Professional Documents
Culture Documents
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification,
and Defeat
Edited by
JESSICA BROWN AND MONA SIMION
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
1
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
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
© the several contributors 2021
The moral rights of the authors 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: 2021931162
ISBN 978–0–19–884720–5
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Contents
1. Introduction 1
Jessica Brown and Mona Simion
2. The Normativity of Knowledge and the Scope and
Sources of Defeat 18
Sanford C. Goldberg
3. The Structure of Defeat: Pollock’s Evidentialism, Lackey’s
Framework, and Prospects for Reliabilism 39
Peter J. Graham and Jack C. Lyons
4. Losing Knowledge by Thinking about Thinking 69
Jennifer Nagel
5. Dispositional Evaluations and Defeat 93
Maria Lasonen-Aarnio
6. Suspension, Higher-Order Evidence, and Defeat 116
Errol Lord and Kurt Sylvan
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Index 285
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
List of Contributors
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
1
Introduction
Jessica Brown and Mona Simion
1.1. Introduction
The notion of defeat has been central to epistemology, practical reasoning, and
ethics. Within epistemology, it is standardly assumed that a subject who
knows that p, or justifiably believes that p can lose this knowledge or justified
belief by acquiring a so-called ‘defeater’, whether evidence that not-p, evidence
that the process which produced her belief is unreliable, or evidence that she
has likely misevaluated her evidence. Within ethics and practical reasoning, it
is widely accepted that a subject may initially have a reason to do something,
although this reason is later defeated by her acquisition of further information.
The notion of defeat has been central to a wide range of different philo-
sophical debates, including, but not limited to:
093/oso/
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
nature of defeat can help settle the issue (see e.g. Frances and Matheson
2019).
(8) Reductionism versus anti-reductionism about testimony: say that a
suspect for murder S tells you that she did not do it. According to both
of the main views in the epistemology of testimony, you are not
justified to believe S. According to reductionism, that’s because you
always need positive, non-testimonial reasons to believe what you are
being told. In contrast, according to anti-reductionism, you are prima
facie justified to believe S, but your justification is defeated. The correct
nature of the nature of defeat will likely go a long way in the direction of
settling the issue (see e.g. Green 2008).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
3
overview of the volume’s chapters and explain how they build on the state of
the art (Section 1.4).
The first and now considered the classic view on the nature of defeat in
epistemology is due to Pollock (1986). According to this view, D is a defeater
of E’s support for p for S if and only if (i) E is a reason to believe p for S, and
(ii) E&D is not a reason to believe p for S (henceforth Pollock’s view).
The account has a lot going for it; first, it promises to cut across normative
domains, in virtue of being framed in terms of reasons; after all, epistemolo-
gists hardly enjoy exclusivity on reasons. With Pollock’s view in play, it is easy
to see how we could generalize it to cover different targets (e.g. actions) and
types (e.g. moral, prudential) of normativity. Second, Pollock’s view makes
good on the intuitive thought that defeaters are actualizers of the possibility of
a positive normative status to be overridden or undercut; what the view says,
in a nutshell, is that defeaters are the kind of things that render a permissible
belief impermissible.
More recently, though, Pollock’s view has come under heavy attack. First, as
Chandler (2013) points out, Pollock’s view is problematic in that it cannot
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The alternative reliable process account (ARP): S’s belief is defeated if there
are reliable (or conditionally reliable) belief-forming processes available to S
such that, if S had used those processes in addition to the process actually used,
S wouldn’t have held the belief in question (Goldman 1979).
One can see how ARP is an elegant reliabilist translation of the Pollockian
thought that defeat is the kind of normative entity that, when taken in conjunction
with the extant epistemic support for the relevant belief, fails to render it justified.
Bob Beddor (2015) is the locus classicus for criticism of ARP; if Beddor is
right, ARP is both too weak and too strong. Against ARP’s sufficiency direc-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
What this cases shows is that ARP is too weak: contra ARP, for my belief
that p to be defeated, it is not enough that I would change my mind about p in
a counterfactual world due to employing some reliable process. What this
objection identifies is that ARP is normatively too weak: just because I would
change my mind in world W, it does not follow that I should change my mind
in world W: defeat is a normative notion.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
5
Job opening: Masha tells Clarence that her department will have a job
opening in the fall. Clarence believes Masha; assuming that Masha is usually
reliable, Clarence’s belief counts as prima facie justified. Sometime later,
Clarence speaks with the head of Masha’s department, Victor, who informs
him that the job search was cancelled due to budget constraints. Now suppose
that Clarence harbours a deep-seated hatred of Victor that causes him to
disbelieve everything that Victor says, what’s more, no amount of rational
reflection would rid Clarence of this inveterate distrust. Consequently, he
continues to believe that there will be a job opening in the fall.
This case shows that ARP is also too strong: just because, in all counterfac-
tual worlds, I would irrationally and stubbornly hold on to my belief, it does
not follow that I should do so. Once again, ARP is not normative enough to do
the job it is supposed to do.
accepted that a subject may initially have a reason to do something although this
reason is later defeated by their acquisition of further information.
However, the traditional conception of defeat has recently come under attack.
Some have argued that the notion of defeat is problematically motivated
(Lasonen-Aarnio 2010, 2012, 2014; Hawthorne and Srinivasan 2013).
According to a second strand of attack, the notion of revisionary defeat is
inconsistent with important desiderata on an account of knowledge or justifica-
tion such as naturalism, externalism, evidentialism, or a rule-based account
(Greco 2010; Lasonen-Aarnio 2010, 2012, forthcoming, Beddor 2015; Baker-
Hytch and Benton 2015; Weatherson, 2019). A further worry concerns how the
defeat of a proposition’s status as evidence is compatible with a Bayesian frame-
work (e.g. Weisberg 2009; Pryor 2013; Greco 2017). Furthermore, those who
adopt a steadfast view on peer disagreement can be understood as denying the
defeating effect of evidence that a peer disagrees with one (e.g. Kelly 2005).
In the light of these attacks, some have suggested that we should diagnose
the intuition that there is something wrong with someone who continues to
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
believe that p in the face of defeating evidence not by the suggestion that the
relevant belief loses its status as justified or as constituting knowledge but
instead in some other way, e.g. by some variety of error theory.
According to Williamson (2009: 315), for instance, when we intuit that
one’s knowledge gets defeated, we are, in fact, confusing between whether one
knows and whether it’s probable on one’s evidence that one knows. Another
popular suggestion is that continuing to believe that p in such circumstances is
to exhibit an epistemic disposition likely to lead to trouble in the long run
(Lasonen-Aarnio 2010; Hawthorne and Srinivasan 2013). According to this
view, in cases of alleged defeat, our intuitions fail to distinguish between
impermissibility and mere blameworthiness generated by irresponsible
behaviour.
Against defeat scepticism, in recent work, one of us (Brown 2018: Ch. 5) has
argued that it involves an unnoticed and unacceptably high cost: once we
explicitly distinguish between contributory (i.e. synchronic) and revisionary
(i.e. diachronic) defeat, it is hard to argue against revisionary defeat. That is
because it is deeply implausible to deny the contributory notion of defeat—i.e.
roughly, evidence against p that is part of the initial body of evidence. For, to
do so is effectively to deny that when evidence affects justification, it is one’s
overall evidence that matters and not merely a part of it. But, once we accept
the contributory notion of defeat, it is hard to deny the revisionary given the
plausible epistemic symmetry between cases of contributory and revisionary
defeat.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
7
see that you’re at 10,500 feet and remember that hypoxia is a risk at altitudes of
10,000 feet and higher. You now have evidence that you might have hypoxia,
and thus you have evidence that you might have miscalculated. Are you now
justified in believing that you can get to the more distant airport? Are you
justified in believing that your evidence supports that claim? (Christensen
2010).
Answering ‘yes’ to both questions is intuitively extremely problematic: it
licenses an extremely form of dogmatism. Answering ‘no’ to the first question
and ‘yes’ to the second doesn’t seem very promising either: even if you can’t
actually bring yourself to believe F, being justified in believing your evidence
supports F prima facie justifies you in believing F. Answering ‘no’ to both
questions is the traditional way to go for champions of defeat, and probably
the most widely spread view in the literature. For one, this is the answer that
suggests itself on both the main views on the nature of defeat (the reasons-
based and the alternative reliable process-based accounts).
One highly debated view in the literature combines defeat scepticism with
answering ‘yes’ to the first section and ‘no’ to the second question. On this
view, then, your second-order evidence fails to defeat your first-order justifi-
cation: you are still justified to hold your first-order belief that you’ll reach the
second airport, but you are no longer justified in believing that your evidence
supports your claim. This view has become known as the ‘level-splitting view’
(Lasonen-Aarnio 2010, 2014; Wedgwood 2011; Coates 2012). Thus, on this
‘level-splitting view’, the pilot is justified to believe that they’re going to reach
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
the second airport and that their evidence does not support this conclusion
(henceforth Rational Akrasia).
Feldman (2005: 110–11) argues that it is impossible for this belief to be both
true and reasonable since the second conjunct undermines the reasonableness
of the first conjunct (see also e.g. Elga 2007; Christensen 2007; 2010; Horovitz
2014; Titelbaum 2015). Furthermore, the level-splitting view has also been
argued to licence problematic theoretical and practical reasoning (e.g. Brown
2018; Horovitz 2014). For instance, if Rational Akrasia obtains, if the pilot
believes that she has enough fuel to reach the second airport, then it seems that
she can exploit the claim that she has enough fuel to reach the second airport
in her practical reasoning. For example, she might reason that since she has
enough fuel to reach the second airport, she should not bother to land on the
first airport. Nonetheless, while ignoring the first airport on the grounds that
she has enough fuel to reach the second one, she ought to also admit that it’s
unlikely on her evidence that she’ll reach the second airport! But acting in this
way seems entirely unreasonable.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
1.3.1 By Mechanism
How does defeat work? Pollock’s distinction between rebutting (also known as
overriding) and undercutting (also known as undermining) remains the
classic answer to this question.
According to Pollock (1986), there are two ways in which defeaters act:
roughly, they either speak against the content of one’s belief, or against the
credentials of its formation procedure. To see how this works, say that my
friend Mary tells me that TRAIN: The train leaves at eight sharp, and thereby
I come to justifiably believe that TRAIN. Now, say that my other good friend,
Alice, sharply disagrees: according to Alice, the train leaves at 8.30. The
thought is that, after receiving Alice’s testimony, I am no longer (fully)
justified in believing TRAIN, in virtue of acquiring evidence against TRAIN
being the case. In this case, Alice’s testimony constitutes itself in a rebutting
defeater for my believing TRAIN.
In contrast, undercutting defeaters speak against the credentials of the
source of my belief. Consider for instance, an alternative scenario in which
Alice tells me that Mary is a compulsive liar about trains’ schedules. After
receiving her testimony, I am no longer (fully) justified in believing TRAIN
based on Mary’s assertion.
On Pollock’s view, a rebutting defeater, then, for a belief that p of S is a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
reason for S for either believing non-p or for believing some proposition q,
where q is incompatible with p. In that, on Pollock’s view, rebutting defeaters
work by overriding the reasons in favour of believing that p. In contrast,
undercutting defeaters don’t speak against p, but rather against the credentials
of the belief formation procedure employed in forming the belief that p. An
undercutting defeater for a belief that p of S is a reason for S attacks the
connection between S’s ground for believing p and p.
One worry that arises about Pollock’s distinction is whether it is plausible
that it is an exclusive one, as the account seems to suggest. Pryor (2013), for
instance, gives an example meant to illustrate that a rebutting defeater can, in
virtue of its rebutting normative power, also constitute itself in an undercut-
ting defeater: to see this, suppose S’s justification for believing p from evidence
E is rebutted, as S acquires new evidence E* that more strongly supports q,
which is incompatible with p. When this happens, Pryor argues, S is left with
the question of why p is false even if E is true. Pryor’s suggestion is that the best
answer available to S is that, in the present context, E is not a reliable indicator
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
9
distinction—which itself entails that defeaters need not have positive epistemic
status—has become a classic in the epistemological literature of the last years.
According to Alvin Plantinga (2000: 364–65), for instance, irrational and
unwarranted beliefs can defeat beliefs that have impeccable epistemic status.
Suppose I believe that I’m made of flesh, blood, and bone. I then come to
believe, due to some cognitive disorder, that my head is made of blown glass.
According to Plantinga, given that I come to hold this second belief I now have
a defeater for the prior belief, even if the defeater was formed by way of
cognitive malfunction. According to Plantinga, this case is one where I have
a ‘rationality defeater’: given that I acquire the second belief, it is no longer
rational for me to hold the first one.
Contra Plantinga, Alston (2002) argues that only beliefs with positive
epistemic status can defeat beliefs that have positive epistemic status, and,
furthermore a belief D can defeat belief A only if D has greater warrant
than A. One of us (Simion 2020) has also argued that the existence of
merely psychological defeat implies the existence of merely psychological
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
the political leader’s entourage. Jill reads the true story in the paper but misses
the false report on television. Many people believe with Harman that Jill
doesn’t know that the political leader has been assassinated. There are, how-
ever, several different views on the market as to what explains this datum.
According to Swinburne (2001) and Pollock (1986), the explanation is social:
the fact that Jill does not have knowledge is the consequence of there being a
true proposition (suggestive of a defect in justification) that is widely believed
in Jill’s society. Others think that the explanation lies with the easy availability
of the relevant defeating information (Sudduth 2018).
We have seen that epistemologies that incorporate doxastic defeaters typ-
ically take them to defeat justification (Alston 1989: 238–9; Bergmann 2006:
155–6) or some species of rationality (Plantinga 2000: 357–66). When it comes
to non-doxastic defeaters, however, things look slightly different: while non-
doxastic defeaters are widely taken to have an impact on knowledge, whether
mere facts out there in the world can defeat justification is a more controver-
sial issue. The epistemic internalist will have to answer ‘no’: after all, by
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
11
This volume brings together recent work to re-examine the very notion of
defeat, and its place in epistemology, and in normativity theory at large.
In Chapter 2, The Normativity of Knowledge and the Scope and Sources
of Defeat, Sanford C. Goldberg investigates the nature and scope of defeat
with an aim to argue in support of the category of normative defeat, as well
as in favour of there being a social dimension to knowledge. To this effect,
the chapter appeals to a prior grasp of the normativity of knowledge itself—
its role in entitling a subject to confidence and in authorizing others to
believe on the strength of one’s epistemic standing—to shed light on the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
not-p, or else having warrant to believe that their warrant for believing that p
are inadequate (where warrant is understood in non-evidentialist terms).
Chapter 4, Losing Knowledge by Thinking about Thinking, by Jennifer
Nagel, puts forth a novel defense of infallibilism in epistemology against the
threat coming from the phenomenon of defeat. Defeat cases are often taken to
show that even the most securely based judegment can be rationally under-
mined by misleading evidence. Jennifer Nagel argues that defeat cases really
involve not an exposure of weakness in the basis of a judgement, but a shift in
that basis. For example, when threatening doubts are raised about whether
conditions are favourable for perception, one shifts from a basis of unreflective
perceptual judgement to a basis of conscious inference. In these cases, the basis
of one’s knowledge is lost, rather than rationally undermined.
Chapter 5, Dispositional Evaluations and Defeat, by Maria Lasonen-Aarnio,
argues that what explains the intuition of impermissibility in putative cases of
defeat resides not in the impermissibility of the target beliefs themselves, but
rather in the believer’s criticizability for manifesting bad epistemic disposi-
tions to believe. Subjects who retain their beliefs in the face of higher-order
evidence that those very beliefs are outputs of flawed cognitive processes are at
least very often criticizable. This, however, on Lasonen-Aarnio’s account, is
not because such higher-order evidence defeats various epistemic statuses such
as justification and knowledge. Instead, she argues that they manifest disposi-
tions that are bad relative to a range of candidate epistemic successes such as
true belief and knowledge. In particular, being disposed to only give up belief
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
13
showing how these distinctive reasons for suspension of judgment flow from
the constitution of suspension of judgment.
In Chapter 7, Reasons for Reliabilism, Bob Beddor motivates and develops
a synthesis between two leading approaches to justification which are
typically developed in isolation from each other; the first one comes from
the reliabilist tradition, which maintains that a belief is justified provided that
it is reliably formed, while the second one comes from the ‘Reasons First’
tradition, which claims that a belief is justified provided that it is based on
reasons that support it. On the view proposed by Beddor, justification is
understood in terms of an agent’s reasons for belief, which are in turn analysed
along reliabilist lines: an agent’s reasons for belief are the states that serve as
inputs to their reliable processes. Beddor argues that this synthesis allows each
tradition to profit from the other’s explanatory resources. In particular, it
enables reliabilists to explain epistemic defeat without abandoning their nat-
uralistic ambitions.
Chapter 8, Knowledge, Action, and Defeasibility by Carlotta Pavese, reviews
some motivations for a ‘knowledge-centred psychology’—a psychology where
knowledge enters centre stage in an explanation of intentional action,—it
outlines a novel argument for the claim that knowledge is required for
intentional action, and discusses some of its consequences, in particular for
the debate on the defeasibility of know-how. As Pavese argues, a knowledge-
centred psychology motivates the intellectualist view that know-how is a
species of know-that. In its more extreme form, this view is committed to an
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
theory must allow for comparative verdicts about which options are more
strongly supported than others, rather than simply which options are required
or permitted. Snedegar rejects a simple and familiar balancing account of the
competition, as well as an account that understands the competition in terms
of giving and answering criticisms of the options, and he introduces a new
account that incorporates a distinct role for reasons against.
In Chapter 12, Perceptual Reasons and Defeat, Mark Schroeder focuses on
the defeasibility of perceptual evidence. If something looks red to you, it is
reasonable to believe that it is red, but if you realize you are wearing rose-
tinted glasses, it may not be reasonable at all to believe this, unless you have
some independent source of evidence. Schroeder compares several models for
how to understand this phenomenon. These models differ in their answers to
two questions: what evidence we get about the external world through per-
ception, and what our having that evidence consists in. According to
Schroeder, some models have the advantage of fitting seamlessly into general
accounts of non-monotonic inference but carry with them a commitment to a
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
15
References
Press.
Chandler, J. (2013). Defeat reconsidered. Analysis, 73(1): 49–51.
Christensen, D. (2007). Epistemology of disagreement: The good news.
Philosophical Review, 116(2): 187–217.
Christensen, D. (2010). Higher-order evidence. Philosophy and Phenomenological
Research, 81(1): 185–215.
Coates, A. (2012). Rational epistemic akrasia. American Philosophical Quarterly,
1(2): 113–24.
Elga, A. (2007). Reflection and disagreement. Noûs, 41(3): 478–502.
Feldman, R. (2005). Respecting the evidence. Philosophical Perspectives, 19:
95–119.
Frances, B. and Matheson, J. (2019). Disagreement. In: Edward N. Zalta (ed.) The
Stanford Encyclopedia of Philosophy (Winter 2019 Edition). Stanford: Stanford
University Press. https://plato.stanford.edu/archives/win2019/entries/disagree
ment/.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
225–39.
Lackey, J. (1999). Testimonial knowledge and transmission. The Philosophical
Quarterly, 49: 471–90.
Lackey, J. (2008). Learning from Words. Oxford: Oxford University Press.
Lasonen-Aarnio, M. (2010). Unreasonable knowledge. Philosophical Perspectives,
24: 1–21.
Lasonen-Aarnio, M. (2012). Higher-order evidence and the limits of defeat.
Philosophy and Phenomenological Research, LXXXVIII: 314–45.
Lasonen-Aarnio, M. (2014). The dogmatism puzzle. Australasian Journal of
Philosophy, 92(3): 417–32.
Lasonen-Aarnio, M. (2020). Enkrasia or evidentialism. Philosophical Studies, 177:
597–632.
Lyons, J. (2009). Perception and Basic Beliefs: Zombies, Modules, and the Problem
of the External World. Oxford: Oxford University Press.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
17
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
2
The Normativity of Knowledge and the
Scope and Sources of Defeat
Sanford C. Goldberg
2.1.
¹ Williamson (2000), to whom we owe this approach, appears to deny that there is much illumin-
ation to be had regarding standards other than knowledge.
² For a discussion, see Goldberg (2018).
³ The term and the claim are Alston’s; see Alston (1993, 2005).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
CONFIDENCE
If S knows that p, then S is entitled to be confident that p.⁴
AUTHORIZATION
If S knows that p, then S is in an epistemic position from which to authorize
others to believe that p.
⁴ Can’t one know that p, despite having misleading evidence as to one’s epistemic position with
respect to p? Such a view appears to be endorsed by several authors in the higher-order evidence debate:
Williamson (2014); Larsonen-Aarnio (2010, 2014); Baker-Hytch and Benton (2015), and others. I will
reply to this thought in the next section.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
20 .
⁵ I thank Jessica Brown for indicating the need to make this point.
⁶ Not everyone is agreed that knowledge is sufficient to warrant action: see e.g., Brown (2008a,
2008b) and Reed (2010). In at least some of these cases we might come to doubt whether knowledge is
sufficient to warrant assertion as well (for which see also the references in fn 9).
⁷ For an overview of the various positions, see Goldberg (2015a, 2015b).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
2.2.
DEFEATCONF
If S is not entitled to be confident that p, then S does not know that p.
DEFEATAUTH
If S is not in a position to authorize others to believe (with confidence) that p,
then S does not know that p.
My reason for calling these the ‘defeat corollaries,’ despite the fact that they
nowhere mention defeat, is straightforward: the presence of defeaters is
⁸ This covers most of the field of those who endorse some norm of assertion. Most, but not all: some
endorse a truth norm of assertion (Weiner (2005); Whiting (2013)); others endorse a variable-standard
norm of assertion and allow that there are cases in which the standard can be satisfied even though the
speaker is not justified in believing what she asserts (McKinnon (2013, 2016) and Gerken (2014)). And
I should acknowledge, too, the arguments supporting the idea that there are cases in which knowledge
itself is not sufficient to warrant assertion (see Lackey (2011)).
⁹ See e.g., Williamson (2002).
¹⁰ The following discussion has been greatly improved by comments from Jessica Brown.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
22 .
typically what offers the best explanation for the truth of the antecedents of
DEFEATCONF and DEFEATAUTH. In particular, given a subject who otherwise
has a good deal of evidence that p but who nevertheless fails to be entitled to be
confident that p (or to be in a position to authorize others to believe that p),
the best explanation for our subject’s failure to satisfy these antecedents is that
there are relevant defeaters in play. In these cases, the subject’s failure to
occupy the normative standing that knowledge brings in its wake—and, by
extension, the subject’s failure to know—indicate the presence of a relevant
defeater. In this way these corollaries are useful when our interest turns to the
presence of defeaters.
Before proceeding to use the corollaries in this way, it is important to
highlight the two distinct types of case in which the corollaries might be
applied. As noted, both DEFEATCONF and DEFEATAUTH can be used to
address whether a subject S knows that p: if S fails to be entitled to be confident
that p, or alternatively if S fails to be in a position from which to authorize
others to believe that p, then (by the lights of these corollaries) S fails to know.
There are two distinct types of case to which these might be applied, corres-
ponding to the two types of scenario in which we might be interested in the
question whether S knows that p. The first type of scenario—what I will call
the knowledge-prevention scenario—involves a context in which S has relevant
evidence but has yet to acquire the belief that p in the first place. Here the
question of whether S knows that p is tantamount to the question whether her
newly acquired belief that p amounts to knowledge; and when the corollaries
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
deliver a negative verdict, the result, in effect, is that defeaters prevented the
acquisition of knowledge. The second type of scenario—what I will call the
knowledge-defeating scenario—arises for a subject who currently knows that
p, but who gets misleading evidence at some point down the road. Here the
question is whether S’s belief that p retains its status as knowledge even after
the acquisition of misleading counter-evidence; and when the corollaries
deliver a negative verdict, the result, in effect, is that defeaters undermined
or defeated the knowledge itself.
It would seem that we can apply DEFEATCONF and/or DEFEATAUTH to
either sort of scenario.
Take a scenario of the first (knowledge-prevention) type. Suppose that,
prior to forming the belief that p, one’s total evidence, including whatever
misleading evidence one has, is such that (i) one isn’t entitled to be confident
that p and (ii) one isn’t in a position to authorize others to believe that p. For
example suppose that, before Lilly enters a room, she is told by an apparently
reliable informant that the room’s walls are cleverly illuminated by colored
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
lighting, and then she walks in and encounters what appears to be a red wall. It
is plausible that if under these conditions (and without further ado) she were
to come to believe that the wall is red, her belief would not constitute
knowledge that the wall is red. In this way the defeat corollaries can be used
to bring out that (misleading) evidence can prevent one from acquiring
knowledge in the first place—even in cases in which, bracketing the misleading
evidence, one has what would otherwise be knowledge-sufficient evidence.
But what of scenarios of the second (knowledge-defeating) type, in which
one already has the knowledge that p but then encounters further relevant
counterevidence (albeit of a misleading variety)? For example, suppose that
Sally saw (what appeared to be) the red wall before getting the (apparently
trustworthy) testimony that she was in a room whose walls were cleverly
illuminated by colored lights. And suppose that the wall really is red (no
lighting is involved). It is plausible to think that in this case, while she
originally acquired knowledge of the wall’s being red (prior to getting the
testimony), she loses her knowledge on getting the testimony. The hypothesis
of lost knowledge can be further supported on the assumption that scenarios
of the first (knowledge-preventing) type are possible. Given this assumption,
one can deny the possibility of scenarios of the second (knowledge-defeating)
type only at the cost of having to insist that the order in which one acquires
evidence matters to whether one knows. The proof of this is that for any case of
the second (knowledge-defeating) type, we can construct a corresponding case
of the first (knowledge-preventing) type in which the misleading evidence
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
came prior to the other evidence bearing on the proposition that p, thereby
preventing the acquisition of knowledge that p in the first place. So, if we deny
the possibility of knowledge-defeating cases while allowing for the possibility
of knowledge-preventing cases, then in effect we are committed to thinking
that the order of evidence acquisition is epistemically significant in this way.
But this is implausible. Consequently, anyone who grants the possibility of
scenarios of the first (knowledge-preventing) type ought to grant the possibil-
ity of scenarios of the second (knowledge-defeating) type.¹¹
Even so, I acknowledge that the claim in the second (knowledge-defeating) type
of scenario, to the effect that knowledge itself is defeasible, is controversial¹²—more
controversial, anyway, than is the claim in the first (knowledge-preventing) type
¹¹ Compare Brown (2018: Ch. 5), where she presents a slightly different argument for the same
conclusion—namely, that if defeaters can prevent one from knowing then defeaters can undermine
one’s prior knowledge as well.
¹² See e.g., Lasonen-Aarnio (2010, 2014); Williamson (2014); Schoenfield (2015); Baker-Hytch and
Benton (2015).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
24 .
of scenario (that misleading evidence can prevent one from acquiring knowledge
in the first place). In order to minimize the degree to which I antagonize those
who reject the possibility of the knowledge-defeating type of scenario, in what
follows, for the most part I will be restricting my attention to scenarios of the
knowledge-prevention type.¹³ This will enable me to make the points I wish to
make without having to take a stand on the somewhat controversial question
whether knowledge itself is defeasible.
2.3.
Before proceeding to use the defeat corollaries to discern the nature and scope
of the phenomenon of defeat, we would do well first to acquire some confi-
dence that both of the defeat corollaries themselves are acceptable. So, let us
apply them to cases that everyone should agree are cases of a lack of knowledge
owing to defeat, in the hope that they deliver this verdict.
PERCEPTION
Percy has been told (by a speaker he has good reason to trust) that he is about
to enter a room in which many of the walls are expertly illuminated by colored
lights to give the impression of being painted. Percy himself has no further
basis for questioning this testimony. Percy then goes into the room and
perceives what appears to be a red wall. (He has no basis for confirming
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹³ This said, I will occasionally present cases of the second (knowledge-defeating) type as well.
Sometimes I will do so because I am describing cases others have presented in these terms (as a threat to
previously existing knowledge); on other occasions I will do so for reasons of ease of exposition.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
side effects is to create false memory impressions. Malia has no basis for ruling
out the possibility that her present memory impression is false. Verdict: she is
not in a position to know (through memory) that she was twelve the first time
she got on a plane.
I take it that a ‘no knowledge’ verdict is obvious in these cases, and that this
verdict reflects the presence of the defeater. Happily, the defeat corollaries
deliver this result. In none of these cases is the subject entitled to be confident,
and in none of them is the subject in a position to authorize others to believe
(with confidence). Thus: Percy is not entitled to be confident that he is looking
at a red wall, nor is he in a position to authorize others to believe that the wall
is red; Dalia is not entitled to be confident that she is looking at deer tracks, nor
is she in a position to authorize others to believe that the tracks are of deer;
Malia is not entitled to be confident that she was twelve the first time she got
on a plane, nor is she in a position to authorize others to believe that she was
twelve the first time she got on a plane. These facts, together with
DEFEATCONF and DEFEATAUTH, can be used to derive the ‘no knowledge’
verdicts. And, if we ask why the antecedents of these corollaries hold, the best
explanation is that the subject has a relevant defeater. These are cases, then, in
which the presence of a defeater undermines the subject’s occupying the
normative standing associated with knowledge, and so (we can conclude)
fails to know.
Of course, it is not particularly interesting or noteworthy to learn that the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
corollaries deliver the desired results in these cases. But that, after all, is
precisely the point: my aim in these examples is to confirm the validity of
the corollaries, not to use them to get interesting results. Now, having been
given some reason to trust their validity, we can go on to employ the defeat
corollaries in cases in which the presence (or absence) of defeaters is not so
obvious. To this end, I will focus on cases in which an otherwise epistemically
well-positioned subject fails to be entitled to confidence—alternatively, in
which such a subject fails to be in a position to authorize others to believe
(with confidence). The upshot will be that in such cases we have good reasons
to postulate the presence of relevant defeaters, as the best explanation for the
subject’s diminished epistemic standing (despite her otherwise strong epi-
stemic position). That we can get this result in cases in which the result itself
is not obvious is proof positive of the utility of the corollaries in helping us
discern the nature and scope of defeat.
Let us now move on to some of these cases.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
26 .
2.4.
There has been a good deal of discussion of certain cases of would-be higher-
order defeat.¹⁴ These are cases in which a given subject has what in fact is
conclusive first-order evidence that p, but also has evidence—unbeknownst to
her, misleading evidence—that she is not in a position to reliably assess her
first-order evidence. One example, based on a case due to David Christensen,
is the following:
LOGIC
Lonny is highly competent at logic. Prior to taking a logic exam, she is told by a
person she regards as sincere and competent that there is a good chance that,
unbeknownst to her, she was recently administered a drug whose side effect is
that it renders one highly unreliable in discerning which would-be proofs are
actual proofs, even in very simple cases. (One effect of the drug is to make even
simple invalid inferences seem perfectly and straightforwardly valid.) While
under these conditions, she examines what in fact is (and what strikes her as) a
fairly straightforward proof of the proposition that p.
Does Lonny know that p, or does the testimony she received serve as a (higher-
order) defeater? While many people who have addressed this matter agree
with Christensen that Sally does not know that p,¹⁵ others do not.¹⁶
My own sympathies are with Christensen. I come at the issue by way of two
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
After reviewing the would-be proof (having previously gotten the testi-
mony), is Lonny entitled to be confident that p?
After reviewing the would-be proof (having previously gotten the testi-
mony), is she in a position from which to authorize others to believe that p?
It seems clear to me that the verdict for both of these is “no.” And either of
these verdicts, together with the relevant defeat corollary, can be used to draw
the conclusion that Lonny doesn’t know that p.
