that
someone should have done the action, provided that they did it for the reasons
that in fact make it right, and understanding them as such. But this does nothing
to show that justifying reasons render actions intelligible. What in this case
renders the relevant action intelligible is the beliefs and desires that led to it and
explain it—the motivating reasons.
The second difficulty is that unjustified, wrong actions are as intelligible, and
for the same reasons, as justified ones. If someone finds a wallet, the action of
returning it to its owner is no more intelligible than that of keeping quiet. If we
had to list in advance, that is, all the intelligible actions which the finder of the
wallet might do, prominent among them might be keeping the money for himself.
One might suggest now that there are good reasons for keeping the money,
even if they are not good moral reasons. The action of keeping the money is as well
grounded in reasons (e.g., in the fact that he has a good use for the money) as is
that of handing it back. Just as justifying theoretical reasons do not show the
beliefs which they justify to be true, but only reasonable, so justifying practical
reasons do not show the actions they justify to be right, but only rational. The idea
that wrong, unjustified actions are as intelligible as right ones, taken as an
objection to Smith, turns out to be a mistake based on an equivocation between
‘justified’ as ‘right’ and ‘justified’ as ‘rational’; in the second sense, wrong actions
are as justified, and hence as intelligible, as right ones are.
Smith however rejects this line of defence. I can think of two reasons one might
have for doing that. The first is that though there are reasons for doing the wrong
thing, the balance of reasons always favours the right one. I don’t accept this in
general. I think that there can sometimes be more reason to do the wrong thing.
The second is that he takes all justifying reasons to be normative truths. Again,
I don’t accept this. According to me, some are normative truths and some are not.
For instance, the distress I will cause her if I plough on is a reason for me to stop,
and the truth that I will distress her is not a normative truth. So I don’t accept
² It may be possible to establish a reason-free intelligibility in another way; Freud springs to mind.
      251
Smith’s view that it is morally right actions that are specially intelligible, rather
than more generally the ones for which there is most reason.
The reason he gives for his claim that right actions are antecedently intelligible
in a way that wrong actions are not is an appeal to the theory of interpretation. Just
as in the interpretation of utterance there is a preference for attributing true beliefs
to the Other rather than false ones, so in the interpretation of action there will be a
preference for attributing true normative beliefs to agents. And if there is this
preference, there will be an associated preference for attributing consonant motiv-
ation to them. This gives us an antecedent preference for seeing our agents as
doing right actions rather than wrong ones, since we should always prefer an
interpretation which has the agent rightly believing that the balance of reasons
favours her action. We can re-describe this preference by saying that justified,
right actions enjoy a greater degree of antecedent intelligibility.
Whether this approach is sound depends on the reason for believing that in
radical interpretation we must understand the Other as largely believing the truth.
As I understand it, the argument is that we have no alternative to identifying the
truth conditions of the Other’s utterances with the events which cause them. The
circumstances of radical interpretation mean that the Other will always be inter-
preted as believing the true if this is possible. Now if this is the argument, it seems
to me that there are several significant differences between the situation of radical
interpretation and the situation we find ourselves in when we are attempting to
assign normative beliefs to other agents. Quite apart from the fact that the
situation is not the radical one, the beliefs we are here dealing with are not
concerned with our immediate surroundings, nor are the contents of those beliefs
‘occasion sentences’, in any recognizable sense of Quine’s phrase. What is more,
the reason for attributing true beliefs to the Other does not extend to require us to
attribute desires consonant to those beliefs. The reason for doing that is quite
different, namely the sense that if we do it we end up with less to explain than if
we don’t.
So I do not accept the reason Smith gives for holding that right actions are
antecedently intelligible in a way that wrong ones are not. But even if what he says
about the requirements of radical interpretation were true, it would not have the
consequences he claims. It is not that wrong actions would be antecedently
unintelligible in some way, but only that the attribution of right ones, so far as
one can, would be the only possible interpretational strategy. At the end of the day,
intelligible wrong actions are intelligible in just the same sense as right ones are.
The same result holds even if we do not claim greater antecedent intelligibility
for morally right actions, as Smith does, but for the actions for which there is the
most reason. (This might include some wrong actions.) Suppose that the require-
ments of radical interpretation force us to attribute to our agents, so far as we can,
the true belief that there is more reason to do what they are doing than to do
anything else. This would not have the consequence that actions of which such
252   
beliefs are true are intelligible in a different sense from those of which the belief is
false. As long as there is sufficient reason for the action, it is intelligible in just the
same sense as would have been the action for which there is more reason.
There remains something that has not yet come to light. I try to tease it out by
offering a picture of the situation which is different from Smith’s, but which
accepts the need to attribute true beliefs wherever possible. According to this
picture we first establish a list of actions which the person engaged in a certain set
of bodily movements might intelligibly be doing. This list evinces no preference for
the right over the wrong. Then we attempt to attribute suitable beliefs and desires
to the agent for each of the most plausible candidates on the list. At this stage there
is a preference for true beliefs, and possibly for consonant motivation. It is not,
therefore, that right actions are more intelligible than wrong ones. The preference
for the right appears at another, later point.
This picture is inadequate, in a way that cannot be brought out by appeals to the
theory of interpretation. The list of intelligible actions does exhibit some prefer-
ence for the right over the wrong, in the following sense. If an action is right for
reasons A–D, and if an agent does that action for reasons A–D, understanding
them as such, her doing so would therefore be intelligible. By contrast, if the action
is wrong for reasons A–D, and the agent does the action for those reasons,
understanding them as such, her doing so is not yet intelligible. To make it
intelligible, we have to provide further elements.
What this means, however, is just that if the action is right for reasons A–D, an
explanation of the action which appeals to the agent’s beliefs in A–D is a good
explanation. In other cases it is not. This does nothing to introduce a sense of
‘render intelligible’ that is not to do with explanation. It only concerns itself with
the question which explanations are good ones. Smith is promoting a picture with
three elements:
1. Justifying reasons, being facts, are categorially different from motivating

ones, which are beliefs and desires.
2. Both sorts of reasons render actions intelligible.
3. Motivating reasons render actions intelligible in quite a different sense from
that in which justifying reasons do.
But this picture turns out to be ungrounded. The sense of intelligibility at issue is
the same both times.
Doesn’t this just show that justifying reasons do serve to explain actions, thus
undermining my earlier argument that, on Smith’s picture, justifying reasons
cannot be motivating ones? No. What has been admitted is that normative truths
about what is a reason for what and what is a better reason than what are relevant
to the question whether a putative belief–desire explanation of an action is a good
one. But no sense whatever has been given to the idea that a truth can motivate—
      253
that the truth figures in the explanation. And this is what we would need if we
were to show that justifying reasons can be motivating ones.
II
So it seems that the account of the distinction between motivating and justifying
reasons does indeed fall foul of the maxim. Now this should not be news, exactly.
For it is what I see as the main message of Bernard Williams’s notorious paper
‘Internal and External Reasons’ (1981). Williams there suggests that external
reasons always collapse into internal ones, conceived of as sub-Humean packages
of beliefs and desires. The pressure which he applies to the notion of an external
reason comes largely from the maxim. He urges that as soon as an external reason
becomes capable of motivating an agent, the agent is one about whom a true
internal reasons statement could be made (1981: 107). External reasons cannot
both remain external and satisfy the maxim.
This is just another way of making the point I have already made, against the
protagonists of the radical account of the motivating/justifying distinction. That
point is that once we admit a Humean (or sub-Humean) account of motivating
reasons, there can be no other reasons than these. At best, justifying reasons will
be a subclass of motivating ones. In this sense, Williams argues that the theory of
motivation (a theory which has an inbuilt tendency towards Humean answers)
expands to cover the whole of the territory covered by the theory of practical
reasons. The general form of this position is that, with Humean reasons in place,
we need no other sort of reasons than Humean ones. Or, more strongly, that with
Humean reasons in place, there is no space left for any other sort of reason.
The most promising reply to this is that there is no reason why there should not
be more than one style of explanation of the same event, without either competing
with the other. Dan Dennett has made us used to the idea that the behaviour of a
chess-playing computer can be explained in three different ways. So why shouldn’t
human actions be explained not only in the Humean way, but also in a quite
different way that appeals to justifying reasons? The answer to this is that this
might perhaps be possible if the second way was quite different from the first, as it
would be if the first was, as we might put it, brutely causal and the second were
quite other—rational and normative. But that is not how things are in our case.
Humean explanation is not presented as brutely causal, but causal and rational at
once, since it is supposed to be a sort of causal explanation that is subject to
normative rational constraints. That being so, it is necessarily incompatible with
the coexistence of another style of rational explanation.
My own view, then, is that the structure of Williams’s position is sound, and,
what is more, that it is sound no matter how we tinker with the details of what
I have so far been calling ‘Humean theory of motivation’. It is common enough to
254   
see references to thinkers who reject the Humean picture of the way in which
belief and desire combine in motivation. The vast majority, of course, accept that
picture. But most of those who dispute it merely tinker with the details. The best
known anti-Humean suggestions are those stemming from Nagel and McDowell.
But all that these two do, at least on the surface, is to suggest that on some
occasions beliefs can play a role other than that which Humeanism traditionally
allows them: moral and some prudential beliefs can lead desires rather than follow
them. Nagel and McDowell do not follow this up by considering what sort of
things beliefs capable of functioning in this way must be, nor by asking whether
this comparatively slight deviation from the Humean norm—a deviation, what is
more, that is explicitly supposed to leave intact the Humean form of explanation
for the vast majority of actions³—gives us anything like a different picture of how
action-explanation functions. The most trenchant anti-Humean position that
I know is the one which I adopted myself in my Moral Reasons (1993), where
I argued that beliefs are capable of motivation without the sort of support from
desires that Humeanism normally supposes necessary. But even this story is
essentially playing the Humean game. The style of action-explanation that is
being debated here is still the Humean style, and it is merely the players whose
interrelations are being reassessed. If I announce that beliefs are not inert,
secondary motivators, but meaty primary ones, all I have done is to tinker with
the details.
It is impossible, therefore, both to accept the maxim and to run the distinction
between motivating and justifying reasons in the ways that sees the former as any
sort of complex of beliefs and/or desires and the latter as facts, normative or
otherwise. Those who accept the distinction in this form do better, therefore, to
deny the maxim. This is what David Brink does (1989: 39–40), and he marks his
denial by talking not of the motivating/justifying distinction, but of the explana-
tory/justifying one—thus distancing himself from any attempt to say that justify-
ing reasons explain. In Brink’s view, justifying reasons explain why an action was
right, and why a fully rational person would have done it, but they never explain
actions done by ordinary mortals.
If the distinction is between explanatory and justifying reasons rather than
between motivating and justifying reasons, we cannot accuse its proponents of
flouting the maxim. For it is explicit that the maxim applies only to explanatory
reasons. This puts pressure on us to say why the maxim is true for all practical
reasons. Here is Smith’s reason for thinking it true:
Motivating and normative reasons do have something in common in virtue of

which they both count as reasons. For citing either would allow us to render an
³ Which is why I think of it as a hybrid theory that claims that Hume is sometimes right and
sometimes wrong.
      255
agent’s action intelligible. This is essential. For there is an a priori connection

between citing an agent’s reasons for acting in a certain way and making her
acting in that way intelligible: that is, specifying what there is to be said for acting
in the way in question. (1984: 95)
Now I think that this would leave Brink unmoved, since he would merely say that
the notion of ‘an agent’s reasons’ refers here only to explaining (motivating)
reasons, and has no need also to refer to justifying ones. So what can we put in
its place? Why should it be true that ‘if there are reasons for action, it must be that
people sometimes act for those reasons, and if they do, their reasons must figure in
some correct explanation of their action’?⁴ The best I can do is to resort to
rhetoric. How could there be this complex structure of reasons favouring and
disfavouring actions, if humans were incapable of registering the fact? And how
could it be possible in general for people to recognize the fact and not take it into
account in practical deliberation? Is there any difference here between practical
and theoretical deliberation? Suppose that there are reasons for and against
different beliefs, as there are for and against different actions. Again it seems
inconceivable that there should be this structure on the theoretical side unless
humans were capable of recognizing it, at least to some extent. And surely it is
inconceivable that we should do other than take the things we recognize to be
relevant to the question what to believe.
In my view, then, we have to accept that justifying reasons are capable of
explaining actions. This returns us to the position we were in before. All available
positions admit in some sense that Hume was right about something. And we
have seen that once we admit this, whatever the details of our Humean theory of
motivation, we end up with no room for other reasons than our Humean ones. So
is there no other possibility? There had better be, if we are to remain realists about
reasons, and suppose that there are matters of fact which constitute reasons for
action. To the extent that we suppose that the fact that an action is wrong is a
reason not to do it, or that her need for help is a reason for us to provide it, to that
extent we appear to be expressing a realist perspective on reasons, and nailing
ourselves to the mast of justifying or external reasons. How can we do this without
falling foul of the constraints already identified? The most obvious manoeuvre is
to abandon the attempt to add justifying reasons to Humean motivating ones, and
to try to move in the reverse direction, understanding motivating reasons in some
non-Humean way, in terms of what we have already laid down about justifying
ones. This approach is the one which seems to bear the most promise of avoiding
the trap we have identified, of so understanding the motivating ones that we leave
no room for justifying ones at all.
⁴ This way of putting the maxim is taken from Williams (1981: 102).
256   
Even the expressions I used in outlining this strategy need to be treated with
care, of course, since the motivating/justifying distinction is itself in question. It
would be best to start in a non-contentious way, trying to understand reasons in
favour of acting, and then writing an account of reasons why someone acted
within the constraints of the picture we have begun to develop. There is nothing
prejudicial about this procedure. It remains to be seen whether it is possible, and
that is what the rest of this paper is concerned with. One possibility, however, is
that at the end of the day we will find in our account no recognizable successor of
the theory of motivation at all. The very idea that there is room for such a theory
will turn out to be a Humean implant, which we had previously failed to notice
as such.
III
What we are after, then, is a realist account of reasons in favour of an action. But
our account must respect certain constraints. One of those constraints has already
been identified; it is the maxim that anything that is a reason must be capable of
explaining an action. Justifying reasons must be capable of being motivating
reasons. The second constraint is the realist nature of the story we are eventually
going to tell about reasons why agents act. This realist story is that justifying
reasons are facts—often normative facts. For instance, the reason why her action
was right was that it promoted something which is a good. However, we must also
admit and somehow capture the fact that no state of the world can motivate unless
it is recognized by the agent. This gives us three constraints:
1. A reason must be capable of explaining an action. (Justifying reasons must

be capable of also being motivating ones.)
2. Justifying reasons are facts or states of affairs—largely, but not exclusively,
normative ones.
3. States of the world and facts cannot motivate unless recognized by the agent.
There is a fourth constraint to be added to these. This is that:
4. The distinction between true and false beliefs on the agent’s part cannot
affect the form of the explanation which will be appropriate to his actions.⁵
It is the last two constraints that raise the real difficulties. They form a large part of
the pressure in favour of taking complexes of beliefs and desires (of whatever sort)
⁵ This constraint is also operative in Williams (1981: 102).

      257
as the sole possible explainers of actions. This works in the following way. Let us
suppose, in line with the first two constraints, that justifying reasons are largely
normative facts, and that these normative facts are capable of motivating. We have
to allow, however, that false conceptions of the normative are also capable of
motivating. People can be led to act by taking an action that is in fact wrong to be
right. The fourth constraint prevents us from saying that where our normative
beliefs are true, the facts believed are what motivate, and when they are false, the
motivation stems from the false beliefs. The third constraint ensures that even in
the favourable case, there is a belief in play; for recognition of the fact amounts to
no more than true belief. Since we cannot appeal to facts on both sides, then, and
since we must find an explanation of the same form each time, we are reduced to
explaining by appeal to beliefs both times, i.e. to the familiar Humean form of
motivation.
This form of the argument from illusion is the argument that takes us from the
apparently innocuous distinction between two sorts of reasons—reasons in favour
of acting and reasons why the action took place—to the full-blown distinction
between reasons as motivating beliefs and desires and reasons as normative facts.
We have been forced to abandon the first constraint in the attempt to satisfy the
other three.
There is a second, quite distinct form of pressure leading to the same conclu-
sion. This pressure arises when we try to understand explanation by reasons as
causal explanation, without adopting anything very unusual in the way of a theory
of causation. If anything that motivates must be a cause, it is clear that justifying
reasons cannot motivate—cannot be anyone’s reason for acting—if we conceive of
justifying reasons as facts. For facts are causally impotent, especially if they are
normative.
I say that we are committed to this view so long as we do not play clever with
the theory of causation. There are moves in the theory of causation which would
undermine the supposed causal impotence of facts, notably ones that deny the
causal primacy of events. And there are others, some of which are owed to Hume,
that amount to denying the appeal of any notion of causal impotence. But I am not
tempted to make this move here. The real question is whether we are committed
to understanding explanation by reasons as causal explanation.
The point at the moment is only that once one insists on seeing reasons-
explanation as causal, one needs to see one’s reasons as capable of being causes,
and facts—especially normative facts—appear to many to be incapable of playing
this role. Normative beliefs (and desires, of course), however, can play it as well as
any other beliefs. So we have this second, causal pressure to make all our reasons-
explanations Humean.
The two forms of pressure come together when we notice that in cases where
our normative beliefs are false, there are no normative facts to do the causing, even
if we did manage to attain a conception of facts as causally potent.
258   
The situation then is that we seem unable to respect all four constraints at once.
We can only respect the third and the fourth constraint by abandoning the realist
picture of justifying reasons as normative facts.
IV
There is however a possibility which we have not so far considered. Our realism
amounted to the claim that people can be motivated by how things are (either
normatively or non-normatively). We assumed that this had to mean that we can
be motivated by normative facts—by the truths we recognize rather than by our
recognizing them. But that might not be quite correct. What is required might be
something less than that. To comply with our realism, what motivates must be
capable of being believed and capable of being true, but need not be considered as
something incapable of being false. We could, that is, respect the fourth constraint
without being reduced to seeing beliefs as the common explainers. For between
beliefs on the one hand, conceived as psychological states, and normative truths,
facts or states of affairs (all of which are incapable of falsehood) lie such things as
what is believed. What motivates agents to action, I propose, and what justifies
their actions in favourable cases, is what they believe, not their believing it. The
belief may perhaps give their reason, or reveal what the reason was, but it will not
actually be that reason.
Let us try, then, to give an account of reasons for action in terms of what is
believed rather than speaking directly in terms of truths. This will mean yielding
in a way to one form of pressure applied by the third constraint. That constraint
pulls us towards the suggestion that belief is what is doing the motivating, so that
truth adds nothing. That suggestion is correct if it means that the truth of what
explains the action adds nothing to the explanation, being all concerned with the
question of justification, but false if it means that the thing which is capable of
both truth and falsehood adds nothing. For how could what the agent believes fail
to add to the explanation?
This position is capable of respecting all four constraints. Justifying reasons are
capable of explaining actions, though of course they can only do so when actually
believed. For them to do so, the first requirement is that what explains the action
was a true conception of how things are, either normatively or non-normatively—
using the term ‘conception’ as a synonym for ‘what is believed’. In such a case a
justifying reason is a motivating reason and the motivating reason a justifying one,
i.e. one which succeeds in showing that the action was right. In unfavourable cases
the action will still be explained as well as in a favourable case, though now not
justified. But it is still the case that that which explains is the right sort of thing to
do some justifying, and would have justified had it been true. Finally, the true/false
distinction has turned out not to affect the form of the relevant explanation at all.
      259
In this way, then, one satisfies all the constraints by making adjustments at both
ends—at the motivating end and at the justifying end. At the motivating end we
have abandoned the view that beliefs and desires are motivating reasons for
action. Instead, we have announced that it is what is believed that motivates. At
the justifying end we have drawn back to some extent from the claim that
justifying reasons are facts—only to some extent, however, because it remains
the case that when what you believe is true, it is a fact. In such cases, what you
believe both explains and justifies what you did. We have drawn back more
severely from the claim that justifying reasons are states of affairs or parts of the
world. Or rather, we can only make this further claim if we adopt either the
metaphysical view that facts are parts or constituents of the world, or the exter-
nalist claim that what is believed can be a state of affairs. Nothing in what I have
said, however, prevents that claim. For it would be wrong to think that if we make
either of those moves, we are again in conflict with the fourth constraint. The
fourth constraint speaks only about the form of the relevant explanation, and we
have (comparatively trivially) been able to satisfy that constraint by announcing
that whether the beliefs are true or false, it is not the belief that motivates but what
is believed. Nothing that I have said, however, does much to establish either the
metaphysical claim that facts are part of the world, or the externalist claim that
what is believed can be a state of affairs. At least I hope not, for I would like to be
able to propound the view that it is what is believed that both motivates and
justifies without needing yet to offer a substantial theory of what these peculiar
objects are.
The real hackles that my account will raise are causal hackles. We noticed that
truth is causally impotent. So surely is what is believed (most obviously when the
belief is false). If it is genuinely what is believed that is doing the explaining, and
not the belief, how can any explanation that appeals to reasons be shown to be
causal? The answer is that it cannot. Explanation by motivating reasons is one
thing and causal explanation another. This is the price of saying that there are
such things as justifying reasons. Whether it is possible that actions should be able
to be explained in a brutely causal way as well as by appeal to reasons, and what
the elements of such a causal explanation might be, I am doubtful. But I will not
argue against the possibility here. If I did, my general line would be that it is not
obvious what, other than rationality constraints, could prevent situations arising
in which wholly unsuitable beliefs and desires are the ones that actually caused the
action. On a brutely causal account, where the players are still supposed to be
beliefs and desires (and not neural changes, for instance), what will rule out the
possibility of my taking the bus being caused by the belief that it might have been
called Henry and my desire to have worked harder when I was a student? Of
course, the action is not rationalized by these two causes. But in my view it makes
no sense to admit this and then still insist that they might have caused it
nonetheless. So the notion of a brutely causal explanation of action seems to me
260   
to make no sense. We might have a brutely causal explanation of a bodily

movement, but that is another matter.
We have perhaps become recently inured to talk of reasons as causes, partly
because of remarks by Donald Davidson. But at best he only succeeded in dealing
with one of the reasons for refusing to think of reasons as causes—the one that
concerns our ability to provide independent specifications of cause and effect.
There is another reason for denying that reasons are causes. This is that there is no
possible metaphysical distinction between the action and the reasons for it, for if
we subtract the reasons from the action, there is not enough left to be an action at
all. We would only have something like a bodily movement, or the initiation of a
change. So we can make no real sense of the idea that one and the same action
could have been done for different reasons, though we can of course understand
well enough the claim that one and the same action-description can be instanti-
ated in different actions, each with its own reasons.
Is there then such a thing as the theory of motivation? I suggested earlier that the
very concept of a theory of motivation is a Humean concept. This is not because it
is a causal concept; the theory of motivation can be conceived teleologically rather
than causally. It is because the theory of motivation is one within which only
broadly Humean answers are possible, once we distinguish between motivating
and justifying reasons in the sort of way that I have tried to dismantle. To suggest,
by contrast, that the agent can be motivated by what he believes is to abandon the
theory of motivation, so understood, completely, and to put in its place the basis of
a more homogeneous and explicitly normative theory of practical reason.
The reasons that explain our acting must indeed be psychologically real, as
Michael Smith puts it, but they need not for that reason be psychological states.
They can instead be what we believe, which is surely psychologically real in the
sense of being psychologically present to us.⁶
⁶ I am grateful to David McNaughton and to Bernard Williams for advice about this paper, and to
Derek Parfit for forceful discussion.
17
On How to Act—Disjunctively
In this paper, I revisit some issues which I addressed briefly in my Practical Reality
(2000a). I argued there against disjunctive accounts of acting for a reason. I still
think that that is the position I want to take, but it is worth looking again at the
issue, in the light of increasingly sophisticated understandings of the nature of
disjunctive accounts elsewhere.
The issue is only interesting if we intend any disjunctive account that we come
up with in the theory of action to be relevantly similar to disjunctive accounts of
experience. It is not very hard to say something disjunctive about action. Of course
not everyone agrees on the form that a disjunctive account of experience should
take; other essays in the volume in which this paper first appeared attested to that.
But there seems to be something of a consensus on the following points, which for
that reason I will here lay out very snappily.
First, perceptual and non-perceptual experiences are not to share any common
element.
Second, perceptual and non-perceptual experiences have no common features
or qualities. This point is worth distinguishing from the preceding point for
reasons that will emerge below.
Third, non-perceptual experiences are to be characterized entirely as episodes
that the experiencer cannot distinguish from perceptual experiences. This negative
characterization is all that can be said in general about the non-perceptual, in
explication of its nature as experience. Non-perceptual experiences not only have
no qualities or features in common with perceptions; they have no common
features or qualities among themselves either.
Fourth, for this reason, and in this sense, the perceptual case enjoys a certain
primacy in our understanding of experience. There are, in the broadest terms, two
sorts of experience, and the second is to be understood in terms of the first (that is,
as indistinguishable from it).
Fifth, perception reaches out to and essentially involves the perceived object.
Perceptual experience is understood relationally. Non-perceptual experience is
not understood relationally (because there may be nothing there), though it is
understood in terms of something that is relational.
262   
Sixth, and for this reason, a disjunctive account of experience would be

empirically false if there were some non-perceptual experiences that are distin-
guishable (by their owner, and in the relevant way) from perceptions.
There are some friendly rivals to the disjunctive account, which I will consider
in a moment. Before that, I want to make a couple of comments on the points
above. First, it seemed to be worth distinguishing potential common elements
from potential common features or qualities. Talk of elements leads one to think
in terms of building blocks. If two items share a common element, at least one of
them must be metaphysically complex, built out of parts. Two items might have
common qualities without being metaphysically complex in that kind of way. If we
conceive of a non-perceptual experience in such a way that the same sort of thing is
present in a perceptual experience, attended of course by the presence of the
relevant object (probably acting as a cause), we are conceiving of the percep-
tual experience as metaphysically complex, and, what is more, as a complex of
just the type that the disjunctivist is in the business of trying to get away from.
The disjunctivist allows that a perceptual experience is a complex, since it
consists partly, but not wholly, of the relevant object, the object perceived. For
disjunctivists, then, there is a right way of being complex and a wrong way.
The right way is an indissoluble way; whatever is involved in the perceptual
experience, in addition to the presence of the object perceived, could not be
present alone. For if one takes the object away, there is not enough left to
stand independently.
Talk of common qualities, by contrast, does not lead one to think in terms of
building blocks. Two experiences could both be representational, for instance,
without this tending to show that they have some common element. (Just as a
picture on the wall and a dream can both be representational without having some
common element—though they may for all that have common qualities.)
I mention this because it seems relevant to a suggestion that I made some while
ago (Dancy 1995a; repr. as Chapter 15 in this volume), one that has not found
favour with leading disjunctivists. This was that we may perhaps start from a
workable disjunctive conception of experience of the favoured sort, like this:
A subject S is enjoying an oasis-experience iff either

(1) an oasis is manifesting itself to S, or
(2) it is for S just as if (1).
In this version we have, as yet, only characterized the second disjunct as like
what it is not. But, I surmised, ‘there may be available a more explicit character-
ization of the second disjunct, and in a totally explicit version . . . it would indeed
be characterized in that better way’ (1995a: 436). Tim Crane responds to this
as follows:
  — 263
[P]art of the point of disjunctivism is to characterize what genuine perception is,

and to show how this can be done without assuming a substantial conception of
the ‘common factor’ between perception and hallucination: that is, a conception
of a common state which can play a significant psychological or explanatory role.
But if Dancy’s challenge were answered, by giving a positive account of the
hallucinatory experience (for example, in terms of qualia), then this would
amount to giving a substantial account of the common factor between
perception and hallucination. And this is what the disjunctivist says cannot
be done. (2005: 3.4.2)
It seems to me, however, that Crane here supposes, unnecessarily, that any
positive characterization of hallucination, any account of its positive qualities, as
it were, would be bound to specify a ‘common state’—a sort of building block
common to perception and hallucination. That thought is surely a mistake, for
two reasons. First, we don’t need to think of a quality as a part or state. Second,
even if we did identify, in the case of hallucination, something capable of func-
tioning as a part, or state, we would not be thereby forced to allow that the
perceptual state itself also had that state as part of it. We could explain the
indistinguishability in terms of a common quality that is not a common part.
If this were all that there is to be said, I would be suggesting that the need to
deny any common element cannot be presented as a reason not to look for a
positive characterization of non-perceptual experience. But Mike Martin has more
to say. In ‘The Limits of Self-Awareness’, he does, admittedly, seem to echo
Crane’s point:
If what marks [the non-perceptual] cases out in the first place is just that they
involve the absence of perception, then one may worry that whatever fixes what
they have in common with each other will apply equally to any case of percep-
tion. That is to say, the further specification of hallucination will be something
which is present not only in all cases of illusion or hallucination but also in the
case of perception. The disjunctivist will then be left in the unhappy position of
conceding that there is a common element to all of the cases. Now if the common
element is sufficient to explain all the relevant phenomena in the various cases of
illusion and hallucination, one may also worry that it must be sufficient in the
case of perception as well. In that case, disjunctivism is threatened with viewing
its favoured conception of perception as explanatorily redundant. (2004: 46)
Here again, I would say that common qualities need not generate a common
element. But Martin had a different, and, I think, a stronger point. He contrasts
two accounts of the non-perceptual experience, modest and immodest. The
immodest account offers a list of characteristics common to all non-perceptual
experiences. The modest account characterizes such experiences as any episode
264   
relevantly indistinguishable from a perception. Already we see an advantage in the

modest account. For the immodest account can hardly dissent from the modest
one. We all have to start from that—from the notion of whatever is, or is
indiscriminable from, a perception. The immodest account goes beyond that,
and in going beyond it, courts trouble. For even if we suppose that the immodest
account succeeds in specifying features sufficient to make an episode an experi-
ence, it is vulnerable to the ever-present possibility of there being some further
type of episode that is also indistinguishable from a perception. The immodest
account may, that is, succeed in offering sufficient conditions for an episode to be
an experience, but is unlikely to be able to offer necessary ones. The modest
account, by contrast, offers only one condition, and it is both sufficient and
necessary.
It will still be the case, on this showing, that if we could conceive of an
experience that was not relevantly indistinguishable from a perception, the dis-
junctive conception would turn out to be inadequate—empirically inadequate, as
it were. Are we convinced, then, that this is impossible? A counterexample would
have to be an episode that was an experience but which we can still distinguish
from a perception. One is tempted to write ‘can distinguish on internal grounds’.
Certain ‘external’ ways of distinguishing are ruled out. Asking one’s friends, for
example, will not count. Nor will appealing to empirical evidence that experiences
of this sort are often, or generally, or even always hallucinatory. More to the point
would be Austin’s claim that there are lots of differences between dream experi-
ences and waking experiences. Presumably what prevents one from appealing to
these is that one cannot do so when asleep. It seems to me, however, that there is
room for a sort of experience that one knows on ‘internal’ grounds alone to be not
a perception. I once suffered briefly from otitis, an inflammation of the inner ear,
which has various consequences. One of them was that as I lay in the bath, gazing
out at my toes and the taps at the other end, it suddenly began to appear that
everything was moving to the left around my point of view as a fixed point, but
also that my point of view was moving with it. It now seems to me that I was
dealing then not with two experiences, each, as it were, incompatible with the
other, but with one experience with contradictory content. Suppose that one
allows this as a description of the relevant phenomenon. The question then is
whether this is a potential counterexample to disjunctivism. Is an experience with
contradictory content one that is distinguishable on ‘internal’ grounds from a
perception? It does not seem to me that we can say ‘no’ in answer to the question
simply on the grounds that it is, somehow, independently established that the
external world cannot contain contradictions.
I certainly don’t think of this as anything approaching a decisive objection to
the disjunctivist account of experience. I mention it here merely to reinforce the
point, which will be important later, that accounts of this sort are liable to
empirical refutation if we are not careful. This is the first point that I want to
  — 265
take forward from the brief survey in this section. The second is that we will be
considering later what form a disjunctive conception of action might take, and at
that point we will need to ask ourselves whether the reasons given by Crane and
Martin for sticking to the pure, unexpanded form of disjunctive account in the
theory of experience apply with equal force in the theory of action. In fact my
feeling is that it will not be so hard there to produce a set of individually sufficient
and jointly necessary conditions.
There are other ways of denying the existence of a common element. Timothy
Williamson (2000: 44ff.) has shown us that, though disjunctivism says very little,
being largely engaged in denial, there might be what he calls a ‘non-conjunctivist’
position that says even less. What non-conjunctivism does, and all that it does, is
to deny the existence of a common element. It says, of some concept, that there is
more than one way in which that concept can get to be instantiated, and that the
various different instantiations have no common element. So no instantiation
consists in a combination of some other instantiation plus some added extra, and
there is no part of, or element in, any instantiation that is also a part of all the
others.
There is a simple model for non-conjunctivism, the case of colour. Something is
coloured if it is red, or blue, or green, or . . . : there need be no end to this list. (That
is another feature of non-conjunctivism, as compared with disjunctivism proper.)
No colour consists in some other colour plus some further element or elements.
And there is no common element to all the colours. We cannot say that to be red is
to be coloured and red (though it is, of course, to be coloured in the red way), and
no other candidate for the common element seems forthcoming.
It is important that one might still say that all colours have something in
common, a common feature; colours, for instance, are a visual phenomenon
(whatever that vexed phrase means). It is just that what they have in common is
not a common element. And it is also important that none of the conjuncts, at
least in the colour case, enjoys the sort of priority that disjunctivism attributes to
genuine perception. But that is, no doubt, partly because we are dealing with a very
different sort of case.
Not only is there a model for non-conjunctivism, but Williamson offers us an
account of knowledge that is neither conjunctivist nor disjunctivist, even though a
part of that account is a list of disjuncts, different ways of knowing that p. For
Williamson, knowledge is the most general factive stative attitude. There are more
specific stative attitudes, seeing, remembering, etc., and if I know that p I have one
of those attitudes to p. But seeing that p is not knowing that p plus some other
element, and Williamson’s account of knowledge does not consist in the (very
266   
open-ended) list of disjuncts, ways of knowing, in the way that according to

disjunctivists the account of experience consists in their two-line list.
Williamson does appeal to the example of colour, and it is tempting to
generalize that example, and say that the basic model here is that of determinate
and determinable. But though I used to think in that way, I now think it is better
avoided. This is largely the result of reading David Stanford’s impressive essay on
the determinate/determinable distinction in the Stanford Encyclopedia. What
I learned from this is that the distinction, first introduced by W. E. Johnson in
his Logic (1921), is one of which it is far from easy to give a proper account. It
seems, then, better to make such use of the colour model as one wishes and not to
generalize it in terms of determinate and determinable.
Of course, once one is in the business of writing lists, all sorts of models are
available. Any case of what one might call ‘plural constitution’ will suit our bill.
Something is a good night out if it is one spent at a good play with a close friend, or
one spent in a pub with other like-minded citizens, and so on. Or we could look at
moral philosophy. An action is wrong iff it is murder, or rape, or torture, and so
on. How about this: something is dangerous if it is a loaded gun or if it is
indistinguishable from a loaded gun (the police will still shoot you for carrying
it). But enough on this score.
With these preliminaries in hand, I now turn to the philosophy of action. The
disjunctive conception that I considered in Practical Reality was not a disjunctive
conception of acting. The most general account of acting that I know is that to act
is to cause a change; I accept this account and see no need to elaborate it
disjunctively. What I did consider was a disjunctive conception of acting for a
reason. There were two reasons for taking the possibility of such an account
seriously. The first was that there is in the case of action the same pattern of
successful and unsuccessful cases as we find in the case of perception, with the
same array of possible responses. In both cases, that is, some philosophers are
tempted to understand the successful case as what is present in the unsuccessful
case plus some appropriate extra element, perhaps a suitable cause. Others, feeling
that this approach distorts our understanding of the successful case, insist on a
very different reading of the relation between the two, and on taking the successful
case as the starting point, the primary point of reference. I will call these the
Cartesian and the non-Cartesian approaches.
More explicitly, the Cartesian approach in the philosophy of action starts from
a case where the agent’s apparent reason for acting is that p, but where it is not the
case that p. (This is only one form of unsuccessful case: there are others.) It
announces that in such cases we are forced to say that the reason for which the
  — 267
agent acts is not that p, but that she believes that p. It then announces that this
forces us to say the same of the successful case, the case where the agent is right in
believing that p. In such cases, it remains the case that the reason for which the
agent is acting is not that p, but that she believes that p. The non-Cartesian
approach (mine) takes the opposite way in. It starts from a case where the agent
is right about whether p, and takes it that in such a case the agent’s reason for
acting is that p; it then insists that this is still the agent’s reason even in cases where
it is not the case that p.
Both approaches are awkward, and both are awkward in the sort of way that is
usually due to bad theory. As Aristotle said, they leave one saying things that
nobody would say unless defending a theory. It is awkward to say that one’s reason
for action is never such a thing as that the sun has come out; the Cartesian view
seems to say this, but it flies in the face of common experience. (In saying this I am
not meaning to appeal to ordinary language.) But it is also awkward to say that
one’s reason is that the sun has come out when in fact it hasn’t; the non-Cartesian
view is in trouble too. It seemed, then, as though an explicitly disjunctive account
might resolve the situation, at the cost, of course, of abandoning both the
Cartesian and the non-Cartesian views. For what a disjunctive account does is
to admit such differences as there may be between successful and unsuccessful
cases, but somehow finesse the point by establishing that both are nonetheless
perfectly good instances of the definiendum—or better, the intelligendum,
since what we are looking for here is not a definition but an explanatory
characterization.
The other motivation for thinking in disjunctive terms was that one was
thereby enabled to accommodate all the necessary differences between successful
and unsuccessful cases without being in danger of breaching an influential maxim
in the philosophy of action: the distinction between true and false should not
affect the form of the relevant explanation.¹ This may not be apparent. The idea
was that both Cartesian and non-Cartesian approaches respect this maxim by
giving the same account of the one sort of case as they give of the other. It is only if
we give different accounts of successful and unsuccessful cases that we are in
danger of breaching the maxim. But a disjunctive conception can get away with
giving different accounts without breaching the maxim. For our intelligendum is
just ‘A acts for the reason that p’. We can say that when A acts, the relevant
explanation of his action will just be that he was doing it for the reason that p,
whether A is right or wrong about that. And so our explanations will have the
same form, whether A is right or wrong about whether p. In either case we will
account for his action in terms of the reasons for which he acted. That we give a
¹ I associate this maxim with Bernard Williams, but in fact it was important to what used to be called
the Strong Programme in the sociology of knowledge, notably in Bloor (1991).
268   
disjunctive conception of this, that is, that things begin to diverge from here on,
involves no rejection of the explanatory maxim.
There seemed, then, to be good reason to look carefully into the prospects for a
disjunctive conception of ‘A acts for the reason that p’. However, in doing this,
I was influenced by my earlier views that a disjunctive conception can offer a
substantial conception of the second disjunct, and that it can offer more than two
disjuncts. If we can say something substantial about a case which is not a
successful one, why should we not suppose that there can be more than one sort
of unsuccessful case? And, once we do suppose this, surely we should enshrine
that possibility in our account. This is exactly what we find in the case of acting for
a reason. There is indeed more than one way in which things can go wrong. The
agent can be wrong about whether p, or wrong about whether, if it were the case
that p, that would be a reason for acting. These thoughts led to my first attempt at
a disjunctive (better: trisjunctive²) conception of acting for a reason (Dancy
2000a: 140):
A Φs for the reason that p iff

either p and that p is a reason for Φ-ing and A Φs in the light of the fact that p;
or it is not the case that p, but A takes it that that p is a reason for Φ-ing,
and A Φs in the light of his belief that p;
or p, but that p is not a reason for Φ-ing, but A takes it that that p is a
reason for Φ-ing, and A Φs in the light of the fact that p.
This was my first account, but there is a mistake in the second trisjunct; this
should have read:
or it is not the case that p, but A takes it that p, and A Φs in the light of his
belief that p.
One might suspect that some sleight of hand is involved in the phrase ‘in the light
of ’, a phrase which I admit is important to my approach to these matters. It is
partly important because it enables me, even in the first trisjunct, to allow (indeed
insist) that A believes that p; for one cannot act in the light of a fact that one does
not recognize to obtain. The point is only that A’s belief that p is not, in the first
trisjunct, specified as A’s reason for Φ-ing. Still, one can, if one wishes, reformulate
the whole thing thus:
² Better in one way, worse in another. The ‘dis’ in ‘disjunctive’ does not mean ‘two’, but ‘apart’, so
that ‘disjoint’ does not mean ‘split into two’ but ‘parted’ or ‘separate’. So my coining of ‘trisjunctive’ is
an etymological sport. But I keep with it because it draws a useful contrast.
  — 269

either p and that p is a reason for Φ-ing and A’s reason for Φ-ing is that p;
or it is not the case that p, but A takes it that p, and A’s reason for Φ-ing is
that he believes that p;
or p, but that p is not a reason for Φ-ing but A takes it that that p is a
reason for Φ-ing, and A’s reason for Φ-ing is that p.
This should be contrasted with what we would write if we were trying to copy a
disjunctivist conception of experience:

either (1) p and that p is a reason for Φ-ing and A Φs in the light of the fact
that p (=A’s reason for Φ-ing is that p);
or (2) it is for A as if (1) were the case.
It is worth listing the respects in which my version is, and is not, a version of
disjunctivism proper. First, it offers no common element.³ Second, it mentions no
common property or feature. Third, it is at least compatible with the claim that the
three trisjuncts are indistinguishable from each other. Fourth, there is a sense in
which the successful case is understood as reaching out to the world. Success as
represented in the first trisjunct, after all, is not an extra that is added to what is
present in the second trisjunct.
But there are further respects in which my version departs from disjunctivism’s
overall aims, beyond the mere fact that we have trisjuncts rather than disjuncts
and so cannot run the pure disjunctive form. First, there is no explicit represen-
tation of the supposed primacy of the successful case. Second, it is not clear that
the relationality involved in the first trisjunct is suitably like that which disjuncti-
vists find in perception. There is a relation between agent and world required for
the successful case, but what remains, if one takes that relation away, is capable of
standing alone. Suppose it were not the case that p. Then the second part of the
first trisjunct would have to be understood subjunctively, and we would be left
with this:
That p, if it were the case, would be a reason for Φ-ing and A’s reason for Φ-ing
is that p.
³ As it stands, at least. But of course the conjunctive form of the various trisjuncts does open the
door, as it were, to thoughts about common elements, and indeed I go on to allow that common to all
three is that A believes that p. But even if this does stand as a common element, it is not that one
trisjunct (say, the first) consists in the presence of another trisjunct plus some further feature, which is
what the disjunctive account of perception is primarily concerned to avoid.
270   
One might think that an element of relationality remains, in the expression ‘A’s
reason for Φ-ing is that p’. Now as I understand that phrase, it is not factive; A can
act for the reason that p even where A is mistaken about whether p. And if it is not
factive, it can remain in place untouched even when it is not the case that p.
Others, however, disagree with me about this and hold that ‘A’s reason for Φ-ing
is that p’ is factive, supposing it to entail that p; and at the moment, in considering
the possibility of a disjunctive account, we are trying to accommodate their
doubts. And one might, for that reason, say that there is an appropriate relation-
ality in the first disjunct. But, as I say, the matter is debatable.
So my trisjunctive account does not really enjoy some of the more interesting
features of true disjunctivism. And I have a further reason of my own for rejecting
it, which we have already seen. It seems to me that the second trisjunct, if we take
it au pied de la lettre, is false. This is, of course, a substantial matter. But the
arguments I gave for this conclusion in my Practical Reality still seem to me to be
good ones. They amounted to pointing out that there is a substantial difference
between acting for the reason that p and acting for the reason that one believes
that p. If my reason for action is that I believe that there are pink rats in my boots,
or that I believe that everyone is out to get me, I will probably be calling a
psychiatrist. If my reason is that there are pink rats in my boots, I will probably
be calling a pest control expert, and if it is that everyone is out to get me,
I probably won’t be calling any people in white coats at all. A rather different
example is my crumbly cliff. Suppose that I believe that the cliff is crumbly, and, so
believing, I am more likely to get nervous and lose my grip. This very fact—that
I believe it to be crumbly—might be my reason for not climbing it. And this is a
different reason from the reason that it is crumbly. If my reason were that the cliff
is crumbly, that reason would vanish if the cliff ceased to be crumbly; if my reason
is that I believe the cliff to be crumbly, that reason persists so long as I have that
belief, whether the cliff ceases to be crumbly or not.
For these reasons (all elaborated in Dancy 2000a: ch. 6), I reject the suggestion,
encapsulated in the second trisjunct, that, where things are not as I suppose them
to be, my reason is not that p, but that I believe that p. So I cannot accept the
trisjunctive account; it seems to me to involve a substantial philosophical error.
My own view now is that one could combine the first two trisjuncts into one,
returning us to a disjunctive theory—of a sort:

either that p is or would be a reason for Φ-ing and A’s reason for Φ-ing is that p;
or whether it is true that p or not, that p would not be a reason to Φ, but
A takes it that that p is a reason to Φ, and A’s reason for Φ-ing is that p.
Another way of writing this would be:

  — 271

either that p is or would be a reason for Φ-ing, A takes it that p and A Φs in
that light;
or whether it is true that p or not, that p would not be a reason to Φ, but
A takes it that p and that that p is a reason to Φ, and A Φs in that light.
On this version the original trisjunction had a mistaken focus, since one of the
driving forces behind it was the possibility that the agent’s belief that p be false. As
things now stand, however, the issue is not the truth or falsity of that belief, but the
light cast on the action by what one believes, whether what one believes is true or
not. Would the supposed fact that p cast a favourable light on Φ-ing if it turned
out to be the case? This gives us:

either A Φs in the light actually cast by what he believes, namely that p;
or A Φs in the light apparently cast by what he believes, namely that p.
But the real point, for me, is that each of these disjuncts effectively ends with the
very thing we were trying to give an account of. We could have shortened the
whole thing and ended up with just this: A Φs for the reason that p iff A takes it
that p and A Φs in that light. Here all pretence of disjunctivism has been
abandoned.
What about the pure disjunctive account given above? Here it is again:

either (1) p and that p is a reason for Φ-ing and A Φs in the light of the fact
that p (=A’s reason for Φ-ing is that p);
The main problem for this account is that it is empirically false. Its seeming to me
as if I am acting in the light of a particular fact is no proof at all that this fact is
really the reason for which I am acting. We just don’t have this sort of access to our
reasons. It might well be that I seem to myself to be acting from pure altruism
when in fact my motivation is not at all something to congratulate myself about.
Things would be no better if we had followed the line of reasoning just before,
and offered this account:

either (1) A Φs in the light actually cast by what he believes, namely that p;
272   
This is still empirically false, and for exactly the same reason. So my present
conclusion is that no disjunctive account of acting for a reason is defensible.
In this final section, I consider suggestions made by Jennifer Hornsby at the

conference held in Glasgow in 2005 of which this volume is largely, but not
entirely, a record, and adapted in her essay in this collection (2008). I discuss
four issues.
4.1
Hornsby’s original proposal was for a form of disjunctivism in the theory of action
that is very different from the ones I have considered so far, thus:
If X Φ-ed because X believed that p, then

either X Φ-ed because X knew that p;
or X Φ-ed because X treated [the proposition that] p as an agent who knew
that p would treat it, although X didn’t actually know that p.
This is in fact a non-conjunctive account which simply borrows Williamson’s

suggested account of belief (2000: 46–8), under which to believe that p is either to
know that p or to treat the proposition that p as someone who knows that p would
treat it. In her published paper it now becomes (in shortened form):
(D) If X Φ-ed because X believed that p, then

either X Φ-ed because X knew that p;
or X Φ-ed because X merely believed that p.
My rejection of the disjunctive account of acting for a reason is no reason for me

to reject all disjunctive accounts of anything whatever in the philosophy of action.
So this new proposal, with its different intelligendum, needs to be considered on its
merits. In doing so, I bear in mind that I had two sorts of complaint about the
accounts which I considered in the previous section. I maintained in some cases
that some part of the account was philosophically misconceived, and in others that
the account as a whole was empirically false. Before looking to see whether this
new account (in its two different versions) is vulnerable in similar ways, however,
it is worth quickly running through the marks of disjunctivism in order to show
that what we are dealing with is genuinely disjunctive. I take the first version first,
  — 273
because the two versions differ significantly in this respect. First, the two disjuncts
have no visible common element. Second, they have no visible common quality or
feature. Third, there is a sort of indistinguishability involved; if X treated the
proposition that p as someone who knew that p would treat it, it must be for X as if
she knew that p. Otherwise X would presumably treat it differently. Fourth, the
first disjunct is world-involving, because of the way in which it treats the concept
of knowledge. Fifth, it enshrines the desired primacy of the successful case. And
finally, there is the possibility of empirical falsehood, since it might turn out, on
further examination, that it is possible to act because one believes that p without
satisfying either disjunct.
My own view, at the moment, is that this is just what happens, and so that as
things stand we need a further disjunct. (And I will argue below that the reasons
for restricting one’s account of experience to two disjuncts rather than some
other number, say three, do not apply to one’s account of intentional action.) It
seems to me that I am perfectly well aware of many things that I do not know to
be true, but on the basis of which I am willing to act. I am not treating them as
someone who knew them would treat them, because, at least to some extent,
I retain an open mind. The fact that I am willing to act on them is, of course, not
itself enough to establish that I am treating them as someone who knew them
would treat them. If this is right, however, the fault lies really with Williamson’s
account of belief.
I now turn to the new version. It has the first and second marks of disjuncti-
vism, and the third if, where X merely believes that p, it is for X as if she knew that
p. It has the fourth, but seems to lack the fifth. Finally, it is not obvious that the
account is empirically falsifiable. A way to reintroduce the possibility of empirical
falsehood would be to provide a substantial characterization of ‘merely believing’;
for the account as it stands seems to say that X acts because X believes that p if
either X acts because X knows that p or X doesn’t know that p but acts because
X believes that p anyway; and this is not terribly informative until we get the
desired account of ‘merely believing’. But once we get that, as in the earlier version,
the possibility of empirical falsehood returns. Hornsby says, however, that her (D)
‘makes no general claim about the relation between knowledge and belief ’. My
own guess, on this point, is that the introduction of any substantial account of
‘merely believing’ in the guise of a substantial relation between believing and
knowing will create a need for a third clause if we are to achieve a set of disjuncts
that are jointly necessary and individually sufficient, and thus avoid empirical
falsehood. But I note that this is an aim that Hornsby views as optional.
So we already have two ways in which the new version differs from the
paradigm case of disjunctivism. It does not enshrine the primacy of the successful
case, because it does not attempt to understand merely believing in terms of some
relation to knowing (which the first version did). And of course both versions, as
they stand, offer only necessary conditions rather than necessary and sufficient
274   
conditions, and as such can only claim to offer a partial understanding of the
relevant intelligendum.
Hornsby suggests that her account does do something that one might ask of a
disjunctive account, namely that it simultaneously brings together and distin-
guishes acting from knowledge and acting from mere belief, in the sort of way that
McDowell’s perceptual disjunctivism simultaneously brings together and distin-
guishes the experience of a perceiver and the experience of someone who has a
mere experience. But this seems to be an absolutely minimal conception of
disjunctivism; McDowell surely goes further than this in his claim that non-
perceptual experience is a case in which it is for the experiencer as if he were
perceiving.
4.2
I turn now to the question whether there are any philosophical complaints to be
made about the way in which the account has been crafted.
Hornsby thinks of her disjunctive thesis as concerned to relate cases in which
the agent acts for an ‘objective reason’ to those in which her reason is a ‘subjective
reason’. At least, those are the terms in which I would characterize the concern.
Hornsby however uses the terms ‘(F)-type reason’ and ‘(B)-type reason’, and I will
follow her in this. The agent’s reason is an (F)-type reason if it is of the form ‘that
p’; such reasons are canonically expressed in the form ‘A reason for X to Φ was
that p’. Subjective reasons, for Hornsby, are expressed in this form: ‘X had a reason
to Φ: X believed that p’. This distinction between (F)-type and (B)-type reasons
involves us in difficult issues. Hornsby starts from two points on which I think
everyone should agree. The first point is that someone’s acting intentionally is
always explained partly by her believing something, or that her believing as she
does is always at least relevant to the explanation. This, I would say, is true, though
it does not show that the agent’s so believing was any part of the reason she had for
acting as she did. The second point is that intentional explanations of action can
be given in two forms, psychologized or non-psychologized. We can say ‘the
reason why he is running is that he thinks the train is leaving’ or we can say
‘the reason why he is running is that the train is leaving’. My own view about these
two forms is that they are effectively equivalent, both being expressible as ‘the
reason why he is running is that, as he supposes, the train is leaving’. Hornsby’s
view is rather different. She wants to say that the sentence ‘X had a reason to Φ:
X believed that p’ plays a distinct explanatory role. Someone who thinks he is late
when he is not late has a reason to run nonetheless. What is that reason? One
might think that it cannot be that he is late, since he is not late. What can it be,
then? The only thing left is that he believes that he is late. If this were right,
Hornsby’s phrase: ‘X had a reason to Φ: X believed that p’ would have to be read,
  — 275
more explicitly, as ‘X had a reason to Φ, namely that X believed that p’, and not as
‘X had a reason to Φ, namely what X believed, that is, that p’. But this is not the
way that Hornsby wants to take it. We are dealing with a (B)-type reason when
something of this form is true: ‘X had a reason to Φ: X believed that p’. But the (B)-
type reason specified here is not that X believed that p. So the phrase ‘X had a
reason to Φ: X believed that p’ is not to be understood as ‘X had a reason to Φ,
namely that X believed that p’. This is the way I was originally tempted to take it,
but Hornsby stresses that the reason X had in these cases is the content of her
belief, that is, that p, not that X believed that p.
It seems, then, that whether one’s reason is (F)-type or (B)-type, it is still that p.
An (F)-type reason must be something that is the case, but a (B)-type one need not
be. One has a (B)-type reason as soon as one comes to believe that p, whether one
is right or wrong on the matter. (Though it seems to be required for this that that
p, if it were the case, would be an (F)-type reason, and one could add further
conditions as well, such as that one is not transgressing any relevant norms in
believing that p.) So apparently something that is not the case can be a reason for
one to act, and a reason for which one acts, so long as one believes it (with any
further conditions, of course). All we should say is that it is a (B)-type reason
rather than an (F)-type one, and canonically expressible in the form ‘X had a
reason to Φ: X believed that p’.
Consider now the standard worry that if we distinguish between (F)-type and
(B)-type reasons in this sort of way we will end up with two sets of reasons to act.
Is that true within Hornsby’s picture? (F)-type reasons are matters of fact, and
they become (B)-type ones once one comes to believe them to be the case. The
remaining (B)-type ones are not (F)-type reasons; they are those among the things
we believe which would be (F)-type reasons if they were the case. (Note that we are
saying here that they would be (F)-type reasons if they were the case, not that they
would be reasons if they were the case; for on Hornsby’s account they are reasons
already.) What then are we to say to someone who is proposing to do something
which there is (F)-type reason not to do, but which she has good (B)-type reason
to do? How should we advise her? We cannot, it seems, deny the reality of her
(B)-type reasons; they are no less real or genuine than her (F)-type ones. But this
seems to lead to two difficulties.
First, it might be that the strength of the (B)-type ones is such as to defeat the
(F)-type ones on the other side, and that does not seem to me to be a good result.
But it looks as if the only way to avoid it is somehow to downgrade the (B)-type
ones. There are two main ways of doing this. We can say that a (B)-type reason is
just a consideration that would be a reason if it were the case—which in this case it
isn’t, and so it isn’t in fact a reason at all. Or we can say that a (B)-type reason
affects the rationality of our action; rationality requires one to act according to
one’s beliefs, true or false. But it does not go to determine what in fact we have
reason to do, which, I think, really means that it is not a reason at all.
276   
The second difficulty is that we outsiders don’t really know how to relate a third
person’s (F)-type reasons to her (B)-type reasons. Once we have worked out what
her (F)-type ones are and what her (B)-types ones are, how do we assess them
against each other? (The previous worry allowed that reasons of the two sorts are
somehow commensurable; the present one questions that.) It is not at all clear
how in a case of conflict (and this would be the normal situation) we should advise
her. We can’t just say ‘go with the (F)-type reasons’ because we are trying to allow
that her (B)-type reasons are equally reasons in some way.
Of course, all this is a topic for a paper in its own right, and one that I have
treated elsewhere (Dancy 2000a: ch. 3). Let me just say that where we have both
X Φ-ed for a reason that X had: X believed that p—and

X Φ-ed for a reason: that p
Hornsby’s approach is close to saying that there are two reasons around, the one
that X had and another, real one. This seems to me very uncomfortable, especially
because, as far as I can see, the second claim entails the first.
4.3
I now turn to a different issue. Hornsby’s second version, in full dress, runs as
follows:
(D) If X Φ-ed because X believed that p, then

either X Φ-ed because X knew that p (and Φ-ed because p);
or X Φ-ed because X merely believed that p.
The added parenthesis introduces Hornsby’s view that when someone Φs because
they merely believe that p, they do not Φ because p, even when they are right about
whether it is the case that p or not. So truly believing that p does not itself put you
in a position to Φ because p; it only does that if your true belief is knowledge.
I confess that I am very suspicious of Hornsby’s intelligendum ‘X Φ-ed because
X believed that p’. It needs to be restricted in a certain way if it is to be even
relevant to the topic of intentional explanation, since there many cases where one
acts because one believes but in which one’s so believing has little to do with any
reason for which one acted (though it may be the reason, or part of the reason,
why one acted). I much prefer the intelligendum with which I started: X Φ-ed for
the reason that p (or something like that). But Hornsby has a quite different
attitude to that proposition. She glosses her first disjunct ‘X Φ-ed because X knew
that p’ as equivalent both to ‘X Φ-ed because p’ and to ‘X Φ-ed for the reason
  — 277
that p’.⁴ This effectively announces that when one merely believes that p, one
cannot act for the reason that p, even if one is right in so believing, and so takes a
stance on the very interesting question whether one can act for a reason which one
does not know to be the case, but merely believes to be so. Here is the example that
Hornsby relies on in suggesting that one cannot do this:
Edmund . . . believes that the ice in the middle of the pond is dangerously thin,
having been told so by a normally reliable friend, and . . . accordingly [Edmund]
keeps to the edge. But Edmund’s friend didn’t want Edmund to skate in the
middle of the pond (never mind why), so that he had told Edmund that the ice
there was thin despite having no view about whether or not it actually was thin.
Edmund, then, did not keep to the edge because the ice in the middle was thin.
Suppose now that, as it happened, the ice in the middle of the pond was thin. This
makes no difference. Edmund still didn’t keep to the edge because the ice was
thin. The fact that the ice was thin doesn’t explain Edmund’s acting, even though
Edmund did believe that it was thin, and even though the fact that it was thin
actually was a reason for him to stay at the edge. (2008: 251)
Hornsby’s point (echoed by John Hyman (1999) and Timothy Williamson (2000:
60–4)) is that action on a true belief in a reason-giving fact is not enough for it to
be the case that one is acting for that fact as a reason. It is only if one knows that
fact that one can be said to be acting for it as a reason. But this seems to me to
involve an invalid inference. One cannot argue that Edmund is not keeping to the
edge for the reason that the ice is thin from the premise that he is not skating there
because the ice is thin. In fact, I would say of both scenarios above (the one
where the friend is right and the one where he is wrong) that in them Edmund is
skating on the edge for the reason that the ice is thin, or that Edmund’s reason for
skating there is that the ice is thin, or that the reason for which he is skating there
is that the ice is thin. I don’t see that Hornsby’s two scenarios do anything to
upset that entirely natural position. And I therefore think it would be a mistake,
so far as that goes, to gloss her first disjunct ‘X Φ-ed because X knew that p’ as
equivalent to ‘X Φ-ed because p’ and to ‘X Φ-ed for the reason that p’.
There are, however, other reasons to think that one can only act on a reason
which one knows to be the case. Suppose that I am meeting my daughter off the
train. I don’t know that she is on the train, and I know that I don’t know this, but
I do believe that she is on the train. So I head off to meet her. What would I say if
asked what my reason is for meeting this train? I might say ‘My daughter is on it’.
⁴ The notion of a reason for which one acts is connected to that of an (F)-type reason at the end of
section 1.3 of Hornsby’s essay (2008: 251).
278   
But I would be more likely to say ‘I think my daughter is on it’, or ‘My daughter
should be on it’, or ‘My daughter is probably on it’. There is some pressure to
avoid offering the bald claim that my daughter is on the train as my reason if I do
not know this to be the case. But what if my response to this is to specify my
reason thus: ‘I think my daughter is on the train’? Are we to say that I have now
shifted my ground to something I know? In line with a certain approach to
Moore’s Paradox, I think that this remark is not to be understood as autobio-
graphical, but as a more guarded version of the claim that she is on the train. It is
not, then, that pressure can be put on me to restrict myself, in giving my reasons
for doing what I do, to things that I take myself to know. The pressure is more to
express, in the way that I give my reasons, the degree of confidence with which
I hold them.
So my own view is that there is not sufficient reason to restrict the phrase ‘X Φ-
ed for the reason that p’ to cases where X knew that p. This releases that phrase for
use as the overall intelligendum.
4.4
I end, as promised, with the question whether Mike Martin’s reason for insisting
that the only permissible characterization of the non-perceptual experience, at
least from the disjunctivist’s point of view, is that it is any episode relevantly
indistinguishable from a perception, applies with equal force in the theory of
action. Martin’s main reason was that only thus are we able to provide a disjunct-
ive account of experience under which the two disjuncts are individually sufficient
and jointly necessary for an episode to be an experience. A positive characteriza-
tion of the second disjunct would create a danger that, even though all episodes
with that character will be experiences, there might yet be episodes of a different
sort that are experiences as well. If so, the disjunctive account as a whole will offer
conditions that are individually sufficient, but not jointly necessary. One might
ask why that is so worrying. Is it merely a fetish with the disjunctive form? If we
need to add a third clause, and turn the whole thing trisjunctive, what is wrong
with that? One answer to this question is that the process might go on indefinitely,
so that prospects of giving a genuine account of experience begin to fade. Here we
might remind ourselves of Williamson’s list of forms of knowledge. There are
many different forms of knowledge (perceiving, remembering, proving, intuiting)
but the list of these does not, for Williamson, constitute any account of knowledge;
that account consists in the claim that knowledge is the most general factive stative
attitude. Perhaps there is the threat that the process might go on indefinitely. But
that is only a threat. There might be ways of averting the threat. In the philosophy
of action, it seems not impossible to achieve that. Take one of the versions of my
original trisjunctive account:
  — 279

either p and that p is a reason for Φ-ing and A’s reason for Φ-ing is that p;
or it is not the case that p, but A takes it that p, and A’s reason for Φ-ing is
that he believes that p;
or p, but that p is not a reason for Φ-ing but A takes it that that p is a
reason for Φ-ing, and A’s reason for Φ-ing is that p.
It seems to me possible that we could have an a priori certainty that no more

clauses will be required. What other sort of case could there be?
A rather similar response seems to me appropriate to a concern of Martin’s that
is voiced in the passage that I quoted earlier, but which I did not pick up at that
point. The relevant parts are these:
If what marks [the non-perceptual] cases out in the first place is just that they
involve the absence of perception, then one may worry that whatever fixes what
they have in common with each other will apply equally to any case of
perception. . . . [This gives us a common element.] Now if the common element
is sufficient to explain all the relevant phenomena in the various cases of illusion
and hallucination, one may also worry that it must be sufficient in the case of
perception as well. In that case, disjunctivism is threatened with viewing its
favoured conception of perception as explanatorily redundant. (2004: 46)
There is a term of art here that is not sufficiently remarked: ‘one may worry that p’.
One might respond to such remarks ‘one may but one need not’. Consider the
first worry. Martin’s passage here might have started: ‘If the only thing that
marks out the non-perceptual cases is that they involve the absence of percep-
tion, whatever else they have in common will apply equally to perception.’ If he
had said this, his claim would have been undeniable. But he doesn’t quite say
this. What he does say is rather that if we start by distinguishing a group of cases
simply as those that do not involve perception, we (might worry that we) will be
unable to find any further feature or element that they, and only they, have in
common. Well: one might worry about this, but it is far from certain that it will
turn out to be so.
Now turn to the second worry. It is interesting that Martin allows that this
worry, which is about explanatory redundancy, arises anyway, independent of the
present issue; indeed, he sees it as posing a serious challenge to disjunctivism, and
gives an answer to it. If that worry arises anyway, and there is an answer to it, this
particular source of worry does not add any difficulty to what we have already.
Further, what are ‘the relevant phenomena’? The disjunctivist maintains that
genuine perception is a world-involving state. Is it that this claim is useless in
the explanation of any ‘relevant phenomena’? If so, this would be bad news for
280   
naive realism. Luckily, there seems to be room to say that there are certain
phenomena that can be explained in either of two ways—though not in both
at once.
I conclude that the reasons given for rejecting any positive characterization of
the non-perceptual should not deter us in the philosophy of action, and may not
even be compelling in the philosophy of perception.
18
Enticing Reasons
In this paper I consider the possibility of a general distinction between two sorts of
reason: those that entice and those that are, as I will later say, more ‘peremptory’ in
style. The main idea here is that some reasons are obviously in the business of
more or less telling us what to do—though this way of putting the point will turn
out not to be quite right, and I use it here only to help you see what I am getting at;
moral reasons are an obvious case in point, but many reasons of prudence do the
same (the reasons for having your brakes checked are peremptory, in my sense).
Other reasons, however, are more to do with making an option attractive rather
than demanded, required, or right. At its most basic, the distinction I am after is
between those reasons that say ‘Come on, I’d be nice’, and those that say
something of a more forceful sort such as ‘You’d better do this’.
I will be trying to pinpoint the real basis of this distinction later, but an initial
suggestion is that there is no underlying requirement that one choose the most
attractive option. There may not even be a further reason of any sort (enticing or
peremptory) to choose the most attractive option, any more than there is a further
reason to do the action that overall one ought to do. It may be wrong to do an
action when there is more moral or prudential reason to do something else: you
should get your brakes checked, you should help the afflicted. But, I will suggest, it
is not wrong (morally or otherwise) to do a less enticing action when there is a
more enticing one available.
1. Reasons against enticers
The most direct attack one could mount on the notion of an enticing reason would
go something like this. John Broome writes ‘if you have a reason to q and no
reason not to q, then you ought to q’ (2000: 80).¹ I take it that on this view the term
‘reason’ means (among other things, perhaps) a consideration which one would be
wrong not to act on in the absence of any opposition; that you ought to q is a
trivial and analytic consequence of your having a reason to q and no reason not to
q. This being so, the idea of a reason that it is not wrong to fail to respond to, in the
absence of opposing considerations, is effectively self-contradictory. But supposed
¹ Broome does not believe this any more, I think; I am quoting old Broome, not new Broome.
282   
enticing reasons are of just this sort. For if there is no rational requirement that
one choose the most attractive option, there will be enticing reasons that it is not
wrong to fail to respond to—not to act on, that is.
This stance seems really to beg the question against enticing reasons. To
announce at the outset that good reasons are considerations it is wrong not to
act on unless they are outweighed or otherwise defeated by other reasons is simply
to contradict those who suggest that some good reasons are not like this. Once the
notion of an enticing reason is presented as a possibility, more than this will need
to be said if that notion is to be rejected as incoherent.
There is in this an interesting question just what is meant by saying that it is
wrong not to act on a winning consideration (or set of considerations, I presume).
On some accounts, ‘wrong’ here means ‘irrational’. But I agree with Tim Scanlon
(1998: 26–8) that someone who fails to act on an undefeated good reason need not
be subject to so severe a charge as that of being irrational. The charge of
irrationality should be reserved for grosser failures with respect to reasons.
There are less dramatic complaints that we can make; among these is that it is
wrong, though not irrational. But it need not be necessary to go even that far.
I would not say that it is irrational, nor that it is wrong, not to act on an enticing
reason when the alternative is to do something for which there is no reason
whatsoever (enticing or non-enticing): it is not irrational just to do nothing at
all. This would not mean that there was no style of criticism whatever that is
applicable in such a case. One might say about a person who does something for
which there is nothing to be said at all, when there was a nice-ish alternative, that
he was plain silly. There is certainly some criticism to be made, then, that lies
within the broad domain of the rational, but it would be of the soft kind that we
would also use of someone who chose a not very nice option when he had a
stunningly nice one available. It is not irrational to do this, and it is not that one
ought not to do it, but it is just silly to pass up such a nice way of spending the day,
or whatever.
My conclusion so far is that this direct attack on the notion of an enticing
reason is a failure. But there are more subtle attacks to be considered. One such is
offered by Joseph Raz. His paper ‘Reason and the Will’ has a section that discusses,
and defends, what he calls ‘the basic belief ’: ‘[M]ost of the time people have a
variety of options such that it would accord with reason for them to choose any
one of them and it would not be against reason to avoid any of them’ (in Raz 2000:
100). Raz introduces a potential distinction between ‘enticing’ and ‘requiring’
reasons as one possible way of defending the basic belief. (Another way, the one
Raz actually adopts, is to appeal to incommensurability.) In my discussion of Raz,
I will initially follow his terminology; but I will eventually suggest that the term
‘requiring reasons’ is a mistake. According to Raz, then: ‘Enticing reasons make an
option attractive, and render following them intelligible. But they do not bear at all
on the reasonableness of not conforming with them. Failure to conform with
  283
enticing reasons is never wrong, unreasonable, or irrational’ (2000: 101). There

follows a two-page discussion of enticing reasons in which Raz tries to show that
the notion is incoherent (2000: 100–2). He starts by understanding the enticing/
requiring distinction as a distinction between two types of reason; ‘reasons are
divided into two types’, he suggests. He then suggests two difficulties for this
division. The first is that ‘the very same considerations which in some circum-
stances seem to be enticing reasons are under other conditions requiring reasons’.
Now this itself would be no difficulty. The idea should not be that there are two
sorts of reason, so that each consideration is either of one sort or of the other, but
not both. Rather, it is that there are two normative relations, the requiring relation
and the enticing relation. These are, we might say, ‘styles’ of favouring (more on
this later). One and the same consideration can stand in either or both of these
relations—though probably it cannot both require and entice us to do one and the
same action at the same time. Which relation it is standing in on a given occasion
will no doubt depend on the context.
Raz’s second difficulty here is that if we start with an enticing reason, and add
further reasons to it, we may end up with a conclusive reason of which the
allegedly enticing reason is a part. He asks, ‘Should we say that enticing reasons
are enticing one by one, but not in combination?’. If we do say this, we undermine
the official purpose of introducing enticing reasons in the first place, which was to
explain the basic belief.
What is the supposed problem here? We are understanding the suggestion to be
that there are two normative relations at issue, enticing and requiring. On that
account, there is no reason to deny that if you change the context by adding
further reasons, the original enticer may be converted into a requirer, without the
possibility of such conversions doing anything to undermine the distinction
between enticing and requiring. I think that Raz may have two possibilities in
mind here, both of which need to be addressed. The first is that the addition of an
enticer to a requirer may result in a combination that is a stronger requirer, maybe
even a conclusive reason. The second is that enough enticers will amount to a
conclusive reason, even though they themselves retain their individual status as
enticers, and do not become requirers. If we allow the latter possibility, he says, we
undermine the supposed purpose of introducing enticing reasons in the first place,
which was to explain ‘the basic belief ’.
To make sense of all this, we need to get clearer on the relation between the
contributory and the overall (which Raz sometimes calls ‘conclusive’). In particu-
lar, we need to distinguish between the conclusive, the overall, and the requiring.
(A decisive reason is probably the same as a conclusive reason.)
A conclusive reason (as I understand and use the term) is a consideration
which, though there may be others on both sides, really is the one that decides the
issue in the present case. The sum total of the reasons in favour is not itself
properly called conclusive; there may be overall more reason to do the action than
284   
not, without any of the contributing reasons being conclusive, and without the
combination of them being conclusive either. In such a case, the matter should be
expressed as I have just done, using the notion of what there is overall most reason
to do, rather than the notion of a conclusive reason.
Sometimes the notion of overall reason is combined with that of a sufficient
reason. It is common to say that where the reasons come down on one side rather
than the other, those reasons are sufficient and the reasons on the other side are
insufficient. But this is to abuse the notion of the sufficient. That notion goes best
with enticing reasons (though I don’t at all mean to suggest that it does not apply
in the domain of Raz’s requirers). Sufficient enticing reason is something that can
be shared by more than one option. An enticing reason (or a set of enticing
reasons) is sufficient if it makes its option worth doing. There may be more than
one thing that is worth doing, as things stand. An action that is worth doing, in
this sense, is one that is above a certain absolute threshold; it is not a comparative
matter, though there are always comparative questions about whether one action
is more worthwhile than another. There can be something to be said for an action
without the action being interesting enough (say) to be worth doing.
One of the most striking differences between theoretical and practical ration-
ality is that if I have sufficient practical reason for more than one option, that is to
say, if more than one option lies above the threshold, I am rationally permitted to
choose any of them, so long as they are roughly equally well supported. By
contrast, if I have sufficient and equally good reason for each of a set of alternative
beliefs, I am not rationally permitted to choose any of them in preference to the
others. This is all part of what Raz is getting at with what he calls the basic belief.²
Our question, however, does not really concern sufficient reasons. There is no
mystery about how several enticing reasons can combine to make a sufficient case
for doing the action they promote; together they make the action worth doing,
though severally none of them would have been able to do so. Our question (Raz’s
question, that is) was rather how it can be that a conclusive reason can consist of
features that stand individually only in the enticing relation to the action. Part of
the answer to this is that an enticing reason can be conclusive (for this to be the
case, there must be no requirers present, of course) or, if only requirers are capable
of being conclusive, a sufficient combination of individually insufficient enticers
should not be thought of as a conclusive reason at all, so that Raz’s problem does
not arise. Though elsewhere he appears to accept the distinction between being
² Note that the basic belief is not concerned only with the sorts of reason that I am thinking of as
enticers. Like enticers, requirers come in stronger and weaker forms; situations call for certain
responses, and different features of the situation may call for different responses. (This notion of
‘calling for’ is one way to introduce a weak form of requirement.) It is possible for there to be a situation
in which several different responses are called for with approximately equal strength, in such a way that
the agent is not rationally at fault in choosing any of them. This shows, it seems to me, that Raz’s
complex account of potential enticers may introduce a feature that is not restricted to enticers, but
which they share with other sorts of practical reasons.
  285
conclusive and requiring,³ Raz seems to conflate these notions in his second
possibility. For he supposes that one can add enticing reasons to enticing reasons
until one gets to a conclusive reason; the non-conclusive reasons together, he says,
‘constitute a conclusive reason’. And he then supposes that if there is a conclusive
reason, it is one that it would be wrong to act in breach of. But, first, it seems to me
wrong to identify what there is overall, or even merely sufficient, reason to do with
what there is conclusive reason to do; there may be overall reason but no
conclusive reason. And, second, what there is overall enticing reason to do will
not amount to a reason that it is wrong to act in breach of. We get this result only
with overall requiring reasons. To suppose otherwise is to beg the question against
the notion of an enticing reason.
It may help at this point for me to present a little map of distinctions and
concepts. First, we have enticing reasons versus the others (which Raz calls
requiring reasons). Here we are dealing with two styles of rational support, linked
perhaps to a distinction in subject matter. Second, there is the distinction between
contributory reasons and overall reason. There are no overall reasons; the proper
phrase, as I see it, is not ‘I have an overall reason to do this rather than that’, but ‘I
have overall reason to do this rather than that’. If one has overall reason to do this,
one certainly has sufficient reason to do it, even if none of the reasons one has is
itself sufficient. One can, however, have sufficient reason to do it without having
overall reason to do it (as when two actions are both pretty good and roughly
equally good). If one has overall reason to do it, one probably does not have a
conclusive reason to do it. A conclusive reason is one whose presence is enough in
the context to determine the issue; there may be more to be said for the action than
this, but saying that more is really redundant in the circumstances. If there is such
a conclusive reason present in the case, it is a bit odd (though not exactly wrong)
to say that there is overall reason to do the action, because talking about the
‘overall’ seems to introduce conceptions of several reasons acting together. Since
there are no overall reasons, no overall reasons are conclusive reasons. Conclusive
reasons and decisive reasons are the same thing.
This leaves us with Raz’s notion of a requiring reason, and the rather similar
notion of a demanding reason. But I will not attempt to map those notions with
respect to those mapped in the previous paragraph, since I will eventually suggest
that we should abandon them altogether. No contributory reason either requires
or demands, I will be claiming. If, then, the notion of a requiring reason is not the
same as that of a conclusive reason, we will have to drop it completely. But I will
continue to use Raz’s contrast between enticers and requirers at least until the end
of the present section of this paper.
³ See Raz (2000: 229), where he says that reasons ‘are conclusive when they prevail over all
conflicting reasons’.
286   
With all this in hand, we can deal with Raz’s other supposed possibility, that by
adding requirers to an enticer we may get ‘a conclusive reason’. Understanding the
notion of a conclusive reason as suggested above, this possibility can hardly arise,
unless the several combined reasons somehow become one. If the idea is merely
that the combination of enticer and requirers may give us overall or sufficient
reason, that seems innocuous.
Of course, Raz did not only suggest that his two possibilities were somehow
internally incoherent. He also said that if one allows them, one prevents the notion
of an enticing reason from being used in support of the basic belief. Should we
accept this? No. There is nothing in the situation, as we are now understanding it,
that prevents a person from having enticing reasons for more than one action,
some stronger than others, in such a way that ‘it would accord with reason for
them to choose any one of [those actions] and it would not be against reason to
avoid any of them’.
There is an issue here about whether Raz is right to speak of a combination
of reasons as a reason. Sometimes he talks about a complete reason; at one
point he speaks of ‘his reason in its entirety’ (2000: 228). In a long footnote in
his paper on particularism (2000: 229–9) he defines a stipulative notion of a
complete reason as consisting of all the facts stated by the non-redundant
premises of a sound, deductive argument entailing as its conclusion a prop-
osition of the form, ‘There is a reason for A to V’. He claims, in support of
this definition, that ‘An examination of the use of expressions such as “this is
a different reason”, “this is the same reason”, and similar expressions, will
show that this notion of complete reason captures an important aspect of our
understanding of reasons’, though he does allow that those expressions are
ordinarily used in ways that invoke differing, context-dependent, standards
of completeness.
It is in fact easy to show that this definition of a complete reason is inadequate.
I presume that the complete reason we are dealing with is a complete reason for A
to V. Here is a sound deductive argument whose conclusion is in accordance with
Raz’s definition above:
Someone in the room has a reason to V.

There are only three people in the room, A, B, and C.
Neither B nor C has a reason to V.
So A has a reason to V.
It seems to me that none of the premises of this argument is a reason for A to V,

and the combination of them is not a reason either, complete or otherwise. All we
have here is a guarantee that there is a reason, which is not at all the same thing as
a specification of the reason that is guaranteed to exist. Further, on Raz’s account
  287
the reason mentioned in the conclusion is not the same reason as the one specified
in the first premise, which is surely peculiar.
Leaving these criticisms aside, my own view about Raz’s stipulative definition of
a ‘complete reason’ is that it distorts the relation between contributing and overall
reason. (This is one reason why it is no defence of the definition that it is
stipulative.⁴) This criticism, if it is correct, will apply equally to versions of the
view that escape the objections in the previous paragraph. One worry here is that if
the only complete reasons are sets of considerations that together operate as a
guarantee for something or other, the status of the contributory begins to appear
odd. For contributory reasons will have to be understood as incomplete, or partial,
or something like that. But there is nothing incomplete about an ordinary
contributory reason. Whatever it is that such reasons do, they do it perfectly
well, and completely. They are not hopelessly trying to do something else,
something that can be done only by several considerations acting together (as
Raz would have it, acting as the several premises of an argument to the conclusion
that there is a reason for A to V). Furthermore, in defining ‘complete reason’ as he
does, he is in danger of making it impossible to understand the relation between
what is contributed by an ordinary, incomplete reason and what is achieved by a
complete one. Finally, whatever it is that a contributory moral reason does, it
must, just in virtue of that contribution, be capable of standing as the ground for
overall duty. If it were the only relevant feature, the only one actually making a
difference here, our duty overall would be to do the action which it favours. But it
is not at all clear, on Raz’s picture, that an incomplete reason is capable of playing
this sort of role; only complete reasons can do that, for him.
We started by suggesting that we have two normative relations, requiring and
enticing. A set of enticing reasons can be sufficient to make the action they
recommend worth doing, fun, exciting, attractive and so on. (These are the sort
of thick concepts we are dealing with in the domain of enticers.) One may have
several such sets present at once. If one has only one, the action it recommends is
the one that there is overall most enticing reason to do. But one is, as it were, in
charge of one’s enticing reasons. It is never irrational or wrong to choose an
enticing action other than the one that is most enticing—even though it can be
pretty silly to do so, and, in this weak sense, contrary to reason.
This seems to mean that though there can be both enticing and requiring
reasons, the requiring ones, no matter how weak, will always win in a head-to-
⁴ There is a further way in which Raz’s definition of a complete reason is prejudicial, which is
significant if the notion of a complete reason is to play any part in a sound response to particularist
remarks about contributory reasons. If, for instance, I just defined as a ‘complete reason’ for an action’s
being wrong the supervenience base for that wrongness, and then announced that, whatever particu-
larism had to say about anything less extensive than that, still, once we reached complete reasons rather
than parts of reasons, generalism was blatantly sound, surely one would not be tempted to take this very
seriously. Particularists would probably say that this sort of a reason is no reason at all.
288   
head fight. If it is contrary to reason to fail to respond to a requiring reason, but

there is no rational requirement that one respond to an enticer, then reason, we
might say, requires that the requirer wins.⁵ This issue is raised by Raz’s second
argument against enticing reasons, to which I now turn. Assume that enticing
reasons cannot defeat requiring reasons, though they can defeat weaker enticers
(not, however, in a way that makes it contrary to reason to yield to the weaker
enticement). Now suppose that Mary has enticing reason to go to the last night of
the play this evening, and requiring reason to visit her aged mother. The reason
requiring her to visit her mother does not require her to do so this evening. Mary
could go any day this week. So as a requiring reason it is not very strong—or
perhaps we should say that though strong it is unfocused. The question that Raz
asks is this: ‘If the enticing reason to go to the play could have justified her
deferring the visit to the next day, does it not show that enticing reasons can
defeat requiring ones?’ (2000: 102). If so, he suggests, there is no relevant differ-
ence between enticers and requirers; it is just a question of strong versus weak. For
wherever an enticer defeats a requirer, it will be wrong (contrary to reason) to do
what the requirer requires. And this will convert the enticer into a requirer.
There is something odd about this example, because Mary’s requiring reason is
not a reason to visit her mother this evening, while the enticing reason only
applies to this one evening. So the conflict between them is not a direct conflict. If
it were a direct conflict, the requirer would certainly win, so long as it is stronger
than negligible. (This is not the same as saying that moral requirers always defeat
other requirers.) But as things stand, Mary will not have acted in breach of the
requiring reason if, by the end of the week, she has made the required visit. So
I don’t really see the difficulty at this stage. Deferring her visit until the next day is
not something that Mary needs to have a reason to do, if she is to escape censure
for doing so.
One might think that the difficulty could be revived in the following way.
Suppose that Mary has an enticing reason for each day of the week, and a
requiring reason to visit her mother on at least one day, unspecified. What is to
happen? The situation seems to me to be like this. On each day, Mary has a choice,
and faces no censure whatever she chooses, until she gets to the last day. On the
last day, assuming that the visit has not yet been made, it must be made, and the
enticer is powerless.
I conclude that Raz’s attempt to show the notion of an enticing reason to be
incoherent is a failure. But it has raised significant issues about exactly how we are
⁵ I don’t think we should be muddled by the apparently parallel thought that enticers ‘entice’ that the
enticer wins. This is just one expression of the idea that all reasons win in their own terms.
Horticultural reasons, for example, defeat moral ones in gardening terms. The point in the main text
is merely that one lays oneself open to the complaint that one has failed to respond to the nature of a
certain reason if one complies with the enticer rather than with the requirer, and one does not lay
oneself open to an analogous complaint if one complies with the requirer.
  289
to conceive of an enticer, to which I now turn. If we want to know what is the real
difference between enticers and others, what resources have we to call on? The
following possibilities have all been thrown up, implicitly or explicitly, in what has
gone before:
1. Lexical domination: any requiring reason can defeat any enticer.

2. Styles of criticism: mishandling of enticing reasons invites special sorts of
criticism, and the use of members of a distinctive family of thick concepts
(e.g., the silly rather than the irrational).
3. No oughts: no matter how strong our enticing reasons, they never generate
an ‘ought’ (or a ‘right’, or a ‘wrong’).
4. Discounting: an agent may discount his own enticing reasons.
5. Permissibility: it is always permissible to choose not to do the action one has
most enticing reason to do.
Which, if any, of these is really the central point? It cannot be lexical domination,
one would say, since even if we allow that any requiring reason can defeat any
enticer, we want to know why, and the answer to this question seems to be given
by something like no oughts or permissibility. Nor, I would think, can the central
point be styles of criticism. For, again, we would want to know what makes this
family of thick concepts distinctive, that is, what is distinctive about the sort of
praise or criticism that they bring along with them. And, again, it seems as if the
answer to this question might be given by no oughts, or by permissibility. So it
seems as if the first two possibilities cannot stand independently, which reduces us
to the last three. But no oughts can be challenged. John Broome suggested to me
that we can have pretty weak (though still overall) oughts. If so, why bother
denying that enticers can generate such a thing? At least where there is a
considerable difference between the strength of one enticer and another, there is
rational criticism (that is, criticism of the rational sort) of someone who fails to
choose to act in accordance with the stronger reason. This is just what I was after
when I said, of such a person, that he would be plain silly. Broome’s view is that
one ought not to be plain silly.
Suppose that we accept this view. This would deprive us of no oughts. It would
not itself unsettle discounting, nor permissibility, but it would show that the latter
notions are going to have to be rather special in this connection. On the discount-
ing front, normally (and I have supererogation in mind here) that which is
discounted is not an ought—nor the ground for an ‘ought’; by discounting it
one exactly prevents the formation of the ‘ought’—that is the whole purpose of
discounting. Surely, we might say, to discount a reason is to prevent it from
contributing to the overall level at which ‘oughts’ are to be found. But the sort of
discounting we are engaged in with enticers would, if Broome were right, be the
discounting of ‘oughts’. One ought to choose the strongest enticer, though it is
290   
somehow rationally permissible not to. This throws the pressure onto permissi-
bility. Normally we would say that if one ought to do something, it is not
permissible not to. But we are now supposing that though (in the weak sense)
one ought to choose the strongest enticer, it is permissible to choose a weaker one.
One ought to do this but it is not wrong not to do it; one ought to do this, but one
has a right not to. These remarks seem to make sense in the present context,
whereas in ordinary discussion of moral reasons they would be thought to be
straightforwardly contradictory. So permissibility for enticers is not the same
thing as permissibility for requirers. This is not impossible. The familiar notion
of permissibility—the one that is defined in terms of ‘oughts’—was created by
people who had not considered the possibility of enticing reasons. If we under-
stand a permissible action as one that it is not wrong to do, and a wrong action as
one that it is not permissible to do, we get a nice interconnection which reminds
us strongly of the relation between necessity and possibility. But if we once allow
that there can be an action that we ought to do but which it is not wrong not to do,
or which it is permissible not to do, that nice interconnection between the various
deontic concepts (ought, right, wrong, permissible) fails.
Note that Broome’s point only concerned ‘ought’. Perhaps one ought to do the
action which entices most strongly (in the weak sense of ought), but it is not
wrong not to do so. It is not only permissibility, then, whose tight link with ‘ought’
has been broken; the link between ‘ought’ and ‘wrong’ has been broken too.
Perhaps no oughts should be renamed no wrongs.
I said that the notion of permissibility at issue here is not the one we are used to.
A further respect in which that will be true is that ordinarily, where something is
permissible, no criticism is attracted by those who choose it. But here, we are
supposing, even though it is, in a sense, permissible to choose a noticeably less
enticing option, criticism is still appropriate. As we might say, you can do this but
it would be pretty silly of you to do so. We might even say that though it is silly,
you have a perfect right to make that choice. This is not itself unfamiliar. We say,
perhaps, that you have a right to gamble all your money away, even though this is
something that you should not (ought not to) do. You have a right to do it, but this
does not mean that you are immune from criticism if you decide to do so.
But what might it mean to say that you have a right to do this, or that it is
permissible to do it, even though you ought not to do it and can be criticized quite
severely if you do? Can these notions really survive if their normal connections
with ‘oughts’ are severed? Let us not forget that most of the difficulties that have
emerged stem from admitting Broome’s claim that there is a sort of weak ‘ought’
operating here; if we had chosen to tough it out, we could have insisted that there
are no ‘oughts’, not even weak ones, but that there is discountability (of the normal
type) and permissibility (again of a normal type). And it does seem to me that
toughing it out is the best way forward here for someone interested in making
sense of enticers. Doing so restores the standard connections between
  291
discountability, permissibility, ‘wrong’, and ‘ought’, whose rupture caused us all

the complications we have just been struggling through.
But even if we do choose to tough it out, there is still the idea that one has a
right to discount, a right to choose the less enticing option. What sort of right is
that? There is one clear difference between the sort of right that we would be
dealing with here and anything worth thinking of as a moral right. This is that
moral rights are rights against other agents, as it is usually though rather unbeau-
tifully put. What this means is that if I have a right to do it, it is wrong of you to do
anything to prevent me. We get something of this with enticers. If I have a perfect
right to do an enticing action (of any sort, now, whether it be the strongest enticer
or not), this can only be because I have no moral reason not to do it, and so
I already have a moral right to do it; it would, therefore, be wrong of you to
interfere. In this sense, my right is a right against others, just like any other. But
that is not the point. The point is that I have a right to choose the less enticing
action even if I ought (in the weak sense) to choose the most enticing. Though
I suppose one could say that you ought not to interfere in order to make me
choose the most enticing action, this hardly seems to be the right point to focus
on. The real point seems to be that certain sorts of criticism are not appropriate.
But this throws us back to our starting point.
Can we now do better? What sort of criticism is inappropriate? I have already
suggested that selecting the less enticing option should not be thought of as
irrational. We would be more likely to think of a person who does this as silly.
Does this amount to anything more than saying that severe criticism of the
rational sort is not appropriate in this type of case? Mild criticism, on the other
hand, is fine. If that were all there is to it, the idea that enticing reasons are
distinctive would be under some threat. Mild criticism of the rational sort is
appropriate in many cases where we are dealing only with requiring reasons.
We might keep going by pointing out that the severity of the criticism should be
somehow related to the disparity between the reasons. If criticism of someone who
selects an enormously less enticing option is still only mild, this would itself be
significant. But I think there is more to be said. The idea is that there is a sort of
blip in the judgement of someone who decides to take one option when he knows
full well that another is more required. Such a person asks himself the question
what is the thing to do, and then fails to attend to his own answer. There is no such
blip in the judgement of someone who prefers the less enticing option. That
person may have decided that the more normative question ‘What is the thing
to do?’ has no answer in the present case. But this does not mean that nothing is
left to determine his choice. He starts by asking ‘What shall I do?’. If he decides
that there is one course of action that is the one to do, that is the way he has to go,
on pain of the blip. But if he decides that nothing is here required of him, he can
continue to ask ‘What shall I do?’ by asking what it would be nice to do, and he can
then act according to his answer to that question. But he is not at odds with his
292   
answer to it if he decides to do something nice enough, but less nice than

something else.
I return to these issues in my final section.
2. The contributory and the overall
If there are such things as enticers, why have they escaped notice? A possible
answer to this question is that we have all been working with a general conception
of what I call a contributory reason, which obscures the possibility of enticers, but
which is in fact mistaken. The locus classicus of this mistaken conception is Jean
Hampton’s posthumously published The Authority of Reason (1998).⁶ When
Hampton talks about reasons, she constantly uses terms that are, one would
have thought, quite inappropriate for the merely contributory. She says that
reasons are ‘directive’ (pp. 51, 85), have ‘obligatory force’ (p. 99), have ‘compelling
rightness’ (pp. 93, 99), are ‘prescriptive’ (p. 87), are or express commands (p. 88),
concern the ‘ought to be’ (p. 92), ‘feel like orders’ (p. 106), and have a ‘compelling
quality’ (pp. 91–2). But all these remarks seem to be more appropriate at the
overall level. An overall ‘ought’, we might say, is a directive; in a perfectly clear
sense (at least for present purposes), it tells us what to do—what we should do. It
expresses a command, maybe, it has a compelling quality, it has ‘obligatory force’
in some sense or other, it feels like an order and directs us to do this rather than
that. The question then is whether contributory reasons do the same as the overall
does. And the answer is surely that they don’t. If an overall ‘ought’ commands, it
cannot be that contributory reasons command as well; there is no such thing as a
pro tanto command. We have reasons on both sides of the question, often enough.
Is it that each reason on either side commands, and then that somehow the sum
total of them commands as well? I don’t think this is a coherent scenario; there are
too many commands floating about.
Raz’s notion of a requiring reason, and of a ‘demanding’ reason, are vulnerable
to similar criticism. For contributory reasons neither require nor demand,
whether they are on the winning or the losing side. Of course they do do
something that is a bit like requiring; though it would be inadequate to say that
they do something like requiring but weaker. Requiring and demanding are things
that, though they come in degrees, happen only at the overall level. So from now
on we will have to abandon Raz’s contrast between the enticing and the requiring.
What is more, the way in which I started this paper, where I talked of reasons that
are in the business of more or less telling us what to do, or of saying ‘You’d better
do this’, will have to go as well.
⁶ Parenthetical page references in this section are to Hampton’s book.