¹⁴ Christensen (2010); Lasonen-Aarnio (2010, 2014); Horowitz (2014); Horowitz and Sliwa (2015);
Schoenfield (2015, 2018); Brown (2018); Skipper (2018); Ye (2018).
¹⁵ See e.g., Horowitz (2014); Brown (2018); Ye (2018); Goldberg and Matheson (forthcoming). (For
his part, Christensen is less interested in the question whether Sally knows that p, than he is in
modeling how the higher-order evidence affects Sally’s warranted credence in p.)
¹⁶ See e.g., Lasonen-Aarnio (2010, 2014); Schoenfield (2015); Baker-Hytch and Benton (2015).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
day, as Elodie reflects on Clancy’s being in San Francisco, the letter from
Clancy is sitting in a pile of mail that is right in front of her, unopened, on the
desk. In fact, Clancy is in San Francisco.
NEWSPAPER (from Harman)
Ying Yue reads the morning paper to learn that the president has been
assassinated, and that there has been a coup. In fact, this is precisely what
happened. But later in the day the generals now in power want to fool
everyone into thinking that no such thing happened. So, they order the
afternoon edition of the newspaper to issue a retraction. The afternoon edition
goes out across the community and is read by virtually everyone else.
However, Ying Yue never sees the afternoon edition (and never overhears
others speaking of its contents). As a result, she continues to believe that the
president has been assassinated and that there has been a coup.
Both Pollock (explicitly) and Harman (implicitly) rely on the idea that there is
a social expectation of the subject in the case in question to avail herself of the
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
28 .
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
30 .
Even so, there are clear cases, and NEWSPAPER was meant by Harman to be
such a case. When this expectation is unmet, and the result is that one fails to
have some crucial evidence which one would have had if one had met the
expectation, we are in a position in which knowledge is at risk. On my analysis,
Ying Yue’s knowledge that there has been a coup would be defeated just in case
the result of expanding Ying Yue’s current belief set so as to include the infor-
mation in question—that there has been an afternoon edition of the newspaper
publishing a retraction—would be a belief set in which Ying Yue isn’t justified in
believing that there has been a coup. The hypothesis of defeat is the best
explanation for why Ying Yue, despite being in an otherwise strong epistemic
position with respect to the existence of the coup from having read the morning
paper, nevertheless fails to be entitled to confidence regarding the coup (or to
authorize others to believe with confidence).
If these analyses are correct, then the defeat corollaries have a significance
that may not be obvious at first glance. They can be used to make clear that the
normativity of knowledge includes a social dimension, and that this social
dimension can render evidence one doesn’t have epistemically significant with
respect to one’s epistemic standing on some proposition.¹⁷ As noted previ-
ously, such an idea might initially seem curious: how can evidence one doesn’t
possess have any epistemic significance whatsoever, let alone prevent one from
acquiring knowledge (or defeat knowledge one already has)? The answer that is
suggested, in part, by the defeat corollaries is that this prospect merely reflects
the social dimension of knowledge’s normativity: to know that p requires a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
2.5.
I just argued that the defeat corollaries can be used to support the idea that
there is a kind of normative (social) defeat. These are cases in which the
acquisition of knowledge is prevented (or existing knowledge is defeated)
owing to the subject’s failure to satisfy legitimate social expectations as to
her epistemic conduct or condition, in contexts in which there is adequate
¹⁷ This point has been emphasized in the ethics literature by Moody-Adams (1994).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
counterevidence that she would have acquired had she satisfied those
expectations. Both UNOPENED LETTER and NEWSPAPER were cases of
this sort; and the application of the defeat corollaries enabled us to discern this
sort of “normative (social) defeat.” However, my analysis may strike some as
unconvincing. It might be wondered, for example, whether, far from estab-
lishing a kind of normative (social) defeat, these cases are better regarded as
undermining the defeat corollaries themselves. In this section, I argue that this
reaction is wrongheaded.
What might prompt one to jettison the defeat corollaries themselves, rather
than having to countenance the possibility of a kind of normative (social)
defeat that is supported by those corollaries?
The obvious way to try to motivate this reaction would be to insist on the
idea (i) that unpossessed evidence can never have any epistemic significance
for the subject who lacks the evidence in question. For this idea is incompatible
with the combination of two further views, (ii) that both of the defeat
corollaries are true, and (iii) that the subjects in UNOPENED LETTER and
NEWSPAPER are such that, given the facts of the case, neither one is entitled
to be confident or to authorize others to believe with confidence. Since the
combination of (i)–(iii) form an inconsistent triad, (at least) one of these must
be jettisoned. So those who insist on (i) will need to reject either (ii) or (iii).
Before considering the move to reject (ii)—the defeat corollaries themselves—
let us first consider the proposal to reject (iii). According to this proposal, the
subjects in UNOPENED LETTER and NEWSPAPER are entitled to confidence
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
in their beliefs, and they are in a position to authorize others to believe with
confidence—despite whatever impression to the contrary that is generated by
our reflection on the evidence not in their possession. Indeed, it is easy to see
how one might think to combine a rejection of (iii) with a commitment to (i).
According to (i), unpossessed evidence is epistemically insignificant. If we
combine this with a rejection of (iii), we reach the conclusion that the only
evidence that is relevant to the question whether the subject is entitled to
confidence—and the only evidence relevant to whether the subject is in a
position to authorize others to be confident—is evidence in the subject’s pos-
session. Those who endorse this approach will argue that if we restrict ourselves
to the evidence in the subject’s possession in the previously mentioned cases
(UNOPENED LETTER and NEWSPAPER), the subject is entitled to be confi-
dent, and is in a position to authorize others to be confident.¹⁸
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
32 .
Such a view inherits the burden of explaining away any intuitions to the
contrary. I can think of two ways to do so. One would be to trace such
intuitions to a conflation of the subjects’ epistemic position before and after
they learn of the facts. After the subjects learn of the facts—after Elodie reads
the letter; after Ying Yue acquires and reads a copy of the afternoon paper
containing the retraction—they are not entitled to confidence in their original
beliefs, nor are they in a position to authorize others to be confident. But this is
consistent with the claim that prior to learning of these facts both are entitled
to confidence and in a position to authorize others to be confident. The other
way to try to explain away the intuition that (iii) holds is to argue that the
intuition rests on conflating the knowledgeable third-person perspective with
the subjects’ own perspectives. It is true that anyone who is aware of the
relevant facts¹⁹—of the contents of the unopened letter, of the existence of the
afternoon news retraction—is not entitled to confidence, and is not in a
position to authorize others to be confident, on the matter at hand. But this
is consistent with saying that nevertheless both Elodie and Ying Yue are so
entitled and so positioned (prior to being apprized of those facts).
I find both of these defensive maneuvers implausible. However, to explain
why, I will be appealing to (ii)—the defeat corollaries themselves. This suggests
that those who want to insist on (i)—i.e., that unpossessed evidence is epi-
stemically insignificant—would do best to reject both (ii) and (iii). Conversely,
if (ii) is plausible, so is (iii), in which case we ought to jettison (i). (That is, we
ought to allow that unpossessed evidence can have epistemic significance for
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
PEDIATRICIAN
Patricia is a pediatrician whose practice is more than two decades old. One
day a patient of hers, Luis, comes in with a relatively rare form of asthma.
Patricia remembers reading about this condition when she was in medical
school (more than two decades ago), and she distinctly remembers the
then-prescribed treatment protocol. At the same time, the American
Academy of Pediatrics (AAP) recommends that pediatricians stay on top of
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
In what follows, I want to argue for two things. First, despite the fact that her
belief as to the best treatment protocol is true, Patricia is not entitled to be
confident, nor is it the case that she is in a position in which to authorize others
to believe with confidence. As a result, Patricia does not know that the best way
to treat Luis is X (where ‘X’ designates what she learned in medical school),
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
34 .
live up to the standards set forth by the AAP (the relevant regulatory entity), as
it is the legitimate expectation of those who rely on pediatricians that they do
so.²⁰ But then it would seem that her failure to do so, when the failure pertains
to evidence she should have had, affects her normative standing—it affects her
entitlement to confidence, her being in a position from which to authorize
others to believe. In short, because she is normatively deficient in matters of
her profession, and because this deficiency highlights a flaw—her lacking
evidence that, given her professional duties, she ought to have had, and
which bear against her belief—she is thereby normatively downgraded in the
ways noted previously: she lacks entitlement to confidence, and is not in a
position from which to authorize others to believe. Both antecedents of the
defeat corollaries hold.
Some might think to question this by insisting that Patricia’s failure to
possess the relevant evidence is merely a professional failure—a matter of
medical negligence, as it were—which as such does not constitute the basis for
any epistemic downgrade of her attitudes. But this reaction would be wrong-
headed. It is here, of course, that the defeat corollaries earn their keep: these
make clear that what many might try to explain away as a merely professional
shortcoming is in fact an epistemic shortcoming—one that deprives Patricia of
knowledge. The evidence that Patricia fails to know just is that she is not
entitled to be confident and that she is not in a position from which to
authorize others to believe. Of course, her belief is true; and it is supported
by what she learned in medical school. But given her ignorance of the evidence
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
she ought to have had, and the fact that this evidence counters her belief as to
the best treatment, she is not entitled to be confident, nor is she in a position
from which to authorize others to believe with confidence.
Might one agree that Patricia fails to know, but argue that this case is best seen
as a Gettier case?²¹ Like Gettier cases, PEDIATRICIAN involves additional
“nearby” evidence which, if the subject had that evidence, she would not be
justified in believing as she does. But the assimilation of this case into the
category of Gettier cases is troubled by what is distinctive in PEDIATRICIAN:
like the subjects in UNOPENED LETTER and NEWSPAPER, Patricia is prop-
erly expected to have had the evidence in question. This is no part of standard
Gettier cases: one is not standardly expected to be aware of when one’s
officemates sell their Ford, to be sensitive to the prospect of entering Fake
Barn County, to realize that sheepdogs look a lot like sheep from a distance, or
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
to be sensitive to the prospect that one’s clock has stopped a short time ago.
The unpossessed evidence in PEDIATRICIAN, by contrast, is evidence that
others properly expect the subject to have had—it was evidence she should
have had. In light of this it would be curious indeed to explain the ‘no
knowledge’ verdict as a matter of (Gettierizing) epistemic luck (rather than a
matter of insufficient evidence).
I submit that the points I am making with PEDIATRICIAN are general.
They hold not only for pediatricians, but for anyone who is a member of a
professional organization whose standards capture good epistemic practice
and on whose members others rely for information (in the expectation that
they conform to those standards): such professionals ought to conform in their
practice to the standards articulated by their professional organization. In
addition, these points hold in cases involving experts as well as people who
play one or another role in a social or work group, as well as in cases involving
participation in other social practices, whether between partners at work,
family members at home, friends, or what-have-you.²² The point is simply
this: when a person who is legitimately expected to have certain evidence (or to
follow a certain protocol in inquiry etc.) fails to do so, where her failing to do
so results in her failing to have relevant counterevidence that she ought to have
had, the person is not entitled to confidence, nor is she in a position from
which to authorize others to believe (with confidence). This is seen not only in
PEDIATRICIAN but also in both of the unpossessed evidence scenarios in
UNOPENED LETTER and NEWSPAPER as well. In sum, the general lesson is
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
2.6.
The time has come to draw the various pieces of the picture into a more
cohesive whole. My general conclusion is that the normativity of knowledge
has more far-reaching implications than is typically assumed. In this paper,
I have tried to capture some of these implications in terms of the two defeater
conditionals themselves; these can be applied to certain cases to help us
discern a normative (social) type of defeat. Failure to recognize this type of
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
36 .
defeat, I submit, is tantamount to a failure to have in one’s sight a state that has
the sort of normativity we associate with knowledge.
I believe that this picture of the normativity of knowledge is far-reaching.
While we may well be able to characterize a condition in fully naturalistic
terms, such that non-human animals can be in that condition, my strong
suspicion is that such a condition would not answer to the sort of thing we
expect human knowledge to be. The defeat conditionals articulate the basis for
this suspicion. One lesson to be learned, if I am right is that even if (following
Ernie Sosa) we designate such a naturalistic condition as a kind of (mere)
‘animal knowledge,’ there is a great chasm separating such a condition
and what we ought to recognize as distinctly and paradigmatically human
knowledge.²³ The chasm is greater than many appear to suppose, since it
involves moving to a domain rife with the normative social expectations we
have of one another as we move about the world acquiring information in our
characteristically socially dependent ways.²⁴
References
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
24(1): 1–21.
Lasonen-Aarnio, M.( 2014). Higher-order evidence and the limits of defeat.
Philosophy and Phenomenological Research, 88(2): 314–45.
McKinnon, R. (2013). The supportive reasons norm of assertion. American
Philosophical Quarterly, 50(2): 121–35.
McKinnon, R. (2016). The Norms of Assertion: Truth, Lies, and Warrant. New
York: Palgrave MacMillan.
Moody-Adams, M. (1994). Culture, responsibility, and affected ignorance. Ethics,
104(2): 291–309.
Pollock, J. (1986). Contemporary Theories of Knowledge. Maryland: Rowman and
Littlefield.
Reed, B. (2010). A defense of stable invariantism. Noûs, 44: 224–44.
Schoenfield, M. (2015). A dilemma for calibrationism. Philosophy and
Phenomenological Research, 91: 425–55.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
38 .
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
3
The Structure of Defeat
Pollock’s Evidentialism, Lackey’s Framework,
and Prospects for Reliabilism
¹ Green (n.d.); Grundmann (2011); Palermos (2011); Matheson (2011); Carter (2015), (2018);
Carter and Navarro (2017); Goldberg (2018); Goldberg and Matheson (2020); Pritchard (2018).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
3
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
3.1.
Pollock provided the seminal treatment of defeat and defeaters.³ Pollock first
defines (epistemic) reasons (to believe) in terms of justified belief.
a stone and that is not an apple. Her new experience defeats her old experience
as a reason to believe that is an apple. Her old experience combined with her
new experience is not a reason to believe that is an apple, but instead a reason
to believe it is a stone, cleverly shaped to look like an apple. Given her new
experience, the rational thing for Carol is to change her mind, if she hasn’t
already.
Given one set of relevant reasons, you can have a good reason to believe
p. Given another, more inclusive set, you might no longer have a good reason
² Please note that our topic here is justificatory defeat. The term ‘defeater’ is sometimes used in
discussions of “defeasibility” responses to the Gettier problem, to refer to an unknown fact that would
defeat justification if known (or justifiedly believed). These “knowledge defeaters”—also sometimes
called ‘factual’ defeaters—are not our concern here.
³ Pollock (1967), (1968), (1970), (1971), (1979), (1986), (1987), (1994), (2001), (2006).
⁴ Pollock and Cruz (1999): 195.
⁵ We ignore here debates about the contents of perceptual states and perceptual beliefs, whether they
involve indexicals, singular terms, or existential generalizations. Our use of the demonstrative is for
convenience’s sake and isn’t intended to take a stand on this issue.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
M and M* range over the kinds of mental states that can serve as reasons for
beliefs. Defeaters are then just reasons functioning in a destructive rather than
constructive mode.⁸
For Pollock, all defeaters are either rebutters or undercutters:
For Pollock, beliefs—doxastic states—are not the only possible defeaters, for
he denies “doxasticism” in epistemology—the view that only beliefs can serve
as reasons. M and M* range over a variety of mental state types, including
perceptual experiences, sensations, feelings, moods, and memories, as well as
doxastic states. Pollock’s view is an evidentialist one: the justificational status
of a belief is determined by the evidence—the reasons—for and against
believing. Pollock’s evidentialism entails that you can’t have defeat without
⁶ Authors often describe the phenomenon of defeat in temporal terms. First you have a good reason
to believe that p. Next, you acquire new information that defeats that reason. Many of our examples
proceed just like this. But nothing about defeat requires this temporal ordering.
⁷ Pollock and Cruz (1999): 195.
⁸ Similar formulations occur in Chisholm (1977): 72–3, (1989): 53.
⁹ Pollock and Cruz (1999): 196. ¹⁰ Pollock and Cruz (1999): 196–7.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
reasons/evidence; the only things that can produce defeat are things that can
produce justification.
Reasons aren’t restricted to beliefs for Pollock, but they are restricted to things
that you might base a belief on. Prima facie doxastic justification is limited to
things that you have based your belief on. Ultima facie justification—even
ultima facie doxastic justification—is importantly different, for it also depends
on other reasons you possess but which haven’t positively or negatively causally
influenced that belief. Pollock is surprisingly inexplicit about ultima facie
justification, but it is fairly clear that he thinks that a belief is ultima facie
justified just in case it is prima facie justified, and the agent doesn’t possess any
ultimately undefeated defeaters for it. To possess a (for example, rebutting)
defeater for the belief that p is to have a reason to believe ~p, i.e., to be prima
facie propositionally justified in believing ~p. This makes ultima facie (doxastic)
justification depend in part on propositional justification in a way that prima
facie (doxastic) justification doesn’t. Thus, while having merely propositional
justification can’t increase your doxastic justification, it can diminish it.
Though rightly influential, there are a few idiosyncratic features of his
formulations worth pointing out.
First, in Pollock’s classic formulations (though this is rectified in Pollock
(2001)), the definition of defeat doesn’t admit of degrees and thus doesn’t
allow for partial defeat: cases where a reason reduces justification without
eliminating it. Second, this problem is exacerbated by the oddly conjunctive
framing of the definitions for rebutting and undercutting defeat: a rebutting
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
d is a rebutting defeater for S’s belief that p iff d is a reason for S to believe
not-p.
d is an undercutting defeater for S’s belief that p iff d is a reason for S to
believe that her reasons for believing p are inadequate.
Finally:
A prima facie justified belief is ultima facie justified iff it’s not defeated.
These definitions make clear the evidentialist presumption that defeat is
simply a matter of having reasons/evidence, and that ultima facie justification
is thereby a matter of what the balance of the evidence—what the whole of the
evidence—supports. They allow defeat to come in degrees, like reasons more
generally. On the evidentialist view, the basis for defeat is the same as the basis
for justification: the subject’s evidence.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
3.2.
Jennifer Lackey has argued that Pollock’s distinction between rebutting and
undercutting defeaters requires supplementation with another distinction
between “doxastic” and “normative” defeaters.
According to Lackey, a doxastic defeater is “ . . . a doubt or belief that is had
by S that indicates that S’s belief that p is either false or unreliably formed or
sustained.”¹¹ By ‘indicates’ she means ‘obviously entails.’ In particular:
¹¹ Lackey (2006b): 438. See also Lackey (1999): 483, (2003): 707, (2005a): 164, (2005b): 638, (2008):
44, 253, (2011a): 317. Lackey uses ‘psychological defeater’ and ‘doxastic defeater’ interchangeably. The
former serves to emphasize that in addition to beliefs, some doubts can function the same way as
doxastic defeaters, but since they are not narrowly classified as beliefs, we need a broader term. Hence
‘psychological.’ Nothing will hang on this point in our chapter. What about perceptual experiences, and
other non-doxastic states? Though in two places she included experiences along with beliefs and doubts
in her list of psychological defeaters (Lackey (2006a): 87, (2006c): 4), otherwise experiences do not
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
appear in her formulations (e.g., Lackey (1999), (2003), (2005a), (2005b), (2008): 44, 157, 204, (2011a):
317, (2014a): 292, (2014b): 75). The more common formulation states her considered view: psycho-
logical (doxastic) defeaters for Lackey are necessarily doxastic states (propositional attitudes) like
beliefs or doubts, never non-doxastic experiences. Non-doxastic experiences, instead, ground norma-
tive defeaters, as we will see. That’s how experiences enter her “ontology of defeat.”
¹² Lackey (2005b): 647, cf. Lackey (2003): 707, (2005a): 164, (2006a): 87, (2008): 44–5, (2011a): 317.
¹³ Lackey (1999): 475, (2003): 707, (2005b): 638, (2005a); 164, (2006a): 87–8, (2006c): 4, (2008): 45,
157, (2011a): 317.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
¹⁴ Lackey (2006a): 8.
¹⁵ For this use of the phrase ‘normative defeater,’ see Lyons (2009); Grundmann (2011); Goldberg
(2018); Pritchard (2018).
¹⁶ Lackey addresses evidence one does not possess but should in only two of her articles, and there
she is using ‘normative defeater’ only to address issues of knowledge, not of justification. Lackey
(2005b), (2014a).
¹⁷ A passage in a recent paper (Lackey (2018): 161) might lead one to think she now has this other
sense in mind in addition to her sense as defined in the text. She gives an example where she says a
subject’s belief is defeated, for the subject lacks evidence that she should have had. She then favorably
cites (Goldberg (2018), who had this other sense in mind. But the example makes it clear that this other
sense is not what she has in mind. For, in her example, the “evidence that the subject should have had”
simply consists in a justified belief that the subject should have formed, given other evidence (in this
case, another justified belief) that the subject already possesses. So, the evidence the subject should have
had is just the proposition that the subject should believe, given her other evidence she already
possesses.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Alice has good evidence from the doctor’s diagnosis that her vision is unreli-
able: she knows he told her that her vision is unreliable, and she believes he is a
reliable informant on matters optometrical. Given this evidence she possesses,
believing the proposition my vision is unreliable is justified for her. Given
Lackey’s definition of doxastic defeaters, this proposition would be a doxastic
defeater if she were to believe it. Furthermore, Lackey insists, she ought to
believe it. That’s why it is a normative defeater. The proposition my vision is
unreliable is then Alice’s normative defeater for her belief there was a car
accident.²⁰
Lackey’s example involves a belief as the evidential basis for the normative
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
defeater. This belief must be justified, given her definition of normative defea-
ters, for unjustified beliefs are not a part of one’s evidence that justifies further
beliefs. Non-doxastic experiences can serve as the evidential bases for normative
defeaters as well, for experiences too are a part of one’s total evidence.
How does Lackey understand the basis of defeat, for both doxastic and norma-
tive defeaters? “The underlying thought” behind both kinds of defeaters, she says,
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
3.3.
then ~p}.
(5) By Lackey’s definition, S’s belief that q is then a doxastic defeater for S’s
belief that p. Absent a defeater-defeater for q, then on Lackey’s view S’s
belief that p is no longer justified.²⁴ For Lackey, S should stop believing p.
(6) But that is counter-intuitive, given that S has very strong evidence for p,
and no evidence against it. (The only consideration that speaks against
believing p is q, but she has, by stipulation, no evidence for q and shouldn’t
believe it.)
²² Lackey also frequently says, in addition to doxastic and normative defeaters, that there is a third
category of so-called “factual” defeaters. Following Lackey, we’ve seen some philosophers to even say of
defeaters that there is a “traditional” three-way taxonomy of “psychological” (doxastic), “normative,”
and “factual” defeaters. Firstly, this is not the traditional view. Secondly, we deny the existence of so-
called “factual” defeaters, for we reject defeasibility analyses of knowledge.
²³ Lackey (2005b): 647, cf. Lackey (2003): 707, (2005a): 164, (2006a): 87, (2008): 44–5, (2011a): 317.
²⁴ Lackey must allow the possibility of cases where such defeater-defeaters are lacking, or otherwise
the possibility of unjustified defeaters is an idle wheel, as they would always be defeated by a defeater-
defeater.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
We think that in such cases the subject should not believe q but should
continue believing p. p is based on strong evidence, formed in a reliable way. q
is neither. How could a belief that is well-justified on strong evidence have the
force of its evidence defeated by an unjustified belief, based on no evidence
at all?
A number of other authors have endorsed—sometimes offhandedly, some-
times deliberately—the idea of unjustified defeaters, i.e., the idea that an
unjustified belief can nevertheless defeat the justification of a prima facie
justified belief.²⁵ But Lackey’s discussion of defeat disguises this very contro-
versial claim as an innocuous terminological point. It’s nothing of the sort.
Notice first that this view of doxastic defeaters is incompatible with any
roughly evidentialist view. On an evidentialist view, nothing can be relevant
to justification but evidence, and unjustified beliefs aren’t evidence.²⁶ On the
evidentialist view, defeat occurs when one’s partial evidence supports believing
p but one’s total evidence (or even just a more inclusive body of partial
evidence) does not. A defeater is then just another piece of evidence. On the
Pollockian theory of defeat, nothing can provide defeat that can’t also provide
justification, but unjustified beliefs can’t provide justification. Hence, an
unjustified belief cannot be a defeater.
Lackey needs an argument that there might be unjustified doxastic defea-
ters. Without such an argument, her entire category of doxastic defeaters is
unmotivated, for the evidentialist already recognizes that justified beliefs can
play the role of defeaters, either rebutting or undercutting. Without an argu-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
²⁸ Lackey (2005b): 639, 647, cf. (2005a): 164, (2006c): 4, (2008): 253, (2014b): 75.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
acquires an unjustified belief that q, where that belief that q combined with the
belief that p then puts the subject into an irrational overall state, then the
subject is required to give up on the justified belief that p to exit the irrational
state.
This, we think, is obviously false. That is not what rationality—even coher-
ence rationality—requires of the subject. What coherence rationality requires
is that the subject give up at least one of the beliefs in the incoherent set, but
not any one belief in particular. The subject could exit irrationality by giving
up q, or some other auxiliary belief. Rationality requires some accommodation.
It does not require giving up the original belief in particular. That’s just the
lesson of the Quine–Duhem thesis. It seems perverse to require that the subject
abandon her justified belief that p when she acquires an unjustified belief that
q, as if two wrongs made a right. One might rather think—and this is just the
point we’ve made already—that if anything, rationality would have required
the subject not to form the unjustified belief that q in the first place. Now that
she has, the best option is to retract it.
Not only is this the intuitively obvious response, and the response that
evidentialism requires, it’s the response that a coherence/responsibilist theory
would endorse anyway. The coherence theory never said to get rid of every
belief that’s implicated in a local incoherence; it says to get rid of (or modify)
the less-well-connected beliefs to maximize coherence among the better-
connected beliefs. We want a large, comprehensive, and coherent belief set,
and the way to get that isn’t to just toss out everything that’s involved in any
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
kind of conflict, but to selectively pin the blame for the conflict on whichever
beliefs are less coherent with the rest of the set, independent of the conflict in
question. The proponent of unjustified defeaters would have us do the oppos-
ite of what the coherence theory recommends: they would have us abandon
the well-connected beliefs to accommodate the poorly connected ones.
There’s another kind of argument one sometimes sees in defense of unjus-
tified defeaters. The proponent will say something along the lines of “given
that she believes her senses are unreliable, S should stop believing that p.”²⁹
This would support a modus ponens argument for the desired conclusion (that
S should stop believing p) if ‘given that’ functioned like ‘if ’: i.e., if the starting
point were “if S believes her senses are unreliable, S should stop believing that
p.” But that’s far from obvious; in fact, it’s the very controversy we’re trying to
resolve. More likely, the plausibility of the ‘given that’ claim derives from its
²⁹ For example, see Bergmann (2006): 165–7. For some discussion of the issue, see Frances (2014):
79–84.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
unjustified because the agent pulled it out of a hat? This suggests a tacit
recognition on Lackey’s part that unjustified beliefs can’t serve as defeaters
after all.
Lackey’s right that incoherence is a problem, and also right that so-called
doxastic defeaters, as she has defined them, introduce incoherence. And we
agree that a subject is required to make some adjustments. But we just don’t
think incoherence as such is the same as epistemic defeat. Defeaters defeat
propositional justification by introducing counterevidence into the subject’s
total evidence, not by merely introducing incoherence at the level of the
subject’s propositional attitudes.
³⁰ See Broome (2013) for similar comments. There is a lively and expanding discussion, partly
growing out of Broome’s work, on questions surrounding the so-called “enkratic” principle, and cases
where a subject’s total evidence might justify the first-order belief that p and the higher-order belief that
one’s total evidence does not justify the belief that p. See for example Worsnip (2018) and Lasonen-
Arnio (2020), among others.
³¹ See Lackey (1999): 485; emphases added; see also the most recent block quotation.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
This means that Lackey is not only wrong about unjustified doxastic
defeaters, but also about why justified doxastic defeaters defeat. They don’t
defeat because they introduce incoherence; they defeat because they negatively
change the subject’s total evidence.
Lackey’s talk about doxastic defeaters is not a harmless and neutral bit of
taxonomy. The idea of unjustified defeaters is flatly incompatible with evi-
dentialism, which holds that only reasons or evidence can provide or play the
role of defeaters; unjustified beliefs cannot play that role. The idea that
“doxastic” defeaters (whether justified or unjustified) defeat in virtue of inco-
herence simply misunderstands the nature of defeat. There is no rationale to
introduce Lackey’s category of doxastic defeaters.
3.4.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Third, there’s no chance that standard cases of defeat will even involve
permissions to believe doxastic defeaters, let alone obligations. In other words,
Lackey’s view makes counterintuitive predictions about standard, straightfor-
ward cases of defeat. Suppose David is indoors and Edwina calls from outside
and says it is raining. David believes her and is justified in this belief. But then
Fiona calls from outside and tells David that it is not raining. David has no
reason to trust Edwina over Fiona or vice versa, but he ignores Fiona’s
testimony and continues to believe it is raining. The evidentialist would say
that David’s belief that it’s raining is no longer propositionally justified, now
that his evidence is equivocal and thus isn’t doxastically justified. Edwina’s
³² Could Lackey get by with mere permissions, redefining normative defeaters as propositions you
are permitted to believe (rather than ought to believe), which would be doxastic defeaters if you did? We
doubt it. While there’s something plausible about the transfer of obligation (if you’re obligated to A,
and doing A would obligate you to B, then you’re obligated to B) this idea loses all plausibility when one
of those obligations is replaced by a permission (if you’re permitted to A and doing A would obligate
you to B, you’re obligated to B).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
testimony gives him a reason to believe it’s raining, but Fiona’s testimony gives
David a rebutting defeater.
What would Lackey say about this case? David has a rebutting defeater, but
he doesn’t believe a rebutting defeater, so on Lackey’s view he must have a
normative rebutting defeater. But this means that he ought to believe a
rebutting doxastic defeater for the belief that it’s raining. That is, he ought to
believe that it’s not raining. But clearly this is wrong. In the face of such
conflicting testimony, David shouldn’t believe anything about the weather. Not
only is he not obligated to believe Fiona’s testimony, he’s not even permitted to
believe it in the circumstances. This kind of case is easily and naturally handled
by a Pollockian theory of defeat, while Lackey’s theory fails pretty decisively.
Why, then, did Lackey ever think there had to be normative defeaters? Her
idea, as we’ve already noted for doxastic defeaters, seems to be that they follow
from the requirements of rationality themselves. If a subject holds a belief
despite a normative defeater, the subject is being irrational and so their belief is
not justified. Here’s a relevant passage:
[We] can ask whether, in forming a certain belief, I have negligently ignored
important counterevidence . . . . beliefs reached [this way] would be irre-
sponsibly held and would not count [as justified for] epistemic responsibility
is the primary sense of justification . . . an essential component of rationality.
So-called normative defeat must then occur when a subject ignores or disre-
gards relevant counterevidence they possess, for the subject is being irrational
and/or irresponsible in so doing.
³³ Lackey (2005a): 171–2. Italics in the original. Bold added for emphasis. Lackey cites BonJour
(1985): 39. See also Lackey (2005a): 178 for another example.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
The first three premises seem to validly entail the conclusions, and so it
looks like we have an argument that makes explicit connections between
rationality, justification, defeat, and so-called normative defeaters. We’ve
thereby made sense of Lackey’s view that when a subject “ignores” or “disre-
gards” counterevidence, that entails “epistemically unacceptable irrationality”
that defeats the subject’s justification for her belief. So, if we are looking for an
argument for why there must be normative defeaters, we have found one.
The argument, however, does not work. Without taking a stand on the first
or second premise, we think it should be clear to everyone that the third
premise is false. One way to regard or acknowledge counterevidence is to
believe what it is evidence for. That is certainly true. But it is not the only way
to regard or acknowledge counterevidence. When subjects like Alice, Carol,
and David have counterevidence, that evidence defeats their prima facie
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
justification for their belief that p, so that their total evidence no longer
propositionally justifies believing that p. How might they acknowledge or
show due regard to their counterevidence? They might do it—and this is
what the evidentialist recommends—by giving up their belief that p. And
that is because their belief that p is no longer propositionally justified on
their total evidence. Alice could and should acknowledge the force of her
counterevidence by not believing there was a car crash. Carol could and should
give due regard to her counterevidence by giving up her belief that is an apple.
David could and should respect his counterevidence from Fred’s testimony by
suspending judgment as to the outside weather.
Insofar as our characters violate a requirement of rationality if they con-
tinue to believe despite their counterevidence, it is not that they are failing to
believe the propositions that their counterevidence supports. It is rather that
they are believing a proposition that their total evidence does not support. The
requirement of rationality at work in all of these cases—a requirement that
supports the requirement to acknowledge one’s counterevidence—it is the all-
too-familiar requirement not to believe what one’s total evidence does not
support. Lackey’s view of normative defeat misunderstands the phenomenon
of defeat and defeaters for it misunderstands not only how counterevidence
defeats, but also what it takes to acknowledge one’s counterevidence. Defeaters
qua counterevidence do not defeat by introducing obligations to believe
propositions supported by the defeating evidence. Defeaters qua counterevi-
dence defeat by changing what propositional attitudes a subject’s total evi-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
dence supports taking. The evidentialist has a handy treatment of all of these
cases without introducing a requirement to believe—or even a permission to
believe—what the counterevidence supports. Just as there are no doxastic
defeaters, there are no normative defeaters.³⁴
Internalist epistemologists tend to be motivated by one or both of two
distinct principles. The first is that your beliefs should always fit your evidence.
The second is that when you believe something, that imposes on you an
obligation to adjust your other beliefs accordingly if doing so is necessary to
maintain coherence. We have seen the first—evidentialist—principle used to
articulate a relatively promising theory of epistemic defeat. We have seen an
attempt to use the second—responsibilist—principle to understand defeat, but
that effort seems to us much less promising.
³⁴ It is worth noting that none of the responsibilists Lackey cites—BonJour, Chisholm, McDowell,
Williams—ever suggest that the only way to acknowledge one’s counterevidence is to believe the
propositions one’s counterevidence justifies believing. As far as we can tell, these philosophers do not
acknowledge her category of normative defeaters.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
3.5.
³⁵ One could, of course, try to claim that there’s a reliable process of responding to higher order
defeaters, or something similar. We don’t think this is at all promising, in part because this is a poor
candidate for a cognitive process in the needed sense (see Lyons (2019)) but also because even if it were,
it would seem to be the kind of belief-dependent process that can only transmit or preserve justification,
not generate or create it (see Goldman (1979)).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
³⁶ See for example Alston (1988); Comesaña (2010); Tang (2016); and Miller (2019).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
This allows us to articulate a general view about defeat that is very much in
the Pollockian spirit. We can use Pollock’s view about the structure of defeat
and simply drop the assumption that all warrants involve reasons. We simply
take what Pollock says in terms of reasons and say it instead in terms of
warrants:
• S has a rebutting defeater for her belief that p iff she has a warrant to
believe not-p.
• S has an undercutting defeater for her belief that p iff she has a warrant to
believe that her warrants for believing p are inadequate.
• S has a defeater for her belief that p iff she has a rebutting or an
undercutting defeater for her belief that p.³⁷
Having a warrant to believe p, we take it, is nothing more than having prima
facie propositional justification to believe p. Thus, all the reliabilist needs to do
to have a theory of defeat is to develop a theory of propositional justification.
This is non-trivial, since reliabilism is first of all a theory of doxastic justifica-
tion, but it’s far from hopeless.³⁸ Pollock himself offers what’s clearly a theory
of propositional justification, where propositional justification is derived from
doxastic justification: a reason, for Pollock, is something that could give you
doxastic justification.
Representing a more pessimistic view, Bob Beddor argues in a recent paper
that the standard reliabilist accounts of defeat fail, and suggests that reliabilists
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
³⁷ For the purposes of this paper, we are assuming that all defeaters are rebutting or undercutting,
which means that you only get defeat where you have warrants, thus that there are no “normative
defeaters” in that other, not-Lackey, sense mentioned above.
³⁸ We’re arguing here that reliabilism needs a theory of propositional justification because it needs it
for a theory of defeat. This stands in contrast both with the views of Kornblith (2017), who thinks that
doxastic justification is the only one of the two that matters, and of Feldman and Conee (1985), who
claim to base their theory of doxastic justification on their theory of propositional justification.
³⁹ See Beddor (2015), and also his Chapter 7 in this volume, which is similar in important ways to
what we’re offering here.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
This case aims to show that, contrary to ARP, the agent’s having an available
reliable process that would have led to non-belief, is insufficient for defeat.
Here’s Beddor’s other case:
S hears a defeater for her belief that p, from a reliable source, but one for
whom S holds an inveterate and irrational distrust. S would never take this
person’s testimony to heart, so she persists in believing p. Intuitively S’s belief
is defeated, even though her distrust means that the process of taking the
testimony on board is not available to her.⁴²
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
This case aims to show that, contrary to ARP, the agent’s having an available
reliable process that would have led to non-belief is unnecessary for defeat.
There are serious objections one could level against the kind of thing Beddor
counts as a process and what he takes availability to be. We won’t pursue these
objections, for we think that Beddor’s cases reveal a genuine flaw with ARP,
whether or not they refute ARP as they’re currently stated.
Beddor thinks that the problem with ARP is that it omits consideration of
reasons. We think the problem with ARP is that it is formulated in terms of
what the agent would have believed, or not believed, had she used another
process. Roughly, an adequate theory of defeat needs to be formulated in terms
of what the agent should have believed or not believed had she used another
⁴⁰ Goldman (1979); Lyons (2009). To be at all plausible, this needs to be interpreted as claiming that
the belief is prima facie defeated, not that it is ultimately defeated.
⁴¹ Beddor (2015): 150–1. ⁴² Beddor (2015): 155.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
process. More precisely, since we’ve seen how badly that fared for Lackey’s
normative defeaters, what the agent would or would not have had (prima
facie) warrant to believe, had she used another process. Counterfactual non-
belief only entails defeat if all the relevant components of the counterfactual
situation go normatively correctly, but this isn’t guaranteed merely by stipu-
lating that the process is reliable. This is because the counterfactual situation
will typically involve all kinds of factors beyond the immediate consequences
of the use of the alternate process. That alternate reliable process might feed
into an unreliable one, so that the agent forms a justified belief and then
concludes something unjustified from it. Or the justified belief might be one to
which the agent would respond in an irrational way that’s in no way sanc-
tioned by reliabilism. Beddor’s cases trade on just these kinds of possibilities,
on the gap between what the agent would have believed (/not believed), and
what the agent should have believed (/not believed), or is licensed in believing
(/not believing). ARP is flawed because it focuses on the way alternative
available processes yield counterfactual non-belief, when it ought instead to
be focused on the way alternative available processes yield warrants.
Does this appeal to warrants, and to what the agent does or doesn’t rightly
believe in the counterfactual situation, amount to a significant concession to
Beddor, or to the evidentialist, or to the opponent of reliabilism? Not at all!
Reliabilism is a normative epistemological theory, after all. Its raison d’être is
to give an account of warrants, of what beliefs are prima facie normatively
appropriate. If it can’t do that, then defeat is the least of reliabilism’s problems.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Here, then, is the sort of thing a reliabilist can and should say about
possession of warrants and thus about propositional justification:
S has (prima facie) warrant for believing p at t iff a cognitive process that
satisfies the general theoretical requirements for prima facie (doxastic)
justification (a) is available to S, and (b) if used at t, taking as inputs only
states that S is already in, does or would likely produce p as output.
Again, this is a sketch and is intended primarily for the purpose of illustrating
that reliabilism has promising resources for accommodating defeat. Nevertheless,
several comments about the proposal are in order.
Notice that the formulation is very general, leaving open what the require-
ments are for prima facie justification. This is for two reasons. First, it allows
the present theory of warrant possession (and consequently of defeat) to be
easily imported into reliabilist theories that differ in other respects. Second, it
makes quite explicit the fact that the present theory of defeat is making use of
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
old theoretical apparatus, not importing new machinery that might not fit the
general reliabilist ethos. It will therefore inherit the virtues (and vices) of that
apparatus. Plug in a naturalistic and reductive and nonevidentialist theory of
prima facie doxastic justification, and you’ll get a naturalistic and reductive
and nonevidentialist theory of propositional justification, and thus of defeat,
and thus of ultima facie doxastic justification.
Clause (b) restricts the relevant outputs of the available processes to those
that would result from the system’s taking as inputs only states S is already in.
This is perhaps already entailed by a reasonable individuation of cognitive
processes (see Lyons (2019)), but it’s worth making it explicit. We don’t want
the fact that S has an unopened encyclopedia nearby to give S warrant for
everything written in the encyclopedia.
Next, the formulation mentions the process producing “p as output,” rather
than producing the “belief that p.” We trust that it is intelligible to say that a
belief-forming process might be producing an output, even when that output
isn’t a belief. A visual process might output the proposition that the lines are of
unequal length, even though the viewer is convinced that it’s an illusion and
thus does not believe they’re of unequal length. More generally, we want to say
that having a process available can give you warrant (/propositional justifica-
tion) for some proposition, even though you don’t believe that proposition,
and perhaps never would.⁴³
Finally, the formulation is vague in some crucial respects. In particular, it
relies on the availability of a process that does or would likely produce p. Some
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
might object that the vagueness connected to availability and likelihood here
are unacceptable. On the contrary, these might just be unavoidable features of
propositional justification, on any reasonable theory. If I justifiedly believe p
and that p implies q, then I’m propositionally justified in believing q; and all
this gives me a defeater for my belief that ~q. But not merely because my
justified beliefs entail q—they entail every necessary truth, many of which I’m
not propositionally justified in believing and which don’t provide defeaters.
Why? Because these consequences aren’t obvious to me, where obviousness is
cashed out by the reliabilist in terms of process availability. Since obviousness
comes in degrees, availability will have to as well, if this move is to have any
hope of working. For similar reasons, we can’t demand of an available process
⁴³ Could a belief-forming process produce as outputs states that aren’t at all belief-like? If an
alternate reliable process produced the desire that p, it doesn’t seem like this would be a defeater for
the belief that ~p (thanks to Beddor, personal communication). Maybe we need to restrict the outputs
to states with assertoric force or which are in some other way sufficiently belief-like. We won’t try to
sort out the details here.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
that it would lead to an output that p. For those logical consequences that are
not screamingly obvious but yet obvious enough to count as propositionally
justified, it’s plausible that the use of the relevant process wouldn’t guarantee
the output of that consequence but would only make it quite likely.
Suppose Chuck believes that he’s never owned a weapon, although if he
thought about it for a while, he’d remember that his parents bought him some
Chinese throwing stars for a birthday back in the 1980s. Is Chuck’s belief that
he’s never owned a weapon defeated? It depends, we think, on how hard it
would be for Chuck to recall this: how long and hard Chuck would have to
probe memory, how robust or fragile the cueing conditions would need to be
in order to retrieve the memory, how likely a memory search would be to
trigger recall, etc. Given that defeat comes in degrees, we might even claim—
with some plausibility—that Chuck’s degree of ultima facie justification is
affected by these considerations: if three seconds of thinking about it, under
pretty much any conditions, would fetch the relevant memory, then Chuck’s
prima facie justification for believing he’s never owned a weapon is signifi-
cantly reduced; if only a great deal of focused concentration, when primed
with ninja movies, would result in retrieval of this memory, then Chuck’s
initial justification is only slightly reduced. In this way, what Goldman (1986)
calls power (a ratio of truth to queries, where reliability is a ratio of truths to
falsehoods) might be directly relevant to propositional justification and defeat,
even if it’s not directly relevant to positive, prima facie justification.⁴⁴
Some people don’t like the idea that possession of reasons, or warrants, or
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
evidence, might come in degrees. In order to avoid this result, Feldman (1988)
goes so far as to claim that the evidence a person possesses at a given time is
restricted to those conscious mental states she’s in at that time. The worry for
this proposal is that it would seriously diminish the body of evidence anyone
possesses at a given time. How many items can be in consciousness at once?
Serious estimates range from one (Dehaene 2014) to 7 +/-2 (Miller 1956), with
most people thinking it’s about 4. Not enough, it seems. We saw above that
Pollock largely elides this important issue; although he articulates a very
detailed theory of ultimate defeat, none of it applies directly to questions of
⁴⁴ These thoughts seriously complicate the treatment of defeater-defeaters, the full treatment of
which we reserve for another occasion. If a regular defeater is a “first-level” defeater, and a defeater-
defeater is a “second-level” defeater, and so on, then higher-level defeaters will tend to be less available
than lower-level defeaters, due to the cognitive difficulty involved in thinking of them and appreciating
their significance, with the result that higher-level defeaters will tend, roughly, to asymptote to
irrelevance. This contrasts with the Pollockian view (see Beddor, Ch. 7, this volume), according to
which higher-level defeaters destroy justification just in case there is an odd, rather than even, number
of levels.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
an agent’s actual epistemic status, unless we add the claim that an agent’s belief
is (prima facie) defeated iff she is in possession of a reason that serves as a
defeater for that belief.
Propositional justification was going to come in degrees anyway, because
reasons and warrants vary in their strength. But strength, availability, and
likeliness of output are orthogonal dimensions of variation: a piece of evi-
dence, for example, can be weakly supportive but fully obvious/available,
strongly supportive but far from obvious/available, etc. The fact that the
current proposed reliabilist theory of defeat reveals and highlights some of
the vagaries surrounding propositional justification is no criticism of that
proposal, not if these vagaries were really already a part of propositional
justification.
This proposal is surely in need of further refinement, and different reliabi-
lists may want to tweak it in different ways. A full treatment of this issue
demands a paper of its own. The point we wanted to make with it here was
this: a roughly Pollockian view about defeat can be smoothly accommodated
in a reliabilist epistemology, without the reliabilist having to embrace any
evidentialist or responsibilist elements that are in any way foreign to the
naturalistic spirit of reliabilism. A belief is prima facie defeated, roughly,
when the agent has available to her a “good” cognitive process (the kind that
would yield prima facie doxastic justification) that would output the content
that that belief is false or unwarranted. The proposal bears some obvious
similarities to the well-known ARP (which we take to be a selling point),
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
although it fixes some serious problems by trading warrants talk for talk of
counterfactual non-belief. At the same time, this shift from reasons to war-
rants serves to highlight what was right about Pollock’s seminal work on
defeat. Although Pollock himself focused on reasons, rather than warrants
more generally, there’s nothing stopping the nonevidentialist from taking on
board the vast majority of the Pollockian view. Having defeaters is having
prima facie propositional justification and having prima facie propositional
justification is having warrants. Having warrants is something that can be
understood in whatever general epistemological framework—evidentialist,
reliabilist, perhaps even responsibilist—you like.
The recent literature has been dominated by two influential discussions of
defeat. One of these has been healthy and productive; the other has clouded
some relevant issues and taken as uncontroversially true something that we
think no one should even believe, but certainly no one should simply assume:
namely, that unjustified beliefs can defeat justified beliefs. The good strand of
the standard view of defeat is the evidentialist strand, but it’s not good because
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
it’s evidentialist. It’s good because it sees defeat as proceeding (only) from
warrants, which is true. Evidentialists make the further claim that only evi-
dence provides warrants. Those of us who deny that can still find a great deal
of common ground with the evidentialist theory of defeat.⁴⁵
References
Prentice Hall.
Chisholm, Roderick. (1989). Theory of Knowledge. 3rd Ed. Englewood Cliffs, NJ:
Prentice Hall.
Comesaña, Juan. (2010). Evidentialist reliabilism. Nous, 44: 571–600.
Dehaene, Stanislas. (2014). Consciousness and the Brain. New York: Penguin.
Feldman, Richard. (1988). Having evidence. In: Philosophical Analysis. D. Austin
(ed.) Dordrecht: Kluwer: 83–104.
Feldman, Richard and Earl Conee. (1985). Evidentialism. Philosophical Studies,
48: 15–34.
⁴⁵ Thanks to Bob Beddor, Adam Carter, Paul Silva, Mona Simion, and especially Jessica Brown for
comments on an earlier draft that led to improvements. Thanks also to Jonathan Kvanvig and Duncan
Pritchard for helpful conversations on these topics. We also thank audiences at the 2019 Bled
Philosophical Conference and at the 2019 Social Epistemology Network Event at Underwood
International College in South Korea for beneficial comments and questions. In particular, we recall
especially helpful comments from Lizzie Fricker, Chris Kelp, Mark Kaplan, Hilary Kornblith, and Brent
Madison.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
4
Losing Knowledge by Thinking about
Thinking
Jennifer Nagel
Knowledge can be real without being armor-plated. We gain and lose knowledge
in many ways in the normal course of life. We learn and then naturally forget,
within minutes or hours, many details of what we have perceived for ourselves
and heard from others. We can also lose knowledge without losing belief; for
example, when the world changes in unseen and unexpected ways behind our
backs. In addition to passive processes of forgetting and becoming outdated,
we can actively switch to suspension of judgment on points where we once had
knowledge. This last type of switching is often routine and unremarkable; for
example, when one decides to suspend belief upon recognizing that one’s
grounds or basis for judgment on a given point may well have weakened
over time. Upon reflection, I have doubts about the accuracy of my mental
map of the city where I used to live two decades ago, a city I used to know quite
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
/9780198847205.003.0004
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
70
popular fallibilist alternative. Section 4.2 aims to explain the appeal of fallibil-
ism, and Section 4.3 examines defeat cases in the light of my theory of
justification.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
¹ Note that this is a necessary, and not a sufficient condition; it may be essential to guessing-
correctly-on-a-Tuesday that the relevant proposition is true, but this will not be enough to qualify it as a
way of knowing.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
72
Roughly stated, the basic idea is that the subject can know something even
though it could have been false. This is not the same as saying that the subject
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
According to fallibilists, false beliefs can have precisely the same epistemic
grounding as instances of knowledge, so epistemic grounding in cases where p
is known cannot be understood in a way which makes the truth of p essential
to one’s epistemic grounding for p.
Not all formulations of fallibilism are given in terms of epistemic grounding;
some speak of evidence or justification. Another standard formulation of
fallibilism is rendered by Reed as follows: “A person S knows that p in a
fallibilist way just in case S knows that p on the basis of some justification j
and yet S’s belief that p on the basis of j could have been false (or mistaken or
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
in error).”² (Reed 2012: 586) The example he offers to bolster his point will
bring out the difference between our views of justifying bases:
For example, I know that the Cubs beat the Dodgers the last time they
played; I know it because my brother told me what happened, and he is
usually reliable about this sort of thing. But, if he had misread the box score
in the newspaper, I still would have believed him. In that case, my belief
would have been held with the same justification, but the belief would have
been false. Because this case is possible, the knowledge I actually have (when
my brother did not misread the box score) is fallible. (Reed 2012: 586)
For simplicity, I will understand the subject of these cases as hearing from his
brother just that the Cubs won, without any further elaboration about his
having read the paper as opposed to having watched the game live or on
television. I will also understand the subject to be taking his brother’s word for
it, in both cases, as opposed to forming his belief on the basis of explicit
reasoning from the premises, “My brother said that the Cubs won,” and “My
brother is usually reliable about this sort of thing.” We generally accept trusted
testimony without such calculated reasoning about its likelihood of being
correct, and if we were to restrict ourselves to such explicit reasoning, it is
not clear that we would gain as much as we do from the reports of others. If
explicit reasoning from exactly the premises just given yields knowledge, it
seems only to yield knowledge of a weaker conclusion like, “It is highly likely
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
that the Cubs beat the Dodgers,” as opposed to knowledge of the stronger
conclusion that the Cubs won. I take Reed’s use of the stronger conclusion in
this example to indicate that he also accepts trusted testimony as a path to
knowledge that is interestingly different from the explicitly reasoned calcula-
tion that could deliver knowledge of the weaker claim. Indeed, the weaker
claim (that it was likely the Cubs had won, given the report of a typically
reliable person), could be true, and arguably even known, even in the case of
the misread loss, so, if the contrast between the cases is to illustrate that the
very proposition known might have been false, we should not weaken it.
I agree with fallibilists like Reed that knowledge can be transmitted through
testimony, even third-hand, as from a brother’s report of a newspaper story.
However, the view I will defend distinguishes different justifying bases for the
² In the face of concerns about knowledge of necessary truths, this provisional formulation is later
amended to “S knows that p in a fallible way just in case S knows that p on the basis of some
justification j and yet S’s belief that p on the basis of j could have failed to be knowledge” (Reed 2012:
587). This amended formulation remains incompatible with my account.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
74
good case and the bad: in the good case, I take the justifying basis to consist in
the entire conduit from the Cubs’ actual victory to the subject’s awareness of
that victory, incorporating the successful perceptual and inferential processes
giving rise to knowledge on the part of the reporter, and the safe transmission
of this knowledge to the brother, who reports what he came to know by
reading the paper to the appropriately trusting subject. It is because the
subject has learned in such a way from a knowledgeable source that his
judgement is epistemically in the clear, normatively appropriate, and fully
justified. In the bad case, there is no such chain of transmission between the
outcome of the game (presumably a loss) and the subject’s formation of a
belief in a victory. Misreading a report is not a way of knowing, nor is trusting
someone who is mistaken. There are some commonalities between the good
and bad cases: for example, in both, the subject trusts someone who is usually
reliable. Thanks to the partial similarities between the knower and the subject
of the bad case, even the bad case subject has some measure of justification (on
which more later), but the justifying bases of the good and the bad cases are
not the same.
On the common fallibilist understanding of these cases, by contrast, the
subjects of the good and bad cases can have precisely the same justification.³
One way to make sense of this would be by identifying justification with what
is available to the subject to report as the basis of his judgment that p, if asked
how he knows that p. What matters here is not the actual history of the
subject’s judgment, but only those aspects of the apparent history which are
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
readily available to the subject for report. This subjective impression of one’s
basis will be called the subjective basis, for short, although we should be careful
to note that the subjective basis is not a basis for the judgment in the causal
³ I say “can have” rather than “always have,” to admit a greater variety of types of fallibilism. For
some forms of internalist fallibilism, the subjects of good and bad cases like the ones just given will
always have exactly the same justification; there is no gap between what one’s justifying basis is and
what it seems to be (e.g., Smithies 2019). But even for externalist fallibilists like Goldman (1979), who
would assign sharply different levels of justification to the perceiver and the hallucinator, running
distinct cognitive processes with different levels of reliability, there are at least some phenomenally
indistinguishable pairs of good and bad cases where the same process is at work across the divide. For
example, if the same retinal stimulation is produced by a convex object lit form above and a concave
object lit from below, generally resulting in a perceptual impression of convexity, given our learned
background expectations of light from overhead, then the same reliable but not infallible process is at
work in the person who sees the normally lit convex object as convex, and the person who sees the
concave object as convex in the presence of subtle lighting from below. These good and bad cases are
equally justified, in Goldman’s later fallibilist reliabilism (although not in his earlier infallibilist causal
view), given his restriction on the individuation of belief-forming processes to what happens “within
the organism’s nervous system” (Goldman 1979: 12). Although for simplicity I focus largely on
internalist forms of fallibilism here, I believe there are ways of extending the arguments of this chapter
to positions like this kind of reliabilism.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
sense of that word, even if it represents some elements in the causal basis.
Between this subjective fallibilism and my objective infallibilism lie a spectrum
of views, including externalist forms of fallibilism such as process reliabilism;
for simplicity, I will focus mainly on the extreme ends of the spectrum in what
follows.
We have a limited view of the causal processes behind our judgments,
even in cases where we gain knowledge. The subject who comes to know
about the victory can trust his brother without being able to report on the
exact track record supporting his sense that his brother is reliable
about “this sort of thing” (and indeed without being able to explain the
exact scope of “this sort of thing”), although there may well be some
intelligent appreciation of the brother’s track record underlying this trust.
Consciousness presents products of cognition, rather than the underlying
processes, let alone the distal features of the processes that secure our
grip on external truths. It is because what is available to consciousness is
so restricted that the bad case (pseudo-testimony) can be mistaken for
the good (testimony): from the subject’s first-person perspective, I can
equally respond to a challenge in either case by saying something like,
“my brother said that the Cubs won, and he is usually reliable about this
sort of thing.”
In my view, the similarity of reports produced in the good and bad cases is
evidence of the disparity between subjective and objective bases: one might
reasonably say the same thing about one’s basis in a case of error and in a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
case of knowledge, and in either case, one has limited first-person access to
what actually makes one’s judgment normatively appropriate. The fact that
one’s judgment is normatively appropriate in the good case means that one
does have suitable first-person access to the key first-order fact (for
example, that the Cubs won), but it is possible to know that the Cubs won
without knowing the full story behind how it is that you know this fact,
down to the relevant characteristics of the reporter who saw the game,
and so forth. Despite failing to capture the full story of how we are
normatively in the clear, subjective impressions of our ways of thinking
nevertheless play an interesting role in our cognitive economy, and are not
simply epiphenomenal to the course of our thinking about first-order issues
over time. In particular, conscious representations of the bases of our
judgments play a vital role in our cognitive flexibility, our capacity to
change how we are thinking about first-order issues, whether such changes
are prompted by anomalies we encounter as individuals, or by challenges we
receive from others.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
76
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
78
testimony, I face the risk of being dismissed as failing to know, given that what
I am reporting will sound unlikely. A better strategy will be to present
whatever evidence I can share to support my claim that the event occurred,
pointing out its causal traces, and so forth, even though my own awareness of
the event was secured through first-hand perception and not inference. Where
a truth is itself known through explicit reasoning, there is little difference
between the strength of one’s dialectical position and the strength of one’s
epistemic position; in other cases, dialectical strength is used as a test for the
presence of knowledge, and there may well be a gap between whether one can
pass this social test for knowledge and whether one actually knows.
There are internalists who make inference essential to justification, seeing
even perception as essentially involving inference from prior evidence. For
example, Stephen Schiffer describes the formation of a perceptual belief (say,
that there is a red cube before him, abbreviated ‘Cube’) in a favorable case
(‘Good’) as follows: ‘in Good I come to be justified in believing Cube by
inferring it from the fact that I am having such-and-such sensory experiences
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
⁴ One might wonder whether this more recent formulation of Williamson’s view is a strengthening
of the earlier view, according to which a perceptual belief that p could be implicitly evidence-based by
being appropriately causally sensitive to the perceptual evidence for p. On the old position, one could
have perceptual evidence for p without knowing that p, for example if one had misleading evidence
about the source of light. However, even in the new position, which highlights situations in which we
just gain the perceptual evidence that p, thereby coming to know that p, we can still distinguish a
separate notion of “perceptual evidence for p,” where the knowledge we gain speaks in favour of p
without entailing knowledge of p. This kind of knowledge might be especially important in cases where
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
80
inspires the view I am defending in this chapter, we are not initially aware of
visual appearances as of the cube, from which some belief about the cube is
then inferred; in the good case, what we are initially conscious of is the cube.
Our first moment of evidence is that there is a cube before us, in the good case.
If we turn introspective, directing attention inward, we are aware that we have
specifically visual experience “as of a cube,” but this is a metacognitive
byproduct of the original unreflective judgment, and not a prior condition of
it. It is true that hallucination would generate a similar metacognitive bypro-
duct, but false that hallucination would make us equally justified. In the bad
case, all we are conscious of is that there is something which appears to be a
cube, but it is part of the badness of the bad case that we will typically judge
that there is a cube present. If we wish to make a judgment which will be
legitimate in both cases, we can restrain ourselves to the more cautious
judgment that a cube seems to be present, but a general strategy of restricting
oneself to such judgments would entirely undermine the value of our world-
sensitive perceptual mechanisms. It is a feature of unreflective cognition that it
does not involve reflection on how we ought to be thinking: in unreflective
cognition such as ordinary perceptual judgment we are focused on the world,
and not on the question of how we should now be making judgments.
Because metacognitive impressions of our judgments play a key role in
defending our epistemic status under challenge, it is not easy to give up the
idea that instances of knowledge always require the backing of prior evidence.
In her defense of fallibilism, Jessica Brown observes that there is something
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
quite counter-intuitive about positions like the one I am defending here. She
defines both fallibilism and infallibilism in terms of a relation obtaining
between evidence and judgment: what infallibilists hold and fallibilists deny
is that, “knowing that p requires evidence which entails that p” (Brown 2018: 4).
She observes that Williamson’s equation of knowledge with evidence (in
Williamson 2000) gives rise to an unusual form of non-skeptical infallibilism:
because any known proposition is automatically included in one’s evidence
set, the fact that every proposition entails itself will ensure that no proposition
is held without entailing evidence (Brown 2018: 32). She then argues that it is
inappropriate to cite a proposition as evidence for itself (Brown 2018: Ch. 3).
Where the function of evidence is to lend support to a contested propos-
ition, citing that very proposition will beg the question at issue, whether the
we have misleading evidence speaking against p which makes us think inferentially about the question
of whether p, rather than forming a naïve perceptual belief. Thanks to Carlotta Pavese for valuable
discussion on this point.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
contest arises from some audience who could have doubts about whether the
speaker knows that p, or whether it arises from the speaker’s own doubts about
her epistemic status with respect to p. So, Brown is right that one’s knowledge
that p is generally not self-supporting in a way which will satisfy those who
have raised doubts about one’s knowledge that p. However, if we can separate
the task of proving one’s epistemic status from the conditions of initially
entering that status, we do not need to insist that every item of knowledge
be backed by prior evidence. If our sensory systems give us appropriate causal
sensitivity to some features of reality, for example, then when these systems
make us aware of those features, we can gain knowledge without prior
supporting evidence.
Brown advances two independent lines of argument against the kind of
strategy I advance at the outset of this chapter, in which ways of knowing
constitute our sources of justification. If I insist that a way of judging that p
only yields full justification when the truth of p is essential to one’s judgment
that p, Brown objects both that our natural ways of individuating belief-
forming processes do not involve this kind of factivity, and also that an
insistence on it will rule out some plausible instances of knowledge, most
notably knowledge gained through inductive inference (Brown 2018: Ch. 2).
On the first point, she grants that English-speakers do use the factive locution
“see that” for successful cases of visual perception, and even that we have some
extended use of it for the detection of mental states (e.g., seeing that another
person is in pain), but she correctly observes that there is no directly parallel
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
⁵ However, it is worth noting that constructions of the form “S was told + interrogative” are factive;
as are constructions like “S learned from X that p.” Thanks to Carlotta Pavese for this observation.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
82
the basis of the metacognitive byproducts they produce in us, and we may be
particularly tempted to do this if we approach epistemological questions from
the first-person perspective: from the inside, there is a conspicuous common-
ality between states of seeing and states of hallucinating. Still, anyone
impressed by Jessica Brown’s concerns about what is reflected in natural
languages might worry that similar concerns also arise when we individuate
belief-forming mechanisms from the inside. Do English-speakers commonly
refer to a way of forming beliefs that is neutral between seeing and seeming to
see? The question only becomes more pressing when we look across languages:
for example, the most comprehensive survey of grammaticalized evidentials
(Aikhenvald 2004) reports no language which has an evidential marking that
is neutral between seeing and visually hallucinating.