  293
My main point here is that if one does think of the contributory as requiring,
compelling, or demanding, the idea of an enticing reason is probably not going to
occur to one at all. And if it does occur, it will probably be only in a distorted form.
Hampton does sometimes speak in ways that look as if they are intended to
capture the enticing. She talks of reasons as having ‘a pull, an attractiveness’
(pp. 91–2), and at one point discusses the nature of a reason to have a haircut
on Sunday afternoon. This looks like an enticer, I think; the idea is not that she
ought to have her hair cut on Sunday afternoon, but that going to the hairdresser
might be an agreeable relaxation. But Hampton distorts this reason in two ways.
First, she has already cast all such reasons as ‘directives’, which is surely part of
what lies behind her thinking of them as commands. A directive sounds pretty like
a requirer to me. Second, she attempts to use Raz’s distinction between mandatory
and permissive reasons to capture the difference between her reason to have a
haircut and her reason to care for her baby. She writes, ‘some reasons command
us, and thereby give us mandates, and others direct us in ways that indicate
permissions, rather than commands’ (p. 90). But this is surely wrong. It is not
that I am permitted to have a haircut on Sunday (though let us suppose that I am,
and that without this permission the enticing reason to do so would be silenced). It
is that there is a reason in favour of doing so, which cannot be simply that it is
permitted nor what makes it permissible.⁷ Where there is permission but no
favouring consideration, there is normally no reason, and all that we have is the
absence of reasons against.
One might reply here that Hampton has just chosen the wrong terminology.
Permissions are indeed not what is at issue in the debate about enticers, but
Hampton’s substantial remarks show that she does not really mean to appeal to
the notion of permission. After all, she says that these reasons ‘merely guide us’
(p. 90)—as opposed, I suppose, to requiring us. This talk of guiding, however, is no
improvement on what we have seen already. It is common to use the notion of
guiding for overall oughts. My course of action is surely guided by thoughts about
what I ought overall to do. (I don’t necessarily go on to do those things, sadly.) Is it
the case that I am guided also by thoughts about contributory reasons? Well, yes it
is, but it cannot be the same sort of guiding as before, on pain of incoherence.
Hampton’s talk of guiding only shows, I think, that though she means to distin-
guish guiding from requiring, or something like that, both of these notions find
their place within a general concept of a directing reason—a directive.
It is within the domain of the directive that the distinction between the
mandatory and the permissive is to be found. The notion of a directive, like that
of the mandatory, is deontic. It is tightly related to oughts. The notion of an enticer
is not deontic, since no oughts are involved (as we eventually decided); this is what
⁷ I don’t mean that the fact that something is permitted can never be a reason to do it, only that it is
not in this case, let us suppose.
294   
makes it significant as a possibility within the theory of reasons, since it is normal

to see people casting the general notion of a reason within the deontic rather than
the evaluative domain. (Both of these domains lie, supposedly, within the more
general domain of the normative.) What about the notion of a permission? Well, a
permission is normally understood as the absence of an ‘ought-not’, and so functions
in the deontic domain—the domain of directives, one might say. And Hampton says
that permissives are like mandatory reasons in having a sort of ‘compelling rightness’
(p. 91). This shows, I think, that the real mistake is to start with the notion of a
directive and then try to put everything under that label; Hampton is, hopelessly,
trying to do everything in terms of force, compulsion, and necessity. Attractiveness,
grounded in enticing reasons, is nothing at all like compelling force, grounded in
mandating reasons—even though there are attractions that one cannot resist. The
absence of a mandating reason is not itself an enticing reason.
If all this is right, Hampton’s attempt to capture what she calls ‘the authority of
reasons’ in terms of anything like an obligatory force (p. 99) is a mistake. If there is
such a thing as a single style of authority belonging to all reasons, we need a quite
different account of it.
One possibility that has not yet been mentioned is that we should try to
understand the authority of reasons by saying that reasons apply to us irrespective
of what we like, what we want, or what appeals to us. But though moral reasons
may be like this, it is extremely contentious to say the same about practical reasons
in general; and to do so would leave no room for enticing reasons in particular.
What is enticing for one person is not for another, and properly so. (This is
another interesting aspect of the enticing.) Perhaps we are going to find ourselves
driven to saying that only some reasons have the sort of authority that Hampton is
talking about. But if we say this, have we abandoned the attempt to isolate the
normative nature of reasons? Is it that enticing reasons are not ‘normative’ at all?
They are, after all, capable of rendering an action intelligible, sensible, worth
doing, and so on. I now turn, therefore, to consider what could be meant by
saying that the very notion of a reason is a normative notion.
3. The normativity of reasons
Raz claims that: ‘The normativity of all that is normative consists in the way it is,
or provides, or is otherwise related to reasons’ (2000: 67). Let us distinguish
between two questions:
1. Is there one normative concept in terms of which all others can be defined,
or at least understood?
2. Is there one normative concept in terms of whose normativity the norma-
tivity of all other terms can be understood?
  295
Raz’s remark, quoted above, hovers between these questions. He doesn’t exactly
say that the normativity of other concepts is related to the normativity of the
concept of a reason. But he doesn’t quite commit himself to the view that other
normative concepts can be defined in terms of a reason either.
It is possible that even if the answer to the first question is no, the answer to the
second is yes. It is possible, for instance, that there are various normative concepts,
forming a local holism, so that each member of the normative family can be
understood in terms of its relations to the others, and none is more basic than any
other. Even so, it might be that the notion of normativity is centrally to be
understood in terms of one member of the family, that of reason.
It is however possible that there is one central normative concept in terms of
which all other normative concepts can be defined and which cannot itself be
defined in terms of others. If so, it would be very tempting to try to understand the
normativity of the other concepts in terms of that of the central concept. And
many people think this is the way things actually are. For many, there is one
central normative concept, that of ‘ought’, and all other normative concepts can be
defined in terms of some relation to ‘ought’, and in this way get their normativity
from that of ‘ought’.
This picture is not Raz’s picture, and there is a weakness in it, which is that it
simply takes for granted (at least as we have expressed it so far) the normativity of
‘ought’. If asked to say what makes ‘ought’ normative, it is hard for us to find an
answer because there is nothing left to say. Raz’s picture, however, has the same
weakness, because he doesn’t really tell us what makes the concept of a reason
normative. I don’t see how one could make progress without saying something
about how to conceive of normativity itself.
The fact that one ought to do this rather than that, and the fact that one has
reason to do this rather than that, bear their practical relevance on their face; they
are explicit answers to the question what to do. Let us suppose that the central
concepts in such facts, those of ‘ought’ and of ‘reason’, are therefore pivotal
members of a special family of concepts, the family of normative concepts. We
need not be sure yet which other concepts are members of this family. (Perhaps
the concept of relevance is one such.) Any fact expressed in ways in which those
concepts play no role may have practical relevance; some such facts are perhaps
necessarily practically relevant. But that such facts are practically relevant is a
further fact about them. As Derek Parfit puts it (1997: 124), a fact that has
normative significance need not for that reason be a normative fact. The difference
here is between facts that might be mentioned in answers to the question what to
do, and the facts that those facts are relevant to the question what to do. To give a
very contentious example of this difference: the fact that this action would make
many more comfortable and none less comfortable could well be mentioned in an
answer to the question whether it is the thing to do, but is not the same fact as the
fact that it is relevant to what to do. On this account, therefore, it is not a
296   
normative fact, though it is a fact of normative significance, one that makes a

difference to what to do.
Some normative facts, then, are more complex than the simple fact that one
ought to do this; they contain that fact, but they also contain what makes that
simple fact the case. Such facts are of this form: that p makes it the case that one
ought (or has reason) to act in way w. These meta-facts are facts about some
matter of fact and about its making a difference to how to act. They constitute
direct answers not only to the practical question what to do, but to the question
why. It is these meta-facts that I think of as the central normative facts, by
reference to which the normativity of all others is to be understood. Each such
fact is the fact that some other fact stands in a certain normative relation to an
action (or a belief or a feeling or a desire . . . ); we have mentioned two such
normative relations, that of ‘being a reason for’ and that of ‘making it the case
that one ought’. Let us hope that these will prove to be enough.
If this is to be our starting point, the question whether we are to conceive of
enticing reasons as normative is not decided by the fact that they do not generate
‘oughts’, for our account of the normative does not require anything of the sort. It
would be enough if enticers can be thought of as relevantly similar to other
reasons. Let us now fix on a term for the ‘other reasons’. We have already decided
that ordinary contributory reasons neither require, nor command, nor direct. But
they are, I would say, still peremptory in style. So from now on I will work with a
distinction between enticing reasons and peremptory reasons. The peremptory
ones were the ones I had primarily in mind in laying out my conception of
normativity. The question now is whether that conception is sufficiently flexible
to allow us to think of enticing reasons as normative. I am going to suggest
that it is.
One way to capture what it is that peremptory contributory reasons do is to say
that they stand in a certain normative relation to action, the favouring relation.
A reason for doing an action is a consideration which counts in favour of that
action; it makes an independent contribution to the case for doing it. As far as this
goes, there is no problem in thinking of enticing reasons as standing in just such a
relation to the actions which they are reasons to do. We will want, of course, to
distinguish the style of favouring done by peremptors from the style of that done
by enticers. But this is just the old question of the difference between the
peremptory and the enticing—a question which needs further consideration but
which should not alarm us.
In particular, no special problem derives from the suspect thought that the
concept of a reason is a deontic concept rather than an evaluative one. (Here I am
just underlining a point that I made in section 2.) We have already seen that
attempts to capture what contributory peremptors do in deontic-sounding terms
like ‘command’, ‘require’, and ‘direct’ are all failures. The only way we can retain
our sense that these peremptors are deontic is to try to characterize what they do
  297
in terms of some relation to ‘oughts’. This is a big topic, but my own view is that
every attempt to do this is a failure⁸—except possibly for one. The one survivor is
the claim that if there is a reason in favour of doing something and no reason
against, one ought to do it. I don’t see a big difference between this claim and the
claim that if there is more reason to do something than to do anything else, one
ought to do it. These things may (both?) be true; and it is true that, on our present
showing, enticing reasons do not generate ‘oughts’ in this sort of way. But if this
relation to ‘oughts’ is all that being deontic amounts to, it is not a further
difference between enticers and peremptors. It is just the same difference again.
Introducing the notion of the deontic is not a step forward.
So far, so good. The difficulty, if there is one, in understanding enticing reasons
as normative derives from the way I tried to explicate the notion of normativity.
Normative concepts are those which play pivotal roles in direct answers to the
question what to do. Now: is the question what to do the question to which
specification of enticing reasons as reasons is an answer? If it is not, enticing
reasons are not normative in the way that peremptory ones are. And there are two
views one could take on this issue. One could say that the question what to do is
one to which one could respond by saying that there is no action here that is the
thing to do, and that this leaves us free to concentrate on which action would be
most fun. Or one could say that in the absence of peremptory reasons, the most
enticing course is the thing to do.
The question ‘What is the thing to do?’ is perhaps not quite the same as the
question ‘What shall I do?’. There is a similar difference, it seems, between the
question ‘What should I decide to do?’ and the question ‘What shall I decide to
do?’. To the latter question, there is no doubt that specification of enticing
considerations as reasons is directly relevant; to the former one, not. But which
of these questions was the one we were thinking of as ‘the practical question’? If it
was the former, there was a covert appeal to an ‘ought’ in setting up the notion of
normativity, and that was something that I was trying to avoid. It must therefore
have been the latter. But to that question, the answer ‘You have enticing reason to
do this rather than that’ is directly relevant, and in this sense the notion of an
enticing reason is as normative as the notion of any other sort of reason.⁹
In thinking of things in this way, we are weakening the notion of the normative.
The question ‘What shall I do?’ is less deontic in style than the question ‘What
should I do?’, but we are understanding the normative in terms of the former
rather than the latter.
⁸ I argue this point in detail in Ethics Without Principles (2004: ch. 2).
⁹ Of course, the answer ‘This would be more pleasant than that’ is not the same answer as any
answer that explicitly uses the concept of a reason. (This is just an echo of Parfit’s distinction between
what is of normative significance and what is normative.)
298   
Nobody denies that enticers speak in favour of the action they commend. They
may speak in a discountable way (that is, one can discount what they say), and
they may commend rather than recommend. But this favouring, or speaking in
favour of, seems nonetheless to be recognizably similar as such to other more
peremptory sorts of favouring. If that is right, my overall suggestion is that we
should weaken our conception of the normative so as to include all forms of the
favouring relation, the non-peremptory as well as the peremptory.¹⁰ Being nor-
mative, in this weaker sense, will turn out to have nothing especially to do with
norms—at least, not in the way that we usually think of norms.
One consideration that supports this weakening of the concept of the norma-
tive is that the debate between naturalists and non-naturalists about the irreduci-
bility or otherwise of the normative in general seems to apply just as much to the
weaker, non-peremptory sort of favouring as to the stronger, peremptory sort. The
reducibility or otherwise of the weaker relation that stands between that which
entices and the action that it favours (or commends) is going to be as interesting
and contentious—from the naturalist’s point of view—as that of the strong
relation of requiring. If it is normativity as such that engages the reductivist
aspirations of naturalism, this is further evidence that we would be wrong to
think of enticing reasons as non-normative.
Another commonality between peremptors and enticers is a functional one:
they are both notions that have a place within the context of practical deliberation,
being considerations relevant to the substantive questions from which deliber-
ation begins, and which it aims to resolve (either ‘What shall I do?’ or ‘What is the
thing to do?’). And we see the same structure both sides, the structure given us by
the distinction between the contributory and the overall; we see the same relation
between considerations favouring one alternative and those favouring another,
and the same relation between considerations favouring one alternative and those
disfavouring it. (There is a difference between speaking against an option and
speaking in favour of something else.)
These common features, which I invoke in order to increase our confidence that
enticers can be thought of as normative, will of course make it harder to show that
enticers are really as different from peremptors as I have earlier tried to suggest.
The more we stress the similarities between the two, the harder it will be to draw
¹⁰ One might wonder whether Broome’s normative requirements would be left out of account, on
this approach, so that there would remain something normative whose normativity we have failed to
capture. The matter will turn on whether we are to think of a normative requirement in this sense as
anything more than a non-detachable requirement on a complex. In his ‘Normative Requirements’,
Broome did think there was more to a requirement than that, but as I understand it he now thinks, as
I always did, that the whole story of these requirements can be told in terms of non-detachability. If so,
the normativity of these requirements will be able to be told in the terms I suggest. For any such non-
detachable requirement will be a reason, or an ought. For instance, the requirement that if one believes
that p and that (p ! q) one should also believe that q amounts to no more than that one non-
detachably ought not to believe that p and that (p ! q) and fail to believe that q.
  299
the sort of distinction I appealed to; in particular, the harder it will be to say that
peremptors generate ‘oughts’ while enticers do not. I now return, therefore, to that
issue.
4. The difference between enticing and peremptory reasons
If there are both enticing and peremptory reasons, there must be a difference
between them. Reasons of both sorts favour, perhaps; but they must favour in
different ways, if one sort of favouring can somehow generate an ‘ought’ and the
other cannot. Up to now, I have been talking about a difference in style. Can we
say any more about this difference?
One might think it would be enough here to say that peremptory reasons are
reasons that one has no right simply to discount; the relevant agent is not in
charge here. One is not ordinarily rationally permitted to discount a peremptory
reason. By contrast, one can discount enticing reasons of one’s own (though not
those of others), without thereby causing them to cease to be reasons.
We have seen something like this phenomenon of discountability elsewhere, in
the theory of supererogation, and indeed more generally in the theory of agent-
relative reasons. In the case of supererogation, I will assume the picture for which
I argued in my Moral Reasons (1993: chs. 8 and 12). This has it that the reason for
me to sacrifice myself is (let us suppose) that three others will die if I hold back.
This reason is non-discountable. If it is not to make it wrong for me not to sacrifice
my life for the sake of the others, some other reasons must be found on the other
side. Naturally we should not ignore such considerations as the loss to me, but this
is obviously less of a reason as it stands than the combined losses to the other
three. The question is how to work it so that, despite this imbalance, reason does
not require me to sacrifice myself; we hope somehow to leave it that I act
outstandingly well if I make the sacrifice, but that I do not act against the reasons
if I fail to do so. If one is to achieve this while working solely within the realm of
reasons, one must, it seems, resort to factoring in some supposedly agent-relative
reasons, all the neutral reasons having been counted in already. Suppose now that
we find some agent-relative reasons capable of justifying my failing to sacrifice
myself. We do not want to say that the balance of reasons now makes the sacrifice
wrong, if I choose to make it. The way in which we avoid saying this is by
announcing that the agent-relative reasons are discountable by the agent.¹¹ The
other reasons, assuming that none of them are enticers, are non-discountable.
Nothing in this picture requires, however, that any of the relevant reasons are
enticers. It may be that the agent-relative reasons, though discountable, are not to
¹¹ In the terms that Thomas Nagel uses in his The View from Nowhere (1986), the agent-relative
reason is non-insistent; all the neutral reasons are insistent.
300   
do with the fun, the amusing, the attractive, and so on. The cost to me may lie not
in the loss of amusement or anything like that, but rather in things to do with the
education, health, or safety of my (soon to be orphaned) children. The whole story
of supererogation given above makes no essential appeal to non-peremptory
reasons. This is what makes the phenomenon of supererogation noticeably
more puzzling than that of discountable enticers, for in the former the reasons
that are supposedly discountable are also, as we are now saying, peremptory.
A discountable enticer seems much less problematic, from a theoretical point
of view, than is a discountable peremptor. Still, the important point is that if there
are discountable peremptors, it cannot be right to understand the difference
between enticing and peremptory reasons in terms of what is and what is not
discountable.¹²
One thing gained by this brief discussion of supererogation is that it seems to
show that normativity and discountability are not themselves at odds, which is of
some help in the defence of enticing reasons. There needs, of course, to be an
explanation of discountability wherever it occurs. But this is no difficulty in the
present case, since our explanation of the discountability of enticers will simply
appeal to the larger idea that agents are entitled to put themselves second. (This is
not much of an explanation, perhaps, though I try to defend it in Moral Reasons:
ch. 12.2.) There is, then, no general argument that enticers cannot be normative
because they are discountable. But we still lack an account of the difference in style
between the enticing and the peremptory.
We might resuscitate here a suggestion that we previously rejected as an
account of the authority of reasons. On those lines, we would say that a peremp-
tory reason applies to us irrespective of our current likes and dislikes, enthusiasms
and affections. Enticing reasons are not like this. Now if this is the difference, it is
(and is only) a difference in ground. The idea would have to be that the difference
in ground comes with a difference in the kind of favouring relation that lies
between that ground and the relevant action. But can we extract a difference in
favouring relation from a (supposed) difference in what comes on the left-hand
side of it—in what is doing the favouring? Not directly, I would think.
The point is not that there are some peremptory reasons which are equally
dependent on subjective conditions of likes, affections, and so on. To make that
point, we might try to appeal to the duties of friendship, for example, or at least to
the sorts of reasons given us by the troubles our friends get into. Relations of
friendship are conditional (in some sense) on affection and liking. So, one might
say, the reasons of friendship are as dependent on subjective conditions as are any
enticing reasons we may have. But I would not accept this picture. We are dealing
with different sorts of dependence. It is true that friendship generates peremptory
¹² In terms of the non-insistent, some peremptory reasons are non-insistent (if the above account of
supererogation is on the right lines).
  301
reasons, and that those reasons would not exist in the absence of the subjective
conditions that make friendship possible. But I would say that they are not
grounded in those conditions in the way that enticing reasons are grounded in
what one enjoys or finds amusing. It is because we find each other pleasant
company that we have become friends, and by becoming friends we acquire
reasons that we did not have before. There is nothing like this sort of thing with
straightforward enticers, where the subjective conditions act more or less as
ground.
The point is rather that one needs to do more work if a supposed difference in
ground is to give one a difference in relation. And it is not obvious where to go.
Suppose that, per impossibile, I could change my likes and dislikes at will. Would
this be any help? No: we would still be facing the fact that the reasons given me by
my likes and dislikes are there for me until I change what I like. So we might say
that I have some reason, so long as woodwork is what I most enjoy doing of a
weekend, to spend my weekends at the workbench; it is irrelevant that this reason
becomes ineffectual as soon as I come to prefer music to woodwork.¹³
What then can we appeal to? Let us take a step back for a moment and ask what
we would take as showing a difference between two favouring relations. We have
already decided that a difference in ground is not enough. Ground is input. What
about a difference at the output level? This looks more promising. Whether the
input is the same or different, if the output is different it looks as if we must be
dealing with a different relation. Now we already have a supposed difference in
output, since we have often claimed that peremptory reasons take us to ‘oughts’,
while enticing reasons don’t. This, then, seems to be the right sort of claim. But it
itself seemed to be in need of explanation. It seemed that this could not be all there
was to be said, for a difference of this sort cannot be a ‘brute’ difference. This is the
present state of play.
But it might improve matters if instead of saying what enticing reasons don’t
do, we say what they do do. Here is a suggestion of that sort. Peremptory reasons
take us to ‘oughts’, and enticing reasons take us to ‘bests’. If this suggestion could
be substantiated, we would have given a sense to the initially attractive idea that
peremptors have a deontic focus, while enticers have an evaluative focus. We
would also have sustained the normativity of enticers as similar to, but not the
same as, that of peremptors.
It would be a mistake to object to this that the reasons in favour of supereroga-
tory acts take us only to bests, not to oughts. It is true that they don’t succeed in
taking us to an ought (always supposing that the notion of supererogation is
coherent). But the idea of taking us to an ‘ought’ that I am appealing to here is not
that of succeeding in generating an ‘ought’; for otherwise there would be no
¹³ In different terms, the reason here is a non-detachable one on a combination: having the
preference for woodwork and spending my weekends at the bench.
302   
defeated peremptory reasons. To take us to an ‘ought’ is, in the sense I am after, to

try to generate an ‘ought’, but not yet to succeed. Similarly, enticers try to generate
a ‘best’, though they may not succeed either, and won’t if something else generates
a ‘better’.
But don’t some peremptory reasons take us to ‘bests’? Yes, they do; but they also
take us to ‘oughts’. In saying what it is that enticers do, we have not avoided the
idea that the main difference between them and peremptors lies in something with
a deontic focus that enticers are not involved in. The point is rather that we can
understand the idea that enticers are reasons by showing that they have a focus
which is like, but different from, the sort of focus that peremptors have. (Here we
can appeal to the commonalities mentioned at the end of the previous section.)
They go to make a choice the best one, but not yet the one which one ought
to take.
At the end of the first section, I suggested, on behalf of the possibility of
enticers, that there is a blip in the judgement of someone who, recognizing an
‘ought’, failed to respect that ‘ought’ in deciding what to do; such a person is one
whose judgement is at odds with itself, I said. But I maintained that there is no
such blip when one decides to do one thing when one s that another would be
more pleasant, and so better. The things I said about normativity in the previous
section only serve to make this picture harder to defend. In broad terms, I was
trying to build the notion of normativity on that of practical relevance. Isn’t there
a blip in the judgement of someone who recognizes that one car would be better
than another but buys the worse one? Certainly, if there are peremptory reasons
around—better brakes, for instance. But what if there aren’t? The way I put it at
the end of the first section was that no blip occurs if one chooses an option that is
nice enough, even if one allows that another would be nicer. This is an appeal to
satisficing. The idea was that there is a level above which no blip occurs whatever
one chooses. So long as an option is enticing enough, one may choose it bliplessly.
But the difficulty for this idea is that the agent seems to be ignoring something
which by his own lights is practically relevant, namely that some other option
would be better. Why isn’t this just another blip—and one not very different in
style from the first one? The difference between ‘best’ and ‘right’ or ‘ought’ doesn’t
seem to get a grip here.
It is hard to be quite sure what is supposed to be explaining what at this point.
I think the real pressure is only to make sure that the two aspects of the position
are consistent. The two aspects are the idea of the absence of blips and the
distinction between ‘bests’ and ‘oughts’. The latter was introduced not so as to
explain the former, but so as to help us accept that enticing is a favouring relation
of a different style from that of the more familiar peremptory reasons. The idea of
the absence of a blip was introduced in order to give some form to the idea that
certain styles of criticism are not appropriate when we are dealing with enticers.
The crucial question is, therefore, whether the person who says that one option
  303
would be best but that he is going to prefer another is vulnerable to criticism of the
very same style as that appropriate when we fail to select the option we know we
ought to select.
The person who asks what is the thing to do, answers his own question, and
then fails to select that option is failing to attend to the results of his own
deliberation. But when I ask ‘What shall I do?’ I am in a rather different case.
The decision that one option is more pleasant, and from that point of view better,
is not itself the decision that this is ‘what I shall do’. So in failing to select that
option my judgement is not at odds with itself. There is, that is, a gap between the
product of evaluative deliberation and choice that we do not find between the
product of deontic deliberation and choice. When I rank options for pleasurabil-
ity, I do so on the way to a decision, but the ranking I emerge with lies at a distance
from the decision that I will make in the light of it. What I want to know is how
much fun each option will be, and in what way, and I can express my discoveries
in terms of what is more or less enticing, and in those terms better or worse. With
that information I will choose, but there is not the same sense that my choice has
effectively already been made as there is when I have already answered the
question what to do. In this way, it seems possible to suggest that the deontic
blip is not present on the evaluative side.¹⁴
Of course, for anything like this suggestion to work, there will have to be the
sort of difference between ‘ought’ and ‘best’ that some positions are officially
opposed to. But we knew already that the possibility of enticers would be easier to
make out on some views than on others.¹⁵
¹⁴ One might say, in reply, that on the evaluative side the blip is that of failing to prefer (choose)
what one has already preferred (judged better); in doing so one is at odds with oneself. But to say this
would be to appeal to the notorious ambiguity of ‘prefer’.
¹⁵ This paper was written for a festschrift in honour of Joseph Raz. I am most grateful to Michael
Ridge for extensive comments on earlier drafts, and also to John Broome, Roger Crisp, Brad Hooker,
Christopher Hookway, Derek Parfit and Philip Stratton-Lake for lively discussion of these issues.
Comments from Jay Wallace made an enormous difference to a penultimate draft that had got out of
control. I am also grateful to the British Academy for the award of a Research Readership, during my
tenure of which this paper was written.
19
On Knowing One’s Reason
Discussion of the relation between knowledge and acting for a reason has become
much more complicated recently. Here are some views that are well worth keeping
apart:
1. You can act for the reason that p without knowing that p, but if you do so
you are in breach of a norm.
2. You cannot act for the reason that p if you do not know that p.
3. You can only act for the reason that p if it is the case that p. Otherwise your
reason will be something else, probably that you believe that p.
4. You can act for the reason that p whether it is the case that p or not, so long
as you believe that p.
5. You can only be guided by the fact that p if you know that p. Otherwise you
are merely treating p as a premise, and to do that you do not need to know
or even to believe that p.
I start with some comments on these views. The first is propounded by Hawthorne
and Stanley (2002). I am completely unpersuaded by it, but it is not my concern
here. The second is mooted by some people interested in the value of knowledge. If
one could not act for the reason that p without knowing that p, knowledge would
indeed have a significant and distinct value. But the same view is also attributable
to Peter Unger (1975) and to Jennifer Hornsby (2008), who have their own reasons;
it can also be attributed to John Hyman (1999; 2006; 2010; 2011), but only in a
certain sense, as we shall see. Hyman now distinguishes between two things that
may be meant when we talk of acting for a reason. The first is that of being guided
by a fact, and this requires knowledge. The second is that of taking a consideration
as a premise in one’s reasoning. He writes: ‘A person can reason from a premise that
he merely believes but does not know. But he cannot be guided by a fact that he does
not know, any more than a cat can stalk a bird it cannot see’ (2011: 366). This is in
fact the fifth view above. The third view is held, among others, by Clayton Littlejohn
(2012); I argued against it at length in my Practical Reality (2000a). It is probably the
mainstream view, despite my efforts. The fourth view is the one I supported in that
book, and which I continue to believe. What I think of as acting for a reason is what
Hyman is trying to capture in his talk of taking a consideration as a premise; but
I will repudiate that reading in what follows. What I am talking about is neither of
the two things that Hyman distinguishes.
   ’   305
For some people, the notion of acting for a reason cries out for analysis—
probably a causal analysis. I am not that much persuaded by the pretensions of
analysis, and would not be much perturbed if no acceptable analysis ever emerges.
Why should everything, or at least all the interesting notions, be analysable? On
the other hand, it is hard to prove that something is unanalysable. I know that
Pamela Hieronymi is on the hunt for a defensible analysis (2011). Perhaps
Hyman’s distinction counts as a disjunctive analysis. Still, there is at least the
option of abandoning the search for an analysis and seeking something else less
demanding, such as an account. And one thing I very much doubt is that some
normative status such as knowledge, justification or rationality will figure in an
account of acting for a reason, if only because one can act for a reason that is no
reason and one can act for a reason that is not the case.
1. Introduction
In an earlier paper (2011), I discussed the relation between the second, third and
fourth views set out above. I was taking it that Hyman’s view was the second view,
and I argued that he had offered no reason in favour of that view that could
influence someone (i.e., me) attracted to the fourth view, which I there called the
Non-Factive View.
The fourth view has two distinct merits. The first is that the sentence ‘His
reason for V-ing was that p’ does not entail p. The notion of acting for a reason is
therefore not a factive notion—unlike the notions of knowing that p, proving that
p, remembering that p (on some accounts) and seeing that p (also on some
accounts). The only way I know of to argue for this claim is by example.
The sentence ‘His reason for giving her the money was that her need was greater
than his—which it wasn’t’ is not contradictory in the way that ‘He knew it but he
was wrong’ is contradictory. The sentence ‘He gave her the money for the reason
that without it she was not going to survive’ does not entail that he was right about
that. Nor does the sentence ‘The reason for which he gave her the money was that
without it she was not going to survive’ entail that she was in such a dire state.
There is a way of supporting this appeal to examples, which might otherwise
look a bit bald. This is to point out that the purposes of intentional explanation of
action do not require showing that the agent was right about anything. What such
explanation does is to reveal the sense the agent saw in acting as he did. It reveals
the sort of supposed appropriateness to the situation that led him to respond in
that way; but to do that, all that we need to do is to show the appropriateness or
fittingness that he saw. We don’t have to show that he was right in seeing things
that way.
So you don’t need to know that p in order to act for the reason that p, and it
doesn’t even have to be true that p for you to act for that reason either. It is,
306   
however, required that you believe that p, or at least take it that p (if that can be
thought of as weaker than believing that p). But this does nothing to show that
where you act for the reason that p when it is not the case that p, your reason
‘really’ is that you believe that p. Belief is required, according to me, but that belief
is not the reason for which you act. This blunt statement needs disambiguation,
thus: it is still what you believe that is your reason for doing what you do, not your
so believing. There are cases where one’s reason is genuinely that one believes that
p, rather than simply that p, just as there are cases where one’s reason is genuinely
that one wants to V, rather than simply so as to V. But these cases are special, and
their special nature is revealing. If someone says ‘I want to get drunk every night;
so I am going to get a job in a bar’, this is a different thought from ‘I want to get
drunk every night; so the last place I should think of working is in a bar’. If
someone says ‘I believe there are pink rats in my boots, so I should call the pest-
control officer’, that is a different thought from ‘I believe there are pink rats in my
boots; so I need to see a psychiatrist’.¹
2. Factivity and explanation
Every statement that we make, in giving an explanation of anything, needs to be

true. But explanations often contain clauses that are themselves capable of truth
and falsity, and if the truth of the whole does not depend on the truth of the
contained part, we think of the context as intensional. There is another sense of
‘intensional’ according to which a context is intensional if substitution of exten-
sional equivalents need not preserve the truth of the whole. It is commonly
recognized that even if substituting extensional equivalents will retain the truth
of the whole, it may not retain explanatory power. What explains the installation
of street lights in a tiny village is a decision of the local council, not a meeting of
the butcher, the baker and the candlestick maker, even though they constitute the
council. Similarly, if what explains my hurry to get to the station is that my wife is
on the train, that is not the same explanation as one that says that a granddaughter
of the man who taught J. L. Austin all his knowledge of Latin and Greek is on the
train, even though it is the same person both times.
It might then appear that the statement that we make, in giving our explan-
ation, remains true under substitution of extensional equivalents, but may lose its
explanatory power and so become, not false as an explanation, but not the right
explanation. But this is not correct. For the explanation ‘his reason for hurrying to
the station is that his wife is on the train’ may be true when ‘his reason for
hurrying to the station is that J. M. Street’s granddaughter is on the train’ is false.
¹ These examples are now familiar. The first comes from Price (2008); the second I got from John
Hyman.
   ’   307
Is explanation factive, then? Not in itself, for explanations can contain clauses
that need not be true for the whole to be true. But we should not forget that we
only explain things that are the case (which includes why things that didn’t
happen didn’t happen). And more importantly, if we say that what explains p is
that q, we are committed to its being the case that p and that q. So the word
‘explains’ is factive, doubly factive, since what explains must be a fact and what is
explained must be a fact as well. We cannot say that what explains his running is
that the train is leaving, except that it is not. Still, we can explain his running by
saying that his reason for running is that the train is leaving, though in fact the
train has already gone.
If we don’t explicitly use the language of explanation, we can still adopt
explanatory devices that commit us. The word ‘because’ is one such. If I re-
express my explanation of his running, saying that he is running because the
train is leaving, I am now committed to the train’s actually being about to leave, in
a way that I would not have been had I merely said ‘his reason for running is that
the train is leaving’. I take these two explanations to be explanatorily equivalent, as
being effectively the same explanation, even though one commits the explainer in
a way that the other does not. This confirms me in my belief that that commitment
is superfluous to the purpose of the explanation.
3. Premises and guides
As I said above, Hornsby’s view is that one cannot act for the reason that p if one
does not know that p, while Hyman’s view is that in one sense one can and in
another one cannot. One cannot be guided by the fact that p unless one knows that
p, but that p can be the premise on which you act even if it is false. I have already
quoted this remark: ‘A person can reason from a premise that he merely believes
but does not know. But he cannot be guided by a fact that he does not know, any
more than a cat can stalk a bird it cannot see’ (Hyman 2011: 366). But Hyman
goes further, suggesting that one does not even need to believe that p to act on it as
a reason.² He notes, correctly, that one can draw inferences from a false premise,
and even from a premise that one knows to be false. But he claims that one can
even deliberate and act on premises (in his sense) that one believes to be false; his
example is that of a jury, who (in the UK at least) can be instructed by a judge as to
some matter of fact (e.g., that the goods were stolen), which in their reasoning they
have to take as given even if they believe it to be false.
² The remark about the cat is revealing. A cat cannot stalk a bird of which it is totally unaware, but
must that awareness amount to knowledge? What is it doing if the bird concerned is a plastic decoy, or
a hologram, or surrounded by fake birds? Maybe one cannot stalk something that does not exist, but
this tells us little about knowledge.
308   
I would say that any reason for which we act is something we are guided by. The
agent adjusts his behaviour to the relevant consideration, and in that sense has it in
view, in mind, and tailors his choice of action to it. We might, I suppose, say ‘This
consideration influenced me, but it cannot have guided me because it turned out not
to have been the case, or because it turns out that though it is true, I did not know it
to be true at the time’. But this would be to introduce a specious distinction between
an influence of the relevant sort and a guide. The supposed link between fact and
guide is similarly specious. The relevant notion is that of a consideration that one
seeks to accommodate in one’s behaviour, or that one does so accommodate.
What Hyman has done, with his distinction between premises and guides, is to
try to have it both ways. I am right (though I did not go far enough) in thinking
that one can act for a reason that is not the case; he is right in thinking that one
cannot be guided by a fact that one does not know. The important point is the
bifurcation in the theory of practical reasoning and action. As he sees it, there is a
significant difference between the two routes to action—so significant, indeed, that
a special role is found for knowledge in the better route, to which knowledge is
essential, not merely accidental. Myself, I would find no difficulty in allowing that
we have a special form of words for the case where the agent acts in the light of
knowledge of a fact. But that would not be to the point. For Hyman, there are two
entirely different relations between agent and reason. In one of them, we are
exercising our knowledge as an ability to be guided by a fact. In the other,
something entirely different is going on; we are reasoning from a premise.
In order to make this out, Hyman needs to downgrade the ‘ordinary’ case where
the agent does not know his reason to be a fact. Calling the reason a premise
certainly fits the idea that one can base one’s responses on something that is not
the case. But it also allows a case where the reasoner draws the relevant inferences
while believing, knowing even, that the premise is false. And it thus takes us too far
away from the realities of practical deliberation, in which people do not decide
what to do on the basis of things they know to be false.
The notion of a premise is not really helpful here, because of its link with
inference. Practical deliberation is not inference from premises, because one
cannot infer an action. And someone who is angry for a reason has not inferred
his anger from a premise (nor is his anger guided by a fact). There are no premises
for action, and the relation between relevant consideration and action is not the
same as that between premise and conclusion. So a jury does not take this or that
supposed matter of fact as a premise. But we might use the verb ‘premise’ in a way
that avoids all those connotations. The jury, perhaps, premise their decision on
this: that the goods were stolen—which they know, or at least believe, to be false.
Note that the jury’s decision is a practical decision, what verdict to bring in;
they are not required to ‘believe’ their verdict, though they are required to believe
that the verdict they bring in is the appropriate one in the circumstances, which
may include the judge’s directions. Some of the force of this example comes from
   ’   309
the implicit suggestion that the jury are reasoning to a belief rather than to an
action; this is what gives the notion of a premise what palatability it has. So what
are we to say about the jury whose deliberations are in part directed by the judge?
Hyman’s conclusion is that one does not need even to believe the premises of one’s
deliberation, that is, the considerations in the light of which one makes one’s
decisions. I would say that the reasons for which the jury come to their verdict in
this case include that the judge has directed them that the goods are to count as
stolen. That is what they would say if asked afterwards why they found the man
guilty. They would not say ‘the goods were stolen’, because that was not their
reason for coming to the verdict they did.
4. Edmund
This takes me to an example which both Hornsby and Hyman use against my view
about the nature of reasons-explanations. As we have seen, despite the differences
between them, they agree that there is a relation in which one stands to a
consideration when one acts for it as a reason. For Hornsby, there is only one
such relation, and it is knowledge. For Hyman, there are two, one of which is the
one I have been talking about, and the other is the sort of guidance that requires
knowledge. My view is that one’s reason for V-ing can be that p even if one does
not know that p, so long as one believes that p, and that there is no other relation
than this at issue when we are lucky enough to know our reason to be a fact. Now
some of these differences could be seen as merely semantic. It could be true that
one cannot be guided by a consideration that is not a fact, if we can construct a
suitably exclusive notion of a guide. And I could perhaps allow that the phrase ‘the
fact that p’ is knowledge-involving, and carry on maintaining that even if one’s
reason for V-ing cannot be the fact that p if one does not know that p, one’s reason
can still be that p.
But matters are not so easily resolved. Here is a now familiar example, taken
from Hornsby, but adapted to avoid any sign of the phrase ‘the fact that’:
Edmund . . . believes that the ice in the middle of the pond is dangerously thin,
having been told so by a normally reliable friend, and . . . accordingly keeps to the
edge. But Edmund’s friend didn’t want Edmund to skate in the middle of the
pond (never mind why), so that he had told Edmund that the ice there was thin
despite having no view about whether or not it actually was thin. Edmund, then,
did not keep to the edge because³ the ice in the middle was thin. Suppose now
that, as it happened, the ice in the middle of the pond was thin. This makes no
³ I would prefer ‘for the reason that’ here.

310   
difference. Edmund still didn’t keep to the edge because the ice was thin. That the
ice was thin does not explain Edmund’s acting, even though Edmund did believe
that it was thin, and even though that it was thin actually was a reason for him to
stay at the edge.⁴
This example is a direct attack on views expressed in my Practical Reality.

I maintained there that to act for a reason is to act in the light of a consideration
that one believes to be the case. I also maintained that to specify the reason for
which someone acted is to give at least a partial explanation of her action. How
then am I to respond to this example?
The example comes in two waves. In the first wave we are not told whether the
ice in the middle was thin or not. But Edmund believed that it was. So, I would say,
that was the consideration in light of which he acted, and we explain the action
(partially, at least) by specifying that consideration. But the point of the Edmund
example is that what explains his action is not that the ice was thin, but simply that
his friend told him that it was. The explanation stops there, as it were. If the friend
had told him that the ice was thin even only partly because he took the ice to be
thin, then we could say that the supposed thinness is part of the explanation of
Edmund’s keeping to the edge. But since the friend was not concerned with
whether the ice was thin or not, we can’t say that.
We can check this by re-running the example in terms of what helps us to
understand Edmund’s action. Edmund’s friend’s advice does help us to under-
stand his action, and that the ice in the middle was thin does not. And nothing
changes, as far as that is concerned, if we introduce the mere fact that the ice in the
middle was indeed thin. That still does not help us to understand his action.
The Edmund example faces us with an inconsistent triad:
1. We explain an action by giving the reason for which it was done.

2. Edmund’s reason for keeping to the edge was that (as he had been told) the
ice in the middle was thin.
3. That the ice in the middle was thin is not (any part of) what explains
Edmund’s keeping to the edge.
If what explains Edmund’s action is not that the ice was thin, what does explain it?
We have two other possibilities: that Edmund believed that the ice in the middle
was thin, and that Edmund’s friend had told him that the ice in the middle was
thin. Hornsby and Hyman opt for the first of these. I am not sure that this sits
entirely comfortably with their views about the relation between knowledge and
action, since if his reason was that he believed that the ice in the middle was thin,
⁴ Adapted from Hornsby (2008: 251); her example is very similar to one given by Unger (1975:
209–10).
   ’   311
this should, according to them, be something that he knew. And it seems to me to

be very odd to prefer these psychologised versions of reasons on the grounds that
we generally know whether we believe something or not. My preference would be
for the other option: that what explains Edmund’s action was that his friend had
told him that the ice in the middle was thin. (This is something that we can all
allow that he knew.) Still, it remains possible that Edmund was adjusting his
behaviour to the supposed thinness of the ice, not to the nature of his friend’s
advice. What for him was the guiding consideration was the thinness of the ice of
which he had been warned, not that he had been warned of it. He might quite
properly have said, if challenged, that he was sticking to the edge because of the
state of the ice in the middle, or, more cautiously (in the sense that it avoids
the word ‘because’), that his reason for sticking to the edge was that the ice in the
middle was thin. Of course, a more sceptical Edmund might have wondered about
his friend’s reliability, and thought that, well, better safe than sorry, sticking to the
edge is the sensible thing. And in that case his reason could have been simply that
his friend had warned him off. (This is like the jury whose reason is not that the
goods were stolen but that the judge has so instructed them.)
But the point is that, no matter how we characterise Edmund’s reason, the
thinness of the ice is not what explains Edmund’s action, and on my account, it
ought to be, if that was his reason.
5. Explanans and explanandum
When I wrote Practical Reality, I worked with an assumption, that where there is
an explanation there must be an explanandum, a thing to be explained, and an
explanans, the thing that does the explaining, and that these two things will be
linked in some explanatory relation. This assumption comes from thinking about
the causal case (which itself should have led me to question it). We explain the
lighting of the match by saying that it was struck, where that the match lit is the
explanandum, that it was struck is the explanans, and there is an implicit
explanatory relation, the causal relation. Turning to the explanation of action,
we seem to find the same structure. His reason for running was that the train was
leaving. The explanandum is his running, the explanans is that the train was
leaving, and the explanatory relation is the ‘in the light of ’ relation; he ran in the
light of this: that the train was leaving.
It then seemed to me that there are plenty of examples of the explanation of
action where the agent’s reason for acting was not in fact the case. Sentences of the
form ‘his reason for V-ing was that p’ can be true even when it is not the case that
p. (This of course is not true of ‘E2 happened because E1 happened’.) This led me
to think that these explanations of action are not factive, since ‘his reason for V-
ing was that p’ does not entail p. And this raised the bewildering question how
312   
something that is not the case can be what explains something that is the case.
Some thinkers are so put off by this question that they resort to different accounts
of the agent’s reason for any case where the agent has made a mistake. (This is the
third view outlined in the preamble to this paper.) The most popular account
is that, in any case where the agent is mistaken, his reason is not that p, but simply
that he believed that p. This has the awkward consequence that in all such cases
the agent makes a double mistake. In acting (as it seems to him) for the reason that
p, he mistakenly takes it both to be the case that p and that he is acting for
that reason; he is wrong on both counts. I, by contrast, remain unwilling to allow
that in all cases where an agent makes a mistake, we need a different account of his
reason for the relevant action. It still seems to me that the agent’s reason is what it
is, whether he is right about that or not.
But still there is the issue about explanation. The Edmund case puts pressure on
the idea that in every case, the reason for which the agent acted is fit to act as
explanans. As I suggested above, there is an Edmund whose reason for keeping to
the edge was genuinely that the ice in the middle was thin, as he had been warned.
There is no theoretical reason to rule out this possibility. But if we do allow that
possibility, we have still to allow that the thinness of the ice is not what explains his
action, even if the ice was in fact thin.
This, however, does nothing to undermine my sense that we explain an action
by specifying the reason for which it was done. (This is the first member of the
inconsistent triad.) If we stick to this thought, the only possible resolution is that
the explanans is not identical with the reason for which the action was done. In
fact, it now seems to me⁵ that I was wrong to seek a separate explanans; the
seeming need to do so stemmed from incautious adoption of the structures visible
in causal explanation. If we avoid that move, we can say that what explains the
action is that it was done for the reason that p, without committing ourselves to
saying that what explains the action is that p. It would remain true, however, that
we explain an action by giving the reason for which it is done. All that we lose is
the idea that the explanans is a proper part of the explanation as a while.
In making this move, we can find support from the nature of purposive
explanation. If our explanation of an action is that it was done in order to save
money, we are certainly not committed to the money having actually been saved.
So there is a sense in which a purposive explanation is not factive—and of course
on the face of it such explanations are neither factive nor non-factive, since there is
no contained proposition capable of being either true or false unless we reconfig-
ure the whole in some artificial way so as to ‘discover’ such a proposition. So what
is the explanans in a purposive explanation? It cannot be ‘in order to V’; this seems
⁵ In coming to this change of view, I have yielded to a barrage of criticism from Davis (2003), Sehon
(2005), Hieronymi (2011) and Sandis (2012).
   ’   313
the wrong sort of thing altogether. Nothing remains but the idea that the explan-
ans is the whole: he did it in order to V.
One thing that persuaded me originally that the explanans in reasons-
explanations is the reason itself was simply that it seems to be that bit that does
the work. If our structure is ‘He did it for the reason that p’, the only thing that can
vary here is the ‘that p’ bit, and when we do vary it we get different explanations.
So, it seemed, it is that bit that carries the explanatory burden. And the same
thought would apply to purposive explanations. The action itself is given as that
which we are trying to explain, and the bit that does all the work is the ‘in order to
V’ bit. But if this does not persuade us to think of ‘in order to V’ as the explanans
in a purposive explanation, the same reasoning should not persuade us to think of
‘the reason that p as the explanans in reasons-explanations.
The reason for which the Edmund I have in mind stays to the edge is that the
ice in the middle is thin, as he has been warned. When we say that his reason for
staying at the edge was that the ice in the middle was thin, we have explained his
action. What we have not yet done is explained why he stayed at the edge for that
reason. Ordinarily, when someone acts for a good reason, we don’t ask why he
acted for that reason unless there is some peculiarity in the case. Perhaps there
were several reasons that he might have acted on, and we want to know why it was
this one. Perhaps he has been known to pay no attention to such reasons, or more
attention to other reasons, in the past.
How do these thoughts relate to the idea that that the ice was thin does not help
us to understand Edmund’s action? When we explain an action as having been
done for the reason that p, the agent’s having been right in believing that p adds
nothing to the force of the explanation. So in Edmund’s case, that the ice in the
middle was (in fact) thin does nothing to explain his action. He would have acted
in the same way had the ice not been thin. So that the ice was thin does not
contribute to the explanation of the action, but in giving his reason, namely that
the ice in the middle was thin, we do explain the action. All this is consistent, once
we allow that though we explain an action by giving the agent’s reason for doing it,
the reason itself, namely that p, need not be the case and does not make the sort of
distinct contribution to the explanation that would enable us to think of it as the
explanans.
On this account, the relevant explanations are non-factive, since they have a
contained clause which does not have to be true if the whole is to be true. But I said
in Practical Reality that the explanans in a reasons-explanation does not have to be
the case, and I now accept that this is wrong. This is an important matter for me,
because I used this mistaken claim as an argument that reasons-explanations
cannot be causal, which is a big conclusion. That conclusion is still available, but
needs to be reached rather differently. What I would now say is that the reason
why reasons-explanations cannot be causal is that they don’t have the structure of
314   
a causal explanation, since there is no discriminable part of the explanation that

stands as the explanans.
6. Two forms of explanation
Hyman is alert to all these issues. His response to them is to say that when the
agent is guided by a fact, that fact acts as explanans. But when the agent is merely
responding to a consideration as a premise, there is no discriminable explanans.
He writes:
If someone is said to have been guided by a certain fact, this means that he took it
into consideration, when he modified his thought or behaviour in some way, or
decided what to think or what to do. This, I suggest, is how we should understand
the idea of a fact being a person’s reason. Furthermore, if we understand the idea
of a fact being a person’s reason in this way, then the explanans can be the
person’s reason. (2011: 361)
Elsewhere, one takes it, there is no discriminable explanans in the explanation ‘he
did it for the reason that p’
Think of the matter in terms of motivation. Hyman offers an implicit distinc-
tion between two forms of motivation: there is motivation by a consideration that
may or may not be the case, but is a different sort of thing from a fact, something
more like a premise, and there is motivation by something that one knows to be
the case, namely a fact. This is where the whiff of disjunctivism appears. If one
does not know the fact to be the case, one cannot be motivated by it, even if one
has perfectly good reason to believe it to be so. In any case short of knowledge,
one’s motivation has an entirely different structure.
In a way, I should be sympathetic to this picture, because I have always found
disjunctivism in the theory of knowledge an attractive resolution of (certain)
sceptical difficulties.⁶ But I don’t see the same attraction here. I don’t see any
reason to believe that where one fails to know the fact to be so, one’s reason cannot
still be that fact. The issue about explanation (whether there is or is not a
discriminable explanans) does not need to be accompanied by a distinction
between two distinct motivational structures. It remains possible that what motiv-
ates the agent is in fact (sorry) a fact, though the agent does not know that to be so.
There is no need to say that, in the absence of knowledge, what motivates is not a
fact but something much more like a premise, something capable of failing to be
the case. If one does not know one’s reason to be a fact, this does not show that it is
⁶ See my ‘Arguments from Illusion’ (1995a; repr. as Chapter 15 in this volume).