However, we do not have to frame the problem from the outset as one of
individuating ways of forming beliefs. Another route in to the problem would
start by distinguishing ways of learning about the world; this path can make it
seem more natural to focus on the distinctions between first-hand perception,
second-hand learning through testimony, and sound inference. Even if similar
problems arise on the side of belief-forming mechanisms individuated from
the first-person perspective, it remains a good question why these factive
modes of learning are not all conveniently and transparently lexicalized across
languages, especially given our reasons to believe that speakers of all languages
are generally monitoring the relevant distinctions. One reason may be that our
conversational contributions do not typically assert their epistemic character;
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
the case where the Cubs lost, it is not true that my brother who said otherwise
told me what happened. But Reed is not wrong to reach for the stronger
construction in the context of a discussion of the testimonial transmission of
knowledge: the tell-wh construction has a good claim to be our more natural
way of thinking of what is happening when we speak with those we trust.⁶
Likewise, when we use an epistemic modal like “must” to indicate inference,
we are not necessarily indicating any diminished commitment to the main
claim we are making, or admitting the possibility of error in our inference
(Von Fintel and Gillies 2010).⁷ In conditions where the epistemic credentials
of our claims are under challenge, it may be dialectically appropriate for us to
retreat to more neutral formulations concerning what we have been told, or
what we gather may well be the case, but this does not mean that our initial
way of thinking about the origins of our judgments is similarly cautious.
Brown’s concern about inductive inference is harder to address. It may seem
odd to suggest that the progression from observing any finite subset of ravens
to concluding that all ravens are black could be a process which securely locks
us onto the truth, so that the truth of this conclusion is essential to the process.
I will have to insist upon individuating processes so that a different process is
carried out when one samples ravens in a possible world containing a few
unsampled albino specimens, versus a world in which all ravens are black, and
one is subject to no risk. These processes might look the same to the raven-
counter, but one of them involves epistemically important luck in avoiding the
white ravens, where the other was relevantly safe. In my view, the justifying
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
basis of a judgment is not strictly a function of the choices for which one may
be responsible, or even those together with the subpersonal processes involved
in cognition; the basis includes environmental conditions as well. In this larger
process, the evidence afforded by individual observations plays a crucial role in
legitimizing the ultimate judgment, but it does not exhaust the normative
conditions responsible for what one comes to learn through induction. Proper
treatment of this issue lies beyond the scope of the present work, but I hope to
have given some sense of the limited role I am assigning to consciously
available evidence in my overall account of justification.
Defeat cases pose an especially interesting challenge to my account, exactly
because they seem to leave the basis of a judgment fixed, while toying with the
⁶ I am grateful to Patrick Shirreff and Carlotta Pavese for discussion on this point.
⁷ This point is controversial, as Carlotta Pavese has reminded me. For a contrary view of ‘must,’ see
Daniel Lassiter (2017).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
84
subject’s representation of that basis. The next section takes a fresh look at
these cases.
Roger White has a sharp statement of the problem that defeat cases pose the
type of view I am defending in this chapter. White is concerned to defend a
‘Cartesian’ conception of evidence, in which ‘the evidence I gain from visual
experience consists in information about how things visually appear to me’
(White 2014: 301). On this conception, one has precisely the same evidence in
seeing or hallucinating. White contrasts his preferred view with ‘Evidence
Externalism.’ according to which ‘the evidence I gain from visual experience
goes beyond information about how things visually appear to me and may
include facts about how things are in my environment’ (White 2014: 301). The
evidence externalist denies that the subject enjoys the same evidence in the
Good case of seeing one’s hand and the Bad Case of hallucinating a hand: the
Good case includes ‘a kind of direct epistemic access to the presence of a hand
before me.’ White then lays out a case in which a person (we will call him
“Hank”) who is actually looking at his hand and is given highly persuasive but
misleading evidence that he is merely hallucinating. The Cartesian view of
evidence has an easy explanation of why Hank seems to lose his justification to
believe that he is looking at his hand, under the pressure of the misleading
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
First, why is it that misleading evidence that I’m hallucinating robs me of this
extra evidence? Nothing appears to have changed about my perceptual state
and connection to the world. Suppose I even go on dogmatically thinking
here’s a hand despite the mounting evidence that I’m hallucinating. I am at
least still seeing my hand. My eyes are still receiving accurate information
from the world and my belief forming mechanisms are operating as usual in
forming a true belief. Isn’t that enough to have direct perceptual access to the
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Hank must consciously entertain the possibility that he himself has ingested a
drug which produces illusions as of hands; to make matters worse, he has also
been given a persuasive argument that this possibility is actual. If we imagine
him as ‘blithely ignoring’ this possibility and dogmatically judging here’s a
hand, this deliberate evasion of evidence is part of the complete belief-forming
process now sustaining his belief (for another version of this argument, see
Van Wietmarschen (2013: 413–15). If this process now involves thinking
dogmatically and disregarding evidence, Hank is no longer making a judg-
ment in a manner which generally leads safely to the truth: sound naïve
thinking has been replaced by questionable reflective cognition.
⁸ Against this kind of move, Ru Ye urges that “how one’s belief is based is a purely psychological
process, a process about how the belief is formed or maintained, and gaining a piece of evidence needs
not change this psychological process” Ye (2018: 7). I would urge that gaining evidence is also a
psychological process, and one which forces changes in the related psychological process of answering a
suitably related question. Ye maintains that defeat is ultimately best understood as a matter of
intellectual responsibility; this may be right, but one could have an account of responsibility which
includes the specification of responsible processes.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
86
Assume, for instance, that Suzy comes to believe that Tuesday is a rainy day
based on looking outside in the morning. She then immerses herself in a
book, ceasing to consciously entertain the belief, and forgetting all about the
rain. At noon she takes a look outside again and undergoes a new mental
process producing a belief that Tuesday is a rainy day. In this case, she may
well believe that Tuesday is a rainy day by using the same method as before,
but not on the same basis as before. (Lasonen-Aarnio 2010: 5)
In Lasonen-Aarnio’s view, the stubborn subject who retains his belief in the
face of misleading evidence is being unreasonable: the disposition that he is
showing in insisting upon this method of belief-formation does not generally
lead safely to truth, if we consider its application across a wide variety of cases.
However, Lasonen-Aarnio argues, in this particular case the belief is still well-
based and safely true: by stipulation our subject is exercising perception in
favorable conditions, and his method of believing on the basis of perception is
at no risk of delivering error in these conditions. The method is a good one,
given the circumstances, although the subject’s adoption of it shows a dispos-
ition which would in most circumstances be problematic (crime does not
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
generally pay, but in this case the subject gets away with something). The
subject’s dogmatic belief actually constitutes knowledge, on Lasonen-Aarnio’s
account, and any intuitions to the contrary are to be explained by appeal to a
general heuristic of taking unreasonable beliefs to fall short of knowledge, as
they typically do.
Lasonen-Aarnio describes the stubborn subject as adopting the method of
believing ‘on the basis of perception.’ If the basis of a belief is the whole belief-
forming process responsible for its production, then arguably this moment of
‘adoption’ should itself constitute part of the basis of the stubborn subject’s
belief, a part which distinguishes this basis from that of the naïve subject.
Taking a more restricted view of the basis as naïve perception, the problem is
rather that this basis is no longer an option for the subject, given the stipula-
tions of the case. It is certainly possible for the dogmatic subject to believe
the same proposition that he initially believed on the basis of perception, to
re-endorse the verdict delivered to him by sight, after this verdict has been
brought into question. However, if perceptual judgment is by its nature
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
88
drug out-of-hand. Consuela is simply told, falsely as it turns out, that there is a
red light shining on the vase. If she understands this report, even if she does
not accept it, then it is a good question whether Beddor can simply stipulate
that the report has no causal impact on Consuela’s credal state. To understand
the report is to contemplate an explanation of one’s current percepts at odds
with the judgment delivered by naïve perception, at which point one would be
already engaged in reflective cognition; in my view, understanding the inter-
locutor will already suffice for basis displacement. The interlocutor does not
need Consuela to accept that the lighting is abnormal in order to raise the
question of whether the lighting is abnormal, a question that cannot be settled
by naïve perception, but one which bears on the legitimacy of trusting her
initial impression of the color of the vase. If Consuela closes the question about
the lighting with some unjustified reasoning about the interlocutor’s unreli-
ability, then again, her belief formation is compromised. Perhaps there is a way
of understanding Consuela’s attitude to the interlocutor as so dismissive that
she does not even consider the question of whether there actually is something
wrong with the lighting; on this way of reading the case, I would venture, her
unreasonable attitude to this misleading interlocutor could insulate her from
defeat.⁹
So far, I have considered cases involving a switch from unreflective to
reflective cognition. Even within the realm of reflective cognition, we can
generate similar problems by shifting to higher-order reflection. Reflective
cognition itself involves sequentially presented contents, where the global
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
⁹ The insulation can only happen under rather special circumstances: perhaps the interlocutor is
Consuela’s little brother who teases her with suggestions she finds too irritating to consider seriously.
Other ways of being dismissive do not leave the epistemic standing of her belief unscathed: if she is
weirdly dogmatic about the color of the vase, for example, then Consuela’s grip on the relevant first-
order truths about the world is no longer safe. Thanks to Mona Simion for helpful criticism on this
point.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
with materials available to her, it is no longer possible for our subject to naively
endorse the initial and correct answer to the puzzle, and it is no longer rational
for her to reflectively endorse this answer either. Because her original reason-
ing was sound, the basis of her original judgment has not been rationally
undermined: one could not rationally demonstrate that this flawless basis was
at fault.
There is a certain elegance in understanding defeat as involving a shift in the
basis of one’s judgment: this strategy saves externalists from needing to add a
special internalist patch to their theory of knowledge. The addition of special
patches to cover defeat cases is a longstanding practice in broadly externalist
analyses of knowledge, and a practice which has long been seen as problematic
for these approaches. In the classical Indian Nyāya tradition, for example,
knowledge was analyzed as a ‘truth-hitting cognitive episode,’ born of a
faultless causal or justificatory source, qualified with the extra condition that
the resulting judgment must be ‘non-dubious in the sense that no reasonable
ground for doubting its truth has appeared’ (Matilal 1986: 135). Bimal Matilal
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
90
notes that the 11th-century skeptic Śrīharśa was already concerned that the
‘no-defeater’ clause seem oddly redundant, given the condition on the quality
of one’s source (ibid). Contemporary Anglo-American externalists still add
similar ‘no defeater’ clauses to their basic externalism, analyzing knowledge as
something like reliably produced true belief in the absence of defeaters (e.g.,
Goldman 1986). Because the other conditions on his analysis of knowledge are
often weaker, they are not always open to Śrīharśa’s worry about redundancy,
but the motivations of this type of hybrid position remain difficult to explain.
Laurence BonJour raises the worry that any theory of this sort will be an
‘untenable halfway house’ between internalism and externalism (BonJour and
Sosa 2003: 32). When externalists consider what makes a belief normatively
appropriate, the defining feature of their position is that they do not insist that
the subject should always have conscious access to positive justifying grounds.
It is curious, then, for externalists to add a requirement excluding the con-
scious availability of negative evidence. Why should conscious access matter
on one side of the ledger and not the other?
If the misleading apparent evidence in defeat cases has a direct impact on
our need for conscious materials, then there is an explanation of the asym-
metry: by raising concerns about the assumptions of less reflective forms of
judgment, defeat cases shift us upwards into more reflective ways of thinking,
in which we will need consciously available materials to proceed rationally and
make safely true judgments. Psychologically realistic externalists do not have
to see the conscious recruitment of reasons as epiphenomenal in our belief-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁰ For very helpful written comments on this chapter, I am grateful to Carlotta Pavese and to Mona
Simion. For discussion, I am grateful to David Barnett, Maria Lasonen-Aarnio, Sergio Tenenbaum, and
Jonathan Weisberg. I would also like to thank the Social Sciences and Humanities Research Council of
Canada for funding my research.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
References
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
92
Monti, M. M. and D. N. Osherson. (2012). Logic, language and the brain. Brain
Research, 1428: 33–42.
Murray, S. E. (2017). The Semantics of Evidentials. Oxford: Oxford University
Press.
Nagel, J. (2015). The social value of reasoning in epistemic justification. Episteme,
12(2): 297–308.
Papafragou, A., P. Li, Y. Choi, and C. Han. (2007). Evidentiality in language and
cognition. Cognition, 103(2): 253–99.
Pomerantz, A. (1980). Telling my side: “Limited access” as a “fishing” device.
Sociological Inquiry, 50(3–4): 186–98.
Reed, B. (2012). Fallibilism. Philosophy Compass, 7(9): 585–96.
Schiffer, S. (2009). Evidence = knowledge: Williamson’s solution to skepticism. In:
Williamson on Knowledge. P. Greenough and D. Pritchard (eds.) Oxford:
Oxford University Press: 183–202.
Simons, M. (2007). Observations on embedding verbs, evidentiality, and presup-
position. Lingua, 117(6): 1034–56.
Smithies, D. (2019). The Epistemic Role of Consciousness. Oxford: Oxford
University Press.
Van Wietmarschen, H. (2013). Peer disagreement, evidence, and well-
groundedness. Philosophical Review, 122(3): 395–425.
Von Fintel, K. and A. S. Gillies (2010). Must . . . stay . . . strong! Natural Language
Semantics, 18(4): 351–83.
White, R. (2014). What is my evidence that here is a hand? In: Scepticism and
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
5
Dispositional Evaluations and Defeat
Maria Lasonen-Aarnio
her that given the altitude and various symptoms she exhibits, Lobelia is suffering
from hypoxia. Such hypoxia impairs one’s reasoning while making it seem
perfectly fine—it is likely, her friend warns Lobelia, that her belief in p is the
output of a flawed process. This time, however, Lobelia’s friend is mistaken:
Lobelia’s original reasoning was impeccable.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
093/oso/9780198847205.003.0005
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
94 -
Lobelia’s belief is no longer safe from error.¹ Nor is it easy to see why it is no
longer produced by a reliable process.² Perhaps more promising would be to
argue for knowledge defeat via justification defeat. A popular way of thinking
about justification is in terms of evidential support. But, it is far from clear
whether Lobelia’s total evidence ceases to support p (e.g., by ceasing to make p
likely): a peculiar feature of higher-order evidence is that it often seems to have
no bearing whatsoever on the relevant first-order propositions.³ And wholly
independently of particular conditions on knowledge, consider the following
verdict: “Lobelia can just see, based on the evidence, that Penelope did it!” To
me this sounds like a perfectly natural thing to say. But, if seeing that is a way
of knowing that, then Lobelia still knows p.⁴ Further, I have argued that there
are principled reasons to think that correct epistemic norms cannot allow for a
systematic phenomenon of defeat by higher-order evidence,⁵ and that there is
a serious tension within views that aim to accommodate it.⁶
My aim here is not to argue that there is no possible theory able to
adequately accommodate and explain defeat by higher-order evidence. The
project of this paper is more positive: to explore a new kind of strategy for
explaining the seeming badness of retaining a belief in the face of evidence that
it is flawed.⁷ Though my main focus will be on defeat, this paper is an
implementation and application of a broader framework. What I see myself
as doing is providing the most promising account of a kind of evaluation
indispensable for making sense of cases ranging from ones involving victims of
massive deceit to so-called Jackson cases—and most importantly for my present
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
doxastic states are formed and retained (the choices made, the actions per-
formed, etc.). Instead of thinking of these ways in terms of methods, processes,
or bases, I will let the dispositions that manifest themselves as one’s ϕ’ing
(coming to believe something, retaining a belief, making a choice) do the work
of identifying these ways. I will simply assume—like I think everyone should—
that there are dispositions of a wide variety of different kinds, and that
inanimate objects and rational subjects alike are constantly manifesting
them.¹¹
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
96 -
focus as evaluators can shift what counts as relevant. Further, I don’t think there is any view in
epistemology that can escape all versions of the generality problem.
¹² Abundant properties come cheap. As a general rule, there is one corresponding to every coherent,
non-paradoxical predicate of a language. They can be wildly extrinsic and unnatural, and any two
things share an indefinite number of them. They are closed under various logical operations, satisfying
various comprehension principles. For instance, if there is the property of being an F, there is also the
property of being either an F or a G, and the property of not being an F. By contrast, from the fact that
F and G are sparse properties, we cannot trivially infer that having either F or G is a sparse property;
that having F and G is a sparse property, or that not having F is a sparse property. For arguments that
we should distinguish between fundamental properties pertaining to a fundamental level of nature
(assuming there to be one) and sparse properties, see Schaffer (2004).
¹³ Corresponding to these different levels of reality are different levels of explanation. Dennett (1969:
93) famously distinguished between personal and subpersonal levels of explanation, taking both to be
psychological explanations. Dennett’s concern was with explanations of behavior, but such a distinc-
tion could be extended to explanations of belief. Following Dennett, we could distinguish between
personal and subpersonal dispositions. Dispositions that work at the personal level are dispositions of
(entire) persons. By contrast, subpersonal dispositions are dispositions of subsystems or parts of
persons. There is much controversy about whether subpersonal explanations are really psychological
in the first place (for a good overview, see Drayson, 2014). I will remain neutral on these issues. The
dispositions I am interested in are psychological, whether or not this means that they must be
dispositions of persons.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
98 -
¹⁶ Cf Sher (2006: 7). Just as we may blame the hurricane for the low voter turnout, we may blame
one’s way of forming a belief for the falsity of the belief, but this kind of blaming is standardly taken to
be is essentially different from e.g., moral blame directed at persons (cf. Sher 2006: ix)—it is not
normative in a broad sense of the term.
¹⁷ This is a widespread criticism; for a recent elaboration in epistemology, see Simion, Kelp, and
Ghijsen (2016).
¹⁸ What about praise? The objects of praise appear to be broader than those of blame. We often
praise ways of acting or believing: we might praise Mahler for his orchestration, but we can also praise
the orchestration itself, just as we can praise the technique displayed by an expert archer’s shot. Indeed,
by praising the orchestration in a Mahler symphony, we bestow praise on the symphony itself, and by
praising the technique displayed by a shot we praise the shot. One reason I am cautious, however, of
expressing my view in terms of praise is that where failing to merit blame is not positive enough,
meriting praise seems excessively positive. A composer’s orchestration might not be praiseworthy, but
not criticizable or negatively evaluable, for it might simply be perfectly fine. By contrast, subjects who
fail to manifest good dispositions in the sense I will outline below are criticizable. Relatedly, it is not
clear, for instance, whether it is appropriate to praise the forming of an ordinary perceptual belief in
good conditions, or other routine acts that require little expertise or effort. Nevertheless, such beliefs are
positively evaluable, being paradigmatic examples of beliefs labelled justified or rational.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
We form beliefs, normally retain them for some period of time, and often
retrieve them from memory. One might, for instance, originally form a belief
about a particular event through an exemplary perceptual process, but the
belief about the event retrieved from memory might involve misremembering
or confabulation. Conversely, one’s memory about a past event might corres-
pond to one’s past experience, but the experience itself may have involved
misperception that manifests bad dispositions, such as cognitive biases. Hence,
our focus should not be exclusively on belief-formation, on acts of coming to
believe. Especially important in connection with discussions of defeat will be
ways of retaining beliefs in light of new evidence. Despite the good credentials
of one’s original belief, the dispositions at play in its retention might be
problematic. Indeed, this will be my diagnosis of what happens in many
putative cases of defeat. Hence, for a belief to be evaluated positively from
the dispositional perspective, the formation of the belief must manifest good
dispositions, and similarly for its retention and possible retrieval from
memory.
Assume that a subject ϕ’s (e.g., comes to believe a proposition p, retains
belief in p, or retrieves a belief in p from memory), and that her ϕ’ing is the
manifestation of some disposition D.¹⁹ What makes D good? Goodness is
always relative to some success: the dispositions at play when one comes to
believe that one’s lottery ticket will lose just based on the odds may be good if
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁹ For the purposes of stating my view, I will assume for simplicity that there is just one disposition.
It is, however, more realistic to assume that one’s ϕ ’ing is the joint manifestation of multiple
dispositions. In that case, the relevant cases will be ones in which all of these dispositions jointly
manifest themselves.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
case it is a manifestation of the best (or close enough to the best) feasible
alternative dispositions. I will take up the different components of the account
in turn. I will then say how the role played by feasibility in my account allows
for taking shortcuts in making dispositional evaluations.
Relevance: I think that dispositional evaluations are highly malleable and
context-dependent: what counts as relevant can easily shift depending on our
focus and context as evaluators. Nevertheless, there are some general struc-
tural points to be made. The first thing to emphasize is that relevance should
not be understood in terms of any relation that a case trivially bears to itself
(to a maximal degree). In particular, relevance is not a matter of relevant
similarity: counterfactual cases are not weighted according to how similar they
are to the actual case. Similarly, relevance is not a matter of what could easily
have occurred. This is because the case a subject is actually in might be deviant
or abnormal and as a result, it—and cases very much like it—may be irrelevant
when evaluating the goodness of a disposition. I won’t here offer a theory of
what such normality consists in, but even lacking a precise account, we have
some initial grasp on what sorts of cases count as normal. Various circum-
stances involving highly misleading evidence, for instance, are abnormal: it is
abnormal to encounter a real-looking fake tree among real ones, to experience
an intricately crafted perceptual illusion with no hints that one is experiencing
it, and so on. The dispositions manifested by a subject who comes to form a
belief on the basis of such an illusion might be good, for across somewhat
normal cases they manifest as knowledge-constituting (or otherwise epistem-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
²⁰ In my treatment of the new evil demon problem, I have argued that what counts as normal often
depends on the kind of world or case that a subject is anchored to—she might in fact be in a case that is
abnormal for her (see Lasonen-Aarnio, Forthcoming A). Hence, “alien cognizers” living in environ-
ments that are abnormal by our lights can be evaluated positively from the dispositional perspective, for
their dispositions fare well in counterfactual cases that are not too abnormal for them.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
then consider other counterfactual cases falling under this type in which the
same dispositions that Lobelia actually manifests manifest themselves.
Scores: the score of a disposition, representing how well the disposition does
across counterfactual cases, is a function of the weighting of these cases
according to relevance, and a value function. I will assume that the score of
a disposition is a weighted average of the values of its manifestations in
counterfactual cases, the weightings being by relevance. (I think this is a
natural view, though it is not the only possible one.)
The nature of the value function will, of course, depend on what the relevant
success under consideration is. For some successes all that matters is whether a
manifestation is a success or failure—coming close to succeeding doesn’t get
assigned any extra value (e.g., penalty shots in football). In other cases, success
itself comes in degrees. Consider, for instance, throwing darts, where one gets
more points the closer the dart lands to the central region. And even if
knowledge, for instance, itself is an all-or-nothing matter, one might think
that a value function should assign less value to a belief that is false than to a
belief that is merely true, but doesn’t constitute knowledge. I return to these
issues below.
It is worth noting that even fixing the success being focused on, the
framework sketched is compatible with the value function shifting depending
on the situation under consideration—being sensitive, for instance, to what is
practically at stake for a given subject. Compare a subject in a low-stakes
situation in which the practical cost of holding a false belief is not great, and a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
Just how close a disposition must be to the best feasible alternatives might
depend on context.
There are various constraints, of different strengths, on how we could be
disposed in the first place, and on what dispositions could be manifested in
specific circumstances. Some are logical, some nomological, some metaphys-
ical. Of particular importance below will be feasibility constraints flowing from
broad features of our cognitive design: as a general rule, feasibility is con-
strained by our cognitive architecture, allowing for some idealizing.²¹ Hence,
feasible alternative dispositions are dispositions that could, in a relevant sense
of ‘could’ determined by context but constrained by our cognitive architecture,
be manifested in the situation under consideration. An alternative disposition
is one that would manifest itself as a relevant doxastic state, choice, or action in
one’s situation. For instance, relevant alternative dispositions in the Lobelia
case manifest in the situation at hand as some doxastic state regarding
proposition p.²² (It is feasible in Lobelia’s situation to manifest a disposition
to scratch one’s head, but this disposition is not an alternative to the doxastic
dispositions Lobelia manifests when retaining her belief in p.)
Of particular importance below will be considerations of feasibility regard-
ing dispositions that discriminate between various cases. For instance, the
rationale for safety precautions that we routinely take is that we cannot be
disposed to only take those precautions in situations in which they are not
redundant. We cannot be disposed to only use safety belts on car rides that end
in crashes, or to buy insurance only for trips that end up involving some sort of
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
calamity. But there are other, perhaps more interesting cases, in which our
dispositions cannot pick up on certain features of the case we are in, even if the
good and bad cases differ in some respects “from the inside”. For instance,
climbers are taught to check their harnesses and knots every time before
relying on them. The aim is to be almost automatically disposed to go through
certain checks in every situation. It would of course be more efficient to only
²¹ Consider the literature on heuristics and biases—for instance, the systematic patterns of fallacious
probabilistic reasoning exhibited by humans. Psychologists tend to think that such errors flow from
features of our cognitive design, which consists of different systems tacked together (e.g., what is known
as System 1 and System 2). If a subject commits the so-called bankteller fallacy, thinking a conjunction
to be likelier than its individual conjunct, then even if she manifests some dispositions typical of human
beings, she is not manifesting good dispositions: there is a better available disposition—indeed, we are
sometimes able to pick up on such errors ourselves.
²² Assume that the subject in fact manifests disposition D₁. Then, in the actual case she manifests D₁,
and no other disposition. Nevertheless, we can ask whether some other disposition that could have been
manifested is better. This is why I talk about situations: when asking what the best feasible dispositions
are in one’s situation, we keep fixed other features about the actual case, but not facts about what
dispositions (if any) one manifests. My account also assumes counterfactual facts regarding what
doxastic state a given disposition would have manifested as in one’s situation.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
double check when one has made some sort of error, but the problem is that in
situations in which one has made an error, and in which double checking
would reveal (or would at least be likely to reveal) the error, the error has
already gone undetected until the procedure of double checking. These are, of
course, often precisely the kinds of situations in which one’s cognitive skills are
somehow compromised due to fatigue, altitude, dehydration, etc. In order to
be disposed to double check only when one has committed an error, one
would have to be sensitive to having committed the error. But how could one
be thus sensitive when the error was undetected in the first place, often due to
compromised cognitive skills?
Cases involving higher-order evidence that one’s doxastic state is flawed are,
I will argue, similar in many respects. The thought will be that given natural
assumptions about human psychology, it is plausible that in a wide range of
somewhat paradigmatic cases involving putative higher-order defeaters, a
subject cannot be disposed to retain belief in light of higher-order evidence
pointing to some sort of cognitive error only when no such error has in fact
occurred. It is not humanly feasible to have retention-dispositions that dis-
criminate between cases in which the higher-order evidence is misleading and
cases in which it is not. As a result, the dogmatic subject’s dispositions
manifest as retaining belief whether or not the higher-order evidence is
misleading. Her dispositions are problematic, for they will lead her to retain
a botched belief—a belief that fails to constitute knowledge, and that fails to be
appropriately based on one’s reasons or evidence—across many relevant
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
counterfactual cases. This is why she is negatively evaluable: there are signifi-
cantly better feasible dispositions.
According to the evaluative standard I have sketched, then, a doxastic state
is positively evaluable just in case it is a manifestation of dispositions that
compare favorably with feasible alternatives. A subject is criticizable if she
manifests dispositions that are is worse (or significantly worse) than feasible
alternatives. I think that evaluations that thus take into account feasibility
constraints arising, in particular, from a subject’s cognitive architecture are the
most promising for capturing a more subjective dimension of evaluation
across both the theoretical and practical realms—a dimension often picked
out by talk of justification, or of a subjective sense of ‘ought’.²³
That is not to say, however, that how good a disposition is in absolute terms
is of no evaluative significance. Assume that a subject succeeds at something
(for instance, at goaling in football, or at knowing). We can now ask: given the
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
after all, plenty of knowledge about how ordinary perception fares across
counterfactual cases.
Moreover, it is important to see that considerations of feasibility alone can
help determine that a subject’s belief manifests bad dispositions. Consider the
following simple case. You observe a subject not putting on a seatbelt. You are
not sure exactly what dispositions she manifests, and just how circumscribed
they are. (Perhaps, for instance, the subject is only disposed to forget when
very distracted.) Nevertheless, you know that certain possible ways to be—
certain possible dispositions to have—are just not ways that ordinary human
²⁴ Virtue epistemologists often talk about success being creditable to a subject: a subject succeeding
in virtue of, or because of her competence. Knowledge, then, is belief that is true in virtue of being
competent. As I am thinking about things, a belief can constitute knowledge, and be evaluated
positively from the dispositional perspective, even if the fact that it constitutes knowledge is largely
explained by somewhat invariant features of one’s situation. I think this is plausibly true in many cases
of testimony—indeed, testimony has been raised as a problem for virtue epistemologists (e.g., Lackey
2007).
²⁵ See Lasonen-Aarnio (Forthcoming B).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
beings could be. It is not feasible to be disposed to not put on a seatbelt only
when the ride won’t end in a crash. So, whatever the dispositions in fact being
manifested are, it is reasonable to assume that they cannot discriminate
between cases in which the safety belt will prove useful and cases in which it
won’t. But dispositions that are thus indiscriminate between good and bad
cases are problematic; it would be better to take the safety precaution and be
systematically disposed to put on a seatbelt. Note also that this judgment does
not depend on actually calculating the precise scores of the no-seatbelt and
seat-belt dispositions, and then comparing these. We simply know that when
crashes do happen, the outcomes for subjects wearing seatbelts tend to be
much better, and that there are no significant harms to wearing seatbelts.
To sum up: we often have a good idea of what kinds of dispositions are
humanly feasible. We often have a good idea of what kinds of humanly feasible
dispositions a subject’s belief (choice, action) could be a manifestation of. If all
of the feasible dispositions are dominated by (significantly) better alternatives,
then this is all we need to know in order to correctly negatively evaluate her
belief (choice, action) from the dispositional perspective.
Let me now finally apply dispositional evaluations to Lobelia’s case.
Consider again the example Lobelia and evidence about hypoxia presented in
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
the very beginning of this paper. Assume that despite the testimony of her
friend, Lobelia retains her original belief. In what follows, I will evaluate
Lobelia’s dispositions relative to the success of knowing, but parallel points
could be made given other kinds of epistemic success, such as true belief.
(Indeed, the reasoning in the section that follows is neutral between veritism
and gnosticism.)
What dispositions does Lobelia’s belief manifest? Some of these are no
doubt excellent: her belief was originally formed in an exemplary way. But
as pointed out previously, we must also evaluate the dispositions at play when
Lobelia retains her belief in light of expert testimony that it was formed by a
flawed process. In fact, we don’t need to know exactly what dispositions
Lobelia manifests, for considerations of feasibility provide a shortcut. Begin
with the following question: could Lobelia be disposed to retain belief in the
face of similar evidence (e.g., expert testimony) about her cognitive malfunc-
tioning only when such evidence is misleading? It is plausible that she could
not, if she is an ordinary human being. Consider, in particular, counterfactual
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
cases in which Lobelia has similar higher-order evidence that is not mislead-
ing, cases in which she holds a belief formed through a flawed process that
nevertheless seems to her to be perfectly fine. Being disposed to only give up
belief in response to such higher-order evidence when that evidence is not
misleading would require Lobelia to have dispositions that discriminate
between cases in which her original cognitive process is fine, and cases in
which it merely seemed to be fine. But such dispositions are not humanly
feasible.
Someone might object that by retaining her belief, Lobelia is acting in a way
that is highly atypical of human subjects. In fact, most of us might be
psychologically incapable of retaining belief in Lobelia’s situation, and for
this reason Lobelia’s case might seem puzzling. But a natural response to
such puzzlement is to start looking for explanations of what might make it
feasible for Lobelia to retain her belief. Perhaps Lobelia has a dogmatic strain
resulting from being overconfident in her own epistemic abilities. Or perhaps
she has a high need for closure, tending to seize on the first piece of evidence
available, and to then freeze her opinion, being more generally indifferent to
subsequent evidence. The fact that Lobelia’s obstinacy is atypical seems to only
strengthen my case, for all such explanations involve problematic dispositions.