   ’   315
not a fact but something of a different order. Hyman writes: ‘In this kind of
explanation, knowledge is transparent: we can look straight through it to the fact.
This is why I said earlier that in this kind of explanation, the explanans can be the
person’s reason’ (2011: 367). But this does not show that in non-transparent cases,
the consideration we have in mind when we act is not a fact. It only shows that it
may not be.
One might worry that something that is capable of not being a fact cannot be a
fact. Only things other than facts are things that may not be facts. So something
that may not be a fact is already not a fact. But there is a modal error here. Suppose
that if you know that not p you cannot be wrong: knowledge is an infallible state of
mind. Still, what we mean when we say that your reason may not be a fact is that,
for all you know, it is not a fact. This does not already show that it is not a fact.
Returning now to explanation: Hyman says that we can explain actions by
citing a mental state that influenced the agent (knowledge or belief), or by citing
the agent’s reason. When we cite the reason, our citation can take one of two
forms. We can say ‘his reason was that p’, or we can say simply ‘p’. The second sort
of explanation is not available when the agent is mistaken, since that p can explain
nothing if it is not the case. And it is not always available even when the agent is
not mistaken; this is the point of the Edmund example. It is available when and
only when the agent knows that p. So when the agent knows that p, we can look
through the agent’s mental state to the fact known, and it is only in those cases that
that p can explain one’s action when one acts for that reason.
But there is a different picture available, which has it that that the reason for
which we act is the same sort of thing whether we are wrong or right, whether we
know or don’t know. That the explanation of Edmund’s keeping to the edge is not
that the ice in the middle was thin (even though that was his reason), shows that
this is not what explained his action, but does not show that it was the wrong sort
of thing to explain his action, in a way that needs to be explained by talking about
knowledge. An explanation can be ‘available’ in the sense of the previous para-
graph, but still not the correct explanation. Further, if the explanation of
Edmund’s action is that his friend told him that the ice in the middle was thin,
this is not because this, at least, is something that Edward knew. It is because if the
friend had not told him, Edmund would not have acted as he did, but if the ice in
the middle had not been thin, he would still have acted as he did.
7. Facts
Hyman’s view is that the fact that p cannot be one’s reason for acting unless one
knows that p. This view seems to me to be eventually less interesting than it
initially appears. If I say that his reason for running is the fact that p, it may be that
I must claim knowledge that p. If so, there is nothing special about acting for a
316   
reason that links it to knowledge. The link to knowledge comes entirely from the
commitment implied by any unconditional reference to particular facts as facts.
The speaker speaks infelicitously (at the least) if he speaks of the fact that p when
he does not know that p. The speaker also speaks infelicitously if he says of
someone else that their reason was the fact that p, if the speaker does not know
that p. But this sort of infelicity does not arise if one says of someone else that their
reason was the fact that p, while denying that they knew it to be the case that p.
This alternative picture depends upon a sort of paratactic reading of ‘his reason
for V-ing was the fact that p’. The reference to p as a fact is a sort of comment on
the part of the speaker, rather like the ‘I suppose’ in ‘his reason for V-ing was
I suppose that p’. There is a difference between these two cases, though, because
the hook for the parataxis, if I may be allowed that expression, in the first case is
simply ‘that p’, while in the second case it is the whole sentence ‘his reason was
that p’. And this raises the possibility of another way of hearing ‘his reason for V-
ing was the fact that p’. We might hear it as a paratactic comment supposedly from
the agent’s perspective. If so, this would be a point for Hyman, in a sense, for it
would then look as if the agent is the one thinking in terms of facts, and so
committed to the relevant knowledge claim. But that the agent is committed to the
relevant knowledge claim does not show that if he does not know, he was not
acting for the reason he thought he was acting for. For that reason does not change
just because the agent thinks of his reason as a fact (and thereby, supposedly, as
something he knows to be so). If I think of my reason as the fact that p, it is not
qua fact that it is my reason. Calling it a fact does not change its nature.
8. Conclusion
I end with some general reflections about reasons and explanation.

We all agree that there is a distinction to be drawn between the idea of a good
reason to act and the reason why, or for which, someone acted. A good reason to
act in a certain way is a consideration that favours so acting. Such a reason can
exist without anyone knowing about it, and also without anybody’s being in the
slightest degree tempted to act on it. But it is there all the same, in uninfluential
obscurity. If one explains anything by citing such reasons, it is some aspect of the
case, such as it is, that can be made for so acting; and maybe thereby one also
explains why someone who did so act might not have been done so completely
unreasonably, had this been their reason. But the notion of a good reason is not
itself an explanatory notion, even it is involved in certain explanations.
When we give the reason why, or for which, someone acted, then we are
necessarily in the business of explaining their action. I would say that the notion
of a reason why something happened is the same notion as that of a reason why
someone acted, and that this notion is essentially explanatory. A reason why
   ’   317
something happened is necessarily a (partial) explanation of that something,

whether it be a mere event or an intentional action. And to give such a reason is
necessarily to give a (partial) explanation.
Matters are different with reasons-for-which. It is, I think, true that to specify
the reason for which someone acted is to offer a (partial) explanation of their so
acting. But this explanatory aspect is a consequence of what is said when one gives
the reason-for-which, not what is said itself. I would say, therefore, that the notion
of a reason-for-which is only derivatively explanatory. The sort of explanation
being offered only makes sense to the extent that we see the action concerned as a
response to the call of a (supposedly) good reason. And, as I have already
suggested, the notion of a good reason is not captured merely by talk of what
plays a salient role in some explanation. Appeal to reasons-for-which plays a
specific role in the understanding, of which the idea of a feature which in the
context calls for a response of a certain type is central, and any explanatory
potential derives from that central idea, rather than being identical with it. An
agent who acts for a reason acts in the light of a certain consideration, and in
coming to learn what that consideration was, we understand the action from the
point of view of the agent. The agent, in acting, had his eye on a certain
consideration, and being told about that (which is not itself an explanation,
even if it is what explains) we learn some part of the explanation of his so acting.
To put it another way: the ground of an action is the (supposed) reason for
which it was done, and to specify that ground is to go some way towards
explaining the action concerned. But the notion of a reason as ground is not itself
an explanatory notion.
In a similar way, the notion of an event (as cause) plays a role in the causal
explanation of events, but is not itself an explanatory notion. As for the notion of a
cause, some understand that as essentially explanatory; Hobbes, for instance,
offered the following as ‘an exact notion or idea of that which we call cause’ thus:
A cause is the sum or aggregate of all such accidents, both in the agents and in the
patient, as concur to the producing of the effect propounded; all which existing
together, it cannot be understood but that the effect existeth with them; or that it
can possibly exist if any one of them be absent. (1839: i, 77)
Hobbes seems here to be thinking of his remark about what cannot be understood
as a proper part of the ‘exact idea of what we call a cause’, rather than as a gloss, so
that the notion of a cause is essentially a rational, or explanatory, notion. Others
think of the causal relation as the cement of the universe, and cementing is not an
explanatory relation.
Of course many actions are explained in other ways than by giving the reason
for which they were done. We might explain an action by mentioning the
proclivities of the agent, by mentioning his upbringing, his inattention, or his
318   
overwork. In most such cases, what is explained is not just the action, but the
action together with the reason for which it was done. When the reason for which
it was done is an evidently good reason, the question why the agent so acted is in a
way already answered. It is when there was no good reason, or the reason-for-
which was not a very good reason, that we tend to seek further explanation. But
that tendency does not show that some explanation has not already been given
when we are told the reason for which the agent acted. When we are told that, after
all, we are told something about the consideration in light of which the agent
thought it worthwhile so to act.
SECTION 4
LEARNING FROM THE
INTUITIONISTS
20
Prichard on Duty and Ignorance of Fact
Introduction
I contributed this paper to a volume titled Ethical Intuitionism: Re-evaluations

(Stratton-Lake 2002) for two reasons, one historical and the other not. The
historical reason is that Prichard’s paper ‘Duty and Ignorance of Fact’ (1932)
caused a complete volte-face in Ross. Ross’s first book The Right and the Good
(1930) expounded the view that Prichard’s paper was intended to undermine, and
Ross was so completely persuaded by Prichard that he included in Foundations of
Ethics (1939) a ten-page summary of Prichard’s argument, often word for word.
(Nonetheless, he managed, as I will suggest below, to make a couple of significant
errors.) This interesting example of the influence that Prichard had over Ross is
worth re-examination. I suggest that Prichard’s arguments were not really as
powerful as Ross supposed.
The non-historical reason is that intuitionism is often thought of as a highly
objective view, under which there are normative (and non-natural) features of the
world in virtue of which certain actions are demanded of us. Certainly that was the
standard intuitionist position. We find it in Broad’s conception of rightness as a
sort of fittingness that can hold between action and situation (1930: 164–5), which
was adopted by both Ross (1939: 55ff.) and Ewing (1947: 132–6). Prichard argues,
by contrast, that whatever may be the case about rightness, agents’ obligations do
not depend on features of the situation but on features of their perspectives. The
arguments he offers would, I think, serve to suggest that reasons, too, are
grounded in features of agents’ thought rather than given them by the nature of
the situations they find themselves in. If sound, then, Prichard’s arguments
support a far less ‘objectivist’ position than the one we standardly associate with
him. This ‘subjectivist’ form of intuitionism remains to be assessed.¹
It is not, however, as if the issue is one that only affects intuitionists. It at least
reminds one of a familiar debate within consequentialism. What divides ‘object-
ive’ consequentialists from ‘subjective’ consequentialists (at least those of the ‘act’
rather than the ‘rule’ variety) is the question whether the rightness of an act is to be
determined by its actual consequences or by the consequences it may reasonably
¹ I make a much more substantial attempt to argue the objectivist case in my Practical Reality
(2000a: ch. 3). Ewing rejects the subjectivist position roundly (1947: 122–5). Carritt also offered an
extended discussion of the issue (1947: ch. 2).
322    
be expected to produce. Of course this is not quite the same as the question
whether moral reasons are grounded in features of the agent’s thought. For what
can reasonably be expected is not necessarily the same as what the agent reason-
ably expects. Still, some of the pressures that drive people towards subjective
consequentialism are also visible in Prichard; I will be suggesting that thinking
about Prichard should help us to resist those pressures.
In his 1932 Henrietta Hertz Lecture ‘Duty and Ignorance of Fact’,² Prichard’s
initial question is, ‘If a man has an obligation, i.e. a duty, to do some action, does
the obligation depend on certain characteristics of the situation in which he is,³ or
on certain characteristics of his thought about the situation?’ (2002: 84). And the
answer that Prichard apparently gives to this question is that the obligation
depends on characteristics of the agent’s thought, not on characteristics of the
situation. Prichard appears, that is, to adopt what he calls throughout ‘the
subjective view’ as opposed to ‘the objective view’. However, a version of ‘Duty
and Ignorance of Fact’ that is printed in a posthumous collection, Moral
Obligation (edited by W. D. Ross in 1947, shortly after Prichard’s death), contains
a sort of appendix that was not included in the original 1932 publication,⁴ and in
which Prichard expresses second thoughts about various aspects of his own paper,
including apparently some dissatisfaction with the initial question referred to
above. It may even be that the right way to think of his final ‘resolution’ of the
difficulty (contained in the appendix) is to take it as an attempt to collapse the
question rather than as the adopting of one horn of the dilemma and the rejection
of the other. I return to this issue at the end of my paper.
Prichard says that the main difficulty with the objective view is a sceptical one:
that, if it were true, then, ‘although we may have duties, we cannot know but only
believe that we have; and therefore we are rendered uncertain whether we, or
anyone else, has ever had, or will ever have, a duty’ (2002: 88–9). To reach this
conclusion, Prichard first appeals to a certain result in the philosophy of action,
namely that the only things that we ‘strictly speaking’ do are things that we can, as
he puts it, do directly. For ‘where we bring about something by causing something
to cause it, the result is not wholly due to us’ (2002: 85–6). Causing that effect,
then, is not something that we have done ‘directly’; the thing we did directly was
the action that resulted in that effect. According to Prichard, it seems, strictly
² This lecture was originally published in the Proceedings of the British Academy (1932), and is
reprinted in Prichard (2002). Page references are given here to the latter version.
³ Later: characteristics of the action; later again: characteristics of the activity.
⁴ I imagine that Ross found these notes among the papers that Prichard left after his death, but
I have no evidence that this is so.
       323
speaking we only have duties to do what we can strictly speaking do, i.e. do
directly. So our duties are all to be expressed in the following form:
When the situation in which a man is contains a thing of the kind A capable of
having a state of the kind X effected in it, and when also it is such that some state
or combination of states Y which the man can bring about directly would, if
brought about, cause a state of the kind X in A, the man ought to bring about that
state or combination of states. (2002: 86)
Prichard adds to this account of the nature of our duties an epistemological

position that he inherited from Cook Wilson and which leads him to claim that
we can never know what our action will cause. Put these two results together and a
blanket form of moral scepticism emerges. We can know that there are duties in a
conditional sense; we can know that if we are in a situation where an action we can
do directly would have certain effects, we ought to act in that way. But we can
never know unconditionally, either of ourselves or of anyone else, that we or they
are in fact bound to act in that way, for our incurable ignorance of the future
effects of our actions means that we can never know that we are in a situation of
that sort.
There is a doubt about Prichard’s thinking here. For his official claim is that on
the objective view we can never know that anyone has ever had or will ever have a
duty. Part of this is in the past tense, and for that part to hold good one of two
things must be the case. Either Prichard’s scepticism about knowledge of conse-
quences must be a fully general causal scepticism, and apply to knowledge of the
effects of some past event or action as much as to those of a present or future one.
Or, more plausibly, he is supposing that I cannot have a duty unless it is possible
for me to know that I have it at the moment of action. Since, he would say, this is
never possible, nobody can know later that I had a duty at that earlier time if it was
not possible for me to know it then.
This is sufficient to establish Prichard’s claim that we are rendered uncertain
whether anyone has ever had or will ever have a duty, so long as we take that claim
in a certain way. Suppose, however, that I know that I have two conditional duties,
with contradictory antecedents. I know that if my action would have a certain
effect, I ought to do it, and if it would not have that effect, I ought to leave well
alone. In this case, despite Prichard’s argument, I know that there is something
that I should do, though I do not know what it is. His claim that we are rendered
uncertain whether we, or anyone else, has ever had, or will ever have, a duty, has to
be understood as meaning that there is no duty that we can be sure that anyone
had, has or will have. So understood, the argument is sound, I think.
Prichard then says that there is only one alternative to the view that our
obligations rest on the facts of the situation, and that is that the obligation depends
on our being in a certain attitude of mind towards the situation (2002: 89); and he
324    
says that the most plausible account of the required attitude of mind is as that of
thinking certain things likely. Here I want to say that there are surely other
possibilities, for all that has yet been said. There are three contrasts that we can
use to generate other options, those between:
(1a) what is the case vs. (1b) what is likely to be the case;
(2a) what is likely given available evidence vs. (2b) what is likely given what
the agent believes;
(3a) what is actually likely (on whatever basis) vs. (3b) what the agent believes
to be likely.
Take the first contrast, between what is the case and what is likely to be the case.
The objectivist who goes for the latter of these has to give some account of the
contrast between the facts and the ‘probable facts’. Relative to what are the
probable facts probable? This takes us to the second contrast. Suppose that we
opt for the view that duty is determined by the probable facts, and that those
probable facts are determined by available evidence. They are not determined by
the beliefs of the agent, which can be false, misconceived, etc., but only by a
selection of the actual facts, those that are in some suitable sense available to—to
whom? We have a choice between saying ‘available in general’ and ‘available to the
agent’. Let us allow, since we cannot keep everything in the air, that we should be
thinking here about availability to the agent. This would give us a position (the
combination of (1b), (2a) and (3a)) that would have some defence against
Prichard’s sceptical worry. If the ground of duty always consists in facts available
to the agent, it would be harsh to insist that we never know that we have a duty, or
indeed what that duty is, even if we still doubt that we know this of anyone else.
Still, we should note that what Prichard has done is to opt immediately for (3b),
the second option of my third contrast, in expressing the subjective view in terms
of what the agent believes to be likely. This makes it plausible to suppose that of
my second contrast, he has opted for (2b), and thinks of likelihood as determined
relative to the beliefs of the agent, rather than in terms of available evidence. The
significance of this is that on Prichard’s view the ground for obligation includes
things that are not the case—false beliefs of the agent’s (understanding ‘beliefs’
here as things believed)—while on the other view it does not. That is, on
Prichard’s view, obligation is determined by what the agent believes would be
likely given what else he believes, whether what he believes is the case or not.
There are all these difficulties, then, in agreeing with Prichard that there is only
one alternative to the objective view, namely the combination of (1b), (2b) and (3b).
But Prichard needs to be handled carefully. He often fails to put his reasoning in at
all, saying such things as ‘When we reflect, we see’. It may be that there are
arguments against these other alternatives that he has just not mentioned yet.
       325
So far, we have had a sceptical argument against the objective view, which
I don’t think many of us will find very persuasive. I will just mention three reasons
for not finding it persuasive. First, I do not think that Prichard has done anything
to persuade us that none of the actions we are obliged to do are things that we can
only do indirectly. He offers some examples of obligatory actions that we cannot
do directly. I ought to pay you back the money I owe you, but I do this by sending
you a cheque, and it is not true that I ought to send you a cheque. The obligation is
to ensure the effect, not to take this means. But what of a case, where, having
affronted you, I ought immediately to apologize? Prichard can only get round this
sort of example by saying that the real obligation is to bring it about that you are
no longer affronted or upset or whatever. Consider his very strained interpretation
of the obligation to honour one’s parents: ‘By “A man ought to honour his
parents” we mean: “A man ought to bring about in his parents the knowledge
that he holds them in honour” ’ (2002: 85). But though I do have this obligation,
I also have the obligation to hold them in honour, which is something that I can
do directly. Second, I do not see any reason to allow that I can never know what
the consequences of my actions will be. Obviously with many actions this will be
true. But don’t I know that when I have torn this paper (i.e., moved my arms in a
certain way while holding the paper), the paper will end up in two or more pieces?
Third, I see no reason to suppose that we can only strictly speaking have
obligations to do what we can strictly speaking do. The ‘strictly speaking’ operator
need not behave in this way. Prichard uses the terms ‘strictly speaking do’ and
‘directly do’ indistinguishably, and says that we ‘indirectly do’ any action whose
result is ‘not wholly up to us’. He would say, for instance, that my persuading you
to come to the party is something that I do not do directly, and therefore I do it ‘in
a less strict sense’ (2002: 85). The idea is that I only really do that for which I am
not beholden to anyone or anything else; the appeal here is to a sort of pure
agency. I directly or strictly speaking only do those things that there is no chance
I should fail to do if I set myself to do them. This is the principle that leads him
later in the paper to announce that the only thing that I can strictly speaking do is
to ‘set myself ’ to act in one way rather than another (and, hence, that one can only
be morally obliged to set oneself to achieve certain things, never actually to achieve
them.) My own view about this is that the things that require the cooperation of
others may be harder to achieve, but still their cooperation may be something that
I do achieve,⁵ something that I can be rightly proud of having achieved, and
something that I may have a duty to achieve. I see no reason to suppose that just
because I cannot do it ‘directly’, I cannot strictly speaking have a duty to do it.
⁵ There is a second principle that only emerges later, which is that if I do B by doing A, I do A in
some more direct sense than that in which I do B. Prichard asks, ‘By directly causing what, did I cause
what I did?’ (2002: 96). This appeal to the by-locution should not be taken as evidence that if I do B by
doing A, my duty can never strictly speaking be a duty to do B. I may signal by waving my hand, but my
duty is to signal (in some way or other), not to wave my hand.
326    
I suspect we are much more likely to be impressed by a different argument

against the objective view which for Prichard appears to be somehow secondary,
since he offers it only to show that much of our ordinary thought is in direct
conflict with that view. (The sceptical argument that he prefers could hardly be
said to be an appeal to what we ordinarily think, and indeed I think that Prichard
prefers it because it rests on a philosophical conclusion, and that he takes that to
be firmer ground for argument.) The two examples that he offers on pp. 93–4 are,
as he puts it, examples of cases where we ought to do something as an insurance.
The first raises the question, ‘Ought we to stop, or at least slow down, in a car,
before entering a main road?’. The second, which Prichard got from Collingwood,
asks whether, if a car is heading towards a fork one branch of which leads to a road
that has collapsed in a landslide, we have a duty to stop the car, assuming that we
do not know which fork the driver is intending to take. The objective view is
committed to saying that we only have such a duty if in fact the driver was
intending to take the dangerous fork, and if he was not, our stopping him is an
action that we have no obligation to do. Similarly, if we do slow down thinking it
likely that there is traffic, we are only entitled to think that it is likely that we ought
to slow down, and if we find there to be no traffic, we should conclude that we had
no obligation after all.
The natural response to this is to claim that the ground of the duty to slow
down is the fact that there might be traffic—the chance of traffic—and that this
ground remains even if there is in fact no traffic. But Prichard’s reply to this is that
‘there are no such things as probabilities in nature. There cannot, e.g., be such a
thing as the probability that someone has fainted, since either he has fainted or he
has not . . . [T]he fact which we express . . . by the statement: “X has probably
fainted” . . . must consist in our mind’s being in a certain state or condition’
(2002: 94).
There are various things to be noticed here. The first is that this is surely a very
dubious and throwaway conception of the meaning of a probability statement,
open to all sorts of objections.⁶ The second is the weakness of Prichard’s argument
against objective probabilities, which consisted merely in the claim that ‘there
cannot, e.g., be such a thing as the probability that someone has fainted, since
either he has fainted or he has not’. For how would the existence of objective
⁶ See Ewing (1947: 125–8) where Ewing objects to this passage on predictable grounds, very similar
to his objections to subjectivist accounts of moral judgement. Ewing’s general view about Prichard’s
paper is that ‘Prichard’s difficulties . . . arise mainly from (a) ignoring the third sense of “right”, (b)
assuming that probability cannot be objective’ (p. 125 fn). The ‘third sense of right’ is that under which
the rightness of an act will depend on what is really probable, given our data. The first sense of ‘right’
takes rightness to be determined by what is in fact the case. The second sense takes it to be determined
by what we judge probable on the evidence available. Ewing allots a certain priority to the first sense of
right, but says he generally prefers to use ‘ought’ in the third sense (p. 128). Ross, taking Prichard to
have destroyed the supposed first sense, opts for the second and a ‘double dose of subjectivity
(1939: 164).
       327
tendencies in nature be incompatible with the Law of Excluded Middle? Perhaps

the problem lies only in objective probabilities about the past, or about the past
and present (though I doubt it). But as far as the future is concerned, the existence
of an objective tendency in nature to produce a certain result (if indeed the idea of
such a thing is coherent in the first place) is perfectly compatible both with that
result’s occurring and its not occurring, as well as with the thought that it must
either occur or not.
A further issue, however, is whether Prichard’s own example is consistent
with his conception of the meaning of a probability statement. Note that
this conception is run in terms of some state of mind of the speaker, not in
terms of some state of mind of the agent; Prichard claimed that the fact which
we express when we say that someone has probably fainted must consist in our
mind’s being in a certain state. As a result, Prichard seems to be unable to give
an account of what could be meant by our saying of the driver that he should
slow down because there might be traffic. It does not mean, on Prichard’s
account of probability statements, that he should slow down because he is in
a certain state of mind. But neither does it in fact mean what officially it should
on Prichard’s account, that he should slow down because we are in a certain
state of mind. Prichard’s account only works if the speaker is identical with
the agent.
But even if Prichard were right on all these points, I think there would still be a
further response available to the objectivist. This is to say that among the reasons
for one to slow down is that one does not know that there is no traffic. If one knew
that there was traffic, what one knows would give one a reason to slow down. In
this case, then, what one knows, rather than the knowing of it, gives one a reason.
But if one does not know that there is no traffic, it is one’s not knowing this,
together with the disaster there would be if one failed to slow down and there were
traffic, that gives one a good reason to slow down.
Someone else’s not knowing something can be a perfectly ordinary, objectivist
reason for action—e.g., for telling them whatever it is that they don’t know. In this
sense, so can the agent’s not knowing something, without this breaching any
objectivist defences. My not knowing something can give me a reason to go and
look it up. Someone else’s knowing something can be a good reason for action.
And so can the fact that one knows something oneself, e.g. that there is no traffic.
(And so, of course, can the fact known.) Further, if this person’s knowing that p is
capable of being a reason, so is her not knowing that p. None of this does anything
to show that reasons emerge from the perspective of the agent, in the sort of way
that Prichard seems to be intending to suggest. It just shows that facts about the
agent are often included among the reasons for or against action, and some of
those facts concern what the agent believes or knows (or hopes or fears or wants or
dislikes). All reasons are given us by facts of one sort or another, and none by our
false conceptions of how things are.
328    
It is with thoughts like these that one begins to wonder whether Prichard’s
original question was a good one. The reasons I have just been talking about are
that she knows that p or believes that q, and they are, let us suppose, her reasons. If
she knows that her friend is in distress, she has special reason to go round and
engage in a bit of support. Is this a reason over and above that grounded in the
distress itself? I think so. If I know that she is in distress and you merely wonder
whether she might be, though both of us have a reason to go round, I may have a
stronger reason than you do. In these cases, that the agent knows, believes or
suspects that p is to be understood as part of the situation. But it could also be
thought of as a characteristic of the agent’s thought. In which case Prichard’s
dilemma fails to get a grip.
It is beginning to look as if we were offered a choice between two views:
1. None of an agent’s reasons is grounded in the agent’s beliefs.

2. All of an agent’s reasons are grounded in that agent’s beliefs, maybe
probability beliefs.
And we think that neither of these things is true. Some reasons are grounded in
the agent’s beliefs, and others are not, and that is an end of it. All are, in the
relevant sense, objective. If we want to say this, of course, we have to be sure that
the way in which we do so does nothing to support the inference that if some
reasons are grounded in the agent’s beliefs, all are.
II
In this section, I consider a potential difficulty for the version of the objective view
that has begun to emerge. This is that it fails to address the tension between two
questions:
1. What ought he to do, given the facts?—the objective question.

2. What ought he to do, given his beliefs?—the subjective question.
The second of these surely appeals to the situation as the agent conceives it, not to
the fact that the agent so conceives. If so, then though the subjective question
makes manifestly good sense, the objective view seems to lack the resources to
address it. Indeed, that view announces that no obligations depend on the agent’s
perspective in the sort of way that the subjective question seems to suppose.
There is a tension between the two questions because the answers to them can
differ; indeed, they can even be opposed. When they are opposed, we want some
account of how to deal with the situation. Could both answers be true? If so, how?
If not, why not?
       329
One way of addressing the tension is to announce that there are two senses of
‘ought’—a subjective sense and an objective sense. This appeal to different senses
in order to resolve a philosophical tension was a common manoeuvre among the
intuitionists; we find it in Ewing, in Joseph, in Broad and in Ross, in one form or
another.⁷ But I find it deeply unsatisfying. One way in which it fails to satisfy is
that it fails to retain any sense of tension or opposition between the answers to the
two questions at all. Of course we gain the advantage of being able to say that
opposing answers to the two questions could both be true. But the loss of any
sense of tension is too high a cost to pay for this.
A second and more popular way of addressing the tension is to suppose that a
univocal ‘ought’ is used to express two distinct sets of constraints. If the ‘ought’ is a
moral one, the agent on this account has two complete sets of moral reasons—or,
better, of moral constraints. First, there are those deriving from the situation.
Second there are those deriving from the agent’s perspective—from the agent’s
take on things. The term ‘rational’ is often used to mark constraints of the second,
‘internal’ kind, in the analogous debate within the theory of ordinary practical
rationality. Derek Parfit writes that there are two questions: what there is most
reason for me to do and what it would be most rational for me to do (1997: 99).
Parfit was thinking about the relation between two questions about ordinary, non-
moral reasons for action. But presumably the same distinction should be available
in the moral case. If so, we have to say something about the relation between these
two questions. But it seems to me hard to suppose that there are just two sets of
moral constraints covering, as it were, the same territory, each with its own
ground. In fact I think that this idea is untenable. The problem, as I see it, lies
in the difficulty of putting together the prescriptions of each. How are we to assess
against each other the claims of objective and subjective duty? This will be no
problem for the agent, of course, since the agent is unable to distinguish in other
than an abstract and general way between how he takes things to be and how they
in fact are. But we can distinguish between these things, as can the agent in
retrospect. And I just don’t see how to put together the demand that he do it,
grounded in the nature of the situation, and the demand that he not do it,
grounded in his beliefs. This is not the normal case where there are reasons of
the same general type in favour and against. There would be no special difficulty if
it were. It is a most peculiar situation altogether.
There is a ‘refined’ version of the approach we are currently considering, which
announces that the two questions are concerned with different objects. The
objective question concerns which action ought to be done. The subjective
question concerns what the agent ought to do. This looks as if it should be some
improvement on the unrefined version above, since we no longer have two sets of
⁷ See Ewing (1947: 120f.; 1953: 145). See also Joseph (1931: 61, 104); Broad (1930: 162–3); Ross
(1939: 46–7).
330    
moral constraints covering the same territory. But I can make no sense of the sort
of distinction between action and agent that would be required to make the
approach work.⁸ The idea, I suppose, is that if there are two distinct objects
here, action and agent, remarks about one are hardly capable of conflicting with
remarks about the other. But agent and action do not seem to be separable objects
in the kind of way that this seems to require. And even if they were, being told that
the action I ought to do is one thing (a remark about the action) and the way in
which I ought to act is another (this is the remark about the agent) seems more or
less incoherent to me. The tension between the answers to our two questions has
been garbled by the attempt to understand it in terms of some metaphysical
distinction between action and agent.
There remain, however, two further moves we can make in trying to under-
stand the tension. The first of these involves an appeal to the distinction between
the deontic and the evaluative. Deontic concepts are such as ‘ought’, ‘right,
‘obligation’, duty’, wrong’, ‘permissible’, ‘required’. Evaluative concepts are such
as ‘good’, ‘bad’ and maybe ‘evil’. But the point here is not so much this sort of
distinction between concepts as an analogous distinction between two types of
thought. The objective question is answered by saying what the agent ought to do,
or, equivalently, what action the agent should take. The subjective question
addresses itself to a different issue: how well should we think of the agent for
acting as she did? The first question is a deontic one, and the second an evaluative
one. Our evaluative answer to this second question need not interfere with our
deontic answer to the first. It may be that we are not to think too badly of the agent
for having acted as she did, even though in fact she did not do what she should
have done in the situation (i.e., she did not do what the situation in fact called for).
Such a case might be one where the agent was misinformed, through no fault of
her own.
It is tempting, though mistaken, to marry this third strategy with the second,
supposing that evaluative thought applies properly to agents rather than to
actions, while deontic thought applies properly to actions. This manoeuvre
inherits all the problems of the second strategy, with some added ones of its
own. These last include the ban on evaluating actions and on asking ‘did he act
rightly?’.
But even if we don’t make this mistake, the third strategy faces difficulties. The
main difficulty is that the two questions seem to be addressed to the same deontic
issue, namely what should the agent do. (They differ on what is allowed to count
as deciding that issue.) The attempt to understand the subjective question as
evaluative rather than deontic really involves the suggestion that its deontic
form is misleading. In this sense, the idea is that there are really no ‘subjective
⁸ I complained about this sort of use of a distinction between action and agent in Moral Reasons
(1993: 13).
       331
oughts’—no deontic question that is somehow relative to the agent’s conception of

the situation. This seems to me to be more a dissolution than a resolution of the
problem.
There is however a fourth possibility that does not make this mistake.
I presented matters as if the objective question runs in terms of a supposed
relation between features of the situation and action, while the subjective question
runs in terms of a different relation between beliefs and actions. But it seems to me
possible to understand answers to the supposedly subjective question in a different
way, seeing them as specifying an objective constraint on a combination of belief
and action rather than as specifying a subjective constraint on action, grounded in
the agent’s beliefs. That there are objective constraints on combinations of this
sort seems to me to be shown by thinking about what we might call ‘hypocrisy’,
namely the ban on thinking that others should not do this sort of thing while
doing it oneself. My view about this is that this ban is genuinely on a combination,
and that it does not permit one to detach a ban on one element once the other is in
place. What we have here, then, is a ban on believing that others should not do it
while doing it yourself, which does not convert (or ‘detach’) into a direct ban on
the belief, given the action, nor into a direct ban on the action, given the belief.
Doing it yourself does not make believing that others should not act likewise
wrong, any more than that belief makes the action wrong. There is nothing wrong
with the belief, nor with the action; what is wrong is having both at once. So we
can hear the phrase ‘believing this, you should do that’ as meaning ‘you should not
both believe this and fail to do that’. But this ‘should’ is an objective one, just the
same sort of ‘should’ as is at issue with the first question.
The idea driving this approach is that it is much easier to hold together simple
objective constraints on actions and complex objective constraints on combin-
ations of beliefs and actions than it is to hold together two distinct types of
constraint, each with its own sort of ground. We get a set of questions, all using
the same sense of ‘ought’, to which our answers may differ. But when they do
differ, this is just another instance of the familiar situation in which there are
different demands on us not all of which we can satisfy at the same time. That is
quite different from the sort of situation generated by the first or second
approaches, where the oughts we are putting up against each other are of quite
different sorts.⁹
I find this way of reading the subjective question persuasive in its own right,
then, and this leaves me asking whether I should add the third approach to the
fourth one, or whether the fourth one should be thought of as adequate all on its
own. I think that the answer to this is that we do need to add use of the deontic/
evaluative distinction to the structural distinction between ‘If he believes this, he
⁹ For further discussion of this issue, see my Practical Reality (2000a: 3).
332    
ought to act thus’ and ‘He ought, if he believes this, to act thus’. For something not
too damning should be able to be said about the agent who acts in good faith and
with good intentions, in a case where that agent is misinformed through no fault
of her own. Just saying that she did the wrong thing seems inadequate. We want to
be able to find something to say, not so as to diminish the sense that what she did
was wrong, but so as to diminish our condemnation of her for having acted
wrongly. Of course the complex objective requirement is one that she did satisfy,
and that is already something. But she might satisfy that requirement even if she
ought not to have believed as she did. Among those who fail the simple require-
ment, then, there are better and worse. The worse is the one who merely satisfies
the complex objective requirement. The better might be one who both satisfies
that requirement and believes irreproachably even though mistakenly. Deontic
conceptions give out at this point, and we need the evaluative if we are to draw the
relevant distinctions.
III
How does Prichard purport to resolve the tension between objective and subject-
ive questions? Officially, he abandons the objective question entirely, so far as
obligation is concerned. There is no objective question about what one ought to
do. Interestingly, however, there is for Prichard a perfectly good objective question
about what it is right to do. In fact, Prichard would read the objective question as
about what it is right to do and the subjective question as about what one ought to
do. The first of these is determined by looking at the situation, and the second by
looking at features of the agent’s thought.
He makes this position possible for himself by arguing that though rightness is
a property of the action, there is no such property as ‘ought-to-be-doneness’.
Obligations belong to agents, not to actions.¹⁰ There is a property of ‘being
obliged’, and it belongs to agents. There is no property of obligatoriness.
Prichard takes this idea to give him the answer to his original question. The
idea is that since obligations are features of ourselves rather than of our actions,
there is no bar on those obligations being grounded in features of our thought
about the situation. This is important because it causes our prejudices against the
subjective view to subside. Those prejudices largely depended on the thought that
our obligations must depend on the fact that our action would have a certain
character, not on our thinking it likely that it would. But once we understand that
obligatoriness, to the extent that there is such a thing at all, is a feature of us rather
¹⁰ This doctrine had already been propounded by Joseph (1931: 61–2).

       333
than of our action, Prichard thinks that our resistance to the subjective view will
subside.
He might be right about this, but he should not be. For the question whether
obligatoriness belongs to the agent or to the action is irrelevant to the battle
between the objective and the subjective view. Prichard himself came to see this
later, as he confesses in the notes that are printed at the end of the later version of
‘Duty and Ignorance of Fact’, and which I consider in the Appendix to this paper.
The point is that even if we allow that obligation depends on features of the agent
rather than on those of the action, we still have to choose between the question
whether those obligations depend on features that the agent actually has or merely
those that he takes himself to have. Locating obligation inside the agent leaves this
objective/subjective choice still to be made. So Prichard’s supposed resolution of
the question with which he started is irrelevant.
Why then was he so impressed with it? The answer lies in his idiosyncratic
reason for supposing that duties must be objective. In a revealing passage, he
writes:
[The subjective view] seems impossible. For an obligation to do some action

seems to be a character of that action; and therefore, it would seem, it must
depend on the fact that the action would have a certain character if it were done,
and not on our thinking it likely that it would. (2002: 99)
The point, of course, is that this metaphysical reason for preferring the objective
view if we can get it is not likely to be our reason. We are likely to think that
assessing obligation entirely through the perspective of the agent is likely to let the
immoral off the hook—a sort of thought that seems to have been entirely foreign
to Prichard, and one that would not be addressed by his ‘discovery’ that obliga-
toriness is properly a feature of the agent rather than of the action. It turns out that
when we begin to read Prichard’s paper and come across his original question on
the first page, we misunderstand its purport. For Prichard, it is a question driven
by metaphysical considerations; for us, by moral ones.
The view that Prichard has now reached is that the objective question, so far as
it makes sense at all, is concerned with which action is right, while the subjective
question is concerned with which is the action that the agent ought to do. But this
could hardly count as a resolution of the tension between the objective and the
subjective question. The difficulty faced by an agent who is told that one action is
right but another is the one that he ought to do has simply not been addressed,
indeed cannot be addressed, and this is surely a reliable sign that something has
gone wrong in the theory. But Prichard’s view seems to me to cause considerable
trouble for the relation between the right and the obligatory in a further and more
general way. For on his own showing rightness remains a feature of the action, and
hence incapable of being grounded in features of our thought, while obligatoriness
334    
is entirely grounded in features of our thought. But whatever the relation between
what is right and what we ought to do, it surely cannot be the case that the grounds
for rightness are utterly distinct from the grounds for obligation, as on the present
account they would have to be. They would have to be distinct because, as
Prichard sees it,
while the truth could not be expressed by saying: ‘My setting myself to do so-and-
so would be right, because I think that it would have a certain effect’—a statement
which would be as vicious in principle as the statement ‘Doing so-and-so would
be right because I think it would be right’—there is nothing to prevent its being
expressible in the form ‘I ought to set myself to do so-and-so, because I think that
it would have a certain effect’. (2002: 99–100; Prichard’s emphasis)
This quotation is the evidence that Prichard retains the standard intuitionist view
that rightness is an intrinsic feature of actions, while he has abandoned the
attempt to define ‘ought’ in terms of ‘right’ or vice versa. This makes it all the
stranger that Ross, in his slavish account of Prichard’s article, never once draws
any distinction between rightness and obligatoriness. (This is one of the two errors
I referred to in my Introduction; the other can be found above in fn. 5.)¹¹
What is more, there are various possible accounts of the relation between
rightness and obligation other than the one that Prichard eventually adopts, and
which do not have the awkward consequence of making the ground for rightness
necessarily distinct from the ground for obligation. Prichard’s view is, effectively,
that one has an obligation to do that which would be right if the world were as one
supposes, even if it is not in fact right. But we might say instead that the ground for
obligation includes the ground for rightness, in the sense that rightness plus
relevant features of the agent, e.g. opportunity, together ground obligation for
that agent. Or we might say that the ground for rightness includes the ground for
obligation, in the sense that obligation is grounded only in accessible right-making
features. This last view has it that some right-making features, i.e. the inaccessible
part of the ground for rightness, cannot stand as part of the ground for obligation.
The former view has it that obligation is grounded in agent-relative features,
among others, but that the sort of agent-relativity concerned is not accessibility.
Finally, there could be a combined view, which took obligatoriness to be grounded
in accessible parts of the ground for rightness plus other relevant features of the
agent. My main point in listing these alternative views is that none of them has the
awkward consequence that Prichard’s view has for the relation between rightness
¹¹ Of course it would be far better if Prichard had given the same general treatment of rightness as he
does here of obligation, and elsewhere in his corpus (largely unpublished) he does indeed seem to take
that more defensible position. But the sentence quoted above from p. 99–100 seems conclusive
evidence that here, at least, he does not take it.
       335
and obligation. Of course these views still have to face the difficulty that the agent
may be told that one action is right and another is the one that he ought to do.
This is the difficulty that I tried to approach in the previous section by under-
standing the subjective question in terms of objective requirements on combin-
ations of beliefs and actions.
There is a further potential difficulty with Prichard’s view, and that is that the
argument that forces us to abandon ought-to-be-doneness may force us to
abandon rightness, as a feature of the action, as well. I mention this worry without
being sure that it is a genuine one. But consider Prichard’s actual argument for
taking obligatoriness to be a feature of the agent rather than of the action: ‘For
since the existence of an obligation to do some action cannot possibly depend on
actual performance of the action, the obligation cannot itself be a property which
the action would have, if it were done’ (2002: 99). If I fail to do the right thing,
there is no right action which I have failed to do. So we can say ‘If I had done that
I would have done a right action’, but we cannot say that what I did was not the
right action in the situation. In this sense, nobody ever does a wrong action.
Perhaps this last point is wrong. My main suggestion, however, has been that,
though there is a significant matter at issue, namely how we are to understand the
relation between the objective and the subjective question, Prichard’s official
resolution of this question leaves matters worse than it found them, quite apart
from being irrelevant. This is harsh criticism, but what I really think is that
Prichard’s general understanding of the issue was defective. He felt that he had
to plump entirely either for the subjective or for the objective understanding of
what is going on in this area. My own position is that there are ways of finessing
the matter so as to retain traces of both views, even if the overall style of the
eventual mixture will be objectivist.
Appendix: on Prichard’s recantation

I consider here Prichard’s later attempt to solve what he sees as the main difficulty for the
subjective view. This matter is confusing (to me at least). Prichard’s first idea, the one in the
original lecture, is that obligations are characters of the agent rather than of the action (or,
rather, of the activity) and so our obligations must depend on some fact about ourselves.
This, as he later acknowledges, and as we saw above, is irrelevant to the issue. For it remains
an open question whether those facts need to be believed by us, and this was what was
supposed to be at issue between the objective and the subjective view.¹²
¹² Ross, in his exposition of Prichard’s paper, ends by claiming that ‘even if we think that there is a
character of rightness that attaches to an activity, it will, on the subjective view as now restated, be a
character which belongs to the activity not because of the activity’s being thought to have a certain
character, but because of its actually being of a certain character, the character of being the setting
oneself to bring about a certain effect. This character it actually has, and there is in principle no reason
336    
The later note appears first in Ross’s 1949 edition of Prichard’s writings, two years after
the latter’s death; it is simply added without explanation at the end of the original paper.
(McAdam conjectures that Prichard intended this note for publication, but I think it more
probable that Ross simply found it in the papers Prichard left after his death, and which are
now collected in the Library of Oriel College, Oxford.) The note offers a quite different
‘resolution’ of the objection to the subjective view that it makes our obligations ‘turn not on
the nature of the situation but on that of our thought about it’ (2002: 100). Consider the
objectivist claim:
The obligation to bring about X must depend on some character that bringing about
X would have, and not on our thinking that it would have that character.
Now substitute ‘willing X’ for ‘bringing about X’, as Prichard suggests in line with his earlier
conclusion that when we act, our activity really consists in setting ourselves, or willing, to
effect a certain change. This yields:
The obligation to will X must depend on some character that willing X would have, and
not on our thinking that it would have that character.
Prichard then says that our thinking X likely to effect something else Y enters into the
character of that activity which we have an obligation to do, on this account. For the
thought that X is likely to produce Y is part of the willing of X. This thinking, then, enters
into the character of the willing, i.e., of that which is obligatory for us. It follows, then, that
the intrinsic nature of the obligation necessarily depends on some features of our thought.
The idea here, I think, is that the character that willing X would have, in virtue of which we
have an obligation to do it, is (partly) that of being thought by us likely to produce Y.
What sort of a resolution is this? In the main text Prichard presents things as if we
should be persuaded that the subjective view is true and the objective view therefore false. Is
he still sticking to that picture? I think he is now suggesting that the contrast between the
objective and the subjective view is a false contrast, for the character that willing X would
have, in virtue of which we are obliged to do it, is that of being thought by us to have a
certain character.
But if so, the suggestion is a failure. If the character that willing X would have, in virtue of
which we ought to do it, is our thinking it to have some character, the character which
grounds the obligation is not the character which we think that willing X would have, but a
character which it has, namely our thinking it to have some other character. And if this is
right, what we emerge with is the objective view in a rather peculiar form, rather than the
subjective view. And the contrast between the two views remains in force.
I think that what has created the confusion here is the apparently tiny difference between
two versions of the contrast between the objective and the subjective view:
1st (better) version: the obligation to will X must depend on some character that willing
X would have, and not on our thinking that it would have that character. (cf. 2002: 92,
also referred to on p. 100)
why it should not be the ground of a further character of rightness’ (1939: 156). But this seems to be an
expression of the objective view, not of the subjective view that Ross is trying to support.
       337
2nd (worse) version: the obligation to will X must depend on some character that willing
X would have, and not on our thinking that it would have some character.
In terms of the second version, Prichard’s resolution amounts to a deconstruction of the

issue. In terms of the first, it does not. And it is the first that we really want. For after all
the question, as I take it, is whether the rightness of an action is grounded in features of the
situation, including the agent—perhaps only those that pass the sort of epistemic filter that
we refer to with the phrase ‘available evidence’—or whether the obligation is grounded in
features of the situation as the agent takes it to be. The idea that so perplexes Prichard, that
it is abhorrent that the obligation to do an action should be (partly?) grounded in features of
our thought, is only disturbing if the phrase ‘features of our thought’ is taken in the latter
sense, as ‘features of the situation as we maybe falsely suppose it to be’. I have argued that,
taken in that sense, Prichard provides no effective argument that obligation is subjectively
rather than objectively grounded.
There is an explanation of what has gone wrong here. I argued earlier that Prichard
misdiagnoses the ground of the apparent impossibility that obligation be grounded in
features of the agent’s thought. It is this misdiagnosis that leads him to see no difference
between the better and the worse version of the initial contrast. He is really only interested
in the question whether the obligation is a character of the action or of the agent (2002: 99).
The idea that causes the trouble is the idea that obligatoriness is (at least capable of being)
an intrinsic feature of the action, and therefore cannot derive, even partly, from features of
other objects. Once Prichard has seen this idea off, he doesn’t care which features of the
agent’s thought might be the ground for obligation. His misunderstanding of the problem
leads him to be blind to the defects of his own resolution.
21
Moore’s Account of Vindictive
Punishment: A Test Case for
Theories of Organic Unities
The final chapter of G. E. Moore’s Principia Ethica contains a fascinating discus-

sion of what Moore calls ‘vindictive punishment’.¹ Moore’s views are worth
extended treatment in their own right. But theorists of punishment have not
given them the attention they deserve, and I can hardly claim that I will be
doing so in this paper. My purpose is rather to treat punishment as a sort of test
case in a debate about the right way to understand organic unities. An organic
unity is a whole, or complex, whose value is not, or at least need not be, identical to
the sum of the values brought to it by its parts. An obvious example is that of a
recipe which seems to be much improved by the addition of a little salt, supposing
the value of the pinch of salt to be far less than the value that it adds or brings to
the recipe. The idea of the value brought to the whole by a part is important,
because it seems possible for a part to have value which it does not contribute to
the whole—which it hugs to itself, as it were. An example here might be a
charming brooch which clashes with the blouse to which it is attached, so that
the value of the whole outfit is less that the value of either blouse or brooch. If this
is the right way to describe the situation, it is easy for a whole to have less value
than the sum of the values of its parts, because some of the latter value may be
withheld, and not passed upwards. But this possibility is not what the friends of
the organic were primarily thinking of. What impressed them was the idea that
even if one only considers the value that the parts bring to the whole, or pass
upwards, the whole may have more value than comes to it from its parts—or less.
There is a debate about the right way to understand organic unities. There are
two views, which have come to be called ‘intrinsicalism’ and ‘variabilism’. Moore
is an intrinsicalist; I am a variabilist. Now it would be nice to begin my discussion
of the debate between these two positions with an example of each, to fix ideas. But
¹ The two extant editions of Principia Ethica, the original edition (1903) and the revised edition
(1993), have different pagination, which is inconvenient. In the references below, I give the relevant
section number, followed by page references to both editions, thus: PE§1, 1/53.
 ’      339
since each position denies the coherence of the other, this cannot be done without
begging central questions. So I will have to proceed with a rather abstract account
of the two views.
Intrinsicalists suppose that parts retain their intrinsic value as they move from
whole to whole; this is not terribly contentious—many would say as much. But
they also suppose that even if the parts do not acquire any new value as they enter
a new whole, still the value of the whole can be increased by their presence more
than can be explained by simple appeal to the value that the parts have there. In
this sense, parts can contribute value that they have not got. In fact, a whole can be
the better for the presence of a part even if the part itself has no value at all, either
there or anywhere else. The salt adds more value to the recipe than it itself has in
that context.
Now in fact Moore held that there are only two sorts of value: intrinsic value
and instrumental value. Thinking this, and thinking that intrinsic value cannot be
affected by changes in context, and thinking that the relation between part and
whole is not instrumental, Moore concluded, very reasonably, that parts cannot
change in value as they move from whole to whole, because that sort of change
could change neither instrumental nor intrinsic value. So when we are trying to
understand the part–whole relationship and organic unities, everything has to be
done in terms of intrinsic value. I say that this was reasonable; but in fact Moore
went a bit further, for he held that intrinsic value is essential value. He writes
elsewhere:
[I]t is impossible for what is strictly one and the same thing to possess that kind
of value at one time, or in one set of circumstances, and not to possess it at
another; and equally impossible for it to possess it in one degree at one time, or in
one set of circumstances, and to possess it in a different degree at another, or in a
different set. (1922: 260–1)
This is not quite so reasonable. The most natural way of understanding intrinsic
value is as that part of the value of an object which is grounded in intrinsic
properties of that object—in what Moore calls the object’s ‘intrinsic nature’ (see,
e.g., 1922: 274). But intrinsic properties include the object’s particular size and
shape, it would seem, and these are surely capable of change. If so, there is no
reason to deny that an object’s intrinsic value can change; what it can’t do, we
might say (though I won’t in fact say this), is change when the intrinsic properties
that are its ground remain constant. But Moore, as the quotation above shows,
seems to go further than that, and in going further, he goes too far.
Variabilists take the view that parts can change in value as they move from
whole to whole. Most variabilists hold this view because they suppose that there
are more sorts of value than the two that Moore was willing to allow. They think
that, in addition to the intrinsic and the instrumental, there are other sorts of
340    
value, which can be called ‘extrinsic’. It is best, I think, to classify the instrumental
as one sort of extrinsic value.² But there are other sorts. The most obvious are
symbolic and sentimental value, which, as far as I can see, are neither instrumental
nor intrinsic. The question then is whether, in addition to extrinsic symbolic and
sentimental value, there is another sort of extrinsic value, ‘value-as-a-part’. If there
is, I can see no reason why such value should not change as the bearer moves from
whole to whole. This gives us constant intrinsic value, supposedly, and variable
extrinsic value. With this weaponry, we are less likely to say that parts can
contribute value that they have not got; for we now have the option of supposing
that all the value that a part contributes is value that it itself has in that whole, even
though it won’t necessarily have that same value elsewhere. Any difference
between the value contributed to one whole and to another will be explained by
appeal to differences in the extrinsic values enjoyed by that part in those two
wholes; that is, by appeal to the differences between its value as a part of one whole
and its value as a part of the other.
Before entering into the substantial debate between intrinsicalism and variabi-
lism, I want to make a couple of comments. The first is that all this talk of parts
and wholes, and of a part contributing value to a whole, is pretty peculiar. But does
its peculiarity vitiate the debate? I don’t think so. However we express the matter,
we think that the value of an object is to be explained by appeal to ‘features’ of that
object, or of the situation in which it stands. These ‘features’ are the characteristics
which make it good, and this ‘making-good’ is something that each of several
features can do at the same time. By this, we don’t mean that each severally
succeeds in making the object good overall. We mean, rather, that each contrib-
utes to the value of the object, or, as we might say, raises the object’s value. This
talk of contribution can be understood less metaphorically in terms of explan-
ation, but there are two ways of doing that. As the intrinsicalist sees things, if the
value of the object could be explained entirely in terms of the values of the various
features, there would be no organicity. But there is organicity, because there are
cases where the value of the object is to some extent explained in terms of the
presence of various features but not in terms of their value, either because they
have no value here (or not enough) or because we do not appeal to their value in
our explanation of the value of the object. What is more, there are no cases where
the value of a part changes as it moves from one whole to another. As variabilists
see things, by contrast, there are no organic wholes of the intrinsicalist’s sort at all.
The only possibility is that, though the explanation of the value of the whole is run
² This suggestion would be disputed by those who hold, with Ross and Prichard, that instrumental
value is no sort of value at all: it is possible to think that to be instrumentally valuable is just to be a
means to something that is of value, without allowing that to be such a means is a way of being valuable.
See Ross (1930: 133).
 ’      341
entirely in terms of the values of the parts, those parts need not have the same
value here as they have elsewhere.
My second comment is that, though the debate about organicity has always
been conducted within the theory of value, so far as I know, it is in fact of wider
significance. It is a debate about the possibility of certain structures, and as such it
has repercussions elsewhere in philosophy. To take the most glaring instance:
there is strong similarity between the present debate, which in the most general
terms can be thought of as a debate about practical purport, and a familiar debate
in the theory of meaning about semantic purport. Must the meaning of the whole
be entirely explained in terms of the semantic contributions of the contributing
parts? Can such parts ‘change’ their meaning as they move from context to
context? Can they contribute meaning that they have not got? Do we need a
distinction between variant (extrinsic) and invariant (intrinsic) meaning, and
must we suppose that in the absence of invariant meaning, there is no meaning
at all? These questions in the theory of meaning may seem peculiar, but I take
them to be strongly analogous to the questions we shall be discussing in the theory
of value.
II
So far I have suggested that there are two approaches to organic unities: the
intrinsicalist and the variabilist. But this does not by itself lead to any substantial
debate. If the question is whether we should understand the relevant phenomena
in terms of changes in the values of parts as they move from whole to whole, or in
terms of a difference between the sum of the values of the parts in the whole and
the overall value of the whole, the answer might be that we need both conceptions.
Some wholes are to be understood in one way, and others in the other. But the fact
is that neither side in the debate is willing to be so conciliatory. Each side comes
with an argument, and the two arguments are designed to show not just that there
are no instances of the opposing conception, but that there could be none, because
that conception is incoherent. In this section, I consider the arguments on
either side.
Moore’s argument is as follows:
Thus we may easily come to say that, as a part of the body [the arm] has great
value, whereas by itself it would have none . . . But in fact the value in question
obviously does not belong to it at all. To have value merely as a part is equivalent
to having no value at all, but merely being a part of that which has it. Owing,
however, to neglect of this distinction, the assertion that a part has value, as a
part, which it would otherwise not have, easily leads to the assumption that it is
also different, as a part, from what it would otherwise be; for it is, in fact, true that
342    
two things which have a different value must also differ in other respects. Hence
the assumption that one and the same thing, because it is a part of a more
valuable whole at one time than at another, therefore has more intrinsic value at
one time than at another, has encouraged the self-contradictory belief that one
and the same thing may be two different things, and that only in one of its forms
is it truly what it is. (PE§22, 35/86–7)
This is a most peculiar argument. The crucial claim is that there is no such thing as
value-as-a-part, in the third sentence. (This claim is structurally identical to the
Prichard/Ross claim that there is no such thing as instrumental value.) It may be
that this claim is not supposed to be supported by the rest of the quotation.³ But if
one is tempted to suppose that there is such a thing as extrinsic value, and that
value-as-a-part is of that sort, what is said here to persuade one otherwise? The
last sentence says that a contradiction follows from supposing that one and the
same thing can have different intrinsic value in different situations. But that is
irrelevant. To make it relevant, I suggest that we remove the word ‘intrinsic’ from
the last sentence, and replace it by ‘non-instrumental’. (It is obviously coherent to
suggest that something can vary in instrumental value from context to context.)
This should not damage Moore’s argument, because he thought that all value was
either intrinsic or instrumental. The question is whether this passage provides
independent support for that view. I take it that, to the extent that there is any
such support, it comes from the claim about supervenience. If you suppose that
something which is a (contributory) part of a valuable whole therefore has
something worth calling ‘value-as-a-part’, you end up with a contradiction,
because two things that have different value must differ in other respects. But
the whole idea was that the part does not change in any way as it moves from
whole to whole; and if not, it cannot change in value either. If it changes in value, it
necessarily ceases to be what it was.
But is it true that two things that have different non-instrumental value must
differ in other respects? No. For all we have yet been told, it is possible for changes
elsewhere to affect something’s value without altering that thing in any other way.
I have argued elsewhere that Moore has adopted too limited a conception of
supervenience. He assumes that an object’s value must supervene on the other
properties of that object; but there is a wider conception of supervenience under
which an object’s value supervenes simply on other properties, which may or may
not be properties of that object. If this wider conception is defensible, as I suppose,
³ The next sentence is a bit ambiguous in tone, because the phrase ‘easily leads’ might be supposed to
imply that the relevant inference is unsound. On that reading, Moore would be rehearsing an objection
which he himself would not accept. But Moore’s general practice in PE tells against this reading. He
always speaks of the whole being better for the presence of the part, and never suggests that the part
might be more valuable here than it is elsewhere. See §§18, 55, 57, 117, 119, 121, 122, at pp. 29/80, 92/
144, 194/243, 197/245–6, 202/250–1, 203/252.
 ’      343
it is not true that ‘two things which have a different value must also differ in other
respects’. And with that, Moore’s argument collapses. This leaves us with the three
or four potential sorts of extrinsic value that I mooted in the previous section:
value-as-a-part, symbolic and sentimental value, and instrumental value.
So Moore’s attempt to destabilise the opposition fails. But what about the
argument in the other direction, the variabilist’s attempt to show that the intrin-
sicalist picture of the organic is incoherent? This argument focuses on the idea that
a part can contribute value that it has not got. (It allows that a part can fail to
contribute value that it has got.) Moore’s view is that the presence of the part can
be important for the value of the whole, vital even, though the part itself remains
totally valueless. But this seems to fly in the face of what one might call the
essential link between values and reasons. It does not matter for the point exactly
how one expresses, or tries to capture, that link. In rough terms, the idea is that
where there is value there are reasons of certain sorts, and vice versa. If something
is of value, one will have reasons to preserve, promote, admire, approve, respect,
or honour it; and if one has such reasons, it must be that the relevant object has
some value, or is valuable in some respect. I have expressed this rather vaguely,
intentionally. There are all sorts of different views about the nature and the
tightness of the link between values and reasons, right up to the ‘buck-passing’
view which understands an object’s being valuable just as its having features that
give us reasons to protect, promote . . . it. We should also bear in mind various
possible exceptions to the rule that values and reasons are essentially connected.
But none of these niggles does anything to undermine the existence of a general tie
between values and reasons; or, if they do, they do not address the case we are
here considering, namely the intrinsicalist suggestion that we might have consid-
erable reason to protect here a feature which itself has no value at all here, because
of the value which its presence generates in the whole of which it is a part.
Whether we say, with the buck-passers, that for us to have those reasons is just
what it is for the feature to be of value, or hold the weaker view that it is incoherent
to deny all value to something which is, in the context, so important, we get the
same result. Something which it is important to protect in a certain context must
be of value in that context. And this rules out the possibility that a part can
contribute value which it has not got.
III
In all this I have been allowing something that I now wish to dispute. This is the
idea, common to intrinsicalists and variabilists, for all I have said so far, that the
intrinsic value of an object must remain unchanged so long as the intrinsic qualities
that are its ground do not change. Intrinsic value, we have been supposing, is to this
extent persistent. So to find such value we have to find something whose value is
344    
completely insensitive to changes in other things, unless those changes lead to

changes in the object’s own intrinsic nature. In this section, I want to suggest that
this picture is at best optional. There is a conception of intrinsic value which allows
such value to vary even when its intrinsic ground remains invariant.
This may come as something of a surprise, but actually I think it should not.
The picture that I am disputing results from unwarily supposing that the only
thing relevant to the generation of value, as one might put it, is the ground for that
value. But it might be that though we hold that ground unchanged, changes
elsewhere will suffice to undermine the ability of the ground to generate the
relevant value, so that the ground will remain but the value vanish. If so, we will
get something worth calling intrinsic value, because its ground consists entirely in
intrinsic features of the object, but which may vary independently of those
features. (Only in the sense that we can retain those features but lose the value,
of course.) There will be variable intrinsic value.
How is this possible? As I see it, the grounding relation itself is one that is
sensitive to potential changes in the surroundings. There can be conditions
necessary for the grounds to do their grounding work, which are not themselves
among the grounds. These features, which I call enabling conditions, are condi-
tions in the absence of which the grounding features still remain, but whose
presence is required for the grounding, that is, for the generation of the value, to
occur. Take the enabling conditions away and the value vanishes, even though the
enablers themselves are not part of the ground; they are doing another job. The
matter is a bit like (but not entirely like, because of the difference between causal
and non-causal relations) the relation between the actors on the stage and the
lighting engineers and the other support staff in a theatre. The activities of the
support staff are required for the actors to play their role; but this does nothing to
turn the support staff into members of the cast. The same structure is to be found
in the theory of value (and, I would say, everywhere where there is normativity).
There are two roles: the grounding role and the enabling role. It is one thing to
play a certain role, and another to enable something else to play that role. But,
given this distinction, we immediately derive a variabilist conception of intrinsic
value. To guarantee the persistence of a value, we need not only the persistence of
the ground but also that of the relevant enablers. But the enablers themselves may
be features of other objects. If so, an object that is intrinsically valuable may lose all
value as a result of changes in other objects, even though its own intrinsic
properties remain unchanged.
My description of this possibility has so far been entirely abstract. The best
example I know, which I owe largely to Frances Kamm, is that of a joke about
someone, where the value of the joke (its humour) vanishes if the joke is told
behind that person’s back. That the butt of the joke is present is required for the
joke to have the value it does, but it does not itself contribute to the humorousness;
it is not among the funny things.
 ’      345
This gives us a conception of intrinsic but variant value. I am mentioning it here

only for the sake of completeness. At most we should say of it what Ross says of
the doctrine of organic unities itself: ‘The importance of the doctrine so far as its
application goes is somewhat doubtful. But its truth in the abstract seems unques-
tionable’ (1930: 72).
Of course, there may remain some objects that have a very special sort of
intrinsic value that is genuinely invariant, so long as their intrinsic properties
remain unchanged. These will be ones such that the enabling conditions for the
relevant value, if any, consist in further intrinsic properties of the value-bearer
concerned. Perhaps there are, indeed, such specially valuable objects. My point is
only that there is no need, so far as the structure of value goes, for any such things.
It might be that there are plenty of things of value, even of intrinsic value, while
this transcendental category remains entirely empty.
One interesting aspect of the distinction between enablers and value enabled is
that the enablers themselves will have value, because of the link between values
and reasons. There is reason to protect something whose presence is required for
the existence of a value. So an enabler for a value will itself be of value. This doesn’t
seem to me to introduce any theoretical problems. It is not as if something in the
theory of organicity is thereby destabilized, for the value of any enabler will not
itself be contributed to the value of the whole. It is of value in the circumstances,
but it does not pass its value upwards to the whole, part of whose value it is
enabling. My present point is only that what one might call ‘value as an enabler’ is
another form of non-intrinsic value, to be put beside the instrumental, the
symbolic and the rest.
IV
After that methodological digression, I return to the main theme. We had two
conceptions of organicity, and two arguments. I purported to dispose of one of
those arguments, leaving the other one master of the field, and with it the
variabilist conception of organic unities. But there is an alternative view, promoted
by Thomas Hurka (2003) among others, which holds that we need both concep-
tions. On this view, both of the arguments must be flawed, but this does nothing to
destabilize the conceptions those arguments were designed to promote. All it
means is that the other conception, the one attacked, remains unimpugned. If
that were so, there would be two possibilities, not just one. And Hurka claims that
this is just as well, because some actual organic unities need to be understood in
one way, and others in the other. Why insist that only one way is possible? There is
no reason, once the two arguments have been seen off.
As an example that fits variabilism, I would offer a mosaic. A mosaic consists of
lots of little stones which were, before the prospect of the mosaic came along,
346    
utterly worthless (let us say). However, once placed in the mosaic, each one (like
the pieces of a jigsaw puzzle) acquires a new importance. It is difficult to get this
part of the story quite right, because we don’t want to award to each part more
than its share in the value of the whole; but I assume that there is some way of
sorting this out. Anyway, Hurka would have no trouble with any of this, and
indeed he offers a clutch of further examples to the same end. His point, however,
is that there are some cases which cannot be handled in the variabilist way. The
rest of the present paper is devoted to considering his main example, which is that
of vindictive punishment. The interest of this example, in the present context, is
that Moore offers in chapter 6 of Principia Ethica an extended intrinsicalist
treatment of punishment as a good which consists entirely of bad parts. The
idea is that variabilists, if they want to turn the same trick, will have to say that if
the whole is good the parts must be good, and that this is so counterintuitive as to
be impossible.
Suppose, for a moment, that we try to understand punishment in a standard
consequentialist manner but without appeal to organic unities. The crime itself is
bad, we may admit; and the punishment is also bad, since it amounts to an
increase of nastiness in the world. If we are to justify making things worse by
adding one bad thing to another, we will have to find some other respect in which,
by making them worse in this way, we are also making them better; and whatever
increase in value we find must be greater than the decrease involved in the
punishment itself. This is very much the picture we find in Bentham, for instance;
Bentham is clear that punishment is itself mischief, as he puts it. It is only worth
inflicting if by doing so we secure a greater good than the punishment is a harm.
Now this is not the only way in which a consequentialist might be thinking
about punishment. It is possible to think of the act of punishment as having
intrinsic value as well as instrumental value. So instead of saying that we will have
to find some way in which, in addition to making things worse, we also make them
better, we might look for some other feature of the act of punishing that makes
that act intrinsically, not instrumentally valuable. Consequentialists are, after all,
allowed to ascribe intrinsic value to actions as well as to consequences of those
actions; and they might seek to take advantage of that freedom in this case. The
problem, however, is to see what feature of the action might be capable of being
good enough to do the trick.
If we look at the matter in terms of organic unities, it all looks different. Moore
approaches it with a distinction between what is valuable or good on the whole
and what is valuable or good as a whole. For something to be valuable on the
whole is for the world to be better for its existence, all things considered. ‘[T]he
value which a thing possesses on the whole may be said to be equivalent to the sum
of the value which it possesses as a whole, together with the intrinsic values which
may belong to any of its parts’ (PE§129, 214/263). But a combination that is bad
on the whole may yet be valuable as a whole, and would be if ‘there arises from the
 ’      347
combination a positive good which is greater than the difference between the sum
of the two evils and the demerit of either singly’ (PE§129, 215/263). So a combin-
ation of two evils is good as a whole if the gain from putting the two parts together
is greater than the loss incurred by the addition of the second evil. But a
combination of two evils might still be such that the two parts are together
worse than the combination is better, so that the world is overall still the worse
for the combination, and that would mean that despite being good as a whole it
remains bad on the whole, not good.
A numerical, and therefore necessarily artificial, example might help. Suppose
we are dealing with units of goodness. The crime scores –5, and the punishment
scores –2; but the combination of crime and punishment together has an inde-
pendent score of +3 (this is its value as a whole). This means that when we punish
the crime we turn an overall score of –5 to one of –4. The situation is still bad on
the whole, because our overall score is a minus one, but it is less bad than it would
have been if we had left the crime unpunished. So we were right to punish, and
would have been wrong not to, even though in punishment we add evil to evil.
So in punishment, one evil is added to another in such a way, we hope, that the
combination is good as a whole; it would be good on the whole if the combination
were more valuable than the disvalue of the punishment. This being so, by adding
the punishment to the crime we would leave things better than they would have
been—though not better than if the crime had not been committed. Moore
concludes: ‘If pain be added to an evil state of either of our first two classes, the
whole thus formed is always better, as a whole, than if no pain had been there;
though here, if the pain be too intense, since that is a great evil, the state may not
be better on the whole. It is in this way that the theory of vindictive punishment
may be vindicated’ (PE§128, 214/263).
The idea then is that we can make things better by adding one evil to another.
Of course in making them better, all we are normally doing is making them less
bad. Crime punished is better than crime unpunished. But this does not normally
mean that a punished crime is overall a good, good on the whole; it is merely less
bad than an unpunished one. And if the punishment is excessive, even this would
not be true; the addition of excessive punishment will generate a combination that
is not even good as a whole, since the disvalue added by the punishment will
exceed the value added by the combination in the light of the appropriateness of
punishment to crime. In our numerical example above, if the punishment comes
in at –4, but the combination of crime and punishment remains at +3, the
punishment just makes things worse, since it turns an overall –5 into –6. This is
genuine mischief.
So what Moore ends up saying is that though punishment is an intrinsic evil, a
combination of crime and punishment may be intrinsically good as a whole. It
won’t be good on the whole (even if it is better on the whole than the unpunished
crime) since we would be right to prefer a world with neither crime nor
348    
punishment to one with crime and punishment. So our ranking is: best, neither
crime nor punishment; second best, both crime and punishment; worst, crime
alone. This is a justification of vindictive punishment that is not run extrinsically,
in terms of causal relations to intrinsically valuable consequences. That is, it is a
justification of punishment that is based on the appropriateness of punishment to
crime, not in terms of the hoped-for social or other consequences of inflicting the
punishment.
This is all rather clever. To see how clever, we should return to the difference
between Moore and the advanced consequentialist who wants to appeal to the fact
that actions can be among the intrinsic goods in order to avoid a merely instru-
mental justification of vindictive punishment. This consequentialist has to find
some feature of the punishing (rather than of the consequences of the punishing)
that makes it good, and good enough to counteract the badness of the pain
inflicted. Moore, by contrast, looks at the relation between punishment and
crime, and thereby succeeds in showing that a consequentialist can adopt a
conception of punishment which strongly resembles that of the retributivist. For
even if we do not allow an independent intrinsic value to the relation of fittingness
or appropriateness which obtains between punishment and crime, as the retribu-
tivist may seek to do, we are at least awarding value to a combination whose parts
fit each other in an appropriate way.
Having distinguished Moore’s views from those of a standard consequentialist,
we might find it worthwhile to contrast them also with those of Ross,⁴ because
there is a sense in which Moore sits in between standard consequentialism, on the
one hand, and intuitionism on the other, at least with respect to punishment. Ross
considers, but rejects, a possible intuitionist understanding of the duty of punish-
ment as an instance of the more general duty of producing as much good as we
can. The good in question would be ‘a certain relative arrangement of virtue, vice,
pleasure and pain’. But Ross objects that, first, the right arrangement for a
particular person would be not what is right in the limited context of that person’s
having committed a crime, but what is right considering the balance of their life as
a whole, which it is quite beyond our ability and that of the courts to determine.
And he adds to this the thought that it is not the business of the state to attempt to
create such a balance in the lives of individual transgressors. States should limit
themselves to protecting the most important rights of individuals transgressed
against. They cannot, therefore, be held to have a general prima facie duty to
punish the guilty, if that is supposed to be an instance of the more general prima
facie duty to do as much good as they can. For states do not have this latter duty;
their duty is only to ‘consult the general interest’ when they look for ways of
⁴ Ross’s views are given in the second appendix to chapter 2 of his The Right and the Good.
 ’      349
protecting the rights of the individual. We need a different account therefore, one
consistent with Ross’s overall theory of prima facie duties.
Ross’s own answer is that the state’s duty to punish is an instance of the duty to
keep one’s promises. The state, he claims, makes promises to the injured person and
his friends, and to society at large. To the injured person is promised compensation
and the satisfaction of knowing that the offender has not gone scot-free. To society
is promised both this satisfaction and a degree of protection against further offences.
This accounts for the prima facie duty to punish, and it does so in a way that,
supposedly, consequentialists cannot borrow, for they necessarily distort the prima
facie duty to keep one’s promises. The right to punish is established by the claim that
those who infringe the prima facie rights of others to life, liberty or property thereby
lose their own prima facie rights to these things, and the state has no such prima
facie duty to spare them as it does to spare the innocent.
Ross claims: ‘There is thus a distinction in kind which we all in fact recognize,
but which utilitarianism cannot admit, between the punishment of a person who
has invaded the rights of others and the infliction of pain or restraint on one who
has not’ (1930: 60). This does not mean that the state may never inflict pain or
restraint on those who have not invaded the rights of others. ‘The interests of
society may sometimes be so deeply involved as to make it right to punish an
innocent man “that the whole nation perish not” ’ (1930: 61). In this, Ross agrees
with the consequentialists, including Moore. It is in the explanation of it that he
differs. All agree that there is something very bad about punishing the innocent.
But when we ask why, the standard consequentialist has to scrape the barrel.
Moore talks about an absence of a certain organic unity, a certain failure of value,
and Ross talks about the infringement of a prima facie right. That prima facie right
is a form of protection of the individual that is absent from Moore’s account.
For present purposes, however, the question is to what extent Moore’s success tells
in favour of his account of organic unities. Is it that no such account is available to
the variabilist? If so, we have here a significant strike against variabilism’s claims
to be the correct way of understanding the organic. And that is how Hurka sees
things. He writes:
A pure variability view must say that in this case the pain is ‘transvalued’ from
bad to simply good, which implies that our emotional response to it should be
simple pleasure. But this is not right: the morally best response to deserved
suffering is somber, mixing satisfaction that justice is done with regret at the
infliction of pain. By making deserved pain still bad as pain, the [intrinsicalist]
view captures this important [truth]. (2003: 606–7)
350    
Hurka has two points here, I think. Suppose that we want to say that the pain
cannot contribute a value it has not got, and that all value belonging to the
whole must somehow be distributed among the contributing parts. It looks as if,
with the purpose of emerging with the result that the complex [crime +
punishment] is better than the crime alone, we will have to say that the pain
inflicted is good—when it plainly is not good. Indeed, its not being good is the
whole point. We want to inflict something bad on our punishee; if our analysis
turns the pain from bad to good, inflicting that pain would hardly be punish-
ment, but rather the donation of an unmerited good. No, Moore was right: we
need to find a way of having it that the addition of something bad can make
things better, without this having the result that the bad thing we have added is
somehow turned good in the process.
The second point concerns the appropriate moral attitude to the punishment of
crime, where that involves inflicting pain. Ordinarily we would suppose that a
mixed reaction is called for, involving, as Hurka says, satisfaction that justice is
done (which is a good) with regret at the pain inflicted. But we can only keep this
mixed attitude going if the pain inflicted is bad, not good.
How can the variabilist reply? Taking the second point first, we should
distinguish between the painfulness of the pain and the goodness or badness
of inflicting that pain. A mixed attitude to punishment is certainly called for;
unalloyed pleasure seems certainly to be out of place. But I suggest that the
mixed attitude can be cast rather differently from the way that Hurka imagines.
What we want is regret at the whole situation, but satisfaction that the proper
penalty is exacted. The painfulness of the punishment is not itself to be
regretted, since it is exactly what is called for in the situation. It is good (or at
least makes things less bad) that the punishee should suffer in this way, given
his crime. And all that this shows is that suffering is not always something to be
regretted. Nobody is suggesting that the pain itself is not nasty, of course; there
is no suggestion that the person receiving it should somehow welcome it. The
suggestion is that we should not be thinking of the pain as a necessary evil, but
rather as something appropriate to the situation, and without which things
would be worse rather than better.
What then for the more structural point, that we are in danger of turning the
pain inflicted into an unmerited good? Here I want to introduce a rather different
distinction. One thing lurking in the back of this debate is the idea that pain is an
intrinsic or essential evil, and that we should reject any theory that converts it into
something which is, on occasions, to be welcomed. The idea should remind us of
recent discussions of the prosperity of the wicked.⁵ Suppose we say that their
prospering is not a good at all, but an evil. In doing that, we destroy the very thing
⁵ See. e.g., Lemos (1994).

 ’      351
we were trying to understand. We started out thinking that prosperity is some-

thing the wicked do not deserve. But we lose that thought if we announce that
prosperity for them is not the good it would be for others. It can only be something
they do not deserve if it is still a good when they get it.
I suggest that we assess this argument by recasting its talk of good and bad in
terms of reasons. There are various theses that I would wish to preserve in any
eventual analysis. First, we have sufficient reason to prevent the wicked from
prospering and no reason to promote their prosperity, and so the prosperity of
the wicked is not a good (because of the link between reasons and values).
Second, prosperity is ‘unfitting’ to the wicked, which is to say that it is wrong, or
bad, that they should prosper (by definition of what it is for something to be in
this sense unfitting). Third, that it is not good that the wicked should prosper
does not show that their prosperity is not a good for them. So, fourth, it is
possible that though their prosperity is a good for them, it is not good that they
should prosper. This conclusion fits the pattern of the reasons. For they may
have sufficient reason to preserve their prosperity, though we have no reason to
do so, having indeed sufficient reason to prevent them from prospering—which
is where we started.
We cannot hope to argue here that prosperity is not a good for the wicked. For
we can define what is to count as prospering for the purposes of the argument
simply in terms of something that is allowed as a good for the wicked. There must
be some such thing, e.g. health or comfort. Health is a more tricky one, because it
is not clear that we have reason to preserve the health of the wicked. Let us allow,
then, that comfort is a good for the wicked, so that they prosper to the extent that
they are comfortable. We are saying that their comfort is something which they
may have sufficient reason to preserve, though we have no reason to do so; indeed,
we have sufficient reason to prevent it where we can. In terms of value, I conclude
that the prosperity of the wicked is a good for them, but not good. It is not good
that they should have this thing that is a good for them.
Mutatis mutandis, I would say that the pain of the punishee is a good,
something we have reason to welcome, inflict, or insist on, even though (and
indeed because) it is not a good for him, and he has reasons to get out of it if
he can.
Returning now to the question of the appropriate attitude: having announced
that we should not regret the pain inflicted, despite its being bad for him, I do not
need to think that the proper attitude to the inflicting of that pain is unmixed
pleasure. The sort of sombre response that Hurka wants to keep in place is still
available, since we must recognize that what we are doing, though good, is not good
for the punishee. One can retain some sense of identification with the recipient and
the pain that is being imposed on him (ouch!), even though one thinks that the
imposition of that pain is all for the good. And this is another source of sombreness,
in addition to regret that the situation has arisen in the first place.
352    
VI
Even if this variabilist response is thought acceptable, there remain two subse-
quent difficulties to be dealt with, or rather one difficulty that comes in two stages.
We have decided that the punishment is good since it is to be welcomed, in the
context of the crime, and that the combination of crime and punishment is better
than the crime alone, i.e. than the crime unpunished. But we are in danger of
having manoeuvred ourselves into a position in which we find ourselves saying
that the crime also is good. There is a symmetry here: just as crime without
punishment is worse than crime with punishment, so punishment without crime
is worse than punishment with crime. Such structures persuaded us that the
punishment, when put together with the crime, makes things better, and therefore
itself acquires positive value; and we should surely say the same of the crime. After
all, our earlier ranking order was incomplete: the crime alone is worse than the
crime + punishment, which is itself worse than what is best, namely having neither
crime nor punishment. But the punishment alone is surely a lot worse than the
crime alone, and so comes at the very bottom of the scale; most of us would prefer
to leave some crimes unpunished if this were the only way to ensure that nobody is
punished for a crime they did not commit. On this basis, adding the crime to the
punishment improves the situation, or at least reduces its disvalue, more than does
adding the punishment to the crime. All the more reason, then, to think that if the
punishment is good in the context of the crime, the crime is good in the context of
the punishment. But then (and this is the second stage), if each is for the better in
the context of the other, won’t the combination have to be positively good, since it
consists of good elements in an appropriate relation? And wouldn’t it be rational
(or even morally required) for someone, knowing that he will end up punished for
a crime that he has not committed, to go out and commit it so as to reduce the
number of unpunished crimes, or at least makes things less bad than they would
otherwise be? And would it not also be rational (and morally required, again) for
us to go about encouraging crimes whose authors we know are going to be
detected and convicted, in order to increase the sum total of value? These
questions are a sure sign that something has gone wrong somewhere.
Moore’s intrinsicalist position is in good shape on both these points. For him,
the pain inflicted is bad, and so is the crime, though the combination of the two is
good as a whole and less bad on the whole than either separately. But is the
variabilist in trouble here?
The obvious way of proceeding is to attack the first stage of the difficulty. Part of
the pressure derives from the fact that it is hard to think that the crime is genuinely
bad when the world is the better for its presence. But in fact it is possible for an evil
to be such that the world is better for its presence, if the world would have been
even better for its absence. It might look as if this is not our case, for we are
 ’      353
imagining starting from the combination of crime and punishment and subtract-
ing the crime—which makes things worse than they were before. On this picture,
things are not better for the absence of the crime. But that is a distorted perspec-
tive. What we should be saying is that though things are the better for the presence
of the crime, they would have been even better for its absence. How could this be?
The answer is that this could occur even if something that with the crime was good
would have been bad without it—so long as without the crime that further thing
would not have occurred at all. And that is, of course, the normal case; without the
crime, there would have been no punishment.
The other thing to bear in mind here is that, on the variabilist picture, by
inflicting the punishment we may perhaps diminish the disvalue of the crime, but
we don’t render it positively good. The combination of crime + punishment is not
a good of any sort; it is merely less bad than either part alone. It would have been
far better to have had neither than both. There is no danger that we ought to go
about maximizing the number of punished crimes, by encouraging crimes whose
authors we know are going to be detected and convicted, in order to increase the
sum total of value.
Perhaps the crucial mistake, though, was to think that because the punishment,
when put together with the crime, makes things better, it therefore itself acquires
positive value, since its presence is to be welcomed rather than regretted; and that
we should surely say the same of the crime. The reply to this should be that a
feature which makes things less bad need not itself be positively good, if that
means that we should seek to have more such things. We should distinguish
between a good feature and a mitigating feature. Add the punishment to the crime
and each becomes less bad than it would otherwise be—in the case of the
punishment, it becomes a lot less bad because punishing the innocent is so
dreadful; in this way it becomes possible that their combination is less bad than
either would be alone.
Variabilists are only condemned to holding the punishment positively good if
they take the crime to remain as bad as before, whether punished or not. For if
they took that line, they would have to find a positive good to set against a
persistent positive evil. But that is not a picture that should recommend itself to
them. What they should say, of course, is that in the context of the crime, the
punishment is much less bad than it would otherwise be—and the same is true in
reverse of the crime, even if not to the same degree. It is quite possible that the
disvalue of the crime is reduced by the addition of something which is still a
disvalue, in a way that leaves the combination still bad, but less bad than the crime
would have been alone. This is the true variabilist picture of vindictive
punishment.
Still, isn’t it the case that if I know that I am going to be punished anyway, I have
good reason (of a certain sort—a reason of justice) to commit the crime, since the
complex [crime + punishment] is more just than the punishment alone? First, this
354    
scenario can occur even for Moore, if the figures come out right—as they can do.
Second, this question is asked from a peculiarly objective standpoint, one encour-
aged by consequentialist talk of how well the world is going. Deontologists would
not find very attractive the idea that I should help the world get along better by
committing a crime. So this ‘consequence’ of organicity is actually a consequence
only of an organic consequentialism, not of organicity as such. What is more, on
any theory we might face the problem whether, if I know that I will be punished
anyway, and that the injustice of this undeserved punishment is greater than
would have been the injustice of the crime, I should commit the crime in order to
reduce the injustice in the world. This question does not even need any sort of
appeal to organicity to get going.⁶
⁶ I am very grateful to many friends and colleagues who have tried to help me with this paper, which
is my final effort on a topic with which I have been struggling for quite a while. Philip Stratton-Lake
deserves a special mention.
22
Has Anyone Ever Been a Non-Intuitionist?
This paper is about an argument of Harold Prichard’s that doesn’t really appear in
his few published works; it is in section IV of the piece entitled ‘Manuscript on
Morals’ in the most recent edited collection of his work.¹ But a version of it is to be
found in the last few pages of chapter 2 of Ross’s The Right and the Good (1930),
and it was this presentation that first attracted my attention. Ross’s version differs
a little from Prichard’s. But the interest of the argument is not merely a matter of
getting things right between Prichard and Ross.
The version that we find in Ross is intended to show that consequentialism
misrepresents its own paradigm case. His claim is that, properly understood,
everything that the consequentialists want to say, or at least can say, about
beneficence is (a) exactly what the intuitionists would say and (b) not structurally
distinct from what intuitionists say about other duties that are not grounded in
consequences. My question here is what that argument is and whether it is sound.
I am not convinced that it has had the attention it deserves.²
In section IV of ‘Manuscript on Morals’ (hereafter MM), Prichard is discussing

what is really at issue between intuitionism and ‘non-intuitionism’, or consequen-
tialism. The non-intuitionist he focuses on is Henry Sidgwick, and Prichard
represents the matter thus: Sidgwick maintains that intuitionism is a view that
nobody has ever held; but in fact non-intuitionism is a view that nobody has ever
held. This is an arresting claim—too arresting, one might say. Still, the idea that
both sides think that nobody really disagrees with them is rather pleasing, and
somehow fits the situation well. For it gives us a sense in which each view aims to
swallow the other up; from its perspective the other view is rendered invisible.
What Sidgwick actually says, however, is the following. He initially character-
izes intuitionism as ‘the view of ethics which regards as the practically ultimate
end of moral actions their conformity to certain rules or dictates of Duty uncon-
ditionally prescribed’ (1907: 96). But he immediately says that no morality has
¹ All references to works by Prichard are to them as printed in the recent collection edited by Jim
MacAdam: Moral Writings: H. A. Prichard (2002).
² There is a short discussion of Ross’s version of the argument in my ‘Wiggins and Ross’ (1998).
356    
ever existed that did not consider ulterior consequences to some extent.
Intuitionism in this sense cannot be ‘extended to the whole range of duty’.
There is, admittedly, an implication here that nobody has ever tried to do so. So
Sidgwick then offers us a version of intuitionism that he says some people have
indeed held, which maintains that certain kinds of acts are unconditionally
prescribed (or proscribed) without regard to ulterior consequences. Sidgwick’s
intuitionist, then, says that some kinds of acts are right or wrong because of their
ulterior consequences, but the rightness and wrongness of other sorts of act is
explained in a quite different way. And this form of intuitionism implies ‘that we
have the power of seeing clearly that certain kinds of action are right and
reasonable in themselves, apart from their consequences; or rather with a merely
partial consideration of consequences, from which other consequences admitted
to be possibly good or bad are definitely excluded’ (1907: 200). One might see a
challenge to intuitionism here, that it draws an indefensible distinction between
those good or bad consequences that are to matter morally and those that are not.
Sidgwick does not present it in this way, but the challenge is one that could be
made and later I will ask how the intuitionist should respond to it.
Sidgwick is aware that his way of setting up the debate between intuitionist and
non-intuitionist suffers from the apparent difficulty of understanding the distinc-
tion between an action and its consequences. His response is that the supposed
difficulty is not that daunting. As he puts it, common notions of different kinds of
act draw a line between the results included in the notion and considered as part of
the act, and those considered as its consequences. To understand this in its turn,
we need to ask about Sidgwick’s conception of an action. Unfortunately, I cannot
find in Sidgwick a clear account of what an action is. What is clear is that for him
an action is a volition plus certain consequences. But the details of this are obscure.
Is it that the consequences of the volition are themselves actions, or are they
merely movements, that is, motions or movednesses, or even just changes? I think
that Sidgwick’s view is probably that the consequences are not themselves actions
(and so are not ‘what is willed’) but are rather changes such as muscular contrac-
tions, in the first place, and further things thereafter (see 1907: 73).
An action, then, is a volition plus consequences, and those consequences will
expand outwards in time. The common notions of an act include some of those
consequences but exclude others. So, to use the standard example, I pull the
trigger, I shoot you, and I kill you. There is one volition here, and the three
‘notions’ describe the situation in different ways, each including consequences not
included in its predecessor. (I won’t address the question whether my pulling the
trigger is the first element in such a series, or whether the first element might not
be my crooking my finger, or even my finger’s crooking as a consequence of a
volition.)
In terms of this picture, the intuitionist maintains that certain types of action,
or common notions of an act, are such that all actions of that type are right (say),
     - ? 357
in a way that is nothing to do with any further consequences—even though there

will of course be further consequences, and some of them will be for the better or
for the worse. This is Sidgwick’s understanding of intuitionism. And such an
intuitionist can perfectly well allow that certain other types of action are right
because of their consequences rather than in their own nature. Some actions will
be right because they lead to the best possible outcome, and there is nothing in
intuitionism (as Sidgwick sees it) that is at odds with that.
Now, as I said at the beginning, Prichard represents Sidgwick as saying that
intuitionism is a view that nobody has ever held. He says that, in Sidgwick’s view,
both in the case of actions of given kinds and of given particular actions, which
we think right or think wrong, every one in order to think of them as right or as
wrong has at least in certain cases first to think of certain consequences of doing
them, so that if ‘intuitionism’ stands for a view according to which right actions
are, and are apprehended to be, right independently of their consequences, it
stands for a view which no one has ever held. (MM: 140)
In saying this, Prichard is thinking of Sidgwick’s first characterization of an

intuitional view. Sidgwick does indeed seem to say that nobody has ever held
such a view. But Prichard doesn’t seem to pay much attention to Sidgwick’s
second characterization, according to which an intuitionist allows, as all moralities
have allowed, that ulterior consequences are relevant to the rightness/wrongness
of some acts, or acts of some types, insisting only that certain other acts or types of
acts are prescribed or proscribed without reference to ulterior consequences.
So far we have two characterizations of intuitionism:
A: Extreme Intuitionism: every action that is right is so, not because of its
ulterior consequences, but in its own nature.
B: Conciliatory Intuitionism: some actions that are right are right because of
their ulterior consequences, others are not right for that reason but right in
their own nature.
Now Conciliatory Intuitionism allows that non-intuitionists are right in their

picture of some duties. And to many critics, this looks like a weakness; recently
we have been hearing suggestions that Ross, for instance, makes himself vulner-
able by allowing that at least three of his prima facie duties are to do with actions
that are duties because of the value of their consequences.³ The problem is
supposed to be that those duties will take over altogether if we aren’t careful,
³ See, e.g., Wiggins (1998).

358    
leaving no room for the non-consequentialist duties of fidelity, gratitude, etc.

Perhaps it would be better not to be a Conciliatory Intuitionist.
I am going to suggest that Prichard is in fact an Extreme Intuitionist, and so
takes the view that Sidgwick says nobody has ever held. But Prichard also
maintains that nobody has ever been a non-intuitionist. In the next section we
see how these things emerge.
II
I will chart Prichard’s progress more or less in his own words. Sidgwick means by
‘an Intuitional view of morality’ a view which maintains (1) that actions of certain
kinds are right in themselves, i.e. in virtue of being acts of those kinds, and
therefore independently of their consequences; (2) that certain specified particular
actions are right in themselves, i.e. as being the particular cases which they are of
acts of one of these kinds, and therefore independently of their consequences; and
(3) that we apprehend them to be right as such.
Prichard says that such a view seems to imply that in order to discover what
actions and kinds of action are right we need never (sic) consider consequences.
(Actually, it doesn’t seem to imply this at all, since it explicitly speaks only of
actions of certain kinds; actions of other kinds might be different. Luckily, this slip
does not affect the course of the argument.) But even an intuitionist might think
giving a neighbour a shilling⁴ right because it would help him out of a difficulty. So
there seems to be consideration of consequences even on an intuitional view. How
then are we to give an intuitionist account of actions that are right because of their
consequences? What would such an account look like?
The answer is that we turn to the philosophy of action. By ‘an action of a given
kind’ we mean the bringing about or ‘originating’ of a state of affairs of a certain
kind S, that kind being indicated by the phrase by which we refer to the action. By
‘a consequence of an act of a given kind’ we mean a state of affairs of another kind
S1 to which the existence of a state of affairs of the given kind S necessarily leads.
(Note the ‘necessarily’ here.)
In fact Prichard had various views about the nature of action at various times.
His first view was the one I have just reported, that to act is to originate, that is, to
cause a change (e.g., 2002: 84–5, where he rejects the view that an act is a conscious
origination of a change). At other times he held that to act is to set oneself to cause
some change, or to will it. This is of course a very different view, since willing a
change might be what causes that change without being the causing of it.
⁴ A shilling was a piece of pre-decimal coinage in Britain, equivalent to 12 pence; 20 shillings made
one pound sterling.
     - ? 359
(Prichard says in a very late piece⁵ that he now thinks that ‘setting oneself to’ is the
wrong term for the mental activity involved in any action (2002: 274).) On this
account the change caused follows from the action but is not itself any part of it,
and the causing of that change is not an action either. What is willed, then, is not
an action. I am not sure, however, that these fine distinctions matter for the points
that I am concerned with in this paper. Though Sidgwick thinks of action as
volition plus consequences, and Prichard thinks of it as the originating of those
consequences, they agree that our terms for different types of action often take us
way out along the train of effects, to things that at best we only ‘indirectly cause’.
Prichard goes on to say that if there is a problem here (and we are still dealing
with problems not for intuitionism but for a certain characterization of intuition-
ism) it derives from the fact that we can refer to an action X which brings about
Y as the bringing about of Y, so that anything which a given action brings about
can be thought of either as a part of that in the origination of which the action
consists, or as a consequence of that action. This applies to all members of the
sequence except the first; the action is the origination of that, and it cannot be
considered as a consequence.
This excursion into the philosophy of action puts Prichard in a position to give
his account of how an intuitionist should understand those actions that are right
because of their consequences. He distinguishes between the wrongness of actions
of a type, and the wrongness of particular actions. He starts with the former, and
claims that the intuitionist can say that when we judge an action to be wrong
because of its consequence, that consequence is not a consequence of the action in
that aspect in which it is apprehended to be in itself wrong. So in all cases in which
we apprehend an act of a certain kind to be wrong, we apprehend an act of some
kind (which may not be the same kind) to be wrong in itself, in a way that is
independent of its consequences.
With a particular action such as giving a neighbour a shilling, even an intu-
itionist (he says) will allow that in order to apprehend the rightness of the act, we
have to consider consequences. But the intuitionist could go on to say that in order
to apprehend that rightness, we have to regard the action as an action of helping
the neighbour out of his difficulties—as we can regard it—and the neighbour’s
getting out of his difficulties is not a consequence of the action so regarded. So in
apprehending the particular action to be right, what we are taking account of is
not a consequence of the action regarded as we are regarding it in apprehending it
to be right. In fact the action regarded in the way in which we have to regard it in
order to apprehend its rightness, is right in itself.
With this, Prichard draws the general conclusion that what Sidgwick refers to as
an intuitional view of morality is a view according to which a kind of action, or an
⁵ ‘Acting, Willing, Desiring’ (1945); repr. in Prichard (2002).