Hence, whatever the precise dispositions manifested by Lobelia are, it is
natural to take her to manifest a kind of obstinacy that is indiscriminate in
problematic ways—obstinacy that manifests itself as retaining a belief formed
by a flawed cognitive process in cases in which her higher-order evidence is
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
not misleading.
Though I cannot defend these further claims here, on my view Lobelia can
still know that Penelope committed the crime. (After all, she can just see that
this is what the evidence points to—there is nothing whatsoever wrong with
her cognitive faculties!)²⁶ Further, perhaps she could even come to know that
the testimony of her friend is on this occasion misleading. She might, for
instance, reason that if she was hypoxic, it would be very unlikely for her to
have come to truly believe p, and yet, she does truly believe p. If Lobelia knows
that her higher-order evidence is misleading, then there is a perfectly good
sense in which she can discriminate the case she is in from those in which she
²⁶ Hence, when it comes to judgments about knowledge, my view is error theoretic: many epistem-
ologists have misdiagnosed cases like Lobelia’s as ones involving knowledge defeat, falsely inferring
failure to know from the manifestation of problematic dispositions. I can here only gesture toward an
error-theoretic explanation, but the rough idea is that in paradigm cases knowledge is accompanied by
good dispositions—indeed, in paradigm cases, good dispositions play an important role in explaining
knowledge.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
is in fact hypoxic, or the victim of some other cognitive defect. Call such
discrimination epistemic. This is compatible, I think, with Lobelia not being
able to discriminate her case and cases in which she is hypoxic in a disposi-
tional sense. It is not feasible, I have argued, for Lobelia to only be disposed to
ignore the kind of higher-order evidence she has in cases in which it is
misleading, and her original cognitive process was in perfect order. In retain-
ing her knowledge, Lobelia is manifesting dispositions that fail to discriminate
between cases in which her cognitive process was flawed and cases in which it
was not. As a result, the very dispositions she manifests lead her to retain belief
formed by a botched process across a wide range of counterfactual cases.
Epistemic access or discrimination does not entail dispositional doxastic
discrimination, or vice versa.
It is worth emphasizing that the mere fact that our doxastic dispositions
indiscriminately issue false beliefs in some counterfactual cases may be of no
significance, for these cases might not be relevant. For example, assume that
I come to believe the there is a tree just outside the window. Now consider
counterfactual cases in which I am looking at a very realistic-looking tree
replica instead. My dispositions cannot discriminate between cases involving
real trees and ones involving very realistic fakes. Nevertheless, my belief may
still be dispositionally good, manifesting the best (or close enough to the best)
feasible dispositions, for recherché cases involving highly realistic tree-replicas
planted outside windows are just not relevant. What is different about putative
cases of defeat by higher-order evidence is that in such cases subjects have
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
fairly strong evidence that their cognitive abilities are malfunctioning or, more
generally, that a doxastic state they are in fails by some normative standard. It
is very natural to evaluate Lobelia’s belief, now retained in light of the higher-
order evidence, by looking at other, somewhat normal cases involving such
evidence. (Indeed, epistemologists tend to talk about cases involving higher-
order evidence as forming one epistemically salient kind.) There is nothing
abnormal about such higher-order evidence not being misleading. In fact, if
anything, it is more normal for expert testimony not to be misleading than for
it to be misleading.
In sum, I have argued that by retaining her belief despite the expert
testimony that it is the output of a flawed process, Lobelia manifests disposi-
tions that fail to discriminate between cases in which her higher-order evi-
dence is misleading and cases in which it is not. As a result, these dispositions
manifest as retaining a belief that is in fact the result of a flawed cognitive
process across a wide range of relevant counterfactual cases—namely, cases in
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
which the higher-order evidence is not misleading. I will now argue that there
are better feasible alternative dispositions.
There are two salient kinds of dispositions that a subject in Lobelia’s situation
could manifest. First, she could retain her true belief (and, on my view,
knowledge), thereby manifesting dispositions that would have her retain
flawed beliefs in other relevant cases involving higher-order evidence.
Second, she could give up her belief by suspending judgment, manifesting
dispositions that in other relevant cases manifest as giving up belief,
whether or not the belief was formed by a flawed process. For ease of
exposition, in what follows I will simply speak of a “retention-disposition”
and a “suspension-disposition”. Many epistemologists have urged that a
subject in Lobelia’s situation ought to give up her belief. I will now argue
that given plausible assumptions about the value of knowledge/true belief and
the disvalue of false belief, the best feasible dispositions manifest in Lobelia’s
situation as suspending judgment. This is so even if we concede, as I think we
should, that Lobelia can retain knowledge.
We will evaluate the retention-disposition by only looking at cases in which
Lobelia retains her belief and evaluate the suspension-disposition by only
looking at cases in which Lobelia suspends judgment. Nevertheless, the rele-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
vant counterfactual cases are otherwise similar: for instance, those assigned
some significant weight, or considered relevant, are ones in which Lobelia gets
higher-order evidence that is similar to the evidence that she acquires in the
actual case. Whichever disposition we are evaluating, we can divide the
relevant counterfactual cases into three classes:
(1) Lobelia’s initial belief is the result of a good doxastic process, p is true,
and the higher-order evidence is misleading.
(2) Lobelia’s initial belief is the result of a flawed process, but
nevertheless true.
(3) Lobelia’s initial belief was false (whether as a result of a good or bad
doxastic process).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
that these weights sum up to 1). Assuming each relevant case to have the same
weight, we can think of the weight assigned to the first class of cases, for
instance, as corresponding to the proportion of counterfactual cases in which
Lobelia’s higher-order evidence is misleading. Moreover, it is plausible that
these weights should be the same whether or not we are evaluating the
retention-disposition or the suspension-disposition: the weighting of counter-
factual cases should depend on issues like how normal it is for expert testi-
mony of the sort that Lobelia receives to be misleading.
In order to score the two dispositions, we need to determine the values of
believing p (Bp) and not believing p (~ Bp) in cases falling into each of the
three classes. I have assumed that if Lobelia is able to retain her belief, then the
higher-order evidence need not destroy her knowledge. I will assume that if
Lobelia retains belief in cases that fall under (1), she continues to know (this
will just make it more difficult, not easier, to argue that the suspension-
disposition is better).²⁷ In the following table, ঋ denotes the value of knowing,
ঔ the value of truly believing, and Àআ the disvalue of falsely believing. Not
believing (for instance, suspending judgment), I will assume, has a neutral
value 0 across the board.
There are different ways of thinking about the value of true belief versus the
value of knowledge versus the disvalue of false belief. According to veritists,
true belief is valuable, and there is no added value to knowing; knowing
inherits its value from the value of true belief.
²⁷ One might object that there are some cases in which p is true, and the HOE is misleading, but
Lobelia nevertheless does not know p, due to being in some sort of Gettier case. I agree but relaxing this
idealization would only make a suspension-disposition better, not worse.
²⁸ See Dutant and Fitelson (Unpublished manuscript).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
All of this, I take it, is rather uncontroversial. I will also make the following
assumption: ঋ < আ. That is, the disvalue of false belief is greater than the value
of true belief or knowledge.²⁹
Whether one is a veritist or a gnosticist, it follows from these assumptions
that if the weight assigned to (3) is at least .5, then not believing does better
than retaining belief across the relevant counterfactual cases. Assume for
simplicity that each individual case is assigned the same non-zero weight.
We can then rephrase the point as follows: if at least half of the relevant
counterfactual cases are ones in which Lobelia’s initial belief is false, then
suspending judgment is a better overall strategy—that is, the suspension-
disposition is better. And it is plausible that Lobelia’s initial belief is false in
at least half of the relevant counterfactual cases, cases in which she gets
evidence akin to the testimony of an expert that her cognitive process is
flawed. Assume, for instance, that in 90 per cent of relevant counterfactual
cases, Lobelia’s higher-order evidence is not misleading: she is in fact hypoxic
(or suffers from some other defect). In those cases, Lobelia’s dispositions will
manifest as retaining a flawed belief. In none of those cases does the belief
constitute knowledge (presumably, even true beliefs that are outputs of flawed
cognitive processes don’t constitute knowledge). Moreover, Lobelia’s belief
will be false in the vast majority of these cases, for flawed processes rarely
happen to produce true beliefs.
One consequence of this approach is that if Lobelia’s higher-order evidence
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
was weaker, retaining belief might manifest good dispositions. Assume, for
instance, that Lobelia is told that at an earlier time, prior to forming her initial
belief, there was an objective chance of 10 percent that she would be hypoxic
by the time she formed this belief. Depending on the value of knowledge/true
belief, and the disvalue of false belief, in this case the best feasible dispositions
might manifest as retaining belief. I think this is a very good result. After all,
we should almost always assign some credence to the hypothesis that any
given belief was formed by a flawed process. Nevertheless, this shouldn’t lead
us to an unending process of second-guessing ourselves or calibrating our
opinions.
Before wrapping up, I want to distinguish my way of evaluating dispositions
from a more instrumentalist one, and to say how my view is not susceptible to
various objections leveled against consequentialist views in epistemology.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
Some dispositions are good when evaluated from the more instrumental
perspective focused on the consequences of their manifestations, but bad
when evaluated from the perspective focused on the manifestations them-
selves. To take an example from Horowitz (2019), consider an arachnophobic
subject who is almost bound to form botched beliefs about a wide range of
topics whenever she believes that there are spiders nearby. She has plenty of
evidence about the proximity of spiders, but she manages to believe that there
are in fact no spiders nearby. For such a subject, a disposition to believe, no
matter what evidence she has, that there are no spiders nearby might be an
instrumentally good disposition, since it allows her to form other good beliefs
(beliefs that are true, proportioned to the evidence, that constitute knowledge,
etc.). Nevertheless, such a disposition is not good in the sense that I have
³⁰ Earlier statements of my view did not clearly distinguish between these, though it was always the
first kind of view I had in mind. For instance, in Lasonen-Aarnio (2010), I talk about “adopting or being
guided by policies, thought of as sets of rules, that have certain consequences for knowledge.”
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
outlined, for it does not tend to have successful manifestations: across a range
of counterfactual cases including the actual one, it manifests as belief that is
false, that fails to be proportioned to the evidence, and that fails to constitute
knowledge.
It should be clear, then, that the view outlined is not susceptible to an
objection often levelled against more consequentialist views—namely, that it
allows for problematic kinds of tradeoffs.³¹ Relatedly, one might object that
evaluating dispositions in terms of the epistemic consequences of their mani-
festations takes us out of the traditional epistemic realm: though failure to get
enough sleep might have bad epistemic consequences, norms telling one to get
enough sleep are not epistemic, but practical ones.³² But of course, the
manifestations of a dispositions to stay up too long at night are not doxastic
states, and hence, not the sorts of things that could constitute epistemic
successes like knowing.
Whether we call dispositions that are good in the more direct way outlined
knowledge-conducive is a terminological choice. Similarly, whether the evalu-
ations I have outlined are consequentialist may be a terminological issue: there
might be a perfectly good sense in which a disposition has good consequences
in virtue of its manifestations tending to be successful. What matters is that
there is an important difference between two ways in which a disposition
could be said to be conducive to success, or to have good consequences for
success. The kinds of dispositional evaluations I have outlined evaluate dis-
positions by looking at how successful their manifestations are across relevant
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
³¹ For a discussion of consequentialist views in epistemology and how they allow for tradeoffs, see
Berker (2013).
³² Cf. Horowitz (2019).
³³ Though this should be very clear by now, it is also worth emphasizing that my account does not
evaluate dispositions in terms of subjective expectations. Consider the success of having accurate
credences, where a credence in a proposition p is more successful the closer it is to 1 if p is true, and
the closer it is to 0 if p is false. Schoenfield (2018) proposes a way of evaluating doxastic plans by asking
what the (subjective) expected accuracy of making the plan—of trying to update in accordance with a
given procedure—is. (Similarly, we could evaluate doxastic dispositions by asking what the expected
accuracy is of credal states produced by those dispositions.) Though our accounts have some structural
similarities, dispositional evaluations are more objective: a subject’s opinions about the goodness of a
disposition might themselves be inaccurate, or she might lack any such opinions. Note also that
Schoenfield’s results (like many results concerning the expected accuracy of update procedures) also
rely on assumptions about evidence that I would reject, like partitionality.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
5.7. Conclusions
³⁴ See Miracchi (2019) and Flores and Woodard (manuscript) for a more detailed discussion of such
cases.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
views and hence, given a range of different ways of assigning epistemic values
to doxastic states.³⁵
References
University Press.
Hurka, Thomas. (2006). Virtuous act, virtuous dispositions. Analysis, 66(1):
69–76.
Lackey, Jennifer. (2007). Why we don’t deserve credit for everything we know.
Synthese, 158(3): 345–61.
Lasonen-Aarnio, Maria. (2010). Unreasonable knowledge. Philosophical
Perspectives, 24(1): 1–21.
Lasonen-Aarnio, Maria. (2014). Higher-order evidence and the limits of defeat.
Philosophy and Phenomenological Research, 88(2): 314–45.
Lasonen-Aarnio, Maria. (2019). Higher-order defeat and evincibility. In:
M. Skipper and A. Steglich-Petersen (eds.) Higher-Order Evidence: New Essay.
Oxford: Oxford University Press.
³⁵ I am grateful to Jessica Brown, Julien Dutant, Giada Frantantonio, Jaakko Hirvelä, Aleks Knoks,
and Niall Paterson for helpful comments on earlier drafts. Many thanks to Chris Kelp for detailed
comments on a closely related talk. Many thanks to Daniel Drucker, Jeremy Goodman, John
Hawthorne, and Tim Williamson for discussions.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 23/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
6
Suspension, Higher-Order
Evidence, and Defeat
Errol Lord and Kurt Sylvan
6.1. Introduction
In regulating one’s intellectual life, one should pay heed to the evidence. Often
this is a matter of considering which facts confirm or disconfirm some
hypothesis one is considering. Although this is hard enough, it turns out
that reckoning with the evidence would be much easier if this were the only
task. For besides evidence that confirms or disconfirms a hypothesis one is
considering (first-order evidence), there is also evidence that bears on our
rationality in inquiring into the hypothesis (higher-order evidence). So, for
example, you might know that you are prone to make mistakes when reason-
ing about the significance of certain statistical information; or you might know
that, given the fact that you’ve been driving all night, you are in less than ideal
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
condition for doing mental math; or you might know that, despite your having
conscientiously reached a conclusion on a difficult subject, someone you take
to be cleverer disagrees. All these facts about one’s competence, condition, or
situation bear on whether one’s stance is rational, but it is unclear at best how
they bear on truth-value of the hypothesis one is considering.¹
As it happens, this basic contrast between first-order evidence and higher-
order evidence raises a host of puzzles and problems. These difficulties have
led many philosophers to adopt some extreme views in two related debates.
The first debate is about the rational impact of disagreement. The second is
about whether it is possible for there to be rational epistemic akrasia (e.g., cases
where one rationally has a doxastic attitude D even though one believes one
shouldn’t have D).
¹ There are possible cases in which higher-order evidence alters first-order relations of evidential
confirmation. But we will be arguing that the distinctive significance of higher-order evidence HOE does
not consist in this kind of impact.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
assumptions fail in the cases that interest us. For this reason, we also have a
positive project.
The positive view in turn has two parts. The first part is the idea that higher-
order evidence provides direct reasons for suspending judgment that typically
leave evidential support relations on the first order intact: instead of destroying
these relations, the reasons for suspension defeat or compete with the epi-
stemic reasons for belief generated by these relations. Secondly, and more
importantly, our framework explains how this defeat is possible by showing
how these distinctive reasons for suspension of judgment flow from the
constitution of suspension of judgment. As a result, our framework is embed-
ded within an account of suspension of judgment that shows how new insights
about its nature lead to a different picture of its rational profile. This frame-
work provides a compelling basis for more moderate positions about disagree-
ment and epistemic akrasia, which show the puzzles about these topics to rest
on more fundamental mistakes about suspension and the relationship between
reasons for suspension, reasons for belief, and evidence.
Our framework for explaining the reasons for suspension that derive from
higher-order evidence might be described as a sort of constitutivist framework.
But we would want to stress two things before having the reader embrace this
description. Firstly, we’re neutral on whether to think about the constitutivism
as a constitutivism of aims (leading to teleology) or a constitutivism of norms
(leading to a non-teleological view). Secondly, to be neutral in a further way,
we are here understanding constitutivism differently than it is normally
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
To appreciate why some have been led to extreme views about the epistemic-
ally appropriate response to disagreement and other sources of higher-order
evidence, it will help to reflect on examples. Consider a now-classic example
that we take verbatim from Christensen (2010: 187):
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
On the one hand, it is tempting to think that the subject’s reaction in this case
is rational, and that it would be irrational to maintain confidence that ‘only
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
From the Conclusive Reason principle and the thought that there is conclusive
reason to believe simple logical consequences of rational beliefs, we get the
conclusion that the subject in Christensen’s example cannot be irrational to
believe that the answer is ‘only (d).’ But from the Level-Bridging principle, we
appear to get the opposite conclusion.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Something, it seems, must go. Three kinds of extreme views have emerged
to explain what should go. On the one hand, some accept Conclusive Reason
but reject Level-Bridging. This option is defended by Coates (2012). He agrees
that if there is a strong undefeated reason to believe that one lacks sufficient
reason to believe that p, then one ought to believe that one lacks sufficient
reason to believe that p. But he denies that if one rationally believes that one
lacks sufficient reason to believe that p, it cannot still be rational for one to
believe that p. Hence, he thinks that rational epistemic akrasia is possible. This
idea is, we think, worth avoiding if possible.
On the other hand, some try to accept both principles but deny that they
conflict by denying that one can have a strong undefeated reason to believe a
false proposition about what the reasons recommend. Titelbaum (2015) and
Way and Whiting (2016) in effect defend this option. Although we respect the
arguments given by these authors, we also think that it would be nice to avoid
the view that mistakes about what the reasons recommend are impossible.
Such mistakes seem possible.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
positions rely on the idea that one sort of evidence is irrelevant. Equal weight
theorists argue that when you disagree, you are required to set aside the first-
order evidence you initially heeded.⁶ Once you bracket this evidence, though,
there isn’t anything left to support belief or disbelief. Thus, suspension is
required.
On the other side, there are theorists arguing that the higher-order evidence
is not relevant at all. According to them, the evidence says what it says, and we
should just follow it. Most of the arguments for this position contend that
⁴ Various definitions of epistemic peerhood are provided in the literature. The rough idea is that
your epistemic peers are those with whom you share evidence and who are just as good as you at
processing that evidence.
⁵ It is important to note that the view we develop below is, in the first instance, about outright
attitudes. There is then a hard and interesting question about the relationship between these attitudes
and credences. We don’t know how to resolve that question. But we would want to resolve it before
saying anything directly about credences. Thanks to Jade Fletcher for pressing for clarification here.
⁶ This is often done by appealing to a principle like David Christensen’s Independence principle (see
Christensen (2007), Christensen (2011), Christensen (2009); and also Elga 2007; Cohen (2013); for
criticism, see Kelly (2013); Sosa (2013); Lackey (2010); Lord (2014)).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
⁷ Most extant opponents of the equal weight view (and other extreme conciliationist views) are in
fact moderate steadfasters rather than extreme steadfasters, with Tom Kelly being the exemplar.
⁸ See Lord (2014) for discussion.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
to consider a further kind of case and a specific version of the reasoning that
has appeared in the literature. Worsnip (2018)’s discussion is, we think,
instructive. Worsnip suggests that in cases where we can rationally believe
false propositions about what the evidence supports, it can be rational to hold
doxastic attitudes that are not supported by the epistemic reasons that this
evidence provides. Now, to get to Worsnip’s conclusion in such cases, we will
need the following assumption:
From this assumption and our two earlier principles, we can infer that in such
cases, it could be rational to hold doxastic attitudes which are not supported by
the epistemic reasons. But if this assumption is false, we cannot make this
inference. We could give a different diagnosis: while the evidence bears a
strong probabilistic relation to a certain conclusion, and favors that conclusion
more than others, there is not yet sufficient reason to jump to that conclusion.
This diagnosis would avoid a tension between our two principles. It would
restore the connection between levels at the cost of weakening the link between
strength of evidential confirmation and overall strength of reasons.
Is this option worth considering? We think so. Reflecting on another hidden
assumption in the discussion of our first example gives strong reason to
consider it. Notice that in Drugs, we cannot generate a tension between the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
This principle is far from obvious, as the literature querying the relationship
between logic and reasons for belief already suggests.⁹ Admittedly, the existing
literature mainly suggests that there cannot necessarily be objective reason to
believe whatever is entailed by one’s beliefs, since these beliefs may be unjus-
tified. But as both of us have more recently stressed in developing accounts of
possessed normative reasons,¹⁰ it is also not plausible that one possesses a
reason to believe p simply in virtue of having a justified belief which entails
that p: one must also be sensitive to the relevant ground of the reason-for
⁹ For the locus classicus, see Harman (1984). ¹⁰ See Sylvan (2015); Lord (2018).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Of course, a story does need to be told about why this doctrine fails in the cases
that interest us. It may, after all, remain tempting to think that this principle at
least generates default reasons that would need to be defeated by competing
reasons, and we need to know more about the mechanics of this interaction.
As Christensen and others have noted, it isn’t enough here to appeal to the
notion of undercutting defeat, since undercutting defeaters are meant to sever
evidence-for relations, and these relations are not severed in the examples that
concern us.¹²
Here, however, we think some further existing literature serves as a useful
guide. Note that we are not the first theorists to reject this doctrine. Defenders
of pragmatic encroachment have already rejected it. Schroeder (2012)’s story is
instructive. He suggests that even if all epistemic reasons for belief are gener-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
ated by the logical and probabilistic relations between the evidence and further
propositions, it does not follow that all reasons against belief have the same
foundation. And if not, then it cannot be obvious that there is sufficient
epistemic reason to believe even straightforward consequences of one’s evi-
dence. For even if there is an impeccable reason for belief, one cannot follow it
if there is an impeccable reason against. Schroeder suggests that stakes gener-
ate such reasons against belief, yielding a deeper explanation of how pragmatic
encroachment works.¹³
¹¹ Gonzalez de Prado Salas (2020)) argues that higher-order evidence casts doubt on the Entailment-
Reasons Principle precisely because it makes it the case that one fails to possess the first-order evidence,
by making one insensitive to the grounds of the reason. This does happen, but we don’t think it is a fully
general explanation.
¹² See, for example, Christensen (2010); Lasonen-Aarnio (2014); and Pryor (2013).
¹³ As a matter of detail, Schroeder does not maintain that the stakes directly provide reasons to
suspend. Instead, he holds that the stakes provide reasons against belief, and that sometimes those
reasons are strong enough that the balance of the reasons for and against belief requires suspension. EL
argues against this in favor of the view that we need to posit direct reasons to suspend in Lord
(Forthcoming B). On pragmatic encroachment, EL then defends similar conclusions to Schroeder.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
one way, and consequentialists in another way. But while the account leaves
some theoretical options open in this respect, it is clearly a principled and
explanatory account.
We suggest that this kind of account provides the best general explanation
of why higher-order evidence generates defeaters. We will turn in the next
section to consider the nature of suspension of judgment and how this leads to
higher-order defeaters, and then return in the section after to explain in more
detail how the account resolves the gridlock and fits in the literature on
Similar moves are made within a contrastivist framework in Snedegar (2017: Ch. 6). We want to be
neutral in this paper on whether contrastivism is true. KS indicates some sympathy for contrastivism in
Fogal and Sylvan (2016). We might be willing to embrace it if it is needed to explain the weight of some
reasons for withholding, as Snedegar suggests.
¹⁴ But again, we think that the higher-order evidence provides direct reasons to suspend, not just
reasons against belief.
¹⁵ See Lord (Forthcoming B) for a more systematic argument that they are properly epistemic.
¹⁶ See, for example, Velleman (2000); Sosa (2007); Sosa (2010); Sosa (2015); Shah (2003);
Wedgwood (2002); Lord and Sylvan (2019).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
defend it. Our goal is merely to show that we have a plausible story about how
higher-order evidence provides direct reasons to suspend.
Those attracted to broadly evidentialist views—i.e., nearly all the main parti-
cipants in our two debates—might doubt the need to use direct reasons to
suspend to understand the rational impact of higher-order evidence. Why
posit direct reasons to suspend if you can get by just with the evidence?
There are several reasons why we need reasons to suspend which aren’t
determined just by evidential confirmation relations and logical relations
between evidence and attitudes.¹⁷ One of these reasons is particularly germane
here: there is no plausible straightforward evidentialist explanation of the full
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Plausibly, the fact that the coffee is laced with the drug has some impact on what
the subject epistemically ought to do. Indeed, it is plausible that he is forbidden
from believing that (d) is the right answer. He ought to suspend judgment.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Furthermore, this intuition remains even if we emphasize that his reasoning was
logically impeccable, and he had proof that (d) is the correct answer.
How might the evidentialist explain this result? One sort of evidentialist
simply can’t. This character accepts what we will call Strong Evidentialism:
The balance of the evidence for and against whether (d) is the correct answer is
clear. Again, we could even imagine that the subject has a proof that (d) is
right. So, if all we have is the first-order evidence, we can’t explain why he
¹⁸ In one way or another, this fact has been pointed out frequently in the literature. See Christensen
(2010); Schoenfield (2015a); Schoenfield (2015b). Related but more general issues are discussed by
Pryor (2013).
¹⁹ See Lord (MS); Lord (Forthcoming B) for further discussion of a similar case that is inspired by
Elga (Ms) and Schoenfield (2015a).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
This view is the sort of view that both conciliationists and steadfasters have in
mind. But here’s the problem: it is wholly unclear how the ‘total evidence’
makes it the case that I should suspend judgment. This result cannot simply be
explained by probability relations. The interaction between first and higher-
order evidence doesn’t affect the probability relations between the first-order
evidence and proposition at issue. Maybe it’s fine to say in these cases that the
total evidence ‘points toward’ suspension. But we want some explanation of
why it points toward suspension.²⁰
Reasons to suspend offer resources to explain how it is that the higher-order
evidence has the requisite effect. It isn’t because it evidentially interacts with the
first-order evidence to yield sufficiently low confirmation or make the needle
on some metaphorical evidential dial move in some direction. Rather, the
higher-order evidence directly provides a reason to suspend, which balances
against the reasons to believe. In this case, it makes sense for me to suspend.
Given what I know about my possible impairment, this is the proper response.
Thus, the reason to suspend is intuitively weightier than the reasons to believe.
Opponents will now likely complain that we haven’t provided an adequate
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
explanation. For we haven’t given a deeper story about why the reason to
suspend is weightier. That’s true, but it wasn’t what we were after. We don’t
think we need to give a general normative theory of weight; this could only be
descried through the correct fundamental normative theory, a matter on which
we wish to remain neutral here. Rather, what we want is a structural explan-
ation of how suspension could be required, and direct reasons to suspend
provide that explanation.²¹ The evidentialist lacks a clear explanation at the
same level, since it is not clear what they can invoke to do the explanatory work.
²⁰ Christensen (2010) suggests that we understand the evidence just in terms of the facts that bear on
the rationality of belief. EL argues in (MS) that this view trivializes the debate between evidentialists and
pragmatists. Better, we think, to conceive of the evidence as the facts that confirm or disconfirm
hypotheses. But doing this necessitates positing reasons to suspend.
²¹ There may be more of a puzzle in cases where one has very strong evidence for p and equally
strong evidence against p. Plausibly, these evidential relations generate correspondingly strong reasons,
and the reason to suspend must be sufficiently weighty to outweigh them both if we want to explain
why one ought to suspend here. Here it may be necessary, as Snedegar (2017: Ch. 6) suggests, to
combine the kind of account we are offering with a contrastivist picture of the reason-relation.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
They cannot exploit the aspect of evidence one usually exploits to do the
work—i.e., to the amount of confirmation or evidential support. By appealing
to direct reasons to suspend, we get additional structural resources to explain
how suspension could be required despite the confirmational facts.
This fact yields strong initial motivation for appealing to reasons for
suspension in the context of theorizing about the rational power of higher-
order evidence. In our view, this motivation is decisive. We can’t get by simply
by with the evidence for and against p. But even if we retreat to Weak
Evidentialism, we need a tool to help explain how mere higher-order evidence
can require suspension. Reasons to suspend furnish this tool.
In the next subsection we will further bolster this point by showing how the
existence of direct reasons to suspend in higher-order evidence cases flows
naturally from a plausible account of suspension of judgment, before turning
to argue that reasons to suspend provide the right tool to prop up a principled
moderate account in the two debates discussed at the outset.
which yields direct reasons for suspension on the grounds that it best captures
key functional desiderata. In fact, we will argue that there are two different
varieties of suspension, and that there are interesting and underappreciated
direct reasons for both varieties.
Let’s start by thinking about the function of belief and disbelief. A helpful
tool in this project is an intellectual agenda, which one could represent by a set
of questions on which one takes a stand or is seeking to take a stand. Belief and
disbelief are ways of determining the status of a question on your agenda.
Determining is a way of taking a settled stance on some issue. When you
determine the status of some claim via belief or disbelief, you settle the
question for yourself, and thereby add some belief to your doxastic corpus.
A functional upshot of this is that when you determine the status of a claim in
this way, the claim is available to you for rational purposes.
Suspending judgment is another way of taking a settled stance. However,
unlike belief and disbelief, when you suspend you do not add any beliefs to your
doxastic corpus. To use Scott Sturgeon’s apt phrase, when you suspend about
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
whether p, you adopt a settled neutrality when it comes to p.²² While you don’t
add any belief to your doxastic corpus, you do adopt some attitude that leaves
the issue on your agenda. This thought captures the datum that suspension is an
honest-to-goodness attitude, not merely the absence of belief and disbelief.
What is the functional upshot of adopting a state of settled neutrality? Or, to
put the question in normative terms, why should one ever adopt such an
attitude when one could just remain unopinionated about p—i.e., not form a
settled stance at all? To answer this question, we need to further investigate the
contours of settled neutrality. Cutting to the chase, we think there are two
important forms of settled neutrality.²³ On the one hand, you can form an
interrogative attitude about whether p.²⁴ As etymology and syntax suggest,
interrogative attitudes seem to be mental states directed at questions.²⁵
Paradigm examples include curiosity and wonder. When you are curious
whether p, you have an attitude directed at the question of whether p is the
case: the question is what piques your interest. In paradigmatic cases of
curiosity, you don’t believe or disbelieve, but you do have some sort of attitude
toward the question of whether p.
The functional role of this kind of state is to dispose you to determine
whether p. For example, it grounds dispositions to be on the lookout for
evidence salient to whether p, to perform actions that will help you settle
whether p, to draw your attention to other knowledge germane to p, and so
on. Hopefully it is obvious why this state often has a rational point. Frequently
we lack the resources to rationally determine the answer to some question we
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
ought to keep on the agenda. For example, in many cases where one’s evidence
is evenly split, it makes sense to keep the question of whether p on one’s agenda
even though neither belief nor disbelief is rational. It would be unfortunate if
this evidential situation forced us to drop the issue off the agenda completely.
Indeed, it is precisely in these sorts of situations that we need a state that
disposes us to settle whether p. Interrogative attitudes fit the bill.