360    
action, which is right, is either right in itself, or else its rightness implies that some
other kind of action or action is right in itself; which is consistent with holding
that some actions or kinds of action are right because of their consequences. And
he concludes by saying that in that case, it is hard to see what a non-intuitionist
view would be. For such a view would have at least to hold that no actions or kinds
of action are right in themselves. But it has been shown that it is impossible to
maintain that some actions or kinds of action are right in virtue of their conse-
quences without maintaining that the actions considered in some other way, and
also certain other kinds of action, are right in themselves.⁶
This appears to give us a third form of intuitionism:
C: Flexible Intuitionism: a kind of action, or an action, which is right, is either

right in itself, or else its rightness implies that some other kind of action, or
some other action, is right in itself. In the latter case it is right because it is
of that other kind, or it is a kind of action that is right because all such
actions are of another kind that is right in itself.
The obvious question now is whether this is really a third form of intuitionism, or
whether it is in fact a form of Extreme Intuitionism, the view that Sidgwick said
that nobody has ever held. If it is the latter, then we do end up where Prichard said
we end up: ‘true’ intuitionism is a position that Sidgwick says nobody has ever
held, but in fact nobody has ever held any other position. And it does seem to me
that we should not so much contrast Flexible Intuitionism with Extreme
Intuitionism as think of the former as showing us a way of maintaining the latter
in the teeth of apparent difficulties. What is more, this fits Ross’s account, namely
that everything that the consequentialists want to say, or at least can say, about
beneficence is (a) exactly what the intuitionists would say and (b) not structurally
distinct from what intuitionists say about other duties that are not grounded in
consequences.
III
At the conference of which this volume is a record,⁷ Tom Hurka objected that this
argument of Prichard’s can be run backwards. For the reasons given, it is going to
be as true that every action that is right is right in virtue of its consequences as it is
that every action that is right is right in its own nature.
⁶ MacAdam cuts the excerpt at this point (MM: 142). But presumably it needs to go on roughly thus:
‘and that actions of the former kinds are in each case right because they are of one of the latter kinds
which is right in itself.’
⁷ Conference on British Moral Philosophers from Sidgwick to Ewing, held at the University of
Toronto’s Centre for Ethics in April 2008.
     - ? 361
This point seems to be correct, as long as we exclude the limiting cases of the
right action of making things go as well as possible, at one end of the chain, and of
the very first action in the chain, which might be right in itself and not because of
‘its’ consequences. Leaving those special cases aside, Hurka’s thought was that
every action that is right in itself is identical with an earlier ‘action’ in the chain,
which is right because of its consequences. The question then is why the ‘collaps-
ing’ argument tells in favour of the conclusion that nobody has ever been a non-
intuitionist rather than in favour of the claim that nobody has ever been an
intuitionist.
Another way of putting the point, which again I owe to Hurka, is that we can
distinguish between two theses here. The first is agreed on all sides. It is that there
is a way to think of an act under which we will think of it as right because of its
consequences if and only if there is another way to think of it under which we will
think of it as right in itself. This is a sort of parity thesis. The second, which is not
agreed, is a priority thesis: the second way of thinking always has priority. The
non-intuitionist accepts the first thesis, and argues that there is no reason to
favour one sort of way of thinking of an action over another. The intuitionist has
to support some sort of priority thesis if the sort of conclusion that Prichard wants
here is going to emerge. What sort of priority thesis might one defend?
One answer appeals to explanatory priority: the way of thinking that reveals the
act as right in itself explains why there is another way of thinking of it under which
the act is right because of its consequences. A different answer is epistemological:
considering the act as described in a way that does not allude to its consequences,
one could not discern its rightness at all; to discern the rightness, we have to
consider the act as the act of causing certain consequences, and so considered it is
right in itself. This seems to be Prichard’s answer; we go forward along the chain
until we meet the way of thinking of the action under which its rightness can be
apprehended. The action, considered as we have to consider it in order to
apprehend its rightness, is right in itself. It is not just as true to say that the action,
considered as we have to consider it in order to apprehend its rightness, is right in
virtue of its consequences. So the collapse tells in favour of intuitionism.
But why does this epistemic fact—if it is a fact—establish the sort of priority
thesis that the intuitionist needs? It is true enough that if we do not mention the
relevant consequences at all, we will be incapable of discerning the rightness of the
act. But that was not what was at issue. The distinction was between thinking of an
act as having certain consequences, and as being right in virtue of those conse-
quences, and thinking of it as the act of causing those consequences, which is right
in itself. In the first way of thinking, we might say (very cautiously, given the
history of these terms) that we are thinking of the act as extrinsically right, and in
the second way we are thinking of it as intrinsically right. The truth, however, is
that the action is of a type that is right in itself, and that is why we can say of it,
considering it as an action of some other type, with certain consequences, that it is
362    
right. Giving one’s neighbour a shilling, with the consequence that he is helped, is
right because it is an action of helping him out of a difficulty. The reverse is not
true: helping him out of a difficulty is not right because it is an action of giving him
a shilling, with the consequence that he is helped. It is that action, but this is not
why it is right. It seems, then, that the epistemic asymmetry really throws us back
on an explanatory one.
Remember Prichard’s conclusion, that the non-intuitionist, to be distinct, must
maintain that no acts or kinds of action are right in themselves. In the terms of
Hurka’s distinction, this is to say that denying the priority thesis commits one to
denying the parity thesis. Have we seen why this should be so? It is so because if
one denies the priority thesis, one loses the explanation of why the action,
considered as an act with certain consequences, is right.
IV
Prichard’s argument rests on the following principle (MM: 139): ‘It is also true
that . . . since in bringing about X, [I am] also indirectly bringing about Y, my
action can also be thought of and referred to as my bringing about Y, and that, so
thought and regarded, Y is not a consequence of my action but something which
in doing it I bring about or originate.’ It is by appeal to this principle that Prichard
is able to get the result he wants, namely that the action which is right in virtue of
its consequences is also right in its own nature, since that same action can be
differently characterized in a way that includes those consequences. Note, how-
ever, that this is different from saying that it is the action of ensuring those
consequences, if by that is meant that it is the action of causing those conse-
quences, rather than the action that causes them.
It seems that Sidgwick would have assented to this principle; but it is worth
considering whether the principle is generally defensible. In this section, I offer
first an indirect and then a direct defence.
First, does the principle inherit the difficulties of what is sometimes called
‘Anscombe’s Principle’ (though I don’t think she was in any way committed to
it), that if I F by G-ing, my F-ing is my G-ing? There are two possible rules to be
considered here, the first being Prichard’s, the second Anscombe’s:
Bringing-about Rule: if my bringing about X has the further consequence Y,

my bringing about X is my bringing about Y
By-locution Rule: if in or by F-ing I am G-ing, then my F-ing is my G-ing.
The latter rule might be false when the former is still true.
     - ? 363
Note that the ‘bringing about’ locution is not itself causal, any more than is the
by-relation. By bringing it about that you receive the money I bring it about that
my debt is paid, but the relation between the two bringings-about is not causal and
the consequences concerned need not be causal consequences. Perhaps ‘bringing
about’ is a generic ‘making it the case that’ relation.
One way of proceeding is to show that the By-locution Rule fails in cases where
the Bringing-about Rule still holds, and then to ask why that is. Suppose that by
refusing I disappointed her expectations. I would hope not to have to say that my
refusing was my disappointing her expectations; the By-locution Rule would force
me to say this. But the Bringing-about Rule approaches things differently. It asks
whether, when I bring it about that the invitation I receive is refused (by refusing
it), and thereby bring it about that her expectations are disappointed, my bringing
the first thing about is my bringing the second thing about—to which the answer
might still be yes. But this example won’t serve the purpose intended. True
enough, we want to say that I did two things: I refused her invitation and
I disappointed her. But then how can it be that the ‘bringing it about’ approach
generates an identity in this case, when the by-relation does not? This question
looks as if it is likely to be persistently relevant. What we need is another example
that seems to have an answer to it.
Suppose that by working hard and getting to know lots of stuff about ancient
history I win the scholarship. Is my working hard my winning the scholarship?
Probably not, I would say; and my reason would be that my working hard seems to
have stopped before my winning the scholarship even began. But my bringing it
about that I know lots of stuff about ancient history might be said to be my
bringing it about that I win the scholarship. The focus on the bringing about and
its nearer and further consequences does seem to make a difference. And this
example seems to have some chance of explaining why, if my bringing it about
that I know these things is my bringing it about that I win the scholarship, this
does not show that by working hard, I was winning the scholarship. But it still
allows us to say that I won it by working hard.
My conclusion so far is that Prichard’s Bringing-about Rule is defensible, or at
least more defensible than the By-locution Rule. But this form of defence is at best
indirect. Direct challenges to Prichard’s principle can take either or both of two
forms. We can deny that where my bringing about X is what brings about Y, this
shows that I am bringing about Y. Alternatively, we can admit this, but deny that
in all such cases my bringing about X is my bringing about Y. For instance, my
success in bringing about X may refute her prediction that I will fail, and so bring
it about that (Y) her prediction is refuted, but I am not sure that I would say that
I am refuting her prediction in so acting, nor (a fortiori) that my success is my
refuting her prediction, or my refutation of that prediction. If not, defence of the
Bringing-about Rule would require us to distinguish between my refuting her
prediction and my bringing it about that her prediction is refuted, and such a
364    
distinction seems to be at odds with what Prichard and Sidgwick say about the
relation between action and consequences.
So there is an issue here about how the Bringing-about Rule stands in relation
to the special conception of action that Prichard shares with Sidgwick, as volition
plus consequences. Would it look less, or more, plausible on a different account of
what an action is? I raise this question in order to have in on the table, not in order
to answer it. My general suspicion, however, is that there are not that many rival
conceptions of what an action is, and that the ones that there are don’t look as if
they would change things much. For instance, the conception of an action as the
causing of a change, which Prichard abandoned but Ross retained, doesn’t look as
if it would make much difference.
Still, I do want to admit that the Prichard/Sidgwick conception of action is
rather peculiar. There is the initial origination, and everything that flows from it,
no matter how tangentially, seems to have an equal claim to be that of which the
initial origination is (at least partly) an origination. There is no attempt here to
distinguish between intended or unintended, foreseen or unforeseen, welcome or
unwelcome consequences. But are we really to think of our actions as moments of
origination followed by we know not what? This is a very strange picture of
action—even if it is shared between Sidgwick and Prichard (and Ewing, I think).
There is also the question that worried Prichard, in his 1932 Henrietta Hertz
Lecture ‘Duty and Ignorance of Fact’,⁸ the question whether I can be said strictly
speaking to bring about anything (and so do anything) which I do not cause
directly. It is true, he thinks, that we make no trouble about allowing that I do the
things that are caused by the things I cause directly; but the only things that are
strictly speaking actions of mine are things that I bring about directly. This picture
doesn’t really fit the way in which Prichard uses the Bringing-about Rule.
Ross does to some extent address these difficulties, at least by implication, since
he asks whether I can be said to cause a change in any case where the occurrence of
that change is due to other agents as well as myself. His answer is that, given the
circumstances (honest and industrious postal workers, etc.) my ‘posting the book
was the one further thing which was sufficient to procure my friend’s receiving it.
If it had not been sufficient, the result would not have followed’ (1930: 45). I don’t
know, however, that Prichard would have allowed himself this sort of flexibility.
⁸ Repr. in Prichard (2002).

     - ? 365
I mentioned earlier the challenge that intuitionism, in its account of the rightness
of certain acts or certain sorts of action, draws an arbitrary distinction between
valuable consequences that are to count and those that are not to count.
Presumably the intuitionist reply to this challenge is that the distinction is not
arbitrary. But why is it not? Not because rightness is not determined solely by
comparative value; even if true, that would be irrelevant. The question is why
some value counts and other value does not count, not whether value is the only
thing that counts. Again, this is not the question of the relation between the good
and the morally good. Let us allow that only actions and people are capable of
being morally good. But the good things we are interested in here are conse-
quences, which are good but not morally good. The question is why some
consequences which are good (but not morally good) do not affect the rightness
of the action when others do. What is the filter that somehow rules out the value of
certain consequences as irrelevant?
We might try appealing to the fact that, if one follows the general line of
Prichard’s 1932 letter to Ross,⁹ the sort of rightness we are dealing with here is
prima facie rightness, rightness in a certain respect. There is very little pressure to
allow this to take us off into the distant future. Suppose, for instance, that an action
is (or would be) a peculiarly violent rape, and we want to say that it is wrong for
that reason. Prichard would think of this as saying that there is some duty not to do
it. Admittedly, this ‘some duty’ is a very strong duty, but so far no more than that;
we are still at the level of the prima facie. So we could say that this feature ensures
that the action is wrong in a certain respect, in a way that is not vulnerable to the
discovery of other features that might speak on the other side. There is no danger of
this being thought of as arbitrary. We are not ignoring the possibility of further
good consequences; they are just irrelevant to what has been established so far.
But again, this is irrelevant. Even allowing that we are dealing with prima facie
rightness, there is nothing here that explains the idea that the value of certain
consequences makes no contribution to the determination of overall rightness and
wrongness. That one thing makes a prima facie contribution does nothing to
explain why something else does not.
The intuitionist position must surely start from the thought that doing the right
thing is not the same as doing the best thing. Valuable consequences are neces-
sarily relevant to the latter, but not necessarily relevant to the former. They would
only be so if our moral duty were to do the best thing possible. But it is not. The
resultance base for rightness is distinct from the resultance base for bestness. But
then we want to know why this is. One answer is that the ground for duty consists
⁹ See Prichard (2002: 286–7). I commented on this letter in more detail in my Ethics Without
Principles (2004: 32–4).
366    
not just in value, but in value + definite relation. This is not to be the idea that duty
has two grounds (so that even in the absence of the definite relation, the value still
contributes, and vice versa) but that in the absence of the definite relation, the
value makes no prima facie contribution at all. I think that this is a possible
answer; it is at least relevant, when the others I have considered in this section
were not. I discuss below the question whether, as some think, it might be
Prichard’s answer.
VI
In the dialectic of this paper, we are now left with Prichard’s claim that nobody has
ever been a non-intuitionist. But isn’t there a way of distinguishing consequen-
tialism from intuitionism that doesn’t generate this result? Surely consequential-
ism is standardly understood to be the view that every action that is right is made
right by the value of its consequences. Intuitionism allows, perhaps, that what is
right is (in some sense) made right by the nature of its consequences, but not by
their value.
In considering this move, it is vital to remember that the discussion in section
IV of ‘Manuscript of Morals’ takes place after Prichard has argued Sidgwick to
what he says is the plausible view that we think it our duty to make mankind,
including ourselves, happy, and we think of this duty as summing up, or as being
the basis of, all our so-called other duties (MM: 137). This is the point, previously
alluded to, at which Prichard’s presentation of the argument differs from Ross’s.
For Ross’s target utilitarian grounds his supreme duty in the value of the conse-
quences of his actions, not in their own nature, and Ross’s own picture of
beneficence seems to agree with this picture.
The effect of Prichard’s argument, then, is to show that consequentialism
cannot be distinguished from intuitionism without appeal to the ‘axiological’
idea that rightness is grounded in value. So there are two aspects to what is called
consequentialism, axiology (as it were) and consequentialism proper. Prichard’s
strategy is to show that there is nothing distinctive for consequentialism to be, if
the contrast we are considering is between duty as grounded in the intrinsic nature
of the dutiful act and duty as grounded in the nature of consequences of that act.
This distinction collapses. There is however a further distinction, between duty as
grounded in value and duty as grounded in something other than value. If we put
the two distinctions together, we get the composite distinction between duty as
grounded in the intrinsic nature of the dutiful act and duty as grounded in the
value of the consequences of that act. This is the way in which the contrast
between intuitionism and non-intuitionism is often conceived. But in MM,
Prichard is keeping the two distinctions apart. First, he argues against the first
distinction as a distinction; it collapses in favour of intuitionism. We have already
     - ? 367
examined that aspect of things. He then considers the second one, and argues (in
the remaining sections of MM, and in many other places) against the first,
axiological horn.
The riposte I introduced at the beginning of this section ignores this aspect of
Prichard’s strategy. It tries to put back together the things that he has so carefully
drawn apart. Consequentialism is indeed standardly characterized in the way
suggested (though often we sloppily say that it is the view that actions are made
right or wrong by their consequences, which is what Prichard would prefer us to
say). But to characterize it in that composite way is to blur what Prichard, which
his characteristic precision, requires us to keep distinct.
VII
But we now hit a scholarly debate about what exactly Prichard’s views were on this
issue. Not everyone agrees with the claim I have just made, that for him duty is
never grounded in value. All would allow, however, that it is very hard to reach a
firm view on the matter. There are no really clear pronouncements in his works,
published or unpublished.
Prichard says more than once that we would not consider a certain feature to be
the ground of a duty if we did not attribute value to that feature. In a very early
piece he allows, ‘that unless the effect of some action were in some way good, there
would be no obligation to produce it, i.e., that the goodness of the thing produced
is a presupposition of the obligation to produce it’ (2002: 2–3) And immediately
afterwards he writes: ‘analysis of any moral principle will show, I think, that it
includes mention of two things, (a) a good thing which the action will produce, (b)
a definite relation in which the agent stands either to another or to himself ’ (2002:
4). (It is worth noting that Prichard is pretty catholic about what is to count as a
‘definite relation’, since he allows common humanity to count as such: ‘the
obligation not to hurt the feelings of another involves no definite relation of us
to that other, i.e. no relation other than that of our both being men, and men in
one and the same world’ (2002: 13). So a ‘definite’ relation need not be a ‘special’
relation.)
The most reliable quotation in this respect comes from ‘Does Moral Philosophy
Rest on a Mistake’:
At best it can only be maintained that there is this element of truth in the
Utilitarian view, that unless we recognized that something which an act will
originate is good, we should not recognize that we ought to do the action. Unless
we thought knowledge a good thing, it may be urged, we should not think that we
ought to tell the truth; unless we thought pain a bad thing, we should not think
the infliction of it, without special reason, wrong. But this is not to imply that the
368    
badness of error is the reason why it is wrong to lie, or the badness of pain the
reason why we ought not to inflict it without special cause. (2002: 10)
And there is a footnote at this point:
It may be noted that if the badness of pain were the reason why we ought not to
inflict pain on another, it would equally be a reason why we ought not to inflict
pain on ourselves; yet, though we should allow the wanton infliction of pain on
ourselves to be foolish, we should not think of describing it as wrong.
How are we to interpret these remarks? In the rest of this paper, I will consider the
merits of two readings, both of which are compatible with all the passages I have
quoted above. The first reading is that the ground for a duty is complex, and
includes elements, or parts, one that concerns value and one that concerns a
definite relation. The second reading is that these things are required, but not as
grounds. (The claim that we are dealing with a presupposition is not one that
interprets itself, after all.)
I start by considering the merits of the second reading. One paragraph after his
claim that ‘any moral principle . . . includes mention of two things, (a) a good thing
which the action will produce . . . ’, Prichard writes that ‘each [moral principle]
stands on its own footing, i.e. has its own special reason, a reason usually conveyed
in the ordinary formulation of it, e.g. a man ought to pay his debts, to provide for
his family, to repay benefits, to serve his state’ (2002: 5). The italicization here is
interesting. It seems intended to draw our attention to something, but does not
look consistent. For instance, if the definite relation is part of the reason, the ‘his’
ought to be italicized in ‘pay his debts’ and in ‘provide for his family’ as it is in ‘serve
his state’. However, the one obvious thing is that none of the reasons is character-
ized explicitly in terms of value. If the ‘ground’ is the reason, then, it doesn’t look as
if the value of the things to be produced is part of the reason, exactly, even if it is
presupposed or required for the thing that is the reason to be the reason that it is.
As it stands, this is not much support for the second reading. But more can be
found in ‘Mistake’, where Prichard says of his own view that it is ‘avowedly put
forward in opposition to the view that what is right is derived from what is good’
(2002: 14). He does not seem to mean merely that, though it is derived from this, it
is derived from it only in combination with certain other things.
More important is the way in which Prichard reaches his own view in ‘Mistake’.
He does so by arguing against two alternative views, both of which hold that if we
ought to act in a certain way, this is because in or by so acting we will be producing
something good. The first view says that the relevant goodness belongs to the
consequences, to something that is not an action but is produced by an action.
The second says that the goodness belongs to the action itself. What is important
in the present context is the way in which Prichard argues against the first view.
     - ? 369
He starts by saying that the view plainly presupposes that what is good ought to
be, because an ‘ought’, if it can be derived at all, can only be derived from another
‘ought’. The point of this is that we cannot derive an ‘ought to act’ directly from
some thought about the goodness of certain consequences; we can only hope to do
so via an intermediate ‘ought’ which is somehow attached to the good conse-
quences, and this will have to be an ‘ought to be’. If those consequences are the
best, they ought to be, and therefore we ought to ensure them. What is more, he
says, the ‘ought’ in ‘ought to be’ must ‘involve just the feeling of imperativeness
and obligation which is to be aroused by the thought of the action that will
originate it’. And he goes on to say that ‘ought to be’ is not proper language; we
can only speak of what ‘ought to be done’. And only the ‘ought’ that applies to
action can come with that sense of imperativeness and obligation.
I think the best sense of this argument is made when we remember that
obligation, for Prichard, is a feature of an agent, not of a situation. So the argument
he is considering (these consequences would be best, so they ought to be, so
I ought to produce them) moves from a premise about some consequences to a
conclusion about an agent without any visible rationale. So we gain nothing by
trying to circumvent Prichard’s point that there is no such thing as an ‘ought to be’
by saying that what is presupposed is just that whatever action has the best
outcome is the one that ought to be done (rather than the one that ought to be).
For him there is no such thing as ‘ought to be done’ either.
Now we philosophers are likely to be interested in the way this argument of
Prichard’s is supposed to work. But Prichard seems actually to be more impressed
by another, and supposedly better, way of refuting the view he is targeting. He
claims that it simply fails to correspond to our actual moral convictions.
Suppose we ask ourselves whether our sense that we ought to pay our debts or to
tell the truth arises from our sense that in doing so we should be originating
something good, e.g. material comfort in A or true belief in B, i.e. suppose we ask
ourselves whether it is this aspect of the action which leads to our recognition that
we ought to do it. We at once and without hesitation answer ‘No’. (2002: 10)
This claim is clearly intended to stand alone, though Prichard reinforces it with an
argument, that our sense that we ought to act justly cannot depend on a recogni-
tion that doing so will lead to some good, since ‘the balance of resulting good may
be, and often is, not on the side of justice’. This argument, unfortunately, is not
persuasive, for two reasons. First, one might respond that there is some good in a
just outcome, even if such outcomes are not always the best available. To say that
duty is grounded in the prospect of value is not yet to say that it is grounded on the
balance of resulting value. Second, one might try to cook the books by allotting
such significant value to the justice of the outcome that it always counterbalances
any badness in the outcome—though I find this very unconvincing.
370    
However this may be, Prichard’s preferred point seems to be that we simply
don’t think that our sense that we ought to act derives from a sense that so acting
would have the best consequences. But this is at best a mild strike in favour of the
second reading. One might suppose that all Prichard means here is that the former
sense does not derive from the latter sense alone.
I reserve my most telling point until last. Summing up the argument in
‘Mistake’, Prichard writes:
The negative side of all this is, of course, that we do not come to appreciate an
obligation by an argument, i.e. by a process of non-moral thinking, and that, in
particular, we do not do so by an argument of which a premise is the ethical
but not moral activity of appreciating the goodness either of the act or of a
consequence of the act, i.e. that our sense of the rightness of an act is not a
conclusion from our appreciation of the goodness either of it or of anything else.
(2002: 13–14)
What is telling about this, in the search for the correct interpretation of Prichard,
is the phrase ‘a premise’ (though Prichard has a rather idiosyncratic conception of
a premise as an activity). Apparently, the thought is not that there are other
premises required as well. The appreciation of the relevant goodness does not
function as a premise at all.
This is the main support for the second reading. On it, Prichard’s view looks
pretty nuanced. Value is required for duty, but not as part of its ground. Obviously
we will want an account of why it is required, and what role considerations about
value play if they are excluded from the ground. But such things may yet be
forthcoming.
What is the support for the first reading? Well, Prichard says that the value is a
necessary condition for the existence of a duty, and that the existence of a duty
presupposes that there is value in the thing that there is a duty to produce. This,
however, is more or less common ground between the two readings. I therefore
turn to a much later passage, which comes in section 8 of an unpublished piece
called ‘Moral Obligation’ (2002: 216–17). Here it looks initially as if Prichard’s
view is very clear. He is attacking the view according to which ‘what renders some
act a duty is its causing something good’, which sounds very much like the view
that the value of the consequences is the ground of duty. He says that in order to
evaluate this view, we should find a case of an action that causes something good,
and ask whether we think of it as a duty for that reason. And he says that we don’t.
So far there is no interpretative difficulty. The difficulty comes from what he says
about the example he gives. He considers two contrasted cases, where the good to
be achieved is an increase in someone’s patience. By doing one thing, I can
increase my own patience, and by doing another I can increase someone else’s
patience. According to the theory being attacked, I should do both of these things,
     - ? 371
and for the same reason, namely that in doing them I will be making someone
better. But Prichard insists that in fact these duties are of different sorts, not the
same sort. It makes a difference who it is that we are making better, just as it
matters whether it is my child or someone else’s when we are considering whether
we ought to help to make them better. Prichard’s criticism of the view he is
attacking, then, is apparently not that the value cannot be part of the reason for, or
ground of, the duty, but that there is something else to be considered (perhaps in
addition to that value) which the view at issue ignores completely. ‘The truth
indeed appears to be that in order to think of some change as one which we ought
to cause, we must think of the change as in some special way related to ourselves,
even if that way consists in its affecting someone other than ourselves’ (2002: 217).
So he does not attack the suggestion that duty is value-grounded, only the
suggestion that the value constitutes the whole of the ground. This gives at least
an impression that he is not concerned to dispute the former.¹⁰
This is the main support for the first reading. By way of comment on it, it is
worth pointing out that the sentence I have just quoted (‘The truth indeed appears
to be . . . ’) can be read in either of two ways. The first way allows that the action
should be done, because of the value to be got, but maintains that nobody has a
duty to do it unless they stand in some definite relation to the relevant change. The
question is, as it were, ‘OK, I allow that he ought to be helped, but why should it be
me that has to do it?’. The second reading maintains that there is no such thing as
an action that ought to be done. No action is a duty as such; there are only actions
that this or that person ought to do. Now I think that Prichard certainly intends
the second reading. First, he doesn’t at all like the idea of a change that ought to be;
and his dislike extends to the idea of a change that ought to be originated,
implemented or achieved. Second, as I have already noted, his overall conclusion
in ‘Duty and Ignorance of Fact’ is that obligations are characteristics of agents, not
of actions.
VIII
Our question is how Prichard conceives of the ground for duty, that is, the reason
to act. He allows that the reason for an action must show the goodness of the
change to be made. But this is not itself to allow that the reason must be that
the change to be made is a good one. Prichard’s reasons (to pay his debts, to
provide for his family, to repay benefits, to serve his state) mention things that are
good, but do not seem to mention their goodness. On the second reading, then, he
thinks that the goodness is not part of the reason; we have sufficient reason to act if
¹⁰ Thanks to Tom Hurka for putting me on to this passage, the relevance of which I would never
have picked up otherwise.
372    
our action would be the repayment of a debt or of a favour. Such changes would be
good, but their goodness does not somehow add to the reason given us by their
nature.¹¹
On the first reading, the value of the consequences (of the change to be made,
the outcome to be achieved) is always part of the ground for duty; but more is
required for a duty to be generated, that more consisting in some definite relation,
and the addition of that relation completes the ground. On the second reading, the
consequences to be achieved are the ground for duty, but for those consequences
to make a duty two more things are required: first, a definite relation and, second,
that those consequences be of value. These two further things are required, but the
explanation of the need for them will, one imagines, not be the same in both cases.
The first reading relies, then, on a distinction between what is the whole and
what is only part of the ground for a duty. Suppose that the ground of a duty is the
reason for doing it. Are we to suppose that really for each duty we have more than
one reason, one concerning value and one concerning a definite relation, or is it
rather that for each duty there is a reason, but that reason is a sort of complex of
parts which make different contributions to the whole? My own view is that the
latter way of understanding the first reading is far better than the former. This is
partly because it doesn’t seem to me that the definite relation plays the same sort
of role in the constitution of a duty as that played by the supposed value (or,
perhaps, by the thing that has that value). Suppose, then, that we accept the latter
account of the way in which considerations of value and definite relation combine
to make a ground for value. In doing so we should remember that not every
definite relation that is required looks as if it is properly thought of as part of the
ground. For instance, my having the ability to do the act, or the opportunity, is
required, perhaps, for me to have a duty to do it. But such things don’t seem to
count in favour of my doing it, nor even to stand as part of what counts in favour
of my doing it, in the same sort of way that the fact that this is my child seems to
count in favour of my diverting my resources in her direction. Now in making this
point I don’t want to revert to the thought that the definite relation constitutes an
independent reason in favour of doing the action. But its contribution is still
different in style from that of ability or opportunity; perhaps it acts as an
intensifier,¹² so that the fact that my action will benefit this child is some reason
for me to do it, and more reason if it is my child than if it is not.
Some definite relations, then, will act as intensifiers, which take a pre-existing
duty and make it more pressing. Others do not function in that way, but are
required for the existence of the duty in the first place. I have a duty to pay my
¹¹ This view is now commonly mentioned as part of the motivation for the buck-passing view of
value. But there is no entailment here. We would not need to think of Prichard as an implicit buck-
passer in order to accept the second reading.
¹² For more on this notion of an intensifier, see my Ethics Without Principles (2004: ch. 3).
     - ? 373
debts; you do not, you only have a duty to pay yours. But there is more than one
way in which a definite relation is required for the existence of a duty in the first
place. Having the opportunity or the ability to help is a definite relation, and one
without which (let us suppose, for argument’s sake) we have no duty to help. But
such things are very different from, say, having been helped by you in the past.
The notion of a definite relation is pretty flexible.
Taking advantage of that flexibility, we can all admit that a definite relation is
required for there to be a duty. And we can argue about whether that relation is a
part of the ground, or plays some other role (perhaps some are and some are not).
The issue that remains—and that is untouched by such things—is whether the
value of the change to be made is taken by Prichard to be an essential part of the
ground. This is the sticking point. Is the reason, properly specified, not just that
this will help my neighbour out of his difficulty, but that doing this would be good
in itself, or will achieve some good? I have already expressed my preferred
interpretation of Prichard. Here is a final quotation to consider:
And I would suggest as a prominent instance of the fallacy involved the attempt
which is often made nowadays (as e.g. I think it is by Professor Moore and
Professor Laird) to maintain a view which implies that we deduce the rightness of
certain actions from our knowledge of what is good taken in conjunction with our
powers of action and existing circumstances. (2002: 47)
IX
I have already mooted one possible argument for the second reading, which was
that once one has specified the relevant consequences, nothing seems to be added
by adding a remark about the value of those consequences. Indeed, one might
think that this addition is worse than otiose, since to require it as a general rule
would be to require a change in our ordinary practice of reason-giving. Like
Prichard, I am always suspicious of philosophical moves that announce that our
ordinary practice of reason-giving is to be understood as shorthand for some more
elaborate practice that the theorist wants to say we are ‘really’ engaged in.
If one wants something more than this, something that looks more like an
argument, one final possibility is that if the goodness of the consequence were part
of the ground, there would be no way of explaining why a definite relation should
be also required for duty. Suppose that I say ‘this is an action that would lead to my
debts being paid and that would be good’. Now why would this goodness not
stand as a reason for anyone? An answer to this question might be: it won’t,
because a definite relation is also required. But this answer gets things backwards.
The definite relation is not there to stop the reason being a reason for everyone. It
seems to be there as the thing that answers the question ‘Why me?’. But if the
374    
thought is that the reason is given by the goodness of the consequences, the
question ‘Why me?’ doesn’t have much point; the answer to it is that I don’t
need to be part of any definite relation for the goodness of the outcome to speak to
me as a reason. The most a relation could do would be to intensify in my case the
reason that the goodness gives to all. (Things would of course be different if the
sort of value we were dealing with is not the agent-neutral sort that Prichard is
thinking of, but rather some kind of agent-relative value; though even for this we
would have to conceive of the agent-relative in the right sort of way.)
This is all very different from the case where I say ‘this is an action that would
lead to my debts being paid’. If we then say I have a duty to do the action, and I ask
‘Why me?’ the answer might be that they are my debts. The definite relation seems
to be able to do some work here.
We have, as it were, two possible ways of reading Prichard’s thought:
A definite relation is required for duty; therefore goodness is not the whole of
the ground.
A definite relation is required for duty: therefore goodness is not part of the
ground.
And I have been trying to suggest that he might have thought the latter.
It has appeared that Prichard has three distinct arguments against consequential-
ism, or non-intuitionism, understood as the view that a right act is made right by
the goodness of something distinct from it, namely its consequences. The first is
that it collapses into intuitionism; the right way to conceive of the relevant act is as
the act of originating those consequences, which is an intuitionist, not a conse-
quentialist picture. The second is that its attempt to ground obligation on value, or
the right on the good, serves to make things worse; if one is going to appeal to the
consequences at all, one should do so directly, not via their value. The third is that
even if one did allow appeal to the value of the consequences, this could never be
sufficient to generate a duty, since the absence of a ‘definite relation’ can prevent
what would be the best consequences from generating the appropriate duty. This
is a hefty barrage.¹³
¹³ Thanks to Robert Adams, John Broome, Brad Hooker, Derek Parfit, Philip Stratton-Lake and
Ralph Wedgwood for very helpful discussion, and especially to Tom Hurka for his excellent comments.
23
More Right than Wrong
In this paper I consider one aspect of the difference between Kant and Ross.¹ It is
often said that Kant has great difficulty in making any sense of the notion of a moral
reason, and that in this respect Ross’s notion of a prima facie duty puts him at a great
advantage—an advantage inherited by those who think that the basic notion in all
this is the notion of favouring, understood as a specific relation that holds directly
between certain features of the agent’s situation and certain forms of response
(mainly action). In his interesting paper ‘The Importance of Moral Rules and
Principles’ (2007), Hill suggests in response that particular facts are reasons because
they are ‘especially salient features of a possible fuller rationale that we might give’.
This is an example of a Kant-inspired account which diverts our attention away from
the supposed relation of favouring, and seeks to explain reasons as parts of a rational
structure, dominated by some grand Kantian Principle, Categorical or Hypothetical,
that generates an overall ought, or duty. The difficulty for this manoeuvre lies in the
phrase ‘especially salient’. Reasons, it might be said, are not just particularly eye-
catching elements in a structure. Unless more can be said than this remark about
special salience, we lose entirely the distinction between the roles played by reasons
and the roles played by other aspects of the supposed ‘fuller rationale’.
In what follows I suggest that those inspired by Ross have much less advantage
on this point than they tend to suppose, for Ross fails to make any sense of the
relation between moral reasons as he understands them (that is, as prima facie
duties) and the rightness of the action that is made right by those reasons. He can
give no account of what it is that his moral reasons do. I then consider the best way
of trying to improve things.
1. Ross on the relation between duty

proper and prima facie duty
Ross was the first person to see that the rightness of an action is built out of certain
features that go to make it right, which he called prima facie duties. Ewing rightly
¹ I claim some credit for having been instrumental in W. D. Ross’s return to philosophical favour
after thirty years of oblivion at the hands of R. M. Hare. I was struck, therefore, when Tom Hill told me
that he had taught Ross every year since his original appointment at Chapel Hill. This is characteristic
of that admirable intellectual independence which we also see in his relation to Kant. My paper was
written as a contribution to a festschrift in Hill’s honour.
376    
said of this that it was ‘one of the most important discoveries of the century in
moral philosophy’ (1959: 126). This was true in 1959, and it remained true forty
years later. What is extraordinary, and very hard to believe nowadays, is that
before Ross nobody had any inkling of this idea. Everyone thought in terms of
principles of absolute duty. If a principle applied at all, it determined the answer to
the question what one’s duty was. Evidence of this fact, if evidence is needed,
comes from the structure of the argument in the final chapter of Ewing’s early The
Morality of Punishment (1929). Ewing argues that all known principles of absolute
duty (that is, the Kantian and the utilitarian versions) are false, and that therefore
there are no moral principles. The result of this reasoning was that Ewing was a
particularist at the beginning of his long career, which consisted of a series of
lurches from one meta-ethical position to another. (His Second Thoughts in Moral
Philosophy (1959) might have been better called Yet Another Change of Mind.²)
Ross offered in The Right and the Good (1930) a way of escaping the need to
choose between Kantianism, utilitarianism and particularism: there are principles,
but they are all principles of prima facie duty. Principles of this form do not
determine the answer to the question what one’s duty is in a particular case; on
that matter they are silent, leaving the issue to judgement. They play their role at a
preliminary stage, and several of them may apply to the same case at the same
time. The role of judgement is to choose between them, and all that the principles
of prima facie duty do is to tell the judger what she is choosing between. The
system (if it can be so called) serves up a list of directly relevant features, but it
does nothing to tell us how significant the various features are in a particular case,
and so does nothing to tell us how to act in response to the different prima facie
duties that we know ourselves to have here. This is no doubt disappointing,
considering the ambitions of traditional moral theory; but Ross held that no
more could be expected or achieved.
Many people now think Ross was right about this, though there are still some
who continue to write as if Ross had never existed. Derek Parfit’s On What
Matters (2011) is an impressive example of this; the general impression (on this
topic) is that if it is not in Sidgwick or in Kant, it can be confidently ignored. But
this prejudice is dangerous, for absolute moral principles are much harder to find
and to defend than are prima facie versions of the same thing. Compare these two
versions of a principle of consent: (1) if a person cannot consent to the way you are
treating him, what you are doing is wrong, and (2) if a person cannot consent to
the way you are treating him, what you are doing is prima facie wrong. The nub of
the matter is that counterexamples to the second of these are much harder to
produce than are counterexamples to the first. So one would think that wary
² This reminds me of Broad’s remark that Ross’s second book, Foundations of Ethics, could well have
been called The Righter and the Better (1940: 239).
    377
theorists would do well to restrict their pronouncements to the prima facie. Parfit
is a contemporary example of someone who is not wary in this way.
So Ross was the first person to see the need for a distinction between what I call
the contributory and the overall levels. And this was indeed an intellectual
triumph. But sadly, this is as far as the triumph went. Pretty well everything else
in Ross’s meta-ethics is wrong—indeed, hopelessly wrong. In saying this, I am not
referring to his intuitionism (by which I mean his pluralism and the associated
epistemology), nor to the list of particular duties that he provides (which is not a
bad first shot, I would say, if such a thing is needed). I am referring to the details of
the account he gives of the contributory, and of its relation to the overall. None of
this can be got to work. So what Ross shows is that we need an account of the
contributory, and we need an account of the relation between what goes on at that
level and the level at which facts about how we ought in fact to act emerge. But
Ross’s own versions of these things are way off the mark.
What Ross rightly wants is the idea of something normative at the contributory
level that is a matter of degree. This is what he is getting at when he talks of one
prima facie duty being more incumbent or more pressing than another. And he
tries to express this idea by talking of prima facie duties. But he immediately
announces that these prima facie duties ‘are not duties at all, but something related
in a special way to duty’ (1930: 20). We need, therefore, to understand the way in
which one of these things can be more pressing than another, when greater
pressingness is not a matter of greater duty. And Ross’s official account of a
prima facie duty just prevents us from doing this. He writes that being a prima
facie duty is ‘the characteristic (quite distinct from that of being a duty proper)
which an act has, in virtue of being of a certain kind, . . . of being an act which
would be a duty proper if it were not at the same time of another kind which is
morally significant’ (1930: 19). But the first and most obvious thing about this
account is that prima facie duty, as characterized, is not a matter of degree. One
cannot have more of being an act which would be a duty proper if it were not at
the same time of another kind which is morally significant; the idea of such a thing
doesn’t make any sense. Ross has left us no room to say that one of these things
can be more pressing than another. The normativity of the contributory has
vanished from view entirely. At best, what Ross offers is a consequence of the
thing he is really trying to get at.
The same can be said of the use of the term ‘prima facie justification’ in
contemporary epistemology. Those who use this term acknowledge their debt to
Ross. But the unfortunate fact is that they also think of a prima facie justification
as having some normative force—as generating a little bit of justification, to be put
in the balance and compared with the prima facie justifiedness of alternative
hypotheses. And this idea is one that Ross’s views completely undermine. If we
take what he says seriously, a prima facie justifier is a feature such that, if it were
the only relevant feature, it would make the belief at issue justified. But this
378    
isolation test is not the way in which the notion of prima facie justification is used
in epistemology, and rightly so.
I said at the outset that Ross was attempting to make sense of the way in which
particular features of an action can, as I put it, go to make that action right. This
notion of ‘making right’ is worth further thought. For Ross denies that any
individual feature can ever succeed in making an action right. An action is
never made a duty proper by one feature alone. A single feature can only make
the action a prima facie duty. If it is a duty proper, it is, he says, made so by all of
its features, not just one, or even some, of them.
The equality of the two angles [of an isosceles triangle] is a parti-resultant

attribute. And the same is true of all mathematical attributes. It is true, I may
add, of prima facie rightness. But no act is ever, in virtue of falling under some
general description, necessarily actually right; its rightness depends on its whole
nature and not on any element in it. The reason is that no mathematical object
(no figure, for instance, or angle) ever has two characteristics that tend to give it
opposing resultant characteristics, while moral acts often (as everyone knows)
and indeed always (as on reflection we must admit) have different characteristics
that tend to make them at the same time prima facie right and prima facie wrong.
(1930: 33–4)
This passage rewards attention. The first thing to say is that the last sentence
introduces a confusion. Later, in chapter 2 of The Right and the Good, Ross
suggests an understanding of prima facie duty that is different from the one
I have already mentioned, and is run in terms of tendencies. But these tendencies
are tendencies to make actions actually right or actually wrong: they tend to make
them duties proper, not to make them prima facie duties. Otherwise the appeal to
tendencies would do nothing to explicate the notion of a prima facie duty. (I
return to this suggestion about tendencies in the next section.)
But it is in the third sentence that the real mistake occurs. Why is the
(supposed) fact that no act is ever, in virtue of falling under some general
description, necessarily actually right, relevant to the question whether an action
can be made right by its possession of a limited number of characteristics? It is
only relevant if one assumes, as Ross did but I do not, that if an action is made
right by having certain features, any other action that has those features must also
be right, no matter how it may differ in other respects. If one makes this
assumption, the idea that an action can be made right by the good it does, or by
any other particular character it may have, is indeed a non-starter.
So Ross, making this assumption, denies the possibility of what I have been
calling ‘right-making features’, and I view this as another abject failure. The sort of
contribution that is made by a particular feature must be compatible with the idea
that such a feature can on occasion be what makes an action the one we ought to
    379
do. Of course this feature will only be able to do it if no other feature succeeds in
making some alternative act a duty proper instead. But this does nothing to show
that when that condition is satisfied, this feature is not what makes the action a
duty. What we should be hoping to provide is an understanding of the thought
that this action is a duty proper because you promised you would do it; your
promise makes it a duty proper and not just a prima facie duty. But Ross fails to
provide any such understanding; indeed, he asserts firmly that duty proper is, as
he puts it (rather exaggeratedly) ‘toti-resultant’: that is, that what makes the action
a duty proper is all its properties at once. Now we know that this is false. If an act is
a duty despite being hard-hearted, it is not a duty partly because it is hard-hearted.
Ross has left no room for the idea of a ‘despite’. And Ross knew that it is false too,
for he admits in a footnote (1930: 33–4) that his account is exaggerated, because
many features of an action will be irrelevant to whether it is a duty. But this is not
enough; he still insists that no act is made a duty by any one consideration, and
this is something we should reject in advance, even though we of course allow that
many duties are made duties by several features acting together, pooling their
normative resources.
2. The tendency account
Ross writes:
We have to distinguish from the characteristic of being our duty that of tending
to be our duty. Any act that we do contains various elements in virtue of which it
falls under various categories. In virtue of being the breaking of a promise, it
tends to be wrong; in virtue of being an instance of relieving distress it tends to be
right. Tendency to be one’s duty may be called a parti-resultant attribute, i.e. one
which belongs to an act in virtue of some one component in its nature. Being
one’s duty is a toti-resultant attribute, one which belongs to an act in virtue of its
whole nature and of nothing less than this. (1930: 28)³
What are we to make of this?⁴ One possible reply is that tendencies do not belong
to particular actions, but to actions of a type. A particular action does not tend to
be dangerous; either it is dangerous or it is not. But perhaps Ross thinks otherwise.
If he does, what is offered here does indeed look like a sort of normative force
coming from below, as it were, which has the great advantage that there is a clear
³ Though Ross adds in a footnote that this last claim, about toti-resultance, is somewhat exaggerated.
⁴ For a discussion of this suggestion, and a reply to some early complaints that I made about it, see
Gay (1985).
380    
account on offer of the relation between the lower and the upper strata—being the
right-maker and the rightness made.
Nonetheless I do not think that this account can be got to work.⁵ Certainly,
tendencies to be right can be stronger or weaker, and so a tendency to be right is a
matter of degree, which helps; but a tendency of this sort is not itself normative.
The normativity of the tendency comes entirely from the normativity of the
rightness that acts of this sort tend to have. It is just not clear that an unfulfilled
tendency is the right sort of thing to generate residual duties, and it should be clear
if an unfulfilled tendency had its own normativity (see the next section for more
on this topic).
It is worth comparing what Ross says about prima facie rightness with what he
says on occasion about prima facie goodness, since he clearly thinks that as far as
the term ‘prima facie’ goes, these two are on a par:
If it were self-evident that the right coincides with the optimific, it should be self-
evident that what is prima facie right is prima facie optimific. But whereas we are
certain that keeping a promise is prima facie right, we are not certain that it is
prima facie optimific (though we are perhaps certain that it is prima facie
bonific). (1930: 36)
Here Ross does not seem to feel any strain in applying his conception of the prima
facie to good-making rather than to right-making. The prima facie optimific is not
something that tends to be best, but something that is already good-making in a
certain respect. Being already good in a certain respect is not the same as tending
to be best (nor is it the same as being something that if it were alone would be
best). Similarly, the prima facie bonific should be something that is actually bonific
in a respect, not something that tends to be so overall. And this is the sort of thing
that Ross should be saying about prima facie rightness—except that he doesn’t.
3. Further difficulties
While I am at it, I might as well mention a further failure of Ross’s account of the
prima facie, that is, of the contributory. One attractive aspect of his approach is
that it seems to make room for a duty that we have, and that we recognize that we
have, but which we think of as defeated by another duty which is stronger. Now
I have already said that Ross in fact offers no workable understanding of how one
prima facie duty can be stronger than another. But at least we might have hoped
⁵ There is a full-scale discussion of the tendency account of prima facie duties in chapter 6 of my
Moral Reasons (1993). Here, I focus only on the use of that account to solve my worries about the
relation between overall and contributory normativity.
    381
that he can give some sense to the idea of a defeated but still real duty (unlike the
friends of absolute duties, who have to say that no real duty can be defeated.)
Unfortunately it turns out that this is not so. For his account of a prima facie duty
as a feature that would decide the issue if it were the only relevant one fails entirely
to make sense of the idea that if a prima facie duty loses out, it can still generate
catch-up duties (such as the duty to apologize). This is because the account is
officially silent about what happens to a prima facie duty when it is not the only
relevant feature. This is exactly the problem with isolation tests. And for that
reason it can say nothing about how a defeated duty behaves, either in general, or
in particular as a generator of catch-up duties.
All these problems stem from the one central failure of Ross’s account, which is
that it fails entirely to make any sense of the normativity of the contributory. But
the idea of the contributory, and the distinction between the contributory and the
overall, are things that we very much want to retain. One feature of the contribu-
tory will occupy us for much of this paper, which is the fact that it comes in
degrees. Rightness, for Ross, is a matter of degree, but this applies only to prima
facie rightness, not to rightness proper. Duty proper is not a matter of degree,
since no duty is more pressing than any other duty. But some prima facie duties
are more pressing than others.
I may say, at this point, that I don’t find it at all helpful to think of Ross’s prima
facie duties in terms of Foot’s distinction between the verdictive and the eviden-
tial.⁶ The verdictive side of this is not too bad. To say of an act that it is the right
one in the circumstances can indeed be understood as the casting of a verdict in
favour of that act. In fact, this verdictive conception is compatible with both views
about the nature of rightness, for one can cast a verdict in favour of an act equally
well by saying that it is the most right act and by saying that it, and it alone, is right
(in the absolute sense). So I will continue to use what is effectively the notion of
the verdictive in what follows. It is the evidential conception of the prima facie that
seems mistaken. A prima facie duty is not to be understood as evidence that the
relevant action is a duty proper. And a defeated prima facie duty is not like
misleading evidence, that is, evidence that leads one away from the true answer.
In my own terms, a reason not to do the action remains in play as that reason even
when defeated by stronger reasons on the other side. It is not shown to be
misleading, or substandard in some way by this defeat. The notion of evidence
just seems out of place here.
After all these criticisms of Ross, it is worth pointing out that in his later
Foundations of Ethics he completely changed his view of the relation between
the contributory and the overall. The earlier view, which we have been examining,
was that the contributory (the prima facie, that is) is to be understood in terms of
⁶ Foot (1978: 182); see also Stratton-Lake (2000: 14).

382    
its relation to duty proper. The later view reverses this, understanding the con-
tributory in terms of a relation of fittingness or suitability in a respect, and the
right action as that action that has ‘the greatest amount of such suitability possible
in the circumstances’ (1939: 53; but see also p. 84, where it is far from clear that
Ross has simply abandoned his earlier views). This view is much closer to the one
that Prichard pressed against Ross in the years between the publications of Ross’s
two books, as we will now see. It is also much more clearly opposed to the
structure we see in Kantian accounts.
The metaphorical distinction between top-down and bottom-up needs careful
handling in the present context. Kant’s position is top-down, in the sense that the
role played by the distinct elements in generating a duty is explained by their
relation to the duty generated at the top. Ross’s first position is top-down in this
sense as well; it is his second view that reverses the direction of explanation and is
therefore bottom-up. The first position does, however, seek to explain duty proper
as generated by normative forces coming up from below (expressed in such terms
as ‘stringent’ and ‘pressing’); and in this second sense of bottom-upness both of
Ross’s views are bottom-up. Kant’s position thus emerges as distinctive in being
doubly top-down.
4. Prichard’s complaint
It is always interesting, when thinking about Ross, to ask oneself about the relation
between his views and those of Harold Prichard. In many ways Prichard was
Ross’s mentor. Ross was enormously influenced by Prichard’s argument that
motive is not choosable, and that therefore that which is our duty cannot include
any motive. It cannot, therefore, be our duty to act from a certain motive. And
Ross followed Prichard’s misguided lurch from objectivism about duty to subject-
ivism.⁷ And we might be under the impression that Prichard agreed with Ross
about the distinction between prima facie duty and duty proper. For Prichard did
write, in his famous ‘Does Moral Philosophy Rest on a Mistake?’, that ‘obligation
admits of degrees, and . . . where obligations conflict, the decision of what we ought
to do turns not on the question “Which of the alternative courses of action will
originate the greater good?” but on the question which is the greater obligation’
(2002: 14, fn.7).
And it is easy to see in these remarks a position similar to the one that Ross
eventually articulated in his book. The same applies to Prichard’s alternative
suggestion to Ross’s terminology of prima facie duties, namely that we should
think of such things as claims. Ross decided not to accept this idea on the grounds
⁷ I have documented these shifts of view in my ‘Prichard on Duty and Ignorance of Fact’ (2002; repr.
as Chapter 20 in this volume).
    383
that if there is a claim, there should be a claimant, but in the case of some duties
there is no claimant to be found (1930: 20). Still, there is something of a hint here
that Prichard was sympathetic to Ross’s position, the only question between them
being one of terminology. But there is compelling evidence that Prichard was in
fact opposed to Ross’s substantive position—though not to his stress on the
contributory. This evidence consists mainly of a letter which Prichard wrote to
Ross in 1932. For the publication of this letter we are indebted to Jim MacAdam’s
edition of Prichard’s works, which contains a fair amount of previously unpub-
lished material (2002: 286–7) The letter reads as follows:
My dear Ross,
I hope you will forgive my taking a sort of pot shot at something in your book.
But I have got it on my chest.
On p. 27 you reduce the 6 or rather 7 prima facie duties enumerated on pp. 21
to 4, viz: your original (1)a, (1)b & (2)⁸ & the prima facie duty of producing as
much good as possible . . . And I have for long wondered that you should feel
comfortable in coordinating the last with (1)a, (1)b and (2), e.g. the prima facie
duty of keeping a promise.
But I have never till now seen clearly what has been puzzling me about this
procedure—though I have always felt it was queer. The queerness I think lies in
this—that whereas e.g. to describe an act as one of keeping a promise or as one of
making reparation is to describe it in respect of a character it has in itself, to
describe [an act] as producing as much good as possible, is only to do this
verbally; it is really to describe it as having a character which it possesses only
in relation to all the other acts the man can do—the character of producing good
in a degree greater than that produced by any of the others.
This difference seems to me vital. The last does not seem to me the character
of an action in the way in which the first 3 are. And the difference seems to me
one which is paralleled in your distinction between some prima facie duty of a
man and his duty sans phrase, and also to be out of place in a list of prima facie
duties. Whether the phrase ‘prima facie duty’ is appropriate for the thing you
refer to in your doctrine or not, the thing referred to is some character which an
action of a certain kind possesses in itself i.e. as an instance of a certain kind and
apart from its relatedness to the actions of other kinds possible to a man—a
character e.g., due to an act being one of keeping a promise.
And I take your view about duty to be that in a given situation the action
which it is my duty to do is that out of all the actions which I can, there is the
greatest prima facie duty to do—so that where some action is what you call ‘my
duty’—or as I would rather put it the act which I am bound to do, or what I ought
⁸ These are the duties of (1)a: promising, (1)b: reparation, (2): gratitude.
384    
to do, it is so in virtue of having more prima facie obligatoriness than any of the
other acts I can do—and so in virtue of a character which the act possesses
only in relation to all the others. (cf. Pickard-Cambridge, note 3, p. 150, Mind,
April, 1932.)
Hence while the basis of a prima facie duty is a character which the action has
in itself, i.e. apart from its relation to the others, the basis of a particular act’s
being ‘my’ duty is not.
Hence, also, it seems to me, given your distinction between a prima facie duty
of mine and my duty, the so-called character of producing as much good as
possible, i.e. really the character of producing more good than any other possible
action, while it might possibly be maintained to be the basis of some action’s
being my duty, can’t be held to be the basis of an action being a prima facie duty.
But, further, strictly speaking I don’t see how it can be maintained even to be the
basis of the thing called an action being my duty,—i.e. your ‘an action being my
duty sans phrase’. For to produce more good than any other possible action is to
have a certain character in a greater degree than certain other actions, and this
cannot give rise to anything but a greater degree of something else; yet on your
view to say of some act that it is a prima facie duty and to say of it that it is my
duty are to make qualitatively different statements.
The truth is that the more I consider it the less I can make sense out of ‘the act
which I am bound to do’—as distinct from ‘an act which I ought to do’—and the
more I get to think that the only fact corresponding to the phrase is ‘the act⁹
which I ought to do more than I ought to do any other’, and that your ‘prima
facie’ duty is really a duty, your ‘my duty sans phrase’ is really that of a man’s
duties which he most ought to do, i.e. that so far as these phrases can be made to
stand for facts these must be the facts.
...
yrs ever
HAP
This letter can be seen as having three sections. In the first Prichard distin-
guishes between the character which an act possesses as being of a certain kind,
and a character which an act possesses in relation to all other acts available to the
agent. (This distinction is different from, but analogous to, Ross’s own distinction
between parti-resultant and toti-resultant.) And Prichard maintains that Ross’s
prima facie duties are all officially of the first sort, and his duties proper are of the
second sort. He then points out that the character of being optimific, which Ross
treats as the basis of a prima facie duty, is in fact of the second sort when officially
it ought to be of the first sort. This by itself would not be damaging, since it only
⁹ The text as MacAdam gives it reads here ‘corresponding to the phrase ‘is the act . . . ’, which renders
the thought unintelligible.
    385
concerns the way in which Ross has formulated his duties of beneficence, non-
maleficence and justice. Ross has offered versions of these in explicitly maximizing
terms, and one could imagine non-maximizing versions of those duties, which
would perhaps make more sense.¹⁰ The idea would be that what makes the
charitable act a duty is not that it does more good than any other act available,
but just the good that it does. Its doing that good is a feature that it has as being of
a certain kind, not in relation to all other acts available to the agent. So this
particular point might be able to be cleared up.
We would still need to find some role for the fact that the act produces more
good than any alternative. If an act is right because of its production of good, is it
right because of the amount of good produced, or because that amount is greater
than any alternative would have produced—or both? Perhaps the answer is both,
and that we are dealing with two different ‘becauses’ here. The ground for the
rightness is the good produced but what enables that amount of good to make this
action right is that no alternative action would have produced more. The com-
parative fact is relevant, but not in the same way as is the amount of good
produced.
This pattern may seem rather contrived, but it is in fact perfectly general. Every
act that is right is so partly because there was no alternative action that would have
taken precedence, or been more of a duty. But this does not mean that all right
actions are made right partly by one and the same fact. The role of the absence of
alternative more pressing duties is that of an enabler, not that of a ground.
With this, I return to Prichard, who has not finished yet. He points out that,
though the act that is one’s duty proper is the act that is the most pressing prima
facie duty, these characteristics are, for Ross, very different—qualitatively differ-
ent, as he puts it. One is a matter of degree and the other is not. And Prichard
doesn’t see how this can work. If we are to understand the relation between
thoughts about prima facie duty and duty proper, we have to abandon the thought
that ‘is my duty’ expresses something that is not a matter of degree, and to allow
that all it can mean is ‘is the most pressing of my prima facie duties’. Far from its
being the case that ‘duty’ in ‘prima facie duty’ means something completely
different from the ‘duty’ in ‘duty proper’, so that prima facie duties are not duties
at all, but something related in a special way to duties, for Prichard there are no
prima facie duties at all. What we call our duty proper is simply our most pressing
duty proper, and there is no room for a distinction between the ground for the
duty proper and the ground for the prima facie duty that it also is. As we might put
it, the ‘right’ in ‘right-making’ is not prima facie rightness but the only sort of
rightness there is, real rightness. And the same applies to ‘ought’, and to ‘duty’.
¹⁰ As David Wiggins suggests (1998).

386    
Ross simply assumed that ‘ought to be done’ has a uniqueness implication, as in

the phrase ‘the act that one ought to do’.¹¹ But Prichard finds in this the same
severance between the supposed matter of degree, as ground, and something that
is not a matter of degree as conclusion. Really all we can make sense of is the idea
of an act which to some extent we ought to do, or which is to some extent our
duty. This notion enables us to ask what we ought to do, understanding that
question as the question which of the acts which to some extent we ought to do is
the one (if there is one) that we most ought to do. But there is no available notion
of duty other than this, and so no conceptual distinction between duty proper and
most pressing prima facie duty.
I end this section with a diagram of Ross’s position as Prichard understands it:
Duty proper
Most pressing prima facie duty
PF1 PF2 PF3 PF4
Ground 1 Ground 2 Ground 3 Ground 4
This diagram is intended to bring out the atomistic idea that degree of pressing-
ness of a prima facie duty is an intrinsic matter, not comparative, since the
pressingness of one prima facie duty will not affect the pressingness of another.
The relation between those independent pressingnesses determines which of the
prima facie duties is the most pressing. Comparison comes in only where we move
up from level 2 to level 3. But to be a duty proper is (for Ross) to have a
characteristic that is toto caelo different from that of being the most pressing
prima facie duty, even though every act that has one of these characteristics also
has the other.
5. Assessment of Prichard’s objection
Prichard needs to offer some defence of the main hinge principle in his letter,
which is this:
For to produce more good than any other possible action is to have a certain
character in a greater degree than certain other actions, and this cannot give rise
to anything but a greater degree of something else.
¹¹ For this assumption, see Ross (1930: 3–4), where he argues that though there can be two
alternative actions both of which are right, only one of them can be what we ought to do, that is,
our duty.
    387
This remark seems to be intended as entirely general. It is that having some

character in a greater degree than other actions can only give rise to something
that is itself a matter of degree. This looks like a ban on threshold properties with a
comparative base of a certain sort. And we need to understand why this ban is
supposed to be justified.
With the ban in place, the situation seems to be like this. That this action did a
certain amount of good could be the ground for a duty proper (if there were such
a notion), since doing that amount of good is not having a certain character in a
greater degree than certain other actions do. But that the amount of good done is
greater than what any other action would have achieved is only capable of ‘giving
rise’ to a greater degree of something else, such as a ‘more ought’ or a greater
prima facie duty, and could not be the ground of any such thing as a duty proper.
In interpreting this ban, we might wonder whether to produce more good is the
same thing as to have a greater degree of good-producingness (or of ‘beneficence’).
What are we to make of Prichard’s ‘having a certain character in a greater degree
than certain other actions’? There seems to be a distinction between being more
productive of good and being productive of more good. And if there is, which of
these is the ban concerned with? It looks as if it should be the former, if either,
since the latter is not obviously a matter of degree. And there is certainly a
distinction between being more impassioned about music and being impassioned
about more music, and only the former seems to fit Prichard’s phrase.
On the other hand, it seems to be the latter (being productive of more good)
that is at issue when we think about which of two beneficent acts we have the
greater prima facie duty to do. If so, Prichard has no complaint against the way in
which Ross treats the duty of beneficence. The property of being productive of
more good is not itself a matter of degree.
To confirm intuitions, suppose that there is only one prize, and it is given to the
person who makes the most cakes. Bernie makes the most cakes, and so deserves
the prize. Are we to suppose that the other cake-makers also deserved the prize,
but deserved it less, so that Bernie is the most prize-deserving, but not the only
prize-deserver? (This seems to be the way that Prichard would have to read the
case.) Does Lucio, who made the fewest cakes, deserve the prize to some extent,
only not so much as Bernie? This seems most improbable. Only Bernie deserves
the prize. But does this case really come under Prichard’s ban on threshold
properties that are not matters of degree but which are grounded in something
that is a matter of degree? No, since making most cakes is not a matter of degree.
There is the now familiar distinction between making most cakes and doing the
most making of cakes. Bernie is not more of a cake-maker than Lucio, even though
he makes more cakes.
My conclusion at this point is that Prichard’s reasoning is not very persuasive.
And in addition, I think that it is at least open to Ross to abandon the claim that
no duty proper is more of a duty proper than is any other. Perhaps he could allow
388    
that of two actions, both of which are duties proper (and which, of course, are
therefore not competing), it would be more wrong not to do one than it would be
not to do the other. If so, the threshold property will end up being a matter of
degree in just the sort of way that Prichard demands.
6. Another shot
So far, then, it has not been shown incoherent to add a conception of absolute duty
to our conception of something that is a matter of degree. As far as that goes, the
structure of Ross’s position remains intact.
Elsewhere, however, also in material that has only recently come to light,¹²
Prichard writes:
To maintain that we ought to do that of two actions which we ought to do more is

to be involved in a contradiction, because it is to imply both that while the one act
is a duty the other is not, and that the latter is a duty though to a lesser degree
than the former. Putting this otherwise, if we think that of two acts there is a
greater obligation to do one of them, we cannot go on to think that we ought to
do that action without implying that there is no obligation to do the other in any
degree whatever. Consequently we gain nothing by maintaining the existence of
degrees of obligation in addition to absolute obligation. (2002: 79)
In this we see Prichard prefiguring the complaint I made in section 1 above, that
Ross ends up trying to say both that prima facie duties are duties and that they are
not. Ross’s official view, of course, is that prima facie duties are not duties at all. He
is not trying to say that they are and are not duties. But there is a sense in which
Prichard is right. As I explained in section 2, Ross is trying to hold on to the idea
that there is something normative to be found at the prima facie level, and he does
this by calling what he finds there ‘prima facie duties’. But this is specious. He
might as well have called them ‘prima facie cuties’ for all that he achieves by it.
This is a different complaint from the general complaint that it is incoherent to
add something that is not a matter of degree to something that is.
So there is something new in what Prichard says here. If we think in terms of
degrees of duty, meaning now by this something that Ross did not succeed in
substantiating, namely that there is something normative that we are thinking of
as a matter of degree, we gain nothing by adding to that a conception of duty that
is not a matter of degree. We gain nothing because in terms of that second notion,
only one action is a duty and all the others are not duties at all. And we would have
¹² This is from a set of student’s notes on a lecture of Prichard’s given in 1928, two years before the
publication of The Right and the Good. These notes read as if Prichard had dictated them.
    389
to say of those actions that they are to some degree a duty and to no degree a duty,
and this is hardly a gain. Prichard could reasonably claim that, if anything, it
makes matters worse than before, because by adding a conception of duty that is
not a matter of degree, we undermine our sense that what is a matter of degree is a
matter of duty (so more or less pressing as a duty) rather than just something
related in a special way to duty. This addition is not merely something whose
relation to what we already had on the table (duty in degrees) is hard to make
sense of—which would have to be admitted anyway; the one positively interferes
with the other.¹³
This returns us to the point made in section 2, that Ross fails to locate
something both normative and contributory at the prima facie level. But it goes
beyond that in claiming that this is no surprise, because every attempt to achieve
that, while insisting on a distinctive non-contributory level of normativity, under-
mines its own starting point.
7. A distinctive argument for deontic buck-passing
Prichard’s view, as I understand it, is a form of deontic buck-passing. We are more

familiar nowadays with evaluative buck-passing, which is the idea that for some-
thing to be of value is for it to have features that give us reasons to respond to it in
certain ways. Deontic buck-passing is the idea that for an action to be right is for
the reasons that favour it to be stronger than the reasons favouring any alternative.
The right act is the one that one has most reason to do.
This formulation in terms of reasons is distinct from anything that Prichard
actually says, and I suspect that he would say that we need more than reasons at
the base. The idea would be that the notion of a reason does not have the right sort
of strength to make duty; we need something more peremptory. And anyway the
characterization of deontic buck-passing in terms of reasons needs a certain
amount of refining, because if it is moral rightness that we are dealing with, we
should probably restrict ourselves to moral reasons, so that for an action to be
right is for the moral reasons in favour of it to outweigh any moral reasons for
alternative actions. And if we take this line, we face the difficulty that nobody
¹³ I should admit that, although the passage I have quoted, and am here discussing, does seem to me
to contain a powerful argument against Ross’s position, that is not Prichard’s purpose in his lecture,
and he does not present things in that way. His purpose is to make sense of the idea of a conflict of
duties, and the suggestion that he is considering at this exact point is whether the way to do that is to
add a notion of obligation that comes in degrees to a Rossian notion of duty proper (which is what
Prichard means when he speaks of what we ought to do). And his conclusion is that this cannot work:
that since we need a notion of what we ought to do, we cannot add to that a notion of what we most
ought to do, or what we ought to some extent to do. I think he should have realized that the same thing
holds in reverse: that we cannot add a notion of what we ought (simpliciter) to do to a notion of what
we ought to some extent to do, without undermining the latter notion. (I am indebted in these
comments to Tom Hurka and Philip Stratton-Lake.)
390    
knows the difference between moral and non-moral reasons; there seems to be no
list of considerations that are somehow moral, and no prior distinction between
reasons of a moral sort and reasons of a non-moral sort. Luckily, Derek Parfit has
shown us how to do the necessary refining (2011: 166–7).¹⁴ We can say that what
it is for an action to be morally right is for those considerations that are morally
relevant in the case to come down in favour of doing it rather than any alternative.
Doing things this way, which avoids explicit appeal to reasons, is probably the
most cautious approach¹⁵ and so the wisest. And perhaps it would be enough to
satisfy any Prichardian sense that our original base properties were too weak to
make a duty.
There remains the worry that the natural way to understand moral relevance is
as relevance to how it is right or wrong to behave, which threatens to take us back
to non-scalar conceptions of the right and the wrong. If we deny ourselves such a
conception of moral relevance, is any other available?
The arguments for evaluative buck-passing are in fact pretty weak, as I think is
now agreed. One argument was that the badness of pain does not add to the
reasons to avoid it that are given us by the way pain feels. What is bad is always
made bad by something, and it is that something that gives us reasons to avoid the
bad, not the badness. And this fact can be explained if we think of the badness
itself as the presence of reasons that give us reasons to avoid the thing that is bad.
Now this is not itself a very powerful argument, even if its starting point were
secure.¹⁶ And there are also doubts about the starting point; it might be, for
example, that the badness of our behaviour gives us reason to regret doing what
we did. As far as this goes, the situation is the same for deontic buck-passing. It
might be that the wrongness of what we have done gives us reason to be ashamed
of ourselves.
But there is another and much better reason to be a deontic buck-passer, which
is that we thereby avoid trying to add something that is not a matter of degree to
something that is, which, if Prichard is right, leads us eventually to incoherence.
Everything is done by the weight of morally relevant considerations, and there is
no independent quality of rightness to add to that.
This reason does not apply to evaluative buck-passing, where everything is a
matter of degree, since we are understanding degrees of value in terms of degrees
of (strength of) reason. So deontic buck-passing is in a noticeably stronger
position than is evaluative buck-passing.
¹⁴ Parfit’s actual distinction is between decisive moral reasons and morally decisive reasons; he
prefers the latter expression, for the reason that I give in the text. I have adapted his distinction for my
own purposes.
¹⁵ Another advantage of this approach is that it allows that not every relevant consideration is a
reason.
¹⁶ This was one of the main points of my paper ‘Should We Pass the Buck?’ (2000b; repr. as
Chapter 14 in this volume).
    391
8. Five worries about deontic buck-passing
I committed myself to deontic buck-passing in an early paper on the topic

(2000b), having also argued against evaluative buck-passing. And that is still my
position. But I have ever since been worried about various things that seem to be at
odds with the deontic buck-passing view that all we need is something that is a
matter of degree—whether we talk of degrees of duty, or more simply in terms of
stronger and weaker moral reasons or morally relevant considerations. There are
three such things to worry about—or perhaps three ways of expressing a sense that
we need something that is not a matter of degree.
The first derives from a distinction between commendation and recommenda-
tion to which I was first alerted by the work of R. M. Hare.¹⁷ Hare held that the
OED rightly says of ‘good’ that it is the most general term of commendation. And
he always maintained that to say that a sort of action is right is to prescribe acting
in that way. But as I see it, prescription is a form of recommendation, not a form of
commendation, no matter in how high a degree. Now of course one can recom-
mend something by saying that it is the best available. But still, to say that it is the
best is no more than to commend it more highly than one commends other things.
One then recommends it for that reason. It is not as if the options that are less
highly commended are also less strongly recommended. Only one of them is at all
recommended. So commendation is a matter of degree, and recommendation is
not. And we need both.
I think the answer to this worry is to point out that the structure of the
evaluative is relevantly distinct from that of the deontic, since to say that an action
is good in some respect is a sort of partial commendation, but to say that there is
some reason to do the action is not to commend it at all. All that we need for
deontic purposes is something that is a matter of degree, with recommendation
coming in only when the balance of duty has been determined. The basic point
perhaps is that commending and recommending are just different speech acts.
Nothing that emerges directly from the differences between these two sorts of
speech act could serve to require a distinction between two sorts of concept, one to
be in operation in acts of one sort and the other in those of the other sort.
But there is another way of making the basic point. (Call this the second worry.)
Consider the phrase ‘You ought to do what you most ought to do’. Even if it is a
truism, this phrase just does not sound like a tautology. If I think I am required to
do the act that is most right, and required not to do any of the others, I seem to
have moved away from any tautology to something substantial. It is not that I am
¹⁷ The only relevant remark I can find in Hare’s work is this: ‘According to the O.E.D. “commend” is
sometimes used with the sense “recommend”; but this use is not common, and it is not in this sense that
the word occurs in The Language of Morals. We normally use “recommend” when a particular choice is
in question, but “commend” when a thing is being mentioned as in general “worthy of acceptance or
approval” ’ (1957: 103n). This distinction is of course quite different from the one I outline in my text.
392    
required to do each of the others severally, only less stringently so as we go down

the list, but most required to do the one that comes out top. So even if we allow
degrees of duty, we need to add an independent concept of something that is not a
matter of degree. Even if the action that is our most stringent duty is what we are
required to do, requirements don’t seem to come in degrees. They select one
action as the one required.
The third thing that has always worried me, as a deontic buck-passer, is the
suspicion that there is a notion of the wrong as that which is simply ruled out, and
that nothing that is ruled out in this sort of way is more ruled out than anything
else that is ruled out. Derek Parfit expressed this with his talk of a notion of what
‘mustn’t-be-done’ (2011: 165–6). This notion is supposedly distinct from any
notion of what we have decisive reason not to do,¹⁸ and Parfit thought of it as
unanalysable. It therefore appears to be something that resists buck-passing
treatment, at least by those forms of buck-passing that are offered as conceptual
analysis.
The opposite of the ruled-out is that which is required—a requirement. And
Margaret Little and Coleen Macnamara (2021) have raised doubts about whether
a requirement can be built out of contributory reasons. Some oughts can be so
built; they call these ‘prescriptive oughts’. But these are merely the acts which one
has most reason (of a certain sort) to do. A requirement, as they conceive it, is
made in a different way. It is something that, no matter what the reasons against,
I just have to do. To fix intuitions, consider an order given by a competent
authority. If my commanding officer tells me to stand on guard, I do not suppose
that his command is just a stronger reason than any reason I had to do something
else, reasons, perhaps, to curl up in front of the fire and watch the tennis, so that
I have to work out whether these reasons are stronger than those. The command
somehow obliterates those reasons. And Little and Macnamara conceive of moral
requirements as functioning in much the same sort of way.
The fourth worry is that there may be oughts without reasons. I tackle this
problem in my ‘Reasons and Rationality’ (2009), and will not pursue it further here.
The fifth point is that we may lose the possibility of tragic dilemmas, which
seem to require something more than two equally strong sets of reasons.
9. Attempt at an answer
Some of these difficulties derive from Prichard’s talk about what we ‘most ought’
to do. Thinking of the issue in terms of reasons, we ought to do what we have most
¹⁸ The notion of a decisive reason, which is certainly available to buck-passers, deserves more
attention than it has so far received. Parfit (2011: 32) understands any set of winning reasons as
decisive; but this is something that I neither recommend nor commend.
    393
reason to do. But this formulation seems to suggest that the only thing we can do
with reasons is to add them up. And though Ross and Prichard do seem to have
thought in this atomistic sort of way about the nature of moral reasoning, others
have not. If we suppose that other relations are possible than the simple one of
combining to make a greater or lesser total, we might come to think that the
supposed tautology ‘you ought to do what you most ought to do’, far from being a
tautology, is not even always true. And if it means that adding up independent
normative contributions will always give us the answer, it is not true.
Similar thoughts may explain Little and Macnamara’s attitude to what they call
requirements. The reasons for fulfilling a requirement are not reasons that simply
turn out to be weightier, in combination, than the reasons to do something else
instead. Requirements are built in a certain way. And on the same lines we might
try to accommodate Parfit’s notion of what mustn’t-be-done.
Taking this line of defence leads to a form of deontology that is not exactly
scalar, but which still does not seem to introduce a separate concept of duty or
obligation in the sort of way that made trouble for Ross. Everything is understood
in terms of the interactions between contributory considerations. The fact that we
have complicated our story about what goes on at that level does not amount to
adding a new upper level of a distinct metaphysical type, which was the source of
the original difficulties.
24
Prichard on Causing a Change
This paper assumes without argument a certain view about the nature of action.
This will limit its interest for those who do not accept that view. But the second
half of the paper discusses a difficulty which arises for them just as much as for
anyone else. So there should be something here for everyone.
The view that I assume is a view about the nature of action. It is an agent-
causalist view, but not the agent-causalist view. Traditionally, agent-causalism has
been the view that actions are events caused by agents. But this is not the view
I assume (and which I in fact accept). The view I assume is one which I first learnt
from an article by Maria Alvarez and John Hyman, ‘Agents and their Actions’
(1998). It holds that to act is to cause a change, and an action is an agent’s causing
a change. The action is not the change caused; that change, which is an event, is (in
Alvarez and Hyman’s terms) the ‘result’ of the action. We need to respect here the
distinction between the causing of a change and the change that is caused by, and
so the result of, that causing. The cause of the change is the agent. So here we have
a form of causation where the cause is not an event, but an agent. The result, the
thing caused, is an event (probably). Is the causing of that result an event? Alvarez
and Hyman say not, and I agree with them, but not for the reason they offer. They
effectively try to prove that this must be so; but I only want to suggest that nobody
should think that a causal relation is an event. Anybody who holds the more
standard view that causal relations are relations between events should avoid
saying that the relation between those events when one causes another is another
event, to be added to the cause-event and the effect-event. Otherwise they will face
the question what is the relation between the cause-event and the causing-event,
and I don’t see a plausible answer to that question—any more than there is a
plausible answer to the question what the relation is between the causing-event
and the effect-event. So I suggest that there is a considerable incentive for all of us
to avoid thinking of causings as events.
To come at the matter another way, causal relations are too thin to count as
events. The sort of thinness I am getting at here might be what is in the mind of
those who think of causal relations as essentially explanatory. An explanatory
relation is not the same sort of thing as a ‘making-happen’ relation. The latter
might be thought of as a sort of forcefulness in nature, and such a relation looks as
     395
if it should be thick, rather than thin. But even if causal relations are not
themselves merely relations of explanation, they will still be thin. On the form
of agent-causalism that I accept, the relation between an agent and an effect
caused by that agent is not itself another event somehow interposed between
the agent and the effect. The exercise of a causal power should not itself be
understood as a happening. For if so, we would want to ask how the agent
manages to get that happening to happen; and if the answer is that she does it
by causing it, we are back where we started, or, rather, off on a regress. And on a
traditional Humean approach, relations of causation are relations of regular
succession, and those relations are not themselves events in addition to the
events in that succession. My conclusion is that we should all avoid thinking of
causings as events.
I said that the view that to act is to cause a change is one that I learnt from an
article published in 1998. But it appears that this view was around 100 years
earlier, since Harold Prichard ascribes it to John Cook Wilson. In his ‘Acting,
Willing, Desiring’ (1945), Prichard writes: ‘We should at first say that to do
something is to originate or to bring into existence, i.e., really, to cause, some
not yet existing state either of ourselves or of someone else, or, again, of some
body’ (2002: 272).¹ And he ascribes this view explicitly to Cook Wilson. But the
relevant paper seems (like so much of Cook Wilson’s oeuvre) to be lost. So we
cannot be sure whether Cook Wilson actually tied himself to the thought that to
do something is to cause some not yet existing state to come to exist. After all,
there is the alternative possibility that some actions, at least, are the causings of a
process rather than of a state; I mention this alternative now because it will
become relevant later. So I think we should not take Prichard’s talk of a state as
committing him to any divisive metaphysics. Still, he mentions this view of Cook
Wilson’s in order to criticize it, and it is that criticism that originally intrigued me.
Prichard continues thus:
Now, so far as I can see, this account of what an action is, though plausible and
having as a truth underlying it that usually in acting we do cause something, is
not tenable.
Unquestionably the thing meant by ‘an action’ is an activity . . . And if we ask
ourselves: ‘Is there such an activity as originating or causing a change in
something else?’, we have to answer that there is not. To say this is, of course,
¹ References to Prichard’s paper are as printed in the 2002 collection edited by J. MacAdam, Moral
Writings: H. A. Prichard.
396    
not to say that there is no such thing as causing something, but only to say that
though the causing a change may require an activity, it is not itself an activity.
(2002: 272–3)
Now in a way I should be sympathetic to this complaint, because it looks as if

Prichard is here supporting the view I have already propounded, namely that
causing is too thin a relation to be an activity. Still, I am here concerned with the
rights and wrongs of this issue, and with those of another remark of Prichard’s in
his paper, together with something that Alvarez and Hyman say on a related topic.
Prichard, having announced that, though the causing a change may require an
activity, it is not itself an activity, still wants to know what activity is required for
causing a change, and he decides that this activity is that of willing. And he
suggests without further ado that:
Where we have willed some movement of our body and think we have caused it,
we cannot have directly caused it. For what we directly caused, if anything, must
have been some change in our brain. (2002: 277)
So it is by willing that we cause the changes we cause, and although we may

thereby manage to cause other changes such as movements of the body, the
changes we cause directly are all changes in our brain—neural changes, as I will
say. This second suggestion is the topic of the second half of my paper. My interest
in it derives from a remark on the same topic by Alvarez and Hyman:
We do not wish to deny that when an agent causes a bodily movement, an

event—perhaps in the agent’s nervous system—causes the same bodily move-
ment; but the agent’s action is not identical with this event. (1998: 230)
I found this remark very puzzling. It seems to allow that the bodily movement can
have two causes, one a neural event—as one event causing another—and one that
is not an event at all but an agent. And the question is whether this suggestion is
both compulsory and coherent. If it is indeed coherent, then it doesn’t matter
whether it is compulsory; but if it is both compulsory and incoherent, we agent-
causalists are in trouble.
The general topic of this paper, then, is how to capture the role of neural events
in one’s account of an action as the causing of a change by an agent. And there is a
question along the way, namely: is there something that drives us to the view that
all our basic activities are acts of the will? Prichard supposes that when one causes
a change, there must be some activity by which one does so, and that this activity
can only be willing. So when I raise my hand, my raising of my hand is my causing
my hand to rise, and such a causing is not itself an activity, but requires another
activity—which he announces can only be an act of the will. Evidently we are not
     397
supposed to press the question how I do that act. We would have thought that
I cannot ‘just cause’ the relevant volitional change in me, because there is ‘no such
activity as originating or causing a change’. Prichard’s own suggestion, that what
we ‘directly cause’ is a change in our brain, seems to be an account of what sort of a
change we can ‘just cause’, since ‘directly causing’ seems hard to separate from
‘just causing’.
There is one terminological issue that needs to be brought out here. I return to
Prichard’s remark first quoted above: ‘Where we have willed some movement of
our body and think we have caused it, we cannot have directly caused it.’ To
understand this remark, we need to disambiguate the potentially ambiguous
phrase ‘some movement of our body’. In the now standard way that we owe to
Jennifer Hornsby, we need to distinguish between transitive and intransitive
senses of this phrase. When I move my body, I cause my body to move. The
first occurrence of ‘move’ in this sentence is transitive, and the second is intransi-
tive. When I move my body, I cause a change in the location of my body; this is
transitive moving. When my body moves, its location changes, and nothing is said
about what causes that change; this is intransitive moving, like the moving of
the moon—mere motion. We say that when I move my body, I cause my body
to move, and this does not mean that when I move my body, I cause my body to
move itself. It only means that I cause a motion of my body. Though my body then
moves, it does not cause that motion—I do. To render this distinction perspicuous
as we go along, we mark it by subscripts: when I move my body, there is me, the
mover, and the body, the thing moved, one thing moving another and this is
transitive moving—bodily movementsT—where the subscript T stands for ‘tran-
sitive’. When the body moves—perhaps because I move it—we are thinking of
bodily movementsI—with the subscript I standing for ‘intransitive’. I will also use
the phrase ‘bodily motion’ for this intransitive sense. So when I move my body,
I cause a bodily motion. Now in which sense should we understand Prichard’s
remark about willing some movement? Am I willing a bodily motion to occur, or
willing to move my body? The point of this question is to raise doubt about the
idea that when I move my body, I do it either by willing some bodily motion or by
willing to move my body. Neither of these things seems to be what I do. I do not
will (one act) to move my body (another act), nor do I will a mere event (a change
in my body’s location). We seem to need a third possibility; and if none can be
found this casts doubt on the whole idea of willing a bodily movement.
Bruce Aune writes: ‘[T]he principle behind Prichard’s claim is . . . “Statements of

the form ‘Thing or person S causes event E’ are to be understood as elliptical
for statements of the form ∃A(A is an activity by S and A causes E)” ’ (1974: 98).
398    
But Prichard does not challenge the idea that agents cause changes in favour of the
view that causing is always done by activities. He only wants to identify the activity
by which agents cause the changes they cause, because he thinks that, though there
is such a thing as causing a change, still causing a change, though it requires an
activity, is not itself an activity. The activity we are looking for is willing.
But we might wonder why Prichard should suppose that causing a change
requires the same activity each time. For he could be right in his claim that causing
a change is not an activity, but requires an activity, but wrong to suppose that it
must be the same one each time. Perhaps I cause my arm to rise by raising it, and
cause my hand to turn (when opening the door) by turning it. So I am suspicious
of Prichard’s apparent assumption that whenever we cause a change, there is one
and the same activity involved each time. If it were, then perhaps the only
candidate activity would be that of willing. But we need reason to believe this
assumption.
What is more, Prichard himself is wary of such assumptions elsewhere. He
famously compared the question ‘Why should we obey the law?’ with the question
‘Why should we read books?’ (2002: 252). It is clear that there are lots of different
reasons to read books, depending on the book (and on us); and similarly there are
different reasons to obey the law, depending on the law and the circumstances
(and on us, again). So why should there not be different activities involved in the
causings of different changes? I see no reason why.
So a Prichardian response to Prichard’s challenge is to allow that there is indeed
no activity of causing a change, but to insist that every causing of a change is an
activity of some (other) sort, and every activity is the causing of a change. So:
walking is causing a change by walking; it is an activity by doing which the agent
causes a change in her location. Writing is causing a change by writing: and so on.
So the general form of action is: causing a change in one way or another. This
reminds me of Jonathan Lowe’s suggestion (2008: ch. 7.3) for the basic form of an
agent-causal statement: Agent A caused event E by V-ing.
Could this be what Prichard meant by saying that though causing requires an
activity, it is not an activity? No, because for him it must be the same activity every
time. Lowe, by contrast, is happy—so far as the form of the causal statement
goes—to allow that it might be a different V-ing for each causing. He argues,
however, that all our basic actions are acts of will. A basic action can be under-
stood for these purposes as a doing which is not done by doing something else. So
I draw your attention by causing my arm to rise; and I cause my arm to rise by
raising it (rather than in some other way). I raise it by an act of will. (I do not
recommend this train of thought, and mention it only because of its relevance to
what is to come later.)
Lowe’s account of the form of an agent-causal statement allows that nothing
can just cause anything. If an action is the causing of a change by an agent, there
must be some way in which the agent causes that change. I cause my arm to rise by
     399
raising it—that is one way—but I could also do it by lifting it with the other arm,
or by getting you to lift it. It is the way in which one does it that gives the causing
its substance, its thickness (to revert to the idea that causing is itself a thin
relation).
This returns us to the idea that causings are not events. If I write, I cause a
change by writing, and the change caused is an event, but the causing of it is not.
Causal relations do not have enough substance, enough thickness, for their
instantiations to be events. So as an agent I cannot just exert a causal power and
get something to happen. I have to act in a certain way if I am to cause a change.
But whether this is compatible with the claim that I can just will, and that by
willing I can do other things, such as raise my arm, is dubious.
If, when I cause a change, I do so by engaging in some activity, are we to say that
it is the activity that is the real cause of that change? This is not the way the story is
supposed to go. The ground idea is that the cause of the change is the agent, and
the activity involved is the agent’s causing that change in a certain way. Is there
some danger that this response to Prichard generates a regress? I doubt it. If it
does, the regress will derive from two thoughts: that one causes B by doing
something else A, and that all doings are themselves causings. But the view
above does not maintain that one causes B by doing something else A. It just
holds that every change caused is caused in a certain way.
Furthermore, if there were a genuine regress here, how is it that willing would
be able to stop it when nothing else can? Willing is an activity (if it happens at all),
and there is no activity by doing which one wills. But why should one not say the
same about causing one’s fingers to move? If I can just will, why cannot I just move
my fingers? To move one’s fingers is to cause one’s fingers to move by acting in a
certain way, namely, by moving them. If we point out that there are various ways
of causing one’s fingers to move, the answer is that in this case one did not move
one’s finger by doing something else of which the motion of the finger is an effect.
So understood, one’s moving one’s finger is a basic action.
The notion of a basic action which I used in the previous sentence is taken from
Alvarez and Hyman. More fully expressed, a basic action is the causing of a change
whose result is not the effect of some other action by the same agent. (Remember
that the ‘result’ of the action is the change of which the action is the causing.) An
ordinary bodily movementI, or motion, can be the result of a bodily movementT,
and such movementsT are basic actions. An ordinary bodily movementT consists
in the movingT of the body by the agent.
As far as this goes, we don’t need to introduce considerations of neurophysi-
ology. But Alvarez and Hyman complicate things enormously by maintaining (or
400    
perhaps one should simply say pointing out) that on their story there are two
causal relations, event-causation and agent-causation, and one and the same event
can be caused both by an agent and by a prior event. If we reserve the term ‘causer’
for an agent-cause, and restrict the noun ‘cause’ to event-causes, a bodily motion
can have both a cause (perhaps neural) and a causer (agent). If that bodily motion
is the result of the agent’s causing (that is, the result of the exercise of a causal
power by that agent) and not the effect of some other action by the same agent, it
is a basic action of that agent’s.
Now when I move my body, the relevant bodily motion is an effect of neural
events. If I cause those events, my causing those events would be my basic actions.
And if I don’t cause those events, how is it that I can be the causer of their effects?
I seem to be unnecessary, and we would be forced to conclude that no bodily
movementT is a basic action.
We might optimistically suppose that the answer to this is that the two causal
relations, event-causation and agent-causation, are so distinct that they do not
compete. If they do not compete, the fact that our bodily motions are the effects of
neural changes in our brain is not in tension with the fact that they are the results
of our causings. The latter causings are ones done by causers rather than by causes.
So we can have both causal relations in play simultaneously without either
interfering with the other. We do not need to deny that those bodily motions
that are the results of our agency are also the effects of neural changes in our
brains. Those motions can be our doings (they are not our actions, of course, but
the results of those actions) even though they have causes other than us.
I fear, however, that this suggestion has only to be understood to be rejected.
It renders agent-causation epiphenomenal. As it represents things, there is the
causal system (the event-causal system, that is) going on remorselessly in its
own sweet way, and we somehow plaster agent-causation on top of that, to get
ourselves into the story as agents. But we might reasonably notice that the bodily
motion is going to happen anyway, once the relevant neural changes are under
way, and there remains nothing really for us to do. Our role as causers is mere
window-dressing.
Matters would be different if we could claim to be the causers of the neural
changes whose effects are the relevant bodily motions. But if we were, those
causings of neural changes would be our basic acts, for the bodily motions are
the effects of those neural changes, and though we can be said to be the causers of
events that are the effects of the results of other causings of ours, our basic acts will
be the causings of neural changes.
I do not want to raise it as an objection to this picture, that our neural changes
have other neural events as causes. This is true, but if we are to retain any sense of
ourselves as agents, rather than as the conduits of extraneous causal forces, we
have to find some point at which we do something—and, I would say, at which we
do something for a reason.
     401
Consider, in this connection, what Alvarez and Hyman have to say. They allow
that the bodily motions that are the results of my basic actions are also the effects
of certain neural events in my brain. And they allow that I do not cause the neural
events to occur. What I do, according to them, is ensure that those neural events
have occurred by performing an action whose result is sure to be the effect of such
a neural event. But what does cause them then? Three answers are possible:
nothing, me, and something else. If something else, we seem to be locked into
the causal nexus. The only way out of this impasse is to allow a plurality of causal
relations, so that though an event is already perfectly well explained causally, there
is room for a second causal explanation of that same event, but of a different sort.
One might hopefully suppose that since the two causal explanations at issue are of
very different sorts, there can be no competition between them. The technical
term for this sort of approach is insouciance.
But even the insouciant might prefer to tie the two causal relations together in
some way, if that were only possible. So the suggestion I am going to consider is
that we think of the agent as causing a causal process, or a series of causally linked
events: the neural at one end causing the overt physical motion at the other, with
the muscular in between. On this account, the basic action is not well understood
as the causing of a bodily motion such as the rising of an arm; it is the causing of a
process of which that motion is the terminal point.
What we cause when we act, on this account, is not a sort of causing to cause.
We do not cause neural events to cause muscular ones, and we do not cause those
muscular ones to cause arms to go up. These things are not up to us. What we
cause is a causal complex, and we cause every part of that complex, but we do not
cause the causal relations that bind it together. It is one thing to cause an event to
occur, one that will have certain effects, and another to endow that event with
causal powers. Similarly, it is one thing to cause a process to occur, and another to
endow some element of that process with causal powers.
An immediate worry about this proposal is that it thinks of the agent as causing
a process, when there are parts of that process of which the agent is more or less
entirely ignorant. In fact, the only thing the agent really knows about is the final
stage. But this epistemological reason for focusing on that final stage and thinking
of the basic action in such terms applies just as well to the muscular as to the
neural. We already knew that we are largely ignorant of the way in which we get
our arm to rise. That the same applies to the neural events which are such an
important part of the process should not disconcert us. The focus on the final
stage had already to be abandoned, despite the fact that the sort of agentual
knowledge that Elizabeth Anscombe tried so hard to understand (1957: §30) is
402    
never of neural changes and only rarely of muscular ones. So what I cause is not
just, or really, the first stage of the neural/muscular/physical process, nor the last
stage, particularly, but the whole thing, of which I may know only the later stage,
and aim only at that.
One’s basic action, on this approach, is the causing of the whole process. And
this may seem to lead to immediate difficulties. Do I cause the neural events which
are the first stage in the process? If so, how do I do that? The answer has to be that
at the most I did that by causing the process of which those neural events were the
first and the bodily motion the last stage. I suppose that one could try to suggest
that I cause whatever neural changes are required for my hand to rise. But I prefer
to stick to the letter of the ‘process view’ and maintain that I cause the whole
process, and that my relation to that process is not identical to my relation to any
of its stages. It is outside in rather than inside out.
Still, there is enough awkwardness about this for us to scout for possible
alternatives. A different account might have it that the occurrence of the neural
events is part of the story of how I cause the bodily motions. Not everything that is
part of that story is itself a doing, or a causing, of mine, or the result of such a
causing. This supposes that my action is really the causing of the bodily motion,
and that is what I am trying to deny; so this new account is definitely different. But
it still faces the familiar problem, which is that if I don’t cause the neural changes,
something else does, and this locks us into the causal nexus.
A better suggestion might be that the neural events are not part of the story of
how I cause the bodily motions; they are part of the story of how those motions get
to happen. It is not about how I do it. It is about how it works that the body
movesI. After all, there is no story of how I raise my hand—I just raise it, that’s all.
But if the neural events are part of how it works, must I not also cause them, in
causing the process of which they are a part? Well, perhaps not. I do not
necessarily cause to happen everything that needs to happen if some event of
which I am the causer is to happen. I do not cause the absence of traffic jams that
is required if I am to arrive in London by lunchtime. And I do not necessarily
cause to happen everything that needs to happen if I am to cause that event—e.g.,
that I stay alive until the end of the process.
But this suggestion too runs up against the question what does cause the neural
events if I don’t. Again we face the fact that if the neural changes are not caused by
me, they must be caused by something else, and the causal nexus looms.
So it seems that the best resort is to stick to the view that I cause the process of
which the neural events are stages—a process which consists in a causal relation
between neural changes at one end and bodily motion at the other. I don’t cause
the neural events in any other sense than this—and by the same token I don’t
cause the bodily motions in any other sense than this either.
It seems to me that the way in which children learn to control their bodily
movementsI might chime rather well with the account I have been trying to
     403
promote. Very young children move around haphazardly. A process occurs at the
end of which they can coordinate hand and eye, and move intentionally. What is it
that they are learning during this process? They are not learning to make happen
those neural events that are necessary for their hand to move towards the toy in
front of them. Perhaps what they are learning is how to make happen a process
which consists in neural changes leading to bodily motions. As I have suggested
several times now, they don’t need to be aware of every stage of that process in
order to acquire the ability to execute it. And they don’t need to be aware of it as a
process either. It is a routine, but not one whose several stages will ever be present
to consciousness. Nobody can say how it is done, but it might be possible to say
what it is that we do.
Places of First Publication
1. ‘The Logical Conscience’, Analysis, vol. 37, no. 2 (January 1977), 81–4.
Section 1
2. ‘On Moral Properties’, Mind, vol. 90 (July 1981), 367–85.
3. ‘Ethical Particularism and Morally Relevant Properties’, Mind, vol. 92 (October 1983),
530–47.
4. ‘Intuitionism in Meta-Epistemology’, Philosophical Studies, vol. 42 (1982), 395–408.
5. ‘The Role of Imaginary Cases in Ethics’, Pacific Philosophical Quarterly, vol. 66, nos.
1–2 (January–April 1985), 141–53.
6. ‘Externalism for Internalists’, Philosophical Issues, vol. 2: Rationality in Epistemology
(1992), 93–114.
7. ‘The Particularist’s Progress’, in B. W. Hooker and M. Little, eds., Moral Particularism
(Oxford: Oxford University Press, 2000), 130–56.
8. ‘Necessity, Universality and the A Priori in Ethics’, in B. Puri and H. Sievers, eds.,
Reason, Morality and Beauty: Essays on the Philosophy of Immanuel Kant (Delhi:
Oxford University Press, 2007), 40–54.
Section 2
9. ‘Two Conceptions of Moral Realism’, Proceedings of the Aristotelian Society, suppl. vol.
lx (1986), 167–87.
10. ‘Contemplating One’s Nagel’, Philosophical Books, vol. 29, no. 1 (January 1988), 1–15
11. ‘In Defence of Thick Concepts’, Midwest Studies in Philosophy, vol. 20: Moral Concepts
(1995), 263–79.
12. ‘McDowell, Williams, and Intuitionism’, in U. Heuer and G. Lang, eds., Luck, Value,
and Commitment: Themes from the Ethics of Bernard Williams (Oxford: Oxford
University Press, 2012), 269–90.
13. ‘Practical Concepts’, in S. Kirchin, ed., Thick Concepts (Oxford: Oxford University
Press, 2013), 44–59.
14. ‘Should We Pass the Buck?’, in A. O’Hear, ed., The Good, the True and The Beautiful
(Cambridge: Cambridge University Press, 2000), 159–73.
Section 3
15. ‘Arguments from Illusion’, in Philosophical Quarterly, vol. 45, no. 181 (October 1995),
421–38.
16. ‘Why There Is Really No Such Thing as the Theory of Motivation’, Presidential
Address to the Aristotelian Society 1994, in Proceedings of the Aristotelian Society,
vol. xcv (1994–5), 1–18.
406    
17. ‘How to Act—Disjunctively’, in A. Haddock and F. Macpherson, eds., Disjunctivism:

Perception, Action, Knowledge (Oxford: Oxford University Press, 2008), 262–79.
18. ‘Enticing Reasons’, in R. Jay Wallace, P. Pettit, S. Scheffler, and M. Smith, eds., Reason
and Value: Themes from the Moral Philosophy of Joseph Raz (Oxford: Oxford
19. ‘On Knowing One’s Reason’, in C. Littlejohn and J. Turri, eds., Epistemic Norms
Section 4
20. ‘Prichard on Duty and Ignorance of Fact’, in P. J. Stratton-Lake, ed., Ethical

Intuitionism (Oxford: Oxford University Press, 2002), 229–47.
21. ‘Moore’s Account of Vindictive Punishment: A Test Case for Theories of Organic
Unities’, in S. Nuccetelli and G. Seay, eds., Themes from G. E. Moore: New Essays in
Epistemology and Ethics (Oxford: Oxford University Press, 2007), 325–42.
22. ‘Has Anyone Ever Been a Non-Intuitionist?’, in T. Hurka, ed., Underivative Duties:
British Moral Philosophers from Sidgwick to Ewing (Oxford: Oxford University Press,
2011), 87–105.
23. ‘More Right than Wrong’, in R. Johnson and M. Timmons, eds., Kantian Reflections on
Morality, Law and Society: Critical Essays on the Philosophy of Thomas E. Hill Jr.
24. ‘Prichard on Causing a Change’, in A. O’Hear, ed., Philosophy of Action (Cambridge:
Cambridge University Press, 2017), 127–38.
Works Cited
Alston, W. P. (1986), ‘Internalism and Externalism in Epistemology’, Philosophical Topics,

vol. 14, no. 1, 179–221.
Alston, W. P. (1978), ‘Meta-Ethics and Meta-Epistemology’, in A. I. Goldman and J. Kim,
eds., Values and Morals (Dordrecht: Reidel), 275–97.
Alvarez, M., and Hyman, J. (1998), ‘Agents and their Actions’, Philosophy, vol. 73, 219–45.
Anscombe, E. (1957), Intention (Oxford: Basil Blackwell).
Audi, R. (1988), ‘Foundationalism, Coherentism, and Epistemological Dogmatism’, in
J. Tomberlin, ed., Philosophical Perspectives, vol. 2, Epistemology (Atascadero, CA:
Ridgeview Publishing), 407–42.
Aune, B. (1974), ‘Prichard, Action, and Volition’, Philosophical Studies, vol. 25, 97–116.
Barnes, J., ed. (1984), The Complete Works of Aristotle: The Revised Oxford Translation
(Princeton, NJ: Princeton University Press).
Bennett, J. (1995), The Act Itself (Oxford: Clarendon Press, 1995).
Blackburn, S. (1971), ‘Moral Realism’, in J. Casey, ed., Morality and Moral Reasoning
(London: Methuen), 101–24.
Blackburn, S. (1992), ‘Through Thick and Thin’, Proceedings of the Aristotelian Society,
suppl. vol. lxvi, 285–99.
Bloor, D. (1991), Knowledge and Social Imagery (Chicago, IL: University of Chicago Press).
Bonjour, L. (1980), ‘Externalist Theories of Empirical Knowledge’, in P. French et al., eds.,
Midwest Studies in Philosophy, vol. 5, Studies in Epistemology (Minneapolis: University of
Minnesota Press), 53–74.
Brink, D. (1989), Moral Realism and the Foundations of Ethics (Cambridge: Cambridge
University Press).
Broad, C. D. (1930), Five Types of Ethical Theory (London: Routledge & Kegan Paul).
Broad, C. D. (1940), ‘Critical Notice of W. D. Ross, Foundations of Ethics’, Mind, vol. 49,
228–39.
Broome, J. (1991), Weighing Goods (Oxford: Blackwell).
Broome, J. (1993), ‘Can a Humean be Moderate?’, in R. G. Frey and C. Morris, eds., Value,
Welfare and Morality (Cambridge: Cambridge University Press), 51–73.
Broome, J. (2000), ‘Normative Requirements’, in J. Dancy, ed., Normativity (Oxford:
Blackwell), 78–99.
Carritt, E. F. (1928), The Theory of Morals (Oxford: Oxford University Press).
Carritt, E. F. (1947), Ethical and Political Thinking (Oxford: Clarendon Press).
Cartwright, N. (1983), How the Laws of Physics Lie (Oxford: Clarendon Press).
Chisholm, R. M. (1964), ‘The Ethics of Requirement’, American Philosophical Quarterly,
vol. 1, 147–55.
Chisholm, R. M. (1977), Theory of Knowledge, 2nd ed. (Englewood Cliffs, NJ: Prentice Hall
Inc.).
Crane, T. (2005), ‘The Problem of Perception’, in E. N Zalta, ed., Stanford Encyclopedia of
Philosophy (Spring) http://plato.stanford.edu/archives/spr2005/entries/perception-problem/.
Crisp, R. (1996), ‘The Dualism of Practical Reason’, Proceedings of the Aristotelian Society,
vol. xcvi, 53–73.
408  
Dancy, J. (1977), ‘The Logical Conscience’, Analysis, vol. 37, pp. 81–4; repr. in this volume,
15–18.
Dancy, J. (1981), ‘On Moral Properties’, Mind, vol. 90, 367–85; repr. in this volume, 21–36.
Dancy, J. (1983), ‘Ethical Particularism and Morally Relevant Properties’, Mind, vol. 92,
530–47; repr. in this volume, 37–50.
Dancy, J. (1985), An Introduction to Contemporary Epistemology (Oxford: Basil Blackwell).
Dancy, J. (1987), Berkeley: An Introduction (Oxford: Basil Blackwell).
Dancy, J. (1988), ‘Supererogation and Moral Realism’, in J. Dancy, J. Moravcsik, and
C. C. W. Taylor, eds., Human Agency: Language, Duty, and Value (Stanford, CA:
Stanford University Press), 170–88.
Dancy, J. (1993), Moral Reasons (Oxford: Blackwell).
Dancy, J. (1995a), ‘Arguments from Illusion’, The Philosophical Quarterly, vol. 45, no. 181,
421–38; repr. in this volume, 229–46.
Dancy, J. (1995b), ‘In Defense of Thick Concepts’, Midwest Studies in Philosophy, vol. 20,
Moral Concepts, 263–79; repr. in this volume, 160–77.
Dancy, J. (1995c), ‘ “For Here the Author is Annihilated”: Reflections on Philosophical
Aspects of the Use of the Dialogue Form in Hume’s Dialogues Concerning Natural
Religion’, Proceedings of the British Academy, vol. 85, 29–60; repr. in T. Smiley, ed.,
Philosophical Dialogues: Plato, Hume, Wittgenstein (Oxford: Oxford University Press),
29–60.
Dancy, J. (1995d), ‘New Truths in Proust?’, Modern Language Review, vol. 90, no. 1, 18–28.
Dancy, J. (1997a), ‘Parfit and Self-Defeating Theories’, in Dancy, ed., Reading Parfit
(Oxford: Blackwell), 1–23.
Dancy, J., ed. (1997b), Reading Parfit (Oxford: Blackwell).
Dancy, J. (1998), ‘Wiggins and Ross’, Utilitas, vol. 10, no. 3, 281–5.
Dancy, J. (2000a), Practical Reality (Oxford: Oxford University Press).
Dancy, J. (2000b), ‘Should We Pass the Buck?’, in A. O’Hear, ed., The Good, the True and
The Beautiful (Cambridge: Cambridge University Press), pp. 59–73; repr. in this volume,
213–25.
Dancy, J. (2002), ‘Prichard on Duty and Ignorance of Fact’, in P. J. Stratton-Lake, ed.,
Ethical Intuitionism: Re-evaluations (Oxford: Clarendon Press), 229–47; repr. in this
volume, 321–37.
Dancy, J. (2004), Ethics Without Principles (Oxford: Oxford University Press).
Dancy, J. (2009), ‘Reasons and Rationality’, in S. Robertson, ed., Spheres of Reason (Oxford:
Oxford University Press), 93–113.
Dancy, J. (2011), ‘Acting in Ignorance’, Frontiers of Philosophy in China, vol. 6, no. 3,
345–57.
Dancy, J. (2012), ‘McDowell, Williams, and Intuitionism’, in U. Heuer and G. Lang, eds.,
Luck, Value, and Commitment: Themes from the Ethics of Bernard Williams (Oxford:
Oxford University Press), 269–90; repr. in this volume, 178–96.
Dancy, J. (2015), ‘Ryle and Strawson on Category Mistakes’, in D. Dolby, ed., Ryle on Mind
and Language (London: Palgrave Macmillan), 9–25.
Dancy, J. (2018), Practical Shape: A Theory of Practical Reasoning (Oxford: Oxford
University Press).
Dancy, J. (2020), ‘Parfit, Derek, 1942–2017’, Biographical Memoirs of Fellows of the British
Academy, vol. xix, 37–57.
Dancy, J. (2021), ‘Three for the Price of Two’, in R. Chang and K. Sylvan, eds., The
Routledge Handbook on Practical Reason (London: Routledge), 214–25.
Dancy, J., and Sosa, E., eds. (1991), A Companion to Epistemology (Oxford: Basil Blackwell).
  409
Darwall, S., Gibbard, A., and Railton, P. (1992), ‘Toward Fin de Siècle Ethics: Some
Themes’, The Philosophical Review, vol. 101, 115–89.
Davidson, D. (1967), ‘Truth and Meaning’, Synthèse, vol. 17, 304–23; repr. in Davidson,
Inquiries into Truth and Interpretation (Oxford: Clarendon Press, 1984), 17–36.
Davis, W. A. (2003), ‘Psychologism and Humeanism’, Philosophy and Phenomenological
Research, vol. 67, no. 2, 452–9.
Evans, G. (1980), ‘Things Without the Mind’, in Z. van Straaten, ed., Philosophical Subjects
(Oxford: Clarendon Press), 76–116.
Ewing, A. C. (1929), The Morality of Punishment (London: Kegan Paul).
Ewing, A. C. (1947), The Definition of Good (London: Macmillan).
Ewing, A. C. (1953), Ethics (Teach Yourself Series).
Ewing, A. C. (1959), Second Thoughts in Moral Philosophy (London: Routledge & Kegan
Paul).
Foot, P. (1978), Virtues and Vices (Oxford: Blackwell).
Foot, P. (1995), ‘Does Moral Subjectivism Rest on a Mistake?’, Oxford Journal of Legal
Studies, vol. 15, no. 1, 1–14.
Frankena, W. (1958), ‘Obligation and Motivation in Recent Moral Philosophy’, in
A. I. Melden, ed., Essays in Moral Philosophy (Seattle: University of Washington
Press), 40–81.
Frankena, W. K. (1963), Ethics (Englewood Cliffs, NJ: Prentice-Hall).
Gay, R. (1985), ‘Ethical Pluralism: A Reply to Dancy’, Mind, vol. 94, 250–62.
Geertz, C. (1973), ‘Thick Description: Toward an Interpretative Theory of Culture’, in
Geertz, The Interpretation of Cultures: Selected Essays (London: Fontana).
Gibbard, A. (1992), ‘Thick Concepts and Warrant for Feelings’, Proceedings of the
Aristotelian Society, suppl. vol. lxvi, 267–83.
Goldman, A. I. (1980), ‘The Internalist Conception of Justification’, in P. French et al., eds.,
Midwest Studies in Philosophy, vol. 5, Studies in Epistemology (Minneapolis: University of
Minnesota Press), 27–52.
Goldman, A. I. (1979), ‘What is Justified Belief?’, in G. Pappas, ed., Justification and
Knowledge (Reidel: Dordrecht), 1–23.
Hampton, J. (1998), The Authority of Reason (Cambridge: Cambridge University Press).
Hare, R. M. (1952), The Language of Morals (Oxford: Clarendon Press).
Hare, R. M. (1957), ‘Geach: Good and Evil’, Analysis, vol. 17, no. 5, 103–11.
Hare, R. M. (1963), Freedom and Reason (Oxford: Clarendon Press).
Hawthorne, J., and Stanley, J. (2002), ‘Knowledge and Action’, Journal of Philosophy, vol.
105, no. 10, 571–90.
Hieronymi, P. (2011), ‘Reasons for Action’, Proceedings of the Aristotelian Society, vol. cix,
407–27.
Hill, T. E. (2007), ‘The Importance of Moral Rules and Principles’, Annual Lindley Lecture,
University of Kansas; repr. in Hill, Virtue, Rules, and Justice: Kantian Aspirations
(Oxford: Oxford University Press, 2012).
Hintikka, J. (1969), ‘Deontic Logic and its Philosophical Morals’, in Hintikka, Models for
Modalities (Dordrecht: Reidel), 184–214.
Hinton, J. M. (1973), Experiences (Oxford: Clarendon Press).
Hobbes, T. (1839), Computation or Logic, Part 1, 6.10, in W. Molesworth, ed., The English
Works of Thomas Hobbes (London: John Bohn).
Hornsby, J. (2008), ‘A Disjunctive Conception of Acting for Reasons’, in A. Haddock and
F. Macpherson, eds., Disjunctivism: Perception, Action, Knowledge (Oxford: Oxford
University Press), 244–61.
410  
Hurka, T. (2003), ‘Moore in the Middle’, Ethics, vol. 113, 599–628.

Hurley, S. L. (1985), ‘Objectivity and Disagreement’, in T. Honderich, ed., Morality and
Objectivity (London: Routledge & Kegan Paul), 54–97.
Hurley, S. L. (1989), Natural Reasons (Oxford: Oxford University Press).
Hyman, J. (1999), ‘How Knowledge Works’, Philosophical Quarterly, vol. 49, 421–39.
Hyman, J. (2006), ‘Knowledge and Evidence’, Mind, vol. 115, 891–916.
Hyman, J. (2010), ‘The Road to Larissa’, Ratio, vol. 23, 393–414.
Hyman, J. (2011), ‘Acting for Reasons: Reply to Dancy’, in Frontiers of Philosophy in China,
vol. 6, no. 3, 358–68.
Jackson, F. (1977), Perception (Cambridge: Cambridge University Press).
Johnson, E. (1921), Logic (Cambridge: Cambridge University Press.
Joseph, H. W. B. (1931), Some Problems in Ethics (Oxford: Clarendon Press).
Kant, I. (1785), Groundwork of the Metaphysic of Morals, trans. M. Gregor (Cambridge:
Cambridge University Press, 1996).
Kovesi, J. (1967), Moral Notions (London: Routledge & Kegan Paul).
Lance, M., and Little, M. (2004), ‘Defeasibility and the Normative Grasp of Context’,
Erkenntnis, vol. 61, 435–55.
Lance, M., and Little, M. (2007), ‘From Particularism to Defeasibility in Ethics’, in
M. Lance, M. Potrc, and V. Strahnovik, eds., Challenging Moral Particularism
(London: Routledge), 53–74.
Lemos, N. (1994), Intrinsic Value: Concept and Warrant (Cambridge: Cambridge
University Press).
Little, M., and Macnamara, C. (2021), ‘Non-requiring Reasons’, in R. Chang and K. Sylvan,
eds., The Routledge Handbook on Practical Reason (London: Routledge), 393–404.
Littlejohn, C. (2012), Justification and the Truth Connection (Cambridge: Cambridge
University Press).
Lovibond, S. (1983), Realism and Imagination in Ethics (Oxford: Blackwell).
Lowe, J. (2008), Personal Agency (Oxford: Oxford University Press).
Mackie, J. L. (1977), Ethics: Inventing Right and Wrong (Harmondsworth: Penguin Books).
Martin, M. G. F. (2004), ‘The Limits of Self-Awareness’, Philosophical Studies, vol. 103,
37–89.
McDowell, J. (1979), ‘Virtue and Reason’, The Monist, vol. 62, 331–50; repr. in McDowell,
Mind, Value, and Reality (Cambridge, MA: Harvard University Press, 1998), 50–73.
McDowell, J. (1981), ‘Non-Cognitivism and Rule-Following’, in S. Holzman and C. Leich,
eds., Wittgenstein: To Follow a Rule (London: Routledge), 141–62; repr. in McDowell
Mind, Value, and Reality (Cambridge, MA: Harvard University Press, 1998), 198–218.
McDowell, J. (1982), ‘Criteria, Defeasibility, and Knowledge’, Proceedings of the British
Academy, vol. lxvii, 455–79; repr. in McDowell, Meaning, Knowledge and Reality
(Cambridge, MA: Harvard University Press, 1998), 369–94.
McDowell, J. (1985), ‘Values and Secondary Qualities’, in T. Honderich, ed., Morality and
Objectivity (London: Routledge & Kegan Paul), 110–29; repr. in McDowell, Mind, Value,
and Reality (Cambridge, MA: Harvard University Press, 1998), 131–50.
McDowell, J. (1987), ‘Projection and Truth in Ethics’, Annual Lindley Lecture, University of
Kansas; repr. in McDowell, Mind, Value, and Reality (Cambridge, MA: Harvard
McDowell, J. (1998), Mind, Value, and Reality (Cambridge, MA: Harvard University Press).
McDowell, J. (2013), ‘Acting in the Light of a Fact’, in D. Bakhurst, B. Hooker, and M. Little,
eds., Thinking About Reasons: Themes from the Philosophy of Jonathan Dancy (Oxford:
  411
McGinn, C. (1983), The Subjective View (Oxford: Clarendon Press).

McNaughton, D. (1996), ‘An Unconnected Heap of Duties?’ Philosophical Quarterly, vol.
46, no. 185, 433–47.
Moore, G. E. (1903), Principia Ethica (Cambridge: Cambridge University Press).
Moore, G. E. (1912), Ethics (Cambridge: Cambridge University Press).
Moore, G. E. (1922), ‘The Conception of Intrinsic Value’, in Moore, Philosophical Studies
(London: Routledge & Kegan Paul), 253–75.
Moore, G. E. (1993), Principia Ethica, rev. ed., ed. T. Baldwin (Cambridge: Cambridge
University Press).
Moya, C. J. (1990), The Philosophy of Action (Cambridge: Polity).
Nagel, T. (1980), ‘The Limits of Objectivity’, in S. M. Macmurrin, ed., Tanner Lectures on
Human Values, 1 (Cambridge: Cambridge University Press).
Nagel, T. (1986), The View from Nowhere (Oxford: Oxford University Press).
O’Shaughnessy, B. (1980), The Will, 2 vols. (Cambridge: Cambridge University Press).
Parfit, D. (1984), Reasons and Persons (Oxford: Clarendon Press).
Parfit, D. (1997), ‘Reasons and Motivation’, Proceedings of the Aristotelian Society, suppl.
vol. lxxi, 99–130.
Parfit, D. (2011), On What Matters (Oxford: Clarendon Press).
Pollock, J. (1971), ‘Perceptual Knowledge’, Philosophical Review, vol. 80, 287–319.
Price A. W. (2008), Contextuality in Practical Reason (Oxford: Clarendon Press).
Prichard, H. A. (1945), ‘Acting, Willing, Desiring’, in Moral Writings: H. A. Prichard, ed.
J. MacAdam (Oxford: Oxford University Press, 2002), 272–81.
Prichard, H. A. (1912), ‘Does Moral Philosophy Rest on a Mistake?’ Mind, vol. 21, 21–37;
repr. in Moral Writings: H. A. Prichard, ed. J. MacAdam (Oxford: Oxford University
Press, 2002), 7–20.
Prichard, H. A. (1932), ‘Duty and Ignorance of Fact’, Proceedings of the British Academy;
repr. in Moral Writings: H. A. Prichard, ed. J. MacAdam (Oxford: Oxford University
Press, 2002), 84–101.
Prichard, H. A. (1949), Moral Obligation, ed. W. D. Ross (Oxford: Clarendon Press).
Prichard, H. A. (2002), Moral Writings: H. A. Prichard, ed. J. MacAdam (Oxford: Oxford
University Press).
Rashdall, H. (1907), The Theory of Good and Evil (Oxford: Clarendon Press).
Raz, J. (2000), Engaging Reason: On the Theory of Value and Action (Oxford: Oxford
University Press).
Resnik, M. (1993), Choices (Minneapolis: University of Minnesota Press, 1993).
Ross, W. D. (1930), The Right and the Good (Oxford: Clarendon Press); repr. with original
pagination in The Right and the Good, ed. P. J. Stratton-Lake (Oxford: Oxford University
Press, 2003).
Ross, W. D. (1939), Foundations of Ethics (Oxford: Clarendon Press).
Sandis, C. (2012), The Things We Do and Why We Do Them (London: Palgrave
Macmillan).
Scanlon, T. M. (1998), What We Owe to Each Other (Cambridge, MA: Harvard University
Press,).
Scheffler, S. (1987), ‘Morality through Thick and Thin: A Critical Notice of Ethics and the
Limits of Philosophy’, Philosophical Review, vol. xcvi, 411–34.
Sehon, S. (2005), Teleological Realism (Cambridge, MA: MIT Press).
Sidgwick, H. (1907), The Methods of Ethics, 7th ed. (London: Macmillan).
Singer, M. (1967), Generalization in Ethics (New York: Knopf).
Slote, M. (1983), Goods and Virtues (Oxford: Clarendon Press).
412  
Smith, M. (1984), The Moral Problem (Oxford: Blackwell).

Smith, M. (1993), ‘Objectivity and Moral Realism: On the Significance of the
Phenomenology of Moral Experience’, in J. Haldane and C. Wright, eds., Reality,
Representation and Projection (Oxford: Oxford University Press), 235–56.
Smith, M. (2013), ‘On the Nature and Significance of the Distinction between Thick and
Thin Ethical Concepts’, in S. Kirchin, ed., Thick Concepts (Oxford: Oxford University
Press), 97–120.
Snowdon, P. (1981), ‘Perception, Vision, and Causation’, Proceedings of the Aristotelian
Society, vol. lxxxi, 175–92.
Stevenson, C. L. (1944), Ethics and Language (New Haven, CT: Yale University Press).
Stratton-Lake, P. J. (2000), Kant, Duty, and Moral Worth (London: Routledge).
Stratton-Lake, P. J., ed. (2002), Ethical Intuitionism: Re-Evaluations (Oxford: Clarendon
Press).
Swain, M. (1978), ‘Epistemic Defeasibility’, in G. Pappas and M. Swain, eds., Essays on
Knowledge and Justification (Ithaca, NY: Cornell University Press), 160–83.
Swinburne, R. G. (1976), ‘The Objectivity of Morality’, Philosophy, vol. 51, 5–20.
Temkin, L. (1997), ‘Rethinking the Good, Moral Ideals and the Nature of Practical
Reasoning’, in J. Dancy, ed., Reading Parfit (Oxford: Blackwell), 290–345.
Thesiger, W. (1959), Arabian Sands (London: Longmans).
Unger, P. (1975), Ignorance: A Case for Scepticism (Oxford: Oxford University Press).
Wiggins, D. (1976), ‘Truth, Invention and the Meaning of Life’, Proceedings of the British
Academy, vol. 62, 331–78; repr. (revised) in Wiggins, Needs, Values, Truth (Oxford:
Oxford University Press, 1987), 87–138.
Wiggins, D. (1978), ‘Deliberation and Practical Reasoning’, in J. Raz, ed., Practical
Reasoning (Oxford: Oxford University Press); repr. (revised) in Wiggins, Needs,
Values, Truth (Oxford: Oxford University Press, 1987), 215–38.
Wiggins, D. (1987), ‘A Sensible Subjectivism’, in Wiggins, Needs, Values, Truth (Oxford:
Wiggins, D. (1998), ‘The Right and The Good and W. D. Ross’s Critique of
Consequentialism’, Utilitas, vol. 10, no. 3, 261–80.
Williams, B. A. O. (1981), ‘Internal and External Reasons’, in Williams, Moral Luck
(Cambridge: Cambridge University Press), 101–13.
Williams, B. A. O. (1985), Ethics and the Limits of Philosophy (London: Fontana).
Williams, B. A. O. (1988), ‘What Does Intuitionism Imply?’, in J. Dancy, J. Moravcsik, and
C. C. W. Taylor, eds., Human Agency: Language, Duty, and Value (Stanford, CA:
Stanford University Press), 189–98; repr. in Williams, Making Sense of Humanity
(Cambridge: Cambridge University Press, 1995), 182–91.
Williams, B. A. O. (1995), Making Sense of Humanity (Cambridge: Cambridge University
Press).
Williamson, T. (2000), Knowledge and Its Limits (Oxford: Clarendon Press).
Wilson, D., and Sperber, D. (2002), ‘Relevance Theory’, in G. Ward and L. Horn, eds.,
Handbook of Pragmatics (Oxford: Blackwell, 2002), 607–32.
Wittgenstein, L. (1969), On Certainty (Oxford: Blackwell).
Index
For the benefit of digital users, indexed terms that span two pages (e.g., 52–3) may, on occasion, appear
on only one of those pages.
a priori knowledge, see knowledge, moral Brink, D. 248, 254–5

absolute conception ch. 12 passim esp. 190 Broad, C. D. 43, 58, 62, 321, 329, 376n.2
act vs. agent ch. 6 passim esp. 80ff., 238ff. Broome, J. 108n.14, 281, 289, 298n.10
acting in the light of a fact 13ff., ch. 19 passim buck-passing conception
acting for a reason ch. 16 passim of the deontic 389ff.
and knowledge 13ff., ch. 19 passim of value ch. 14 passim, 343, 372n.11
action
agency, conceptions of 364–5, ch. 24 passim Carritt, E. F. 37, 321n.1
basic 399 Cartesianism in the philosophy of action 237–8,
as the causing of a causal process 402ff. 266ff.
additivity 99 Cartwright, N. 110n.17
adverbial theory 229 causal theory of knowledge, see knowledge, causal
aesthetic analogy, see colour, experience of theory of
aesthetic concepts 211 causes vs. ‘causers’ 400
aesthetic value 180n.5 causing
agent-causalism ch. 24 passim as an event? 394
agent-causation 395 as an explanatory relation or as making
agent-relative vs. agent-neutral 98, 156–7, 223, happen? 394–5
299, 374 centralism 178, 194
Alston, W. P. ch. 4 passim, 78ff. chess (example) 35
alternatives 101ff. Chisholm, R. M. 53, 55, 56–7
Alvarez, M., and Hyman, J. 394, 399ff. claims 382–3
amusing (example) 185, 198 cognitivism in the theory of motivation 248
Anscombe, G. E. M. 362, 401 colour
anthropocentricity 130–1, 179, 182, 194 as a disposition 179ff., 188ff.
apprehension 179 experience of 136ff.
argument(s) from illusion ch. 15 passim, 257 commendation vs. recommendation 391
Aristotle 12, 175–6 common element vs. common qualities 262
attenuating, see intensifying/attenuating concept vs. conception 204ff.
Audi, R. 78 concepts, variable-attitude vs. multi-attitude 162
Aune, B. 398 consequentialism
Austin, J. L. 51, 239, 264, 306 objective vs. subjective 321–2
autonomy 154ff. vs. deontology 221
Ayer, A. J. 239 vs. intuitionism ch. 23 passim
contingency of moral truth ch. 8 passim esp.
Bakhurst, D. 162 115–18
basic belief (Raz) 282ff. Cook Wilson, J. 323, 395
Bennett, J. 113n.20 Crane, T. 262–5
Bergstrom, L. 103n.11 crime and punishment, see punishment
Berkeley, G. 5 Crisp, R. 98
Blackburn, S. 7–8, 26n.1, 59n.1, 72n.3, ch. 11
passim, 187n.13, 195, 197, 208 Davidson, D. 2, 260
Bonjour, L. 75n.1 defeaters, rebutting/undercutting 31–4
414 
degrees of duty ch. 23 passim fittingness 348, 382

Dennett, D. 253 Foot, P. 2, 110n.16, 222, 381
deontic vs. evaluative 215–17, 293–4, Frank, D. 192
302–3 Frankena, W. F. 37, 48ff., 247
Descartes, R. 155 Fricker, M. 211
description vs. evaluation ch. 11 passim, 200–4 full ordering of values 101ff., 107ff.
desire, and motivation 9
determinate vs. determinable 65–6, 137, 241, Garrard, E. 103n.11
244–5, 266 Gay, R. 379n.4
disablers, see enablers/disablers Geertz, C. 160n.1
discounting 289, 299 generalism 39–40, 46–7, 70–1, 107
disentangling argument 130–1, 200ff. and expressivism 170ff.
see also entanglers in the theory of reasons 92–4, 112–13
disjunctive acccounts generalizations 43ff., 50, 95, 147–50
of acting for a reason ch. 17 passim Gibbard, A. 7–8, 161, 173, 197
of perception 241ff. Goldman, A. I. 54, 75n.1
disjunctivism in general 8–9 good
dispositional conceptions of value, rightness, attributive vs. predicative uses of 214
colour chs. 9, 12 passim the good case 235–8
dispositions, experiences of 7, ch. 9 passim on the whole/as a whole 346ff.
domains 175ff., 199ff., 207 guides, reasons as 308ff.
Edmund (example) 277, 309ff. Hampton, J. 292ff.

Elgin Marbles (example) 63–4 Hare, R. M. 2–4, 30–1, 48–9, 60, 204, 391
enablers/disablers Hawthorne, J., and Stanley, J. 304
for explanation 109ff. Hieronymi, P. 305
for justification 87ff. Hill, T. E. 375
for value 3–4, 6, 12, 345ff. Hintikka, J. 16
vs. ground 385 Hinton, J. 8
entanglers 198ff. Hobbes, T. 317
enticing vs. peremptory reasons ch. 18 passim holism 12
see also reasons local 295
epistemology and ethics 6–7, chs. 4, 6 passim in the theory of choice 101ff.
epistemology, moral 34ff. in the theory of reasons 92ff.
essential contestability 163–4, 168, 171ff. in the theory of value 97ff.
Evans, G. 136n.4, 179, 182, 195 weak, moderate vs. extreme 91–2
evidential vs. verdictive conception 219–20, 381 Hornsby, J. 272–8, 304, 307–14
Ewing, A. C. ch. 1 passim, 98n.4, 100n.7, 215–17, Humeanism
321, 326n.6, 329, 364, 375–6 in the theory of causation 395
explanation of action in the theory of motivation ch. 16 passim
causal? 259ff. esp. 248
non-guaranteeing 110–11 Hurka, T. 345–6, 349–50, 360ff.
explanations, complete 110 Hurley, S. 100n.8, 178, 200
expressivism Hyman, J. 304–10, 314–16
and generalism 170ff.
vs. cognitivism ch. 11 passim, esp, 162ff., 168ff. identity of indiscernibles 5, 27
externalism vs. internalism imaginary cases ch. 5 passim
in epistemology and in ethics 6, ch. 5 passim independence thesis 86, 237ff.
in the theory of moral motivation 236ff. Indifference of Independent
Alternatives 101ff
fact vs. value 52ff. indistinguishability arguments ch. 15 passim
favouring relation 111ff., 255, 283, 296ff., 375 esp. 231ff
see also reasons intensifying/attenuating 6, 372
fear, fearfulness 175, 183ff., 188–9, 199–200 internalism, see externalism vs. internalism
 415
interpenetration 199 moral motivation (internalism vs.

intuitionism externalism) 180ff.
ethical 23, 35–6, 37ff., chs. 4, 20, 22, 23 passim moral principles ch. 3 passim
marks of 193ff. as reminders 71ff.
methodological vs. epistemological ch. 12 moral properties, as dispositions, observability
passim of 34ff.
vs. ethical naturalism 51ff moral relevance ch. 3 passim, 390ff.
vs. utilitarianism 37ff. moral sense theory 54ff.
intuitive induction ch. 3 passim, 117 motivation 9ff.
isolation test 379, 382 by desire or by belief? 9ff.
and justification 256–7
Joseph, H. W. B. 214 by propositions? 10
justification motivation, mine 8
of act vs. of agent ch. 6 passim motive vs. action 83–5
objective vs. subjective 79ff. Moya, C. 230ff.
justified true belief conception of knowledge 53 multi-attitude concepts 162
Murdoch, I. 2
Kamm, F. 344 ‘mustn’t-be-done’ (Parfit) 392
Kant, I. 13, 76–7, ch. 8 passim esp. 118–22, 375–6
Kirchin, S. 211 Nagel, T. 133n.2, ch. 10 passim, 254, 300n.11
knowledge, causal theory of 53ff. Neurath, O. 155
knowledge, moral neurophysiology 396ff.
a priori? 35–6, ch. 8 passim esp. 118–22 non-cognitivism 8, 51–2, 133, ch. 11 passim
objectivity of ch. 10 passim see also expressivism
Kovesi, J. 30 non-conjunctivism 241ff., 265ff., 272
no priority view 183ff., 194, 199–200, 203ff.
Laird, J. 374 normative requirements 298n.10
Lance, M., and Little, M. 198n.2 normativity 297ff.
lewdness 161–7, 174–7, 197–8, 211
Little, M., and Macnamara, C. 392–3 objectification, Hegelian vs. absolute ch 10
Littlejohn, C. 304 passim
Locke, J. 131–2, 135, 139–43, 179, 181 objective vs. subjective 79ff., chs. 9, 10 passim
Lovibond, S. 133n.2 obligation
Lowe, J. 113n.21 a characteristic of agents or of actions? 370
subjective vs. objective ch. 20 passim
Mackie, J. L. 133, 142, 185 and value 367ff.
Martin, M. 263–4, 278–9 Open Question Argument 213
McDowell, J. 2, 8, 9, 11, 13ff., 37, 41n.3, 54, 59, opportunity 374
68n.2, chs. 9, 12 passim, 135n.3, 236, 241, organic unities 24, 99ff., ch. 21 passim esp. 349ff.
245, 254, 274 intrinsicalist vs. variabilist conceptions of 338ff.
McGinn, C. 138n.6 O’Shaughnessy, B. 230ff.
McNaughton, D. 92n.1, 239n.3 ought, two senses of? ch.1 passim
Megone, C. 182n.10 oughts, objective vs. subjective 80ff., 329
meriting 142ff.
vs. eliciting 143, 166, ch. 12 passim esp. 180ff. Parfit, D. 85–7, 104n.12, 219–21, 224, 295,
mitigation 354 297n.9, 329, 376–7, 390, 392–3
monism/pluralism 38ff., 52ff. parti-resultance 65–7
Moore, G. E. 24, 51, 54ff., 58, 99ff., 213–17, vs. toti-resultance ch. 2 passim esp. 22ff., 34ff.,
ch. 21 passim 378ff., 384
on intrinsic value 339ff. see also rightness; supervenience
moral epistemology 34ff., 44ff. particularism, argument for chs. 3, 7 passim
moral imagination 62 particularist epistemology chs. 3, 4, 8 passim
moral knowledge ch. 8 passim Pavarotti (example) 168
a priori or a posteriori? 122ff. perception, theories of 229–30
416 
peremptory reasons, see enticing vs. peremptory sufficient/decisive/directive/overall/conclusive/

reasons requiring vs. variable 283–6, 292ff.
phenomenology, moral 133ff. see also favouring relation
Pickard-Cambridge, W. A. 384 recommendation vs. commendation 391–2
Platts, M. de B. 2 regret 25
pluralism, ethical 38, 47, 51–3 relevance
Pollock, J. 56 default 197–8
polyadicity of rightness and goodness 223ff. practical 197–8, 206, 209ff.
practical relevance 91 variable 197
practical shape 6 relevance, forms of, see defeaters, rebutting/
premises 309ff., 314 undercutting; enablers/disablers;
pressingness of prima facie duties 377 intensifying/attenuating; reasons
Price, A. W. 306n.1 relevant respects 31
Price, R. 41n.3 relevant similarity, principles of 73–4
Prichard, H. A. 37, 41n.3, 54, chs. 20, 22, 23, 24 resemblance between perceptual state and
passim object 136
on action 322–3, 395ff. Resnik, M. 103n.11
on right vs. ought 332ff., 384ff. response, epistemic vs. volitional 179ff.
prima facie duty resultance, see parti-resultance;
Ross’s definitions 24–6, 41ff. supervenience
in epistemology 56, 378–9 rightness
prima facie reasons 56ff. parti-resultant vs. toti-resultant
primary vs. secondary qualities ch. 9 passim ch. 2 passim
esp. 131ff. right-making and goodness ch. 14 passim
probability, objective 327–8 Ross, W. D. 3, chs. 2, 4, 23 passim, 41–8, 51, 54,
projection, projectivism 184–5 55–60, 64–5, 92, 98n.4, 117, 121, 213ff.,
property vs. attitude ch. 11 passim esp. 165–6 222–3, 321–2, 326n.6, 329, 334, 335n.12,
propositions, as what we reason from 336
and to 12 Ryle, G. 160n.1
prosperity, see wicked, prosperity of
psychologism in the theory of motivation 9ff. salience 50n.1, 375
punishment ch. 21 passim Scanlon, T. 98, 193, ch. 14 passim, 282
purport, practical and semantic 341 Scheffler, S. 160n.2
scope distinctions ch. 1 passim, 15–18,
Quine, W. V. O. 35, 251 secondary quality analogy 181ff.
sensibility theory 194ff.
Rashdall, H. 37, 48, 50 shapelessness, natural 52ff., 61, 130, 160, 163, 171
rationality and thick concepts 170ff. Sidgwick, H. 37, 39, 54, 58, ch. 22 passim
Raz, J. 282–8, 292–5 silliness 82, 93, 282, 287–91
real properties ch. 9 passim esp. 129–30 Singer, M. 30
realism Slote, M. 98
moral ch. 9 passim Smith, M. 119, 202, 203–4, 211, 250–5, 260
in the theory of motivation 258ff. Snowdon, P. 8, 136n.5, 241n.5
reasoning, theoretical/moral/practical 12ff. Sosa, D. 198
reasons Sperber, D., and Wilson, D. 124n.3
contributory vs. overall 293ff. Stevenson, C. L. 160
default 198n.2 subjective, see objective vs. subjective
and desires 93ff. supervenience 3, 5, 26ff., 32ff., 48–9, 59–60, 72,
enticing vs. peremptory ch. 18 passim esp. 297ff. 99–100, 177–8, 209, 287n.4
epistemic vs. constitutive 94 does not generate moral principles 43, 49, 59,
as guides 307ff. 72, 100
holism vs. generalism in the theory of 92–4 local vs. global 342–3
mandatory vs. permissive 293 Swain, M. 56
motivating vs. justifying ch. 16 passim Swinburne, R. G. 30
 417
Taylor, C. C. W. 1 incompatibility of 192

Temkin, L. 109n.15 instrumental 213, 340n.2
tendencies 25, 378ff. intrinsic vs. variable 99–100, 338ff.
Thesiger, W. 103 objective vs. subjective 156ff.
thick and thin concepts 7–8, chs. 11, 13 passim variability
tolerance, objective 156 of attitude 164, 167ff.
toti-resultance, see parti-resultance of relevance 198, 209, 211
transitivity 28, 101, 109 Väyrynen, P. 202
trying 230ff. verdictive vs. evidential conception 219, 381
Unger, P. 304 wicked, prosperity of 351ff.

universalizability 3, 5, 27ff. Wiggins, D. 2, 8, 50n.6, 54, 58, 59, 133n.2,
vs. supervenience 48, 60 161n.3, 163–4, 166nn.6–7, 168, 183,
Urmson, J. O. 1, 51 199–200, 205, 239, 357n.3, 385n.10
utilitarianism 37–8 will, willing 397ff.
Williams, B. A. O. 125, 160n.1, ch. 12 passim,
valence, default 198 203, 206, 208ff. 234n.2, 239n.4, 243n.6
value Williamson, T. 160n.1, 233n.1, 245n.7, 265–6,
agent-relative vs. agent-neutral 98, 121n.2, 272–3, 277, 278
223, 374 Wittgenstein, L. 37, 49, 68
holism of 97ff. world-guided vs. action-guiding 190–1

Jonathan Dancy - Practical Thought - Essays On Reason, Intuition, and Action-Oxford University Press (2021)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Jonathan Dancy - Practical Thought - Essays On Reason, Intuition, and Action-Oxford University Press (2021)

Uploaded by

Copyright:

Available Formats

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 9/6/2021, SPi

Introduction: No More Answers? 1

SECTION 1: TOWARDS PARTICULARISM

SECTION 2: MORAL METAPHYSICS

9. Two Conceptions of Moral Realism 129

SECTION 3: ACTION AND REASONS

18. Enticing Reasons 281

SECTION 4: LEARNING FROM THE INTUITIONISTS

Places of First Publication 405

Introduction: No More Answers?

be wrong to do what one believes to be wrong, so long as one is mistaken about

epistemology again—though I edited, with Ernest Sosa, the ﬁrst edition of A

This appeal to an argument whose home ground is ethics, to address an issue in

But if the perfect manifestation of practical rationality is acting in the light of a

On pp. 144–5 of his Ethics, A. C. Ewing writes:

It is a recognised principle of ethics that it is always our duty to do what we when

Ba: a believes that

(1) BaOda ! Oda

If (2) improves on our original representation of Ewing’s original intuition that

16   

(3) BaOda · ~Oda

in which case the revised Ewing formula

(4) O(BaOda!da) · BaOda · ~Oda

(α) BaOda, ~Oda, P~BaOda, O(~BaOda v da), da

   17

18   

to act according to their conscience, without validating everybody’s opinions and

We know a priori that if an action has a moral property it has it in virtue of

It is clear that it is in virtue of my thinking the act to have some other

22     

But he also writes:

If this is an argument, Ross is moving from the anti-naturalistic view that no

view is that rightness is not toti-resultant, he should by his own reasoning

24     

Murder is prima facie wrong

properties in virtue of which it is a duty proper are some in virtue of which it is

(A) An action is a prima facie duty if it is an action which tends to be a duty

26     

Even if the notion of toti-resultance gives an unsatisfactory account of the relation

(S) If a property M is supervenient upon a class of properties C, then: M is not

(S¹) If a property M is supervenient on a class of properties C, then: M is not

Let us restrict ourselves for the moment, to considering the supervenience of

¹ (S) and (S¹) are taken from Blackburn (1971: 106).

of non-moral properties. These deﬁnitions bring to light several difﬁculties with

28     

3. Resultance and universalizability

30     

Hence to give a reason in support of the judgment that a certain individual, A,

Sometimes, as in Swinburne’s case, the view I am denying is called the universal-

[A]ny singular descriptive judgment is universalisable . . . in the sense that it

I think, however, that it is possible to ﬁnd in the notion of a relevant respect

32     

The other interpretation of the phrase ‘relevantly similar’ is to be found in the

4. Moral resultance and supervenience contrasted

cannot hold a candle to

34     

established separately by the identity theory; supervenience is an addition to this

It is not an attribute that its subject is just directly perceived in experience to

36     

This paper is about the non-existence of moral principles. Its conclusion is a

38     

The utilitarian monist enjoys what is in many ways a healthy philosophical

40     