Thus, interrogative attitudes allow us to take a settled stance without adding
some claim (or its negation) to our doxastic corpus. They do this because they
are directed at questions. We can thus take a settled stance about whether p,
and this stance can keep the issue of whether p on the agenda without
determining the answer. Furthermore, by adopting an interrogative attitude,
²² See Sturgeon (2010). ²³ For much more, see Lord (MS); Lord (Forthcoming B).
²⁴ See Friedman (2013a), Friedman (2013b), Friedman (2017).
²⁵ It is controversial whether the fundamental objects of interrogative attitudes are questions. But as
we have previously indicated, the surface appearances certainly make them appear to be question-
directed: verbs for interrogative attitudes take questions rather than that-clauses as complements, and
the word ‘interrogative’ is derived from the Latin word for questioning. For deeper arguments,
however, see Friedman (2013a).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
²⁶ In other work (Lord Forthcoming B), EL argues that anti-interrogative attitudes are also called for
in cases where you owe it to yourself (e.g., cases of so-called grit a la Morton and Paul (2018)) or others
(e.g., cases where you owe it to your friends to think well of them) to become insulated from evidence.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
learning about the laced coffee. After all, he was asked to form a view about
the question. So, he should be on the lookout for ways to determine the
answer. But given what seems to him to be his likely impairment, it is too
risky to rely on the first-order evidence. Forming an interrogative attitude is
the thing to do.
Having sketched our framework, the first thing that we want to note is that it is
on track to vindicate conciliationism of some sort.²⁷ Of course, the claim that
We put these cases to one side because of the controversy surrounding them. For other reasons to adopt
anti-interrogative attitudes, see Rosenkranz (2007).
²⁷ It’s not for nothing that we say, ‘on track.’ For there are important exceptions. For example, it is
not plausible that there will be a reason against belief every time there is disagreement. For there can be
undercutting defeaters of the reason to suspend provided by the fact about disagreement. A second
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
embraced is, after all, the right one. More specifically, he should wonder
whether he has sufficient evidence for this conclusion. If our view is right, it
follows that he should suspend judgment, since suspension of one sort just is
the holding of some interrogative attitude. Hence if some interrogative attitude
point to stress here is that it matters here how you define conciliationism. We don’t think our view
necessarily provides a vindication of conciliationism about confidence or credence. It’s an open
question how reasons to suspend impact our credences. So, it might be that facts about disagreement
don’t always provide reasons to lower our confidence, at least on some ways of understanding what it
means to lower one’s confidence. So, what we think is vindicated is conciliationism about belief: facts
about disagreement provide reasons against belief. If facts about disagreement provide higher-order
evidence and higher-order evidence provides reasons to suspend, then facts about disagreement will
provide reasons to suspend. These are reasons that compete with the reasons to believe, and thus are
reasons against believing.
²⁸ Perhaps there is some purely objective sense in which the underlying logical truths here are
obvious, but this sense is not the relevant sense for understanding rationality. Indeed, as far as we can
see, no logical truth L is so obvious from an objective point of view that it couldn’t make sense to
wonder whether L is true. Otherwise wondering about the possible correctness of various non-classical
logics would be necessarily irrational. But it just doesn’t seem right to claim that one becomes irrational
merely by taking non-classical logics seriously in meta-logical inquiry. Graham Priest may be wrong,
but he is not irrational; similarly, for people who seriously engage with his work.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
is required, one kind of suspension is required. Yet the following more specific
principle seems no less plausible than our original level-bridging principle:
Since the subject ought rationally to wonder whether he has sufficient reason
to believe p in Christensen’s case, it follows that he ought rationally to
suspend on whether his evidence supports p. Given the bridging principle,
it follows that the subject ought rationally not to believe that the answer is
‘only (d).’ Yet he intuitively continues to possess conclusive evidence for this
conclusion. So, if Logico-Evidentialism is preserved, the tension between
Level-Bridging principles and Conclusive Reason is preserved. But this is
contrary to the steadfast view at issue. Hence—barring revisions—the view
implies that Christensen’s subject would be irrational to wonder about
whether his attitude was rational.
We say that it isn’t irrational to wonder in such cases. We offer a simpler
solution: reject the background assumption of Logico-Evidentialism. This
allows us to keep our principles and the data without contempt for curiosity.
And again, rejecting Logico-Evidentialism is not an extreme option. It is what
the nature of suspension of judgment together with the systematic links
between reasons for belief and reasons for suspension recommend.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Other views encourage the same contempt for curiosity. Consider Worsnip
again. Although Worsnip doesn’t try to resolve the puzzle by keeping all the
principles, he will embrace modified versions of these principles which will
still condemn curiosity unless Logico-Evidentialism is rejected. In particular,
although Worsnip denies that reasons and rationality go together, he doesn’t
deny that there is some other normative status that co-travels with the reasons.
One such status we can call the ‘should’ of all-things-considered reason. Here is
a modified principle that Worsnip would accept:
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The only way out for Worsnip is to agree with Titelbaum about the ‘should’ of
all-things-considered reason and deny that one can have strong undefeated
reason to believe false normative propositions of the relevant sort. But this also
requires gratuitous condemnation of intellectual curiosity and wonderment.
Our view again gives a happier resolution.
Although we think extreme steadfast views are too quick to condemn
wonderment, we don’t recommend jumping to the opposite extreme.
Sometimes it is irrational to wonder whether p. If it is plain that p from the
perspective you occupy, it is less clearly acceptable to wonder whether p. For
example, if it is clear from your perspective that p (e.g., if you can easily see
that p), wondering and wondering whether p would seem to manifest inatten-
tion or some other lapse of reason. If one avoids such lapses, one can easily go
ahead and get the knowledge from the plain facts. And, if that knowledge is
obtained, one would do oneself an injustice by allowing the disagreement to
obscure one’s plain view of the facts. One deserves better.
There is a spectrum of cases between the case of wondering whether p when it
is evident that p and wondering whether you selected the right answer in a
Drugs-like case. At what point does wondering become sensible? Common sense
suggests that it is later than extreme conciliationists say, but earlier than extreme
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
steadfast views say. This was why the steadfast conciliationism introduced in
Section 6.2 was commonsensical. Now, we can see how our view can vindicate
common sense. Higher-order evidence provides reasons to suspend; hence,
conciliationism. Sometimes those reasons to suspend are not weighty enough
to defeat the reasons to believe provided by the evidence; hence, steadfasting.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The options can all be viewed as answers to the question of what funda-
mentally determines whether it is epistemically appropriate to inquire into
whether p, where inquiry is understood as seeking to help determine whether
p.²⁹ These options are constrained by obvious truths. It isn’t epistemically
appropriate to inquire into whether p in our sense if it is already clear that p or
already clear that ~p, at least where clarity entails knowledge or being in a
position to know (as we assume). It also isn’t epistemically appropriate to
inquire into whether p in our sense if it is clear that it couldn’t be determined
whether p.³⁰ These obvious truths give us minimal necessary conditions on
epistemically appropriate inquiry: S’s inquiry into whether p is epistemically
appropriate only if (a) the truth-value of p isn’t evident to S, and (b) it isn’t
evident to S that S couldn’t help to determine whether p by her inquiry.
What else is necessary? Strong liberalism would say nothing else is neces-
sary. Unfortunately, this view may seem to give a thumbs-up to foolhardy
inquiry. If it is unlikely that S could even help to determine whether p, it may
seem a fool’s errand for S to inquire into whether p with any earnestness.³¹
Barring rejection of this intuition, it seems we need other conditions on
epistemically appropriate inquiry.
We can now see some important large-scale divisions. On the one hand, we
can make a broad distinction between consequentialist views and non-
²⁹ We say ‘help determine’ to indicate that one’s contribution to inquiry may be part of a larger
inquiry—say, the inquiry of a research team, a social group tasked with answering certain questions
(e.g., political figures interested in whether certain acts would be constitutional), or even a whole
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
academic discipline or intellectual tradition. It may be clear that one couldn’t alone determine whether
p in one’s life even if it is clear one could partly help some relevant group could determine whether
p. Here it may make sense to inquire. But if it is evident that one couldn’t even help advance an inquiry,
it does seem to us foolhardy to inquire into whether p. (Thanks to Justin Snedegar for making us clarify
our beliefs here.) It is worth stressing that there is an important contrasting notion of reflection on an
issue that doesn’t assume that the issue could ever be rationally decided by an individual or a group. It
doesn’t, we assume, make sense to seek to determine whether p if it is obvious that p can’t be
determined. Does it follow that it makes no sense to think about p? This seems extreme and indeed
anti-philosophical: even if we know we can’t settle whether p, it may be worth thinking about what it
would take to settle whether p, and what would need to be true for p to be true. Even if no intuitive
evidence could ever decide between whether natural properties are tropes or universals, it doesn’t
follow that the nature of naturalness isn’t worth contemplating. For this reason, we might call the wider
notion of inquiry contemplation. It can even make sense to contemplate what it would be for a
proposition to turn out true even if one knows it is false: one might deny that gods are actual but
still find theology a worthy study in speculative metaphysics. We leave the constitutive norms of
contemplation (if there are any) for other theorists to determine. It does, however, seem plausible to us
that ideas can be worth contemplating even if we know we will never be able to determine whether they
are right. (Indeed, ideas can be worth contemplating even if they are obviously false: consider Leibniz’s
delightful ontology of monads, for example.)
³⁰ In this case, the only sort of attitude that is rational is an anti-interrogative attitude (although,
sometimes one should form no attitude at all).
³¹ Again, in these cases, it is tempting to think one should form an anti-interrogative attitude. These
reasons aren’t as strong as the reasons to form anti-interrogative attitudes when we know p is
undeterminable, but they might be very strong.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
fertile fields. Such a consequentialist may, however, believe that some matters
have great enough intellectual importance that even if they are not particularly
likely to be determined in good time, spending more time on them is worth
the risk.
Impurist consequentialism allows further values that are not purely intel-
lectual to encroach on the significance of some issue. Such a consequentialist
might instead propose a third condition on epistemically proper inquiry along
the following sorts of lines: inquiry into whether p is epistemically proper only
if it is likely enough that p can be determined in good time, compatibly with a
plan for conduct that maximizes the production of practically important
knowledge.
These two kinds of consequentialism have different implications about the
conditions under which higher-order evidence can defeat belief. To see why,
let’s first consider more officially what our framework would say about when
suspension of judgment is rational. According to our framework, suspending
judgment about whether p is rational if and only if it is rational to have either an
interrogative attitude toward whether p or an anti-interrogative attitude toward
whether p. More briefly, given the foregoing points: suspending judgment
on whether p is rational if and only if it is epistemically proper to inquire into
whether p, assuming the issue shouldn’t just be tabled. For simplicity, we can
bracket issues that should be tabled by restricting attention to live issues (to
borrow some language from William James (1896). It is then epistemically
proper to suspend judgment on some live issue if and only if inquiry into the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
defeats, as we noted earlier. What we have just said suggests a deeper potential
reason for signing up to their story: the nature of suspension of judgment
together with the systematic connections between the weight of reasons for belief
and the weight of reasons for suspension provide independent reasons to think
that certain sorts of consequences might decide the sufficiency of the evidence.
Nonetheless, we think it would be too quick to assume that the nature of
suspension of judgment and data about higher-order defeat force a conse-
quentialist treatment; indeed, one of us is a diehard Kantian. We will conclude
with some thoughts about the compatibility of non-consequentialism with
our framework, and how a non-consequentialist might offer a different story
about higher-order defeat and explain away the apparent evidence for
consequentialism.
Epistemic non-consequentialists will deny that fundamental epistemic val-
ues such as accuracy are merely ‘to be promoted,’³³ and for this reason oppose
³³ See, e.g., Sylvan (2018); Sylvan (2020); Littlejohn (2011); Gibbons (2013).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
³⁴ Perhaps the subject should adopt an anti-interrogative attitude instead, but this also seems a bit
extreme: more plausibly, one can and should just go with what repeated witnessing of truth makes
plain.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
³⁵ For other kinds of objections that we think would generalize, see Sylvan (2020).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
properly epistemic reason to love truth. At some point, we should all learn to
stop querying and embrace the truth.³⁶
6.6. Conclusion
We must leave the task of deciding between these options for future work: it is a
task worthy of many papers and books. What we would like to emphasize
in conclusion is that our wondrous structural resolution of the puzzles indi-
cates that wider controversies—most centrally, consequentialism vs. non-
consequentialism and purism vs. impurism—must be settled in order for us
to know exactly how weighty the higher-order evidence will be in a given case.
As far as the general rationale for accepting higher-order defeat goes, we don’t
think much more can be said beyond what we’ve said earlier. This fact is
heartening for those of us who are more interested in theory than application:
for what we’ve seen suggests that some of the apparent gridlock in the literature
on disagreement and higher-order evidence owes simply to the fact that these
topics are applied topics, and could never be fully resolved without the help of a
systematic normative theory (barring epistemic particularism).
References
³⁶ Thanks to Jade Fletcher, Mona Simion, Justin Snedegar, and Daniel Whiting for detailed
comments, and to audiences at the University of Southampton and New York University for questions
which shaped the paper.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Cohen, S. (2013). A defense of the (almost) equal weight view. In: J. Lackey and
D. Christensen (eds.) The Epistemology of Disagreement: New Essays. Oxford:
Oxford University Press.
Elga, A. (2007). Reflection and disagreement. Nous, 41: 478–502.
Elga, A. Ms. Lucky to Be Rational. Unpublished paper (written 2008), Princeton
University.
Fogal, D. and Sylvan, K. (2016). Contextualism about epistemic reasons. In:
J. Ichikawa (ed.) The Routledge Handbook of Epistemic Contextualism.
London: Routledge.
Friedman, J. (2013a). Question-directed attitudes. Philosophical Perspectives, 27:
145–74.
Friedman, J. (2013b). Suspended judgment. Philosophical Studies, 162: 165–81.
Friedman, J. (2017). Why suspend judging? Nous, 51(2): 302–26.
Gibbons, J. (2013). The Norm of Belief. Oxford: Oxford University Press.
Gonzalez de Prado Salas, J. (2020). Dispossessing defeat. Philosophy and
Phenomenological Research, 101: 323–40.
Harman, G. (1984). Logic and reasoning. Synthese, 60: 107–27.
James, W. (1896). The will to believe. Reprinted in The Will to Believe and Other
Essays in Popular Philosophy (1956). New York: Dover.
Kelly, T. (2013). Disagreement and the burdens of judgment. In: D. Christensen
and J. Lackey (eds.) The Epistemology of Disagreement: New Essays. Oxford:
Oxford University Press.
Kolodny, N. (2005). Why be rational? Mind, 114: 509–63.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Sylvan, K. (2015). What apparent reasons appear to be. Philosophical Studies, 172:
587–606.
Sylvan, K. (2018). Veritism unswamped. Mind, 127: 381–435.
Sylvan, K. (2020 ). An epistemic non-consequentialism. Philosophical Review 129:
1–51.
Titelbaum, M. (2015). Rationality’s fixed point (or: In defense of right reason).
Oxford Studies in Epistemology, 5: 253–94.
Velleman, D. (2000). On the aim of belief. In: D. Velleman (ed.) The Possibility of
Practical Reason. Oxford: Oxford University Press.
Way, J. (2010). The normativity of rationality. Philosophy Compass, 5: 1057–68.
Way, J. and Whiting, D. (2016). If you justifiably believe you ought to ϕ, you ought
to ϕ. Philosophical Studies, 173: 1873–95.
Weatherson, B. (2019). Normative Externalism. Oxford: Oxford University Press.
Wedgwood, R. (2002). Practical reason and desire. Australasian Journal of
Philosophy, 80: 345–58.
Worsnip, A. (2018). The conflict of evidence and coherence. Philosophy and
Phenomenological Research, 96: 3–44.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
7
Reasons for Reliabilism
Bob Beddor
Reliabilism holds considerable appeal. First, it captures the intuition that there
is an important connection between justification and truth: a reliable process is
standardly defined as a process that produces a high ratio of true to false
beliefs. Second, it explains how justification reduces to non-epistemic proper-
ties. Justification is explained in terms of reliability, which is explained in
terms of truth and falsity.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹ The locus classicus of reliabilism about justification is Goldman (1979). For further development
and defense, see Goldman (1986, 2012); Kornblith (2002); Lyons (2009).
² The reductive goals of reliabilism are clearly announced by Goldman, who writes: “I want a theory
of justified belief to specify in non-epistemic terms when a belief is justified” (1979: 90). As Kim (1988)
notes, one motivation for seeking a reductive account is that epistemic properties supervene on natural
properties. A reductive account offers to explain this supervenience.
³ For an overview of the major challenges and replies, see Goldman and Beddor (2015).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
05.003.0007
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
prima facie reasons. Defeat occurs when this support is undermined by the
acquisition of further reasons. This approach to defeat has been elaborated in
great detail by John Pollock, who develops a rigorous system for computing the
defeat statuses of an agent’s beliefs on the basis of their prima facie reasons.⁴
Despite these advantages, the Reasons First framework faces difficulties of
its own. As standardly developed, the Reasons First framework lacks the
explanatory benefits that make reliabilism attractive. It leaves unexplained
the intuitive connection between justification and truth, and it does not reduce
epistemic properties to non-epistemic properties. After all, Reasons Firsters
take the notion of a reason to believe as a primitive. But this is clearly an
epistemic notion.
Given this tradeoff, we might hope for a theory that combines the attractions
of both traditions. This paper develops one such theory: “Reasons Reliabilism.”
⁴ See Pollock (1987, 1992, 1994, 1995, 2001). For a useful overview of the development of Pollock’s
framework, see Prakken and Horty (2012).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
The theory follows the Reasons First tradition in explaining defeat in terms of
the notion of a reason to believe, which is taken to be the most fundamental
normative notion. But, unlike most Reasons First views, my theory does not
leave this notion unanalyzed. Rather, it goes on to explain this notion in
reliabilist terms. Simplifying slightly, I identify an agent’s reasons for belief
with the states that serve as potential inputs to their reliable processes. What
emerges is a more reasonable form of reliabilism, one that preserves the best of
both traditions. In particular, it provides an elegant treatment of defeat while
remaining faithful to reliabilism’s reductive project.
Of course, I am not the first to advocate an ‘impure’ or ‘hybrid’ version of
reliabilism. In recent years a number of other authors have proposed syntheses
of reliabilism and evidentialism.⁵ However, I argue that extant evidentialist-
reliabilist hybrids lack one of the chief advantages of the view advocated here:
namely, its ability to explain defeat. Much like the ‘pure’ Reasons First
framework, extant hybrids struggle to provide a theory that is both reductive
and predictive. They also problematically single out a privileged class of
beliefs—those entailed by the agent’s evidence—as immune to defeat. The
synthesis developed here fares better on both counts.
Seeing Red: Lori is gazing at a wall, which appears red. Consequently, she
comes to believe : The wall is red. Just then, a generally reliable
⁵ See Alston (1988); Henderson et al. (2007); Comesaña (2010, 2018); Goldman (2011); Tang
(2016); Pettigrew (2018); Miller (2019). Most of these syntheses have been motivated by considerations
other than defeat, though Miller (2019) is an important exception.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
acquaintance, Sal, mentions to Lori that there are hidden red lights angled
towards the wall.⁶
Reliabilists have not left this question unanswered. The standard reliabilist
strategy is to explain defeat in terms of counterfactuals about what the agent
would have believed, were they to have used some alternative reliable process.⁹
More precisely:
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
⁶ For discussion of this sort of case, see a.o., Chisholm (1966); Pollock (1995); Lasonen-Aarnio
(2010a).
⁷ Perhaps, some may suggest, once Lori receives Sal’s testimony, her belief in is no longer the
result of vision alone. Rather, it’s the result of a complex process: using vision while disregarding
testimony that vision is locally unreliable. Arguably, this complex process is unreliable. However, I think
there is reason to be skeptical of this ‘typing maneuver.’ In order for it to handle all cases of defeat, we
would need to take on board a substantive commitment: in every case of defeat, there is some way w of
typing the agent’s belief-forming process on which it comes out unreliable. And in order to ensure that
this approach is not ad hoc, we’d need to give some independent motivation for typing the belief-
forming process using w. For further development of this concern, see Beddor (2015a): 147–8, where
I argue that this typing maneuver stands in tension with promising solutions to the generality problem.
For related criticisms of the typing maneuver, see Lasonen-Aarnio (2010a): 4–7; Baker-Hytch and
Benton (2015): 45–7.
⁸ Adding a ‘No Defeaters’ condition may also help the reliabilist deal with other challenges. Take
Bonjour’s (1985) famous example of Norman the clairvoyant. Norman has a reliable clairvoyant faculty
but has no reason to suspect that he does. One day his clairvoyance causes him to believe that the
president is in New York. Bonjour contends that Norman’s belief is not justified, even though it is
reliably formed. As Goldman notes (1986: 112), one strategy for handling this case is to maintain that
Norman’s belief is prima facie justified but defeated.
⁹ This proposal dates back to Goldman (1979), and has been recently defended by Lyons (2009,
2016). Close cousins are defended in Grundmann (2009) and Bedke (2010).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Two Testimony Seeing Red: as before, Lori believes the wall is red, based on
its appearance. And, as before, Sal comes along and mentions that the wall is
illuminated by red lights. But now another reliable acquaintance, Anne, comes
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
According to ARP, Lori’s belief in is defeated, even after receiving the
evidence of Sal’s mendacity. After all, Anne’s testimony is misleading, and so
trusting Sal’s testimony continues to be a reliable process. And this process
remains available to Lori. (Lori could, after all, simply disregard Anne’s
testimony.) But this is the wrong verdict. Intuitively, Anne’s testimony defeats
the defeater provided by Sal’s testimony. In doing so, it reinstates Lori’s
justification for believing .
A second concern for ARP was raised by Fumerton (1988) but has not
received much attention in the subsequent literature. The worry is that ARP,
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
when properly unpacked, smuggles the notion of ultima facie justification into
the analysis of defeat. As a result, the reliabilist account of justification fails to
be reductive; worse still, it is circular. Fleshing out this worry requires some
stage setting.
Many reliabilists opt for a theory that distinguishes between inferential and
non-inferential beliefs. To motivate this complication, consider an inferential
process such as deducing the consequences of what one already believes. This
process is not reliable or unreliable simpliciter; it is only conditionally reliable.
In order to handle beliefs formed through conditionally reliable processes,
Goldman (1979) officially formulates reliabilism as a recursive theory:
Recursive Reliabilism
Base Clause: if (i) A’s belief B results from a belief-independent process that
is unconditionally reliable, and (ii) B is undefeated, then B is ultima facie
justified.
Recursive Clause: if (i) A’s belief B results from a belief-dependent process
that is conditionally reliable, (ii) the inputs to this process were ultima facie
justified, and (iii) B is undefeated, then B is ultima facie justified.¹⁰
The recursive clause requires that the conditionally reliable processes operate
on ultima facie justified beliefs. Despite this, the theory is still reductive. After
all, we can use the theory to explain what it is for these input beliefs to be
ultima facie justified. Either these inputs are themselves inferential or they are
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
not. If they are, we appeal once again to the recursive clause; if not, we appeal
to the base clause. Either way, we eventually arrive at some foundational beliefs
whose justificatory status can be explained using the base clause. Now, the only
epistemic notion that appears in the base clause is the notion of being
undefeated. Assuming ARP provides a reductive account of this notion,
Recursive Reliabilism is reductive.
Where, then, lies the problem? Fumerton observes that, according to ARP,
conditionally reliable processes sometimes function as defeaters: a belief is
defeated if there is some conditionally reliable process available to the agent
which, had it been used, would have resulted in the agent no longer holding the
belief. But a conditionally reliable process cannot lead someone to abandon a
belief all on its own; it can only do so if it is fed certain inputs. This raises the
question: what epistemic status do these input beliefs need to have? Fumerton
¹⁰ See also Lyons (2013), who argues that distinguishing between inferential and non-inferential
beliefs can help reliabilists handle the new evil demon problem (Cohen 1984).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
suggests that just as Recursive Reliabilism required that the inputs be ultima facie
justified, so too should ARP. And so, ARP really amounts to the following:
But, if this is the proper way of understanding ARP, then reliabilism’s reduc-
tive project is in trouble. After all, Recursive Reliabilism qualified as reductive
because the base clause purported to tell us what it takes for a foundational
belief to be ultima facie justified without using any epistemic terms in the
analysans. But if we use ARP Unpacked to explain what it is for a belief to be
undefeated, the base clause will itself rely on the notion of ultima facie
justification.
Perhaps, some may reply, this just shows we should reject the assumption
that the inputs to a conditionally reliable process need to be ultima facie
justified. Perhaps the input beliefs need only be prima facie justified; or
perhaps they do not even need this slender epistemic merit.¹¹ To make trouble
for this reply, consider a variant of Seeing Red where Lori unjustifiably
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
believes that the wall is illuminated by red lights, but continues to believe
anyway. Clearly, her overall set of beliefs is epistemically defective. But
does this defect render her belief in unjustified? To answer this, it will help
to consider a slightly different question: Should Lori abandon her belief in
? I think not. After all, she has no good reason to abandon it. Instead, it’s
her belief about the lighting that should get the boot. But if Lori should retain
her belief in , it becomes hard to maintain that this belief is defeated. (If it
is defeated, then she should abandon it!) This intuition motivates the idea that
only ultima facie justified inputs can serve as defeaters. But if we hang on to
this idea, the circularity problem remains.¹²
¹¹ A number of authors have defended the idea that unjustified beliefs can function as defeaters, e.g.
Lackey (1999); Bergmann (2006). Goldman himself flirts with this view in places (1986: 62, 111).
¹² One way out of this problem would be to reinterpret ARP as providing a theory of prima facie
defeat rather than ultima facie defeat. So, in our variant of Seeing Red, Lori’s belief is is prima facie
defeated, but this defeater is itself defeated by whatever considerations make her belief about the
lighting conditions unjustified. But this response is unavailable to proponents of ARP, since—as we’ve
seen—ARP lacks the resources to handle defeater defeat.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
¹³ The formal epistemology literature offers a natural place to look for a measure of reliability along
these lines. See the discussions of credal scoring rules in Joyce (1998); Moss (2011); Pettigrew (2016),
among many others.
¹⁴ In my earlier paper, I also offered a counterexample to the necessity of ARP for defeat. In the
proposed counterexample, Clarence reliably forms a belief that p; a reliable interlocutor later tells him
that p is false. But Clarence irrationally disbelieves everything this interlocutor says. Moreover, no
amount of reflection or counseling would ever uproot this deep-seated mistrust. Intuitively, Clarence’s
belief is defeated. But ARP seem unable to deliver this result: there is no process available to Clarence
which would lead him to trust his interlocutor; hence there is no process available to him which, if used,
would lead him to abandon his belief in p (cf. Baker-Hytch and Benton 2015: 53). However, in this
case—unlike Thinking About Unger—it now seems to me that a modification of ARP will suffice. The
key is to reformulate ARP in terms of dispositions rather than counterfactuals: A’s belief that p is
defeated iff there is some alternative reliable process available to A which, when fed A’s current states as
input, is disposed to lead A to cease believing p. Defenders of ARP could then propose that Clarence has
a general testimony-believer process available to him. This is a process that, for any testifier (or, perhaps,
any testifier that he has no good reason to distrust), produces a relatively high credence in their
testimony. This process is generally reliable. Moreover, it is disposed to lead Clarence to cease believing
p, when fed the experience of receiving his interlocutor’s testimony as input. It’s just that this
disposition is masked by Clarence’s mistrust of his interlocutor.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
ST Sal’s testimony
α
Lori’s wall illuminated
VE RL
visual experience by red lights
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
ve doesn’t
wall is red R U reliably indicate R
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
This formulation invites two questions. First, what does it mean for a belief to
be the result of an inference branch? For starters, the belief must be supported
by the sequence of reasons represented by the inference branch. But this is not
sufficient: the agent must actually have gone through this reasoning and hold
the belief on this basis.
Second, what does it mean for an inference branch to be ultimately
undefeated? This question is harder, and Pollock’s answer proceeds in stages.
The first stage is to give an account of what it means for one reason to be
defeated by another. According to Pollock, there are two species of defeat:
rebutting defeaters and undercutting defeaters. A rebutting defeater for a node
n is a prima facie reason to think that n is false. By contrast, an undercutting
defeater for n targets the inferential connection between n and the reasons that
support it. Seeing Red is like this: Sal’s testimony that the wall is illuminated
by red lights is not itself a reason to believe that the wall is not red, but it is a
reason for thinking that the building’s appearance does not give good grounds
for thinking that the wall is red. While there are different ways of fleshing out
the notion of an undercutting defeater, for our purposes we can define an
undercutting defeater for n as a prima facie reason for thinking that the
considerations that support n do not reliably indicate its truth in the agent’s
present circumstances.¹⁶
If we follow Pollock in assuming that these are the only two species of
defeat,¹⁷ we can venture the following disjunctive definition of when one
inference branch defeats another (cf. Pollock 1992):
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁶ The ‘in the present circumstances’ qualification is important. After all, Sal’s testimony does not
provide a reason for doubting that reddish wall appearances usually indicate the presence of a red wall.
¹⁷ This is a controversial assumption; some have suggested that cases of higher order evidence
constitute a distinct species of defeat (e.g., Christensen 2010). My own inclination is to think that when
higher order evidence functions as a genuine defeater, it does so by indicating that the agent’s actual
(or believed) grounds for their belief do not reliably indicate the truth of this belief. If this is right, then
defeat by higher order evidence is really just a type of undercutting defeat. But this is a debate that will
need to be deferred to another occasion.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
This gives us a definition of when one inference branch defeats another. But
what we really want is a definition of when an inference branch is ultimately
undefeated. The simplest option would be to say that an inference branch is
ultimately undefeated just in case there is no inference branch that defeats it.
But this delivers the wrong results in cases of defeater defeat.
Recall Two Testimony Seeing Red, in which Anne testifies that Sal is a
compulsive liar. As in Seeing Red, Sal’s testimony supports the node, The wall
is illuminated by red lights (), which supports the node, Lori’s visual
experience doesn’t reliably indicate the truth of , which undercuts .
Hence the simple account predicts that branch α is ultimately defeated, and
hence that Lori’s belief in is unjustified. But this is wrong. After all, Anne’s
testimony provides a prima facie reason to believe that Sal is a liar, which
provides a prima facie reason to believe that Sal’s testimony does not reliably
indicate , which undercuts . As noted in Section 7.3.1, it thereby reinstates
Lori’s justification for believing (see Fig. 2).
For this reason, Pollock opts for a somewhat more complicated account of
what it takes for an inference branch to be ultimately undefeated. Here we will
Anne’s testimony at
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
β
Sal is a
Sal’s testimony st cl
compulsive liar
α
Lori’s wall illuminated
ve rl uʹ
visual experience by red lights
st doesn’t
reliably indicate rl
ve doesn’t
wall is red r u reliably indicate R
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
In At A Level
To get a feel for this proposal, let’s walk through how it applies to Two
Testimony Seeing Red. While all three inference branches depicted in Fig. 2
are in at level 0, only γ is in at every level, since only it lacks a defeater. Since γ
defeats β, the latter is out at every level 1. And so, while α is out at level 1
(since it is defeated by a branch that is in at level 0), it is back in at level 2, and
remains in at every level thereafter (see Table 1). So, α is ultimately undefeated.
Hence Lori’s belief in qualifies as ultima facie justified, as desired.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Level α β γ
0 in in in
1 out out in
2 in out in
n>2 in out in
Pollock’s framework thus offers a promising way of handling some of the cases
that created trouble for ARP—in particular, cases of defeater defeat. Why not
just jettison reliabilism in favor of the Reasons First program?
There are a number of reasons why one might be dissatisfied with Pollock’s
theory as it stands. Some of these are issues of detail which Pollock himself
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
sought to address in later work. For example, Pollock later refines his character-
ization of what it takes for an inference branch to be ultimately undefeated
refinements that are mainly motivated by self-defeating inferences. He also
complicates the account by adding additional structure in order to represent
the strengths of reasons for belief.¹⁸ Other concerns focus on specific applications
of Pollock’s framework to various philosophical puzzles, such as the lottery
paradox.¹⁹ I will set these worries aside, since they are not directly relevant to
our purposes. Rather, I want to raise two more fundamental concerns.
One concern is that Pollock’s account does not accommodate the intuitions
and impulses that motivate reliabilism. First, it does not capture the intuition
that there is an important connection between justification and truth. Suppose
a belief is based on undefeated reasons that support it. Why should we expect
the belief to be connected with the truth in any interesting way? Pollock’s
framework provides no answer. Second, and more critically, Pollock’s frame-
work does not satisfy the reductive impulse behind reliabilism. Pollock
explains ultima facie justification and defeat in terms of the notion of a
prima facie reason for believing. But this is surely an epistemic notion.
Those who want an analysis of justification in non-epistemic terms will be
left empty-handed.
Of course, some Reasons Firsters might retort that we should never have
hoped for a reductive analysis in the first place. Reductive analyses of other
epistemic phenomena have a spotty track record, the analysis of knowledge
being a case in point. Why think that the prospects for a reductive analysis of
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁸ See Pollock (1994, 1995) for his refined treatment of self-defeating inference branches. See Pollock
(2001) for discussion of strengths of reasons.
¹⁹ See Lasonen-Aarnio (2010b).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
What states play this role? The clearest candidates are perceptual experiences.
For example, Lori’s visual experience of a red-looking wall is a prima facie
reason to believe . Why? Because she has a reliable process that takes the
contents of her perceptual experiences as input and produces a belief in those
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
²¹ Here the details will depend on how we individuate perceptual processes. Is veridical perception a
different process from non-veridical perception? If yes, then Reliable Reasons predicts that whether
Lori’s experience provides a genuine reason for believing depends on whether she is hallucinating.
For our purposes, we can refrain from taking a stand on this issue; I intend my reliabilist account of
reasons to be compatible with various positions on how to individuate processes, perceptual processes
included.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
agent’s reasons supervene on their non-factive mental states, and that agents
have privileged access to their reasons for belief. From this internalist per-
spective, the account offered here risks mangling the concept of reasons
beyond recognition.
But is the ordinary conception of reasons really internalist? Compare the
dialectic here with debates over the nature of justification. Prior to 70s, most
epistemologists embraced some form of internalism about justification. But
from this the fact alone it does not follow that the folk notion of justification is
internalist, as reliabilists will take pains to insist. Similar remarks apply to
reasons. It is questionable whether the folk concept of reasons is committed—
even implicitly—to technical claims about the supervenience of reasons on our
non-factive mental states.
A more plausible view is that the folk concept of reasons is exhausted by
various platitudes connecting reasons with other epistemic notions—e.g.,
‘A justified belief is supported by reasons’ (Section 7.4). Such platitudes are
consistent with both internalist and externalist construals of the relevant
notions, as well will now see.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
inference branch defeats another. And they can go on to analyze what it is for
one inference branch to defeat another in terms of prima facie reasons. But
what it takes for there to be such a reason is now explained in reliabilist terms.
Call the resulting combination of Reliable Reasons and Pollock’s frame-
work, ‘Reasons Reliabilism.’ I now argue that this synthesis preserves the
primary advantages of both frameworks.²²
In Section 7.5, I raised some grounds for dissatisfaction with a ‘pure’ version of
Pollock’s Reasons First framework. The first concern was that a pure Reasons
First framework fails to accommodate the intuitions and commitments that
lend plausibility to reliabilism. The synthesis advocated here fares better in this
respect. First, it captures the intuition that there is an important connection
between justification and truth. On the approach defended here, for an agent
to have a reason to believe p is for them to have a reliable—hence truth-
conducive—process that is disposed to produce a belief in p. Second, our
synthesis remains faithful to reliabilism’s reductive ambitions. Rather than
resting content with a non-reductive analysis of justification in terms of the
notion of a reason to believe, Reliable Reasons shows how this notion can be
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
²² My proposed synthesis bears comparison to the theory advanced in Graham and Lyons
(Forthcoming)—a proposal which was developed independently of the theory offered here. Graham
and Lyons suggest a theory of defeat built around the notion of prima facie warrants, which they define
in terms of available cognitive processes which would likely produce certain outputs if they were used.
In many respects, their approach is complementary to the account offered here. However, there are
some important differences. First, their theory is structured around the notion of warrants rather than
reasons—indeed, Graham and Lyons reject the need to appeal to reasons in a theory of justification. By
contrast, it is one of the contentions of the present essay that once we naturalize reasons, they pose no
threat to the reliabilist. Second, their account is not explicitly reliabilist: they define prima facie
warrants in terms of available processes that satisfy the requirements of prima facie justification,
without taking a stand on how best to cash out the latter requirements. Third, Graham and Lyons’
analysis of warrants takes a counterfactual form. This raises the worry that it will be subject to versions
of the counterfactual fallacy-style worries that beset ARP (e.g., the sort of worries raised in
Section 7.3.3). By contrast, I’ll be arguing shortly that Reasons Reliabilism avoids these concerns.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Reliable Reasons answers this question. What all reasons have in common is
that they serve as the inputs to reliable (or conditionally reliable) processes.
The upshot: Reasons Reliabilism avoids the main concerns with a pure
Reasons First approach.
I’ll now argue that my proposed synthesis also preserves the main advantage of
Pollock’s framework—specifically, its superior treatment of defeat. To make
this point, let us revisit the difficulties facing ARP and see how Reasons
Reliabilism avoids them.
ultimately undefeated, since the only inference branch that defeats it is out
at every level 1.
However, Reasons Reliabilists do not stop there. They supplement this
formal representation of Lori’s reasons with a reliabilist account of where
the nodes come from, and why the various support and defeat links hold.
According to Reliable Reasons, the reason why the experience of receiving
Anne’s testimony is a prima facie reason to believe that Sal is a compulsive liar
is that there is a reliable process (believing reliable interlocutors) available to
Lori that, when fed this experience as input, is disposed to produce a belief that
Sal is a compulsive liar. Similar remarks apply, mutatis mutandis, to the other
nodes depicted in the graph.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
of . And so, according to Reliable Reasons, Harry doesn’t have either a
rebutting or an undercutting defeater for his belief in .
Perhaps, some may suggest, the problem re-emerges if we amend the case.
Meet Elijah the eliminativist. Elijah thinks trees do not exist, and he wants
others to share this belief. Suppose that Harry has a highly reliable Elijah
Predictor, which is part of an Elijah Emulator: if he were to predict that
Elijah would advise him to believe p in his current situation, this would in
²³ At the same time, Reasons Reliabilism respects Fumerton’s claim that a conditionally reliable
process can only serve as a defeater if it is applied to ultima facie justified inputs. Proof: Suppose A’s
inference graph contains some node n, and suppose that A unjustifiably believes some proposition q
that, when fed into a conditionally reliable process, is disposed to produce a belief in d, where d is either
of the form, ¬n or n’s immediate ancestors do not reliably indicate n. Since A’s belief that q is unjustified,
it follows (from Justified Belief as Undefeated Reasoning) that either (i) A’s belief in q is not the result of
any inference branch, or (ii) it is the result of some inference branch, but one of the branch’s nodes is
ultimately defeated. If (i), then A doesn’t even have a prima facie reason to believe q, and so (by Reliable
Reasons) q is not a prima facie reason to believe d. If (ii), then q is a prima facie reason to believe d, but q
is ultimately defeated. Either way, n is not ultimately defeated.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
turn lead him to believe p. And so, if Harry were to apply his current
experiential states to his Elijah Predictor, he would be led to believe ¬.
Does Reasons Reliabilism predict that this gives Harry a rebutting defeater
for his belief in ?
An initial point: given this way of describing the case, it is doubtful whether
Harry’s Elijah Emulator is reliable. After all, it systematically misleads him
about the presence of trees, which hardly bodes well for its reliability!
However, the objector might try to circumvent this point by tempering
Elijah’s eliminativism. Just stipulate that, much like Shmunger before him,
Elijah has entirely correct beliefs about all sorts of topics—astronomy, geog-
raphy, physics, whatever. It is only when it comes to trees that Elijah is an
eliminativist. And so, Harry’s Elijah Emulator is overall reliable, even though it
is unreliable on arboreal matters.
Suppose we grant all of this. Then Reasons Reliabilism does indeed predict
that Harry has a rebutting defeater for his belief in . However, this need
not worry us, provided that this defeater is itself defeated. Consider: why,
exactly, would Harry be unjustified in using his Elijah Emulator in his current
situation? Presumably, because he has good reason to think that his visual
experience of trees reliably correlates with the presence of trees. Where does
this reason come from? Presumably from his past experiences, which support
the generalization that having a visual experience representing x is a reliable
indicator of the presence of x. Plausibly, there is a reliable process available to
him that, given these past experiences as input, is disposed to produce a belief
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
that his current tree-like experiences reliably indicate rather than ¬.
If so, these past experiences constitute an undercutting defeater for his rebut-
ting defeater for believing (see Fig. 3). And so, his belief in counts as
ultima facie justified, as desired.
Could proponents of ARP co-opt this response? No. After all, the key move
here is to diagnose this variant scenario as a case of defeater defeat. But we’ve
seen that ARP lacks an adequate treatment of defeater defeat.
Tree in ea doesn’t
front of Harry
t ¬t u
reliably indicate ¬t
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
²⁴ This is at least true of process reliabilists in the tradition of Goldman (1979, 1986). The work of
Tyler Burge offers a very different sort of reliabilist approach in which reasons play an important role.
See e.g., the papers collected in Burge (2013).
²⁵ See the references in fn. 5.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Two Component View: A’s belief that p is ultima facie justified iff both:
Reliable Process Condition: A’s belief that p is the result of a reliable belief-
forming process,
Evidential Support Condition: A’s total evidence supports believing p.²⁶,²⁷
As it stands, the Two Component View is not reductive. After all, the
Evidential Support Condition packages together two epistemic notions:
²⁶ The Two Component View could be complicated in a number of ways. For example, we might
hold that the two conditions are not entirely independent. Rather, we might follow Comesaña (2010) in
holding that the reliable process needs to take evidence as input. (According to Comesaña, this helps
with both the generality problem and Bonjour’s case of Norman the clairvoyant.) For my purposes,
I will set such complications aside.
²⁷ Hybrid views that impose some version of an Evidential Support Condition include Tang (2016);
Comesaña (2018); and Miller (2019).
²⁸ Note that the challenge here is not just to give an account of evidence in non-epistemic terms. It’s
to give an account of what it is for an agent to possess evidence in non-epistemic terms. See Beddor
(2015b, 2016: Ch. 1) for discussion of some difficulties on this front.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Perhaps, some might suggest, this just shows that we need to supplement
the Two Component View with a reductive analysis of these notions. To do so,
proponents of the Two Component View could try taking a page from the
Reasons Reliabilists. According to Reliable Reasons, prima facie reasons are
the inputs to reliable and conditionally reliable processes. Why not say the
same about evidence? An agent’s total evidence, on this view, is the total set of
states of the agent that can serve as potential inputs to the reliable processes
available to the agent.
This would give us a reductive analysis of (i). What about (ii)? According
to one common approach, evidential support should be understood in
probabilistic terms: a body of evidence e supports believing p just in case
the probability of p given e is sufficiently high. Putting these two suggestions
together, we get:
• s₁ . . . sn are all the states of A that can potentially serve as inputs to A’s
reliable belief-forming processes,
• t is some threshold.
are justified by the evidence. But clearly this is to give up any reductive
ambitions.²⁹ Perhaps, then, we should follow Tang (2016) in taking Pr to
reflect objective probabilities. This would result in a reductive theory, but it
gives rise to a further concern: Is the theory predictive? And do the predictions
vindicate our pretheoretic judgments about defeat?
To flesh out this concern, go back to Seeing Red. On the view under
consideration, to determine whether Lori’s belief is defeated by Sal’s testi-
mony, we check the objective probability of conditional on Lori’s total
post-Sal-testimony evidence. But how do we check this? Perhaps via intuition,
but it is questionable whether we have clear-cut intuitions about such prob-
abilities. And things only get worse when consider defeater defeat. In Two
Testimony Seeing Red, how do we determine the objective probability of
conditional on Lori’s post-Sal+Anne-testimony evidence? The worry, then, is
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
that while this version of the Two Component View may be consistent with
our intuitions about defeat, it does not yet predict these intuitions.
While this is hardly the final word on the matter, these considerations
highlight the hurdles that arise when we try to develop the Two Component
View in a way that is both fully reductive and predictive.
The second advantage of Reasons Reliabilism over the Two Component View
stems from a structural difference between the two frameworks. One hallmark
of Pollock’s framework is that nothing is exempt from defeat. For any prop-
osition p, it’s easy to concoct a defeater for p—just imagine a reliable source
tells you either ¬p or that your reasons do not support p. By contrast, the
Evidential Support Condition singles out a class of propositions that get a free
pass from defeat. Let me explain.
It’s a familiar observation that whenever some part of an agent’s evidence
entails p, their total evidence also entails p, hence the probability of p condi-
tional on their total evidence is 1. So, probabilistic approaches to defeat predict
that one cannot have a defeater for something entailed by any part of one’s
current evidence:
At least two sorts of cases suggest that Limited Indefeasibility runs contrary to
intuition. First, consider cases where an agent has a defeater for a belief in a
necessary truth:
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Some might regard necessary truths as a special case, to be dealt with via
independent means.³⁰ Still, a second class of counterexamples remains: cases
where one has a defeater for some proposition that is itself part of one’s evidence.
According to the view of evidence under consideration, one’s evidence consists in
various states that serve as inputs to reliable processes. But why think that these
states enjoy some special exemption from defeat? Consider:
Kilian’s total evidence includes his happiness for his brother, since this state
serves as the input to a reliable process (introspection). And this experience
entails .³¹ Limited Indefeasibility thus predicts that his justification for
this belief is undefeated. But this seems wrong. Even though the psychiatrist is
mistaken, her testimony still provides a rebutting defeater for his belief.³²
Reasons Reliabilism fares better here, since it is not committed to Limited
Indefeasibility. Even if p is entailed by one of your reasons, you could still have
a reliable process that is disposed to deliver either a belief that ¬p, or a belief
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
that your basis for believing p does not reliably indicate its truth. In Logical
Luck there is a reliable process available to Tom (trusting the testimony of
³⁰ For example, some might take a page from Stalnaker’s (1999) strategy for handling the problem of
logical omniscience. According to this proposal, our intuitions about Logical Luck are not really
tracking Tom’s justification for believing L, but rather Tom’s justification for believing some contingent
proposition associated with L—for example, the proposition: SL is true, where SL is some sentence that
expresses L. However, even if this strategy can be made to work, it is a mark in favor of Reasons
Reliabilism that it has no need of such maneuvers.
³¹ It entails it both in the sense that its content (trivially) entails , and in the sense that the fact
that Killian has this experience entails . So regardless of whether we understand evidential
support in terms of probabilities conditional on the contents of an agent’s states or in terms of
probabilities conditional on the fact that the agent is in these states, the counterexample goes
through.
³² Faced with this counterexample, one option would be to try to complicate the Evidential Support
Condition. Perhaps in order for A’s belief that p to satisfy the Evidential Support Condition, we should
also require that every sufficiently similar body of evidence would also support believing p. (Cf. Miller
(2019), who suggests quantifying over ‘partial psychological duplicates’ to handle this sort of case.)
According to this diagnosis, Killian could have been in an internally indistinguishable mental state that
only included the appearance of happiness. Had he been in this state, his total evidence arguably would
not have supported . However, going this route raises a difficult question: can we give a
principled and reductive characterization of when a body of evidence is “sufficiently similar”?
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
is the most promising one to date, at least when it comes to handling defeat.³³
7.9. Conclusion
For most of their history, the reliabilist tradition and the Reasons First
tradition have been developed in isolation from each other. In this paper,
I’ve argued that an integration of the two approaches proves mutually bene-
ficial. The synthesis developed here avoids reliabilism’s difficulties with defeat,
³³ Another view that bears some resemblance to Reasons Reliabilism is the Reasons First virtue
epistemology developed in Sylvan and Sosa (2018). According to Sylvan and Sosa, facts about what an
agent is justified in believing are determined by facts about what she has sufficient epistemic reason to
believe, which are in turn determined by facts about her competent attractions to assent to various
propositions. While there are a number of similarities between the two approaches, there is also a
crucial difference. Sylvan and Sosa do not offer their Reasons First virtue epistemology as a reductive
approach. Rather, they take the notion of a ‘competent attraction to assent’ to be a normatively loaded
notion—a notion that cannot be reduced to talk of reliable processes.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
References
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
(Forthcoming).
Thomas Grundmann. (2009). Reliabilism and the problem of defeaters. Grazer
Philosophische Studien, 79(1): 65–76.
David Henderson, Terry Horgan, and Matjaž Potrč. (2007). Transglobal
evidentialism—reliabilism. Acta Analytica, 22: 281–300.
James Joyce. (1998). A nonpragmatic vindication of probabilism. Philosophy of
Science, 65(4): 575–603.
Jaegwon Kim. (1988). What is “naturalized epistemology”? Philosophical
Perspectives, 2: 381–405.
Hilary Kornblith. (2002). Knowledge and its Place in Nature. Oxford: Clarendon
Press.
Jennifer Lackey. (1999). Testimonial knowledge and transmission. The Philosophical
Quarterly, 49(197): 471–90.
Maria Lasonen-Aarnio. (2010a). Unreasonable knowledge. Philosophical
Perspectives, 24: 1–21.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Robert Stalnaker. (1999). Context and Content. New York: Oxford University
Press.
Kurt Sylvan and Ernest Sosa. (2018). The place of reasons in epistemology. In: Star
(ed.) The Oxford Handbook of Reasons and Normativity. Oxford: Oxford
University Press.
Weng Hong Tang. (2016). Reliability theories of justified credence. Mind, 125
(497): 63–94.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
8
Knowledge, Action, and Defeasibility
Carlotta Pavese
8.1. Introduction
One can intentionally do something only if one knows what one is doing while
they are doing it. For example, one can intentionally kill one’s neighbor by
opening their gas stove overnight only if one knows that the gas is likely to kill
the neighbor in their sleep. One can intentionally sabotage the victory of one’s
rival by putting sleeping drugs in their drink only if one knows that sleeping
drugs will harm the rival’s performance. And so on. In a slogan: Intentional
action is action guided by knowledge.¹
This essay reviews some motivations for a ‘knowledge-centered psychology’—
a psychology where knowledge enters center stage in an explanation of inten-
tional action (Section 8.2). Then it outlines a novel argument for the claim that
knowledge is required for intentional action (Section 8.3) and discusses some of
its consequences for the debate about know-how. Section 8.4 argues that a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹ I am grateful for comments to Adam Carter, Clayton Littlejohn, and Mona Simion.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
780198847205.003.0008
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
A long tradition in the philosophy of mind assigns beliefs a central role in folk
psychological explanations of intentional behavior (e.g., Stich 1978; Fodor
1987; Lewis 1976; Stalnaker 1984; Humberstone 1992). More or less explicitly,
this tradition confines psychological explanations to an explanation of
attempts. Consider the usual example of a psychological explanation, where
one’s belief that there is water in the fridge and one’s desire to drink it together
are supposed to explain one’s attempt to grab a bottle of water from the fridge.
Success happens when the belief is true—when there is indeed water in the
fridge. If one’s belief is true, then one will succeed at finding a bottle; if one’s
belief is false, one will not succeed at finding water. The dominant thought
behind a belief-centered psychology is the idea that, as far as the psychological
explanation of behavior goes, whether the world makes the belief true (e.g.,
whether there is water in the fridge) is irrelevant: what is to be explained is the
fact that one attempted to get water from the fridge, whether or not one has
succeeded. And one’s belief that there is water in the fridge, together with one’s
desire to drink it, suffices to explain one’s attempt, whether or not the belief is
true.²
Let us get a bit clearer about the underlying assumption of a belief-centered
psychology. Let a condition be something that obtains or fails to obtain at a
case, and let a case be a centered possible world or situation—an ordered triple
of a subject, a time, and a location. Some conditions are mental, such as that
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
I feel pain or believe that it is raining. Other conditions are non-mental, such
as that I have broken my leg. Beliefs are mental conditions and so are
attempts—e.g., my trying to get on the bus is a mental condition.³ Like beliefs,
attempts are non-factive mental conditions: an attempt to ϕ does not entail
successful ϕ-ing, for an attempt might be successful or might fail. As a non-
factive condition, an attempt does not encompass those external aspects of the
world that make for an agent’s success.
The assumption that psychological explanation should be confined to
explaining attempts relies on the idea that actions are decomposable into
mental conditions and non-mental conditions—into attempts, on the one
hand, and into those external conditions that makes for the agent’s
success, on the other. Call this the decomposability assumption and suppose
it is true. If so, we can appreciate one of the main motivations that underlie
a belief-centered psychology. A psychological explanation of intentional
² See in particular Stich (1978). ³ This terminology follows closely Williamson (2000: Ch. 3).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Gibbons 2001):
Mary the Bomber: Mary intends to kill her uncle by setting off a bomb by a
bomb in his house and then, after moving a safe distance away, pressing the
large red button on the remote-control device. She does not know much
about how these things work and thinks that pressing the button will cause
the bomb to detonate but has no idea about the details of this process. Her
belief is true and justified and here is what happens. A satellite, launched by
the National Security Agency and designed to prevent bombings of just this
kind, intercepts Mary’s transmission which causes the satellite to send a
warning to the intended victim. But, because of an unfortunate choice of
frequency, this causes the bomb to detonate.
Mary killed her uncle and caused the bomb to detonate and did intend both
things. But she did not do either of these things intentionally. Hence, the
intentionality and mentality of this action is not reducible to the intentionality
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
and mentality of the attempt. This residual ‘mentality’ of actions calls for an
explanation—and presumably an explanation that reduces the intentionality
of the action to some mental state of the agent. And for this sort of explan-
ation, belief alone cannot suffice.⁴ For intentional action is a factive mental
condition: if one intentionally ϕs, then (trivially) one ϕs. And, if the mental
condition to be explained is factive in the way intentional action is factive, then
its explanation calls for a factive condition. Non-factive (attempts) might be
explainable by non-factive (i.e., beliefs). But factive (intentional actions) ought
to be explained by factive (i.e., knowledge).
Mary the Bomber can be accounted for by a knowledge-centered psych-
ology (Gibbons 2001). Mary does not really know that she can provoke the
explosion by implementing her plan. That is why her success is too coinci-
dental to count as intentional. More generally, the prediction of a knowledge-
centered psychology is that if one’s belief is Gettiered, then one cannot act
intentionally on that belief. This prediction is borne out. To see this, consider
two more examples. The next is also from Gibbons (2001) but slightly revised:⁵
Cindy and the Lottery: Cindy mistakenly believes that someone rigged a
lottery in her favor and that she will be handed the winning ticket at the
ticket store. On this basis, she believes of a particular ticket that is being
handed to her, that if she buys it, she will win. She buys the ticket and wins.
So, her belief that she will win the lottery by buying that ticket is true. It is
even justified. Buying a winning ticket is a perfectly reliable way of winning a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
lottery. Still, intuitively, she did not intentionally win the lottery.
Here again, Cindy intends to win the lottery and attempts to do it by buying a
ticket that she believes truly and justifiably to be a winning ticket. This case
differs from the previous one, for here no deviant causal chain plays a role in
explaining her success (i.e., in making her ticket win). And yet, Cindy’s victory
is not intentional: fair lotteries cannot be intentionally won. That suggests,
once again, that the intentionality of actions is not exhausted by the inten-
tionality of attempts. But if so, we need a sort of psychology that differs from a
belief-centered psychology in that it is not only tailored to explaining attempts.
⁴ Cf. also Levy (2013) and Williamson (2017) for similar remarks about the non-decomposability of
actions. Gibbons (2001) also argues that psychological explanations should explain intentional suc-
cesses and not merely attempts.
⁵ Revised in order to overcome an objection raised by Cath (2015) to Gibbons’ original example. See
Pavese (2018) for a detailed discussion of Cath (2015) on Gibbons. Cath’s objection does not extend to
Cindy and the Lottery as presented in the text, nor does it extend to the next example from Pavese
(2019)—i.e., Daniel and the Barn.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Daniel and the Barn: along the road to Larissa, Daniel is instructed to stop at
any barn that he finds and to await further instructions. However, the road to
Larissa passes through fake-barn county. Daniel passes the first barn-looking
construction but does not stop for he thinks “There is not enough shadow for
me to wait comfortably.” At the second barn-looking construction, Daniel
stops and parks. As it turns out, only the second barn-looking construction
was a real barn.⁷
In Daniel and the Barn, Daniel ends up stopping at a barn and intended to
do so. However, intuitively, he did not intentionally stop at a barn. In fact,
he would have easily stopped at a fake barn, had the shadow been present
there.
In all Mary the Bomber, Cindy and the Lottery, and Daniel and the Barn,
the success of the agent does not count as intentional. This fact can be
explained on a knowledge-centered psychology. If one possesses knowledge,
then one’s belief cannot be lucky (Sosa 1999; Williamson 2000).⁸ In all of these
examples, the subject’s belief turns out to be true by luck and the same sort of
luck undermines the intentionality of their actions.⁹ A knowledge requirement
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
on intentional action can explain why the luck of their successes can under-
mine the intentionality of Mary’s, Daniel’s, and Cindy’s successes: by under-
mining their knowledge. In contrast, it is unclear that a belief-centered
psychology has the resources to explain why these actions are not intentional.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
But could one act intentionally on the basis of extremely high probabilistic
evidence that however falls short of knowledge?¹⁰ Consider:
Mary, the Bomber, and the Lottery: Mary intentionally plays a trillion ticket
lottery such that the bomb fails to go off and fails to kill her uncles only if she
wins the lottery (and let’s suppose the satellite stuff is left out of the story).
Like in standard lottery cases, arguably Mary does not know that she will lose
the lottery (Williamson 2000). But, the objection goes, it seems as though
Mary could intentionally kill her uncle by simply intentionally playing that
lottery. This is the case even though (like in Gettier cases) she doesn’t know
that playing the lottery will detonate the bomb.
In response, granted, Mary does not and cannot know that playing the
lottery will detonate the bomb. But there is another thing that she does know—
i.e., that it is sufficiently likely that playing the lottery will detonate the bomb,
for she knows that it is sufficiently likely that she will lose the lottery. In fact, as
I will explain later in the essay, it is independently plausible that probabilistic
knowledge of this sort is central to an explanation of intentional action (Pavese
2020). To see this, consider Davidson’s (1971: 50) example:
believe that I will succeed. In this case, I might succeed in this endeavor while
failing to know that my signature would be legible on the last page because
I do not have sufficient confidence for full belief.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The claim that intentional action requires knowledge has been recently chal-
lenged. It is instructive to consider the challenge for it points towards a more
principled argument for thinking that knowledge is required for intentional
action.
Cath (2015: 11) argues that one can have intentional action without know-
ledge, upon considering cases like the following:
¹¹ Setiya (2012) puts forward an example of the subject believing their hand to be paralyzed who
tries nonetheless to clench their fingers. This case too can be handled by appealing to probabilistic
knowledge, for although one does not know that one will succeed by intending to succeed, one knows
that one might succeed by intending. Cf. Pavese (2020). For a different sort of diagnosis of Setiya’s case,
cf. Pavese (2018).
¹² Gibbons (2001) emphasizes that the role of knowledge in explaining intentional action provides a
novel argument for the claim that knowledge is a mental state—a claim famously defended by
Williamson (2000) and Nagel (2013). Although both Williamson (2000) and Nagel (2013) provide
arguments for the role of knowledge in explaining behavior, neither focuses specifically on the role of
knowledge in action theory. For a comparison, see Pavese (2019).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
justified true belief about how to fly, but that justified true belief does not
amount to knowledge.¹³
Cath uses this example to argue that one can have know-how even though
one’s relevant belief is Gettiered. Cath claims that intuitively Bob can inten-
tionally fly. And for some action ϕ, if one can intentionally ϕ, then one knows
how to ϕ (Williamson and Stanley 2001; Hawley 2003; Setiya 2012; Cath 2015;
Pavese 2018, 2020). Therefore, Bob must know how to fly, even though he
does not know the relevant instructions.
Cath thinks that this diagnosis is supported by comparing the case of Bob
with the case of Joe, who is a near perfect counterpart of Bob except that his
belief is not Gettiered: his simulator operated correctly and did so non-
accidentally; his instructor intentionally gave him the correct advice, and so
on. If Joe were to try to fly a plane in normal circumstances, he would typically
succeed in so doing and his successful actions would be unquestionably
intentional actions. And it is an implicit stipulation of the flight simulator
case that if Bob were to try to fly a plane in normal circumstances then he
would be just as likely to succeed as Joe. Cath contends that not only would
Bob succeed as often as Joe but, like Joe, his actions would appear to have all
the standard kind of properties that are thought to distinguish merely suc-
cessful actions from intentional actions. Hence, according to Cath, it is
plausible that Bob’s successful actions of flying, like Joe’s, would be perfectly
under his control or guidance as he performs them. But is Cath (2015) right in
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
assuming that, in Bob the Pilot, Bob’s performance is under his control?
Both intuitive and theoretical considerations suggest that the answer to this
question ought to be “No.” Compare Joe and Bob. Note that, strictly speaking,
Joe knows what he is doing while he flies the plane: he knows that he is
following instructions that are conducive to successful flying. In contrast, by
assumption, Bob does not know that. Hence, he does not know what he is
doing while he flies the plane. But consider how unintuitive it is to ascribe
intentional action to one who lacks knowledge of what they are doing:
Awful: Bob intentionally landed the plane, but he did not know that he was
landing the plane by following the given instructions.
¹³ This example is initially due to Stanley and Williamson (2001) who actually give it as an example
showing that know-how is incompatible with epistemic luck. Poston (2009) also discusses it and so
does Stanley (2001: Ch. 8).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
by step:
¹⁴ Greco (2016) considers a different sort of argument for the claim that knowledge is necessary for
explaining action—one that relies on the nature of explanation as counterfactually robust. See also
Pavese (2018).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Note that Sufficient does not require that for one’s action at @ to be not-too-
lucky, one succeed at performing it when one tries it in every sufficiently close
world.¹⁸ This would be too demanding a requirement: as Austin (1961: 218)
put it, “a human ability or power or capacity is inherently liable not to produce
success, on occasion, and that for no reason.” If Austin is right, many actions
might be under the control of their agent and even manifest their skills and
¹⁵ It is worth noting that some recent work takes moral luck to be a species of a larger genus of luck,
of which there are other species as well, such as epistemic luck. Such an approach does not build in the
idea that luck is opposed to control. See Pritchard (2006) and Coffman (2015) for similar approaches.
Thanks to Mona Simion for the discussion.
¹⁶ It is, of course, a difficult and partly a context-sensitive matter, what counts as “under one’s
control.” But, for our purposes, we do not need to settle this question, the idea being that if a certain
success counts as too lucky, it is not sufficiently under the control of the agent to count as intentional.
¹⁷ Beddor and Pavese (2018: 6) for a defense of this sort of condition on lucky actions.
¹⁸ Although strictly speaking only attempts might fail to succeed, it will simplify my exposition to
talk as if one’s action might fail to succeed in some worlds where it is tried.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
still fail in some close worlds, for no particular reason.¹⁹ Rather, according to
Sufficient, an action is too lucky if it fails in more than just some close worlds
when tried—if it fails in many close worlds (for that is what “many” means:
more than just some).
My argument for Premise (c) is that actions based on Gettiered beliefs are
doomed to satisfy Sufficient, and hence are doomed to be too lucky in the
relevant sense. Here is why. If an action is based on a true belief about how to
bring it about, then it might nonetheless fail when tried in some close worlds.
In fact, even if the action is based on knowledge, it might still fail in some close
worlds. That might be because in those close worlds, something interferes with
the agent’s basing their action on the relevant belief. For example, an archer’s
shot might fail in a counterfactual circumstance where the archer gets dis-
tracted by a passer-by and hence fails to appropriately base their shot on their
knowledge about how to shoot. Or it might fail also because, although their
shot is appropriately based on their knowledge, the world does not cooperate
in the way required for success. For example, the archer’s shot might fail in a
circumstance where a fluke happens, and an unexpected gust of wind inter-
feres. Or it might fail because their hand was for no reason slightly less firm
than they expect it to be a moment ago. In those counterfactual circumstances,
the belief on which the agent’s performance is based might still be true—it is
still true that one must do this and this to shoot the target under certain
conditions (e.g., in non-windy conditions and when one’s hands are firm). It is
just that there the world does not cooperate in the way required for success.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
This concludes my argument for Premise (c). With Premise (a) to Premise
(c) in play, the conclusion deductively follows actions based on Gettiered
beliefs cannot be intentional. Intentional action ought to be guided by
knowledge.
specifies the means to ϕ (cf. also e.g., Audi 1986; Bratman 1987; Ginet 1990;
Harman 1976; Velleman 1989, 2007; Mele and Moser 1994):
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
to ϕ. What does it mean for some means to be means for one to ϕ? Not that,
for some way ψ of ϕ-ing, one will ϕ by ψ-ing: that is far too strong, for one
might intentionally ϕ even though one has some doubts about whether one
will succeed (Goldman 1970; Harman 1976). Recall, for example, Davidson’s
(1971: 50; 1980: 91–4) carbon copy example, discussed earlier.
Should the relevant knowledge be that, for some means ψ of ϕ-ing, one
would in most cases succeed at ϕ-ing by ψ-ing? This is also too strong: one
might intentionally ϕ even though one might fail in most circumstances, as the
baseball player who fails at batting nineteen times out of twenty may none-
theless intentionally bat the one time they succeed. That suggests that the
relevant knowledge is that, for some means ψ of ϕ-ing, one could ϕ by ψ-ing.
But what does it mean that one could ϕ by ψ-ing, if not that one is sufficiently
likely to ϕ by ψ-ing, where what counts as sufficiently likely may vary from task
to task? This gives us:
²⁰ I am stating Intellectualism as the view that know-how requires knowledge of the means for
action, rather than as a fully reductive claim. Stating Intellectualism as a fully reductive claim would
require talking about practical modes of presentation, which I cannot discuss here. See Pavese (2015,
2017, 2019a).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
(Intellectualism about Know-How): s knows how to ϕ only if, for some means
ψ for s to ϕ, s knows that it is sufficiently likely for them to succeed at ϕ-ing
by ψ-ing.
Hence, if one unpacks the clause means for one to ϕ, a plausible upshot is that
both (Intellectualism about Know-How) and (Intentionality/Knowledge)
should be stated through a probabilistic language: they both require knowledge
that one is sufficiently likely to succeed at ϕ-ing through certain means.
Following Moss (2018), one might know that it is sufficiently likely for oneself
to succeed at ϕ-ing by ψ-ing in virtue of possessing a sufficiently high credence
that one will succeed at ϕ-ing by ψ-ing. Because possessing this credence does
not require that one grasp the concepts of likelihood or probability, Pavese
(2020) argues that this rendition of (Intellectualism about Know-How) over-
comes the challenge of over-intellectualization that Setiya (2012) and other
authors have raised against it.
Hence, the knowledge that (Intentionality/Knowledge) requires for inten-
tional action is the same that intellectualists require for know-how. I suggested
that we can use a knowledge-centered psychology to argue for intellectualism
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
about know-how. How would such an argument go? Start from (Know-How/
Intentionality), endorsed by intellectualists and anti-intellectualists alike (Ryle
1949; Stanley and Williamson 2001; Stanley 2011; Hawley 2003; Hornsby
2004, 2011; Setiya 2012; Pavese 2018):
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The intellectualist picture provides the best explanation for why (Know-How,
Intentionality, Knowledge) should hold. According to this explanation, (Know-
How, Intentionality, Knowledge) is true not just out of a coincidental aligning
of propositional knowledge and know-how in intentional action. Rather, its
truth is grounded in the very nature of know-how.
By mostly appealing to a linguistic argument when motivating their views
(Stanley and Williamson 2001; Stanley 2011), intellectualists have sold the
view short.²¹ The chief motivation for the view does not come from linguistics:
it comes from the sort of action theory that a knowledge-centered psychology
recommends.
²¹ For worries concerning the linguistic argument on behalf of intellectualism, see Brown (2013).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Ana and the Grenade Factory: Ana and Marıa work in a grenade factory
during the Spanish Civil War. They are thoroughly instructed when hired,
with examples and practical explanations. By controlled trial and error, they
learn their job, and both continue working at the factory for years, believing
they are making working grenades. However, one day each comes to realize
that the other is making grenades in an importantly different way, and they
identify the origin of the problem: as it turns out, the instructions were
ambiguous and allowed for two different interpretations. The instructors
were not aware of this, and there is nobody above them now who may say
who is right. Given that the grenades may only be used in battle, which is very
far away, neither Ana nor Marıa knows whose grenades actually work, and so
there is no way to find out who is making them the right way. As a matter of
fact, Ana got the instructions right (she produces grenades in way w, which is
the correct way); she is very successful in producing grenades that later work
perfectly. It is Marıa who got something wrong (she makes them in w’, the
possible interpretation of the instructions that the instructors did not fore-
see), and her grenades are always duds. Unaware of this, both have reasonable
doubts they did not have before, but they have to keep on working.
‘MISLEADING’), Ana might know, for some means to make grenades, that
it is a means to make grenades; but her knowledge is defeated as soon as the
misleading evidence is acquired. On the other hand, they think that Ana still
knows how to make grenades after receiving MISLEADING. If they were right,
this would be a case where know-how stands undefeated whereas the corres-
ponding knowledge is instead defeated. They conclude (2018: 669):
²² The argument assumes that knowledge can be easily defeated by high-order evidence. This
assumption is controversial and is not granted by prominent epistemologists (Aarnio 2010, 2014).
Let us play along, however, and see that the putative challenge rests on other false assumptions about
the nature of know-how.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Carter and Navarro’s argument hinges on two claims. The first claim is that
the relevant propositional knowledge is defeated in this case; the second is that
Ana still knows how to make grenades, upon receiving MISLEADING. Let me
grant the former claim and focus on the latter.
Why think Ana still knows how to make grenades, upon receiving
MISLEADING? The intuition that she does is not nearly as robust as they
seem to think. And the only argument Carter and Navarro (2018: 666) provide
in support of their intuition is the following:
The claim that Ana preserves her know-how along all the variations of the
case is supported by the fact that she is still able to make grenades profi-
ciently, and the doubts she acquires do not seem to imperil this ability in any
relevant sense.
Carter and Navarro (2018) are going far too quickly here. Granted, Ana still
preserves some ability that is relevant to grenades-making. What is much less
obvious, and as I argue ultimately incorrect, is to assume that Ana preserves
the sort of ability that goes together with know-how.
As the discussion in the previous section already suggests, know-how does
not just go together with any ability. It goes together with the ability to
intentionally perform a certain task. For example, knowing how to make
risotto does not merely go together with the ability to make risotto but with
the ability to intentionally make risotto. For if one had the ability to make
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
risotto but lacked the ability to intentionally make it, one would not count as
knowing how to make risotto. This point is well-known in the literature at
least since Hawley (2003) and is accepted by both intellectualists and anti-
intellectualists (Ryle 1949; Setiya 2012; Pavese 2017). For example, the clumsy
person has the ability to fall and tumble, as they reliably do so. But only the
clown has the ability to intentionally do that. As another illustrative example,
Susie may have the ability to irritate Ben, for she would succeed at irritating
him if she tried. But suppose she falsely believes that it is the smell of the
smoke, rather than the noise she makes whenever she smokes, that irritates
Ben. In this case, she does not intentionally irritate Ben: her success is too
coincidental to count as intentional. Because of this, it seems that she does not
know how to irritate Ben. On the bases of similar examples, intellectualists and
anti-intellectualists alike endorse the claim that know-how goes together with
the ability to intentionally perform the task.
Suppose it is true that know-how goes together with the ability to intention-
ally perform a task. If so, it is independently plausible that there is an important
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
sense in which, upon receiving MISLEADING, Ana does not know any longer
how to make grenades. For upon receiving MISLEADING, when asked to make
grenades, she will be at a loss. Not only that: she will also refuse to teach others
how to make grenades. And she will even stop performing at the workplace,
until she is told that she has been making grenades correctly all along. Suppose
she were forced to reproduce whatever process she initiated before
MISLEADING. She would unknowingly succeed at making grenades. But the
success would be too out of her control to count as intentional. She is still able in
some sense to make grenades but in an important sense she now lacks the ability
to intentionally make grenades. If so, then she also lacks know-how.
If the reader is not yet ready to grant this conclusion, it is because, actually,
things are more complex, and some additional distinctions are called for.
Ascriptions of abilities of the form “s can intentionally perform a task” are
opaque, for as it is well known in action theory, “intentionally” is an inten-
tional operator (cf. Davidson 1971; Goldman 1970). For example, Lois might
intentionally kiss Superman but not intentionally kiss Clark Kent. Because
of the opacity of intentionality reports, it is paramount to distinguish between
(de re ability) and (de dicto ability):
(de re/ability): there is some task t that is in fact the task of making grenades
such that Ann has the ability to intentionally perform t.
(de dicto/ability) Ana has the ability to intentionally make grenades.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
While (de re/ability) ascribes Ana a de re ability, (de dicto/ability) ascribes Ana
a de dicto ability.
Now, with this distinction in play, consider again Ana’s situation upon
receiving MISLEADING. (De dicto/ability) is now false: Ana does not have a
de dicto ability any longer. For one to have the relevant de dicto ability, one
needs to be able to make grenades on demand (to be in a situation such that, if
asked to make grenades, Ana would do so). Ana does not have that ability:
were she asked, after receiving MISLEADING, to make grenades, she would
now be at a loss. If know-how goes with the ability to intentionally perform a
task, then, to this distinction between a de dicto ability and a de re ability, there
corresponds the distinction between de re know-how and de dicto know-how:
(de re/KH): there is a task t that happens, unknown to Ana, to be the task of
making grenades, such that Ana still knows how to perform t.
(de dicto/KH): Ana knows how to perform grenades.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
(de re/K): there is a task t that is in fact, but unknown to Ana, the task of
making grenades such that Ana knows for some way w that w is a way to
execute t.
(de dicto/K): Ana knows for some way w that w is a way to make grenades.
Upon receiving MISLEADING, (de dicto/K) becomes false: Ana loses the
relevant de dicto knowledge. However, Ana arguably still also preserves the
relevant de re knowledge (ascribed by (de re/K)). MISLEADING only defeats
(if anything) her de dicto knowledge—i.e., the knowledge that the procedure
she was implementing was for making grenades. Her de re knowledge, instead,
is not at all defeated by MISLEADING: Ana continues to have it, as she might
continue to know what procedure she was following before receiving
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
MISLEADING, when she was intending to make grenades. And so, by intel-
lectualism’s lights, she might continue to know how to make whatever she was
making when she thought (correctly, it turns out!) that she was making
grenades: because Ana retains her de re knowledge, by intellectualism’s lights,
Ana can retain her de re know-how as well as her de re ability.
Let me end by considering two possible responses.
I have argued that Ana might lose her de dicto know-how upon receiving
MISLEADING, while possibly retaining her de re know-how. Could not
Carter and Navarro (2018) reply that Ana does retain he de dicto know-how
and her de dicto ability all along but upon receiving MISLEADING, she simply
cannot act on those, because of her new doubts? Compare: many Olympic
gymnasts know how to do the fancy tricks they do even though many—in the
heat of the competition—have doubts about whether they can do them
successfully.
In response, the analogy with the Olympic gymnasts is misleading. In the
case of Olympic gymnasts, it is plausible that despite their doubts, they still
know how to perform their fancy tricks (de dicto). After all, they can still
intentionally do their fancy tricks (de dicto) outside of the heat of the compe-
tition. Their ability to intentionally act on that knowledge is not lost but only
‘masked.’ This can be explained on the current picture: These athletes retain
their (de dicto) knowledge all along and simply cannot access it in some
circumstances. By contrast, Ana has lost her de dicto knowledge and unless
she regains it, there is no circumstance where she can still intentionally make
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
grenades (de dicto). In Ana’s case, then it is utterly implausible that her ability
to intentionally act is simply masked.
A second possible response goes as follows. Maybe, Ana does preserve
her de dicto know-how and her de dicto ability upon receiving
MISLEADING. What she lacks is knowledge that she does know how and
that is what explains the lack of de dicto ability. If one embraces this position,
one commits oneself to replacing (Know-how/Intentionality)—a claim that, as
we have seen, both intellectualists and anti-intellectualists agree upon—with a
considerably stronger claim that intentionally ϕ-ing requires knowledge that
one knows how to ϕ. This stronger claim is rather implausible. For one thing,
some non-human animals can certainly intentionally act, though lacking the
concept of know-how. For another, suppose one learns how to make grenades
but has not had the occasion to form a belief one way or another about
whether one has indeed learned. If one is asked to produce one and one
tries, surely one can still intentionally do it (even de dicto!), even though one
does not know that they knew how to make them.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
8.6. Conclusions
References
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
9
Undercutting Defeat
When it Happens and Some Implications
for Epistemology
Matthew McGrath
¹ See for instance, Pollock (1970, 1975, 1987); Pollock and Cruz (1999).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
perjury—and the proposition I believe—that the official did this.² The category
of undercutters is often thought to represent an important advance in epis-
temology, for it promises to help us explain the defeating power of consider-
ations that don’t bear on the truth of the target proposition, such as evidence
that a testifier is unreliable or evidence that one is hallucinating.
Although there is disagreement about the details, the distinction between
rebutting and undercutting defeat—defeat by striking against the target prop-
osition vs. defeat by striking against the connection between a ground and the
target proposition—is now part of the received wisdom in analytic epistemol-
ogy. Not only that, but many epistemologists think Pollock’s account of how
these defeaters work is roughly on the right track. Recently, however, cracks
have appeared in the consensus. While not questioning the existence of
undercutting defeat, Scott Sturgeon argues that Pollock misconceives it:
undercutting operates differently from rebutting in that it occurs only in
conjunction with certain higher-order contributions, i.e., with beliefs about
(or justifications to believe propositions about) the basis on which one does or
would believe.
I will argue, contrary to Sturgeon, that in the case of inferential justification,
undercutting defeat takes basically the shape Pollock suggests it does, not
needing contributions from higher-order beliefs or justifications. However,
I agree with Sturgeon that for noninferential justification, the Pollockian
account is in trouble. I try to explain why there should be this difference.
My interest, however, is not merely in determining what makes for a good
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
² In principle, the same defeater could be at once a rebutter and an undercutter, if it strikes against
both the target proposition and the connection between the reason and that proposition.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Next, rebutters:
For undercutters, they give two definitions, one for doxastic reasons and
another for nondoxastic reasons:
We can simplify these definitions, even by Pollock’s and Cruz’s own lights.
They go on to tell us that undercutting defeaters are reasons to believe the
appropriate subjunctive conditional is false (Pollock and Cruz 1999: 196).
Their suggestion, I take it, is that it’s because undercutters are reasons to
doubt or disbelieve such conditionals that they manage to be defeaters. This in
turn suggests that the clause that “M* is a defeater for . . . as a reason for S to
believe Q” is redundant in the previous definitions. The same goes for rebut-
ters. Thus, I attribute to them the following simplified definitions of under-
cutting defeat:
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Rebutters:
If M is a defeasible reason for S to believe Q, M* is a rebutter for this reason if
and only if M* is a reason for S to believe ~Q.
Undercutters:
(The doxastic case): if believing P is a defeasible reason for S to believe Q, M*
is an undercutter for this reason if and only if M* is a reason for S to doubt or
deny that P would not be true unless Q were true.
(The nondoxastic case): if M is a nondoxastic state that is a defeasible reason
for S to believe Q, M* is an undercutter for this reason if and only if M* is a
reason for S to doubt or deny that he or she would not be in state M unless
Q were true.
³ I omit exposition of Pollock’s discussion of defeated defeaters, as well as his full account of
justification in terms of status-assignments assigning statuses of “defeated” and “undefeated” to
elements of arguments (see Pollock and Cruz 1999: 197–200).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Sturgeon thinks that the milk taster remains justified in her conclusion about
the milk, in virtue of the same non-doxastic reason that justified her
previously. No defeat occurs. Nevertheless, her belief that she is undergoing
olfactory hallucination meets the (simplified) Pollockian definition of an
undercutter. Putting things in terms Pollock’s framework: although the
olfactory experience o is a reason to believe the milk is OK, and the belief
that one is suffering an olfactory hallucination (H) is a reason to believe that
having o doesn’t guarantee the milk is OK, the combination <o, belief in H>
is still a reason to believe the milk is OK—i.e., one can become justified in
believing that the milk is OK by believing the milk is OK on the basis of
being in this combination of states, and the Milk Taster case is a case in
point.⁷ Sturgeon concludes that Pollock’s account of undercutters is too
weak—it implies that there is an undercutter for a reason in this case
when there is no defeat at all. However, Sturgeon thinks that there is defeat
if we modify the case so that the milk taster has a correct belief about the
source of her belief about the milk. This suggests to him that we need to
revise Pollock’s account of undercutting defeaters by giving an essential role
to beliefs about one’s mental states, in particular to beliefs about the sources
of one’s belief.
More generally, let U be the claim that source S is untrustworthy about
whether P and BOS-P be the claim that your belief in P is (or would be or likely
would be) based on source S. Suppose you believe U and believe P. Sturgeon
proposes that in such a situation:
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Your belief in U undercuts your belief that P iff you believe BOS-P.⁸
⁷ As noted earlier, more information is needed about what it is to become justified in believing
something on the basis of being in a certain combination of states. It would be too much to demand
that the combination be used as one’s grounds for the belief. When defeat occurs, it’s not because it
would be wrong to use the combination of states as one’s ground in any intuitive sense. I wouldn’t think
of using the combination of John told me that p and John was insincere in his remark to me on whether p
as my ground to believe p; rather, I might use John told me that p as my ground to believe p despite
being also in the state of believing that John was insincere in his remark to me on whether p. I will read
‘on the basis of being in state M’ in such a way as to allow for such possibilities. To believe p on the basis
of being in state M, then, as I will understand it within Pollock’s framework, is to be in state M and to
believe p using some component state of M (proper or improper) as one’s ground.
⁸ As Casullo (2018) points out, it’s implausible that the mere having of higher-order beliefs does
much epistemological work. What counts is justification to have such beliefs. We could reformulate
Sturgeon’s proposal so that it concerns justification to believe rather than belief. But as I mentioned, my
main concerns don’t hinge on such tweaks, so I will not bother with reformulation. Mellis (2014: 438)
usefully refines Sturgeon’s proposal, bringing in the concept of a justificatory process in addition to the
concept of a source of justification.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Finally, to handle undercutting defeat of reasons for beliefs one already has, we
would need to adjust the definition of ‘reason’ slightly as well:
Presumably the issue for the milk taster isn’t becoming justified but remaining
justified, and Sturgeon’s thought is that if you have a belief about the identity
of the source of your belief that p, together with a doubt or disbelief that
having that source guarantees P, you can’t remain justified in believing P.
Undercutting for Sturgeon is thus a matter of believing both:
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
⁹ Thus, if Sturgeon is right about undercutting, Pollock’s account is not only too weak; it is also too
strong. An alternative view is that undercutting happens iff Pollock’s conditions are satisfied and one
believes one source is M. On this view, Pollock’s account would not be too strong. I don’t have the space
to examine this view in any detail. But note that even if this view is correct, undercutting doesn’t
happen solely in virtue of having reason to believe something that strikes against the connection
between one’s reason and the target proposition—higher-order contributions are also part of the
explanatory story.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
¹⁰ We saw an example of this previously. You falsely think you are relying on taste, which you think
(and even know) to be unreliable, but in fact you’re relying on olfaction: you rely on an olfactory reason
O. Then your belief that you are using the unreliable source of taste is a higher-order defeater of O as a
reason to believe P, but it is not a reason to believe not-P (and so not a rebutter) and it does not it strike
against the connection between O and P (and so is not an undercutter).
¹¹ Note that higher-order defeat does not boil down to the same thing as “defeat involving defeaters
with higher-order content.” Perhaps some defeaters with higher-order content, such as I have strong
evidence against p are rebutters, and perhaps others can be undercutters. “Higher-order defeat,” as I am
stipulatively using this terminology, refers to defeat by virtue of being a reason to believe that one’s
belief is unreliably based.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Since Sturgeon sees no role for higher-order beliefs or justifications in the case
of rebutting defeat, Casullo thinks Sturgeon must give the externalist answer.
Casullo’s claim is then that there is no reason to think that if the externalist
position is correct here it wouldn’t also be correct for cases of undercutting
defeat. He concludes that the two sorts of defeat work in the same way—they
both work with—or without—the help of higher-order beliefs/justifications.
Casullo, as I read him, doesn’t strictly take a stand on whether rebutting
defeat needs higher-order help. But it is implausible to think that it does. If
I have evidence for P, and I then gain equally good evidence for not-P, I don’t
need to have higher-order beliefs or justifications about these matters in order
for the different justificatory forces to weigh up in one direction or another.
This is intuitive as it stands. But we can add as well that if higher-order
contributions were necessary, worries would arise about how creatures that
don’t grasp these higher-order propositions could have rebutting defeaters
(e.g., young children or higher animals).¹²
There is also reason to worry that if rebutting defeat requires higher-order
contribution, the same would hold for a single source’s justificatory power. We
could ask: in order for a source to justify a subject’s belief to a certain degree,
must the person believe that the belief is justified (by that source) to that
degree? It is difficult to see why the facts about how justificatory forces weigh
up should require beliefs about those forces while the existence of a single
justificatory force doesn’t. But it is implausible to think that in order for a
source to provide justification for a belief the person would have to believe (or
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹² Using a distinction from Audi (1993), we can distinguish two ways that a justificatory fact—such
as my having justification from a piece of evidence or my justification being defeated—could depend on
higher-order beliefs (or justifications). One way is negative dependence: the justifications obtain only if
you do not have certain negative higher-order beliefs (justifications), such as a belief that your basis is a
poor one. But this is not the kind of dependence at issue. At issue is positive dependence: in order for
justificatory facts to obtain one must have a higher-order beliefs (justifications) about those facts.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
We have seen that what Sturgeon says about the Milk Taster case is plausible:
Pollock’s conditions for undercutting are met but there is no defeat. I agree
with Sturgeon that the failure is not a minor one: there are no defeaters of
nondoxastic reasons that do their work by attacking the connection between
one’s actual reason and the proposition believed. The closest things are defea-
ters that work by giving one a reason to think one’s belief is unreliably based.
But Sturgeon draws a much broader conclusion: there are no defeaters—even
for doxastic reasons—that work in this attacking-the-connection way. I now
argue that this broader conclusion is false: we find just such defeat in the case of
inferential justification, i.e., of defeat of doxastic reasons.
Consider an example of undercutting defeat described by Sturgeon (similar
to one discussed by Pollock and Cruz):
¹³ What if higher-order beliefs are present? Suppose, despite believing respondents decided their
answers by coin flip (U), and despite believing that roughly 87 percent of Texans are going to vote
Republican (G) on the basis of the belief this is what 87 percent of the Texan respondents told you (T),
you think you believe G on the basis of some other belief, H, which you think makes G likely true. In
such a case could you become or remain justified in believing G based on <belief in T, belief in U>?
I think not. These false beliefs about your basis even if justified do not seem to restore your actual basis
as a reason to believe G by somehow defeating your defeater (U) for that reason. Intuitively, you’re on
the hook, epistemically, for the basis you have; and you can’t get off the hook by having beliefs (even
justified beliefs) about having some other basis.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
• If T, then G
◦ If 87 percent of Texas respondents told me they will vote Republican,
then roughly that percentage of Texans are going to do so.
• It wouldn’t be that T unless G
◦ It wouldn’t be that 87 percent of Texas respondents told me they will
vote Republican unless roughly that percentage of Texans are going to.
• T indicates in the circumstances that G
◦ 87 percent of Texas respondents telling me they will vote Republican
indicates in the circumstances that roughly that percentage of Texas
are going to do so.
In fact, the pollster gains reason to believe these are false. Notice that each of
these conditionals seems to be a plausible way of spelling out what T guaran-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
teeing G amounts to, and similarly for T supports G.¹⁴ Assume, then, that
the taking condition is a genuine condition on inference. Then there is an
explanation available for why you cannot become justified in believing G based
on <believing T, believing U>, as follows:
¹⁴ Isn’t guaranteeing much stronger than mere support? As the terms are used in ordinary parlance,
yes. I use ‘support’ here only because it is the standard term used for the taking condition on inference.
However, its ordinary meaning is not sufficiently strong for the purposes of giving an account of
inference. I might take the presence of some clouds in the sky to support rain, but I certainly won’t infer
it will rain based just on such slim support. By contrast, if I take the particular clouds to establish,
ensure, imply, indicate—guarantee—rain then I will make the inference. If I don’t so take them, I may
well not make the inference. There are complexities here about whether ‘guarantee’ is too strong, but
I don’t think they matter to our discussion, since the undercutters at issue are reasons not merely to
deny that there is a guarantee but also to deny that there is strong support. To keep things simple, then,
I use ‘support’ and ‘guarantee’ interchangeably.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
have led to the inference, as is the case for associations. The taking condition is
not perhaps the only way to explain these facts, but it is one way.
But the taking condition brings with it a host of problems as well. What is
taking? Is it believing? If so, then presumably for inference to transmit
justification from premises to the conclusion, the taking state must itself be
justified, which raises at least the threat of a regress.¹⁶ If taking isn’t believing,
¹⁵ One could alternatively recast the previous explanation in terms of remaining justified in believing
G on the basis of the combination <believing T, believing U>. In place of the language of “being
appropriate to infer G from T” one could substitute “being appropriate to continue to base inferentially
one’s belief that G on one’s belief that T.”
¹⁶ However, one might claim that both this linking belief and its justification are derivative from the
disposition to make the inference and the transmissibility of justification available through the inference.
For an example of someone who argues that the linking belief is derivative from the reasoning, consider
Broome (Forthcoming): “my account of reasoning entails that, if you reason from premises to conclu-
sion, you implicitly have the linking belief. You believe the conclusion as a result of a reasoning process
starting from your believing the premises. This shows you implicitly believe that the conclusion follows
from the premises. This implicit belief is constituted by your disposition to derive the conclusion from
the premises by a process of reasoning.” If one accepts this view of linking beliefs, it’s natural to claim
that their justification derives from the justificational properties of the available inference.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
conclusion was true—or if you prefer that T supports G—is plausible on its
own without assuming the taking condition. In fact, its plausibility provides at
least some reason to accept the taking condition, insofar as the taking condi-
tion would help explain why inference would be epistemically inappropriate in
such situations.
A key background assumption in these two explanations is that inference from
premise(s) to a conclusion is something that itself can come up for assessment as
epistemically appropriate or not, ex ante and ex post. This seems plausible.¹⁷ You
can be epistemically inappropriate to infer even when you know the premises,
and epistemically appropriate to infer—to draw the inference—even if your
beliefs in the premises are epistemically inappropriate or unjustified. Once
you’ve inferred, if you weren’t epistemically appropriate to infer, we can criticize
your actual inference as epistemically inappropriate (ex post).
¹⁷ See also Balcerak Jackson and Balcerak Jackson (2013: 115). They focus on normative evaluation
of inferences people draw, and so on ex post rather than ex ante epistemic appropriateness.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The conclusion of the previous section raises a question. Given that there is
genuine undercutting defeat in the Polling case and in other cases of inferential
justification, why wouldn’t the same hold for cases of noninferential justifica-
tion such as the Milk Taster case? Why, in particular, wouldn’t the explan-
ations discussed in the last section apply just as well to non-inferential cases?
This section aims to answer these questions.
Is there a taking condition on experience-to-belief transitions (of the sort
that can result in non-inferential justified belief)? This might seem implausible
on its face (how could a taking be required for noninferential transitions;
wouldn’t that make them inferential?¹⁸), but it’s worth looking into the details.
If a taking condition holds for these transitions, then either the taking state
must be explanatory of the transition or it needn’t be. Let’s look at both
possibilities in turn.
A taking condition for experience-to-belief transitions cannot be explana-
tory of these transitions. Consider Boghossian (2014) on inference. He thinks
that it’s because you take your premises to support your conclusion that you
“draw” your conclusion. But this clearly depends on your already believing the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
premises. If you take P to support Q but don’t believe P, you will not be in a
position to infer Q. In the same way, your taking some fact about your
experience to support Q, or even your taking your experience itself to support
Q, can’t by itself explain why you come to believe Q. Takings are relevantly like
commitments to conditionals, and for a commitment to a conditional to
explain belief in the consequent, one needs a commitment to the antecedent,
so that the content of this state together and that of the taking state “lock
together” to explain the transition to the belief in Q. Thus, one must be
committed to that which one takes to support Q—the fact about the experi-
ence or the experience itself. Plausibly, a belief is needed for commitment to
¹⁸ Balcerak Jackson and Balcerak Jackson (2013) claim that where there is no gap between the
contents of states involved in a transition, it is implausible to posit inference. One might think that in
genuinely perceptual justification, the transition is from an experience with the content P to a belief
that P.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
the antecedent. I must believe this fact about the experience obtains or that the
experience itself exists.¹⁹
What if a taking state is required for the relevant transitions but isn’t
explanatory of them? John Broome (Forthcoming) is sympathetic to the
view that if one reasons from beliefs in P1, . . . , Pn to a belief in Q then one
has a disposition to reason from the former to the latter, and that this
disposition is an implicit belief that Q follows from P1 . . . Pn. This implicit
“linking belief” doesn’t help explain inference; rather, the nature of inference
guarantees that one has such a belief. Could something like this hold for
experience-to-belief transitions? I don’t think so. The relevant disposition
exercised in a noninferential experience-to-belief transition doesn’t seem, by
itself, to constitute such a linking belief.
Why not? It’s instructive to consider why Broome himself denies that every
instance of reasoning requires a linking belief. He gives the example of
reasoning involving transitions between intentions, such as instrumental
reasoning in which one starts with an intention for an end and a belief
about the means to that end and concludes with an intention concerning the
means. Unlike the case of inference involving only beliefs, in which the
contents for the linking belief would not go significantly beyond the contents
of the beliefs involved, adding only whatever content is needed for the
conditional, the contents of the linking beliefs corresponding to the instru-
mental reasoning with intentions would contain new material going far
beyond the contents of the intentions/beliefs involved and beyond the con-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
tents needed for a conditional; the linking belief would have to be about
intentions—it would have to be a belief that if I had such and such intention,
then . . . . The same goes for experience-to-belief transitions; the linking belief,
here too, would need to be about experiences. But it isn’t plausible that your
having the disposition to make experience-to-belief transitions ensures that
you have a linking belief with such a rich content. Not only that, but as
Broome notes, it’s implausible to think that young children have such linking
beliefs, but not implausible to think that they engage in the relevant instru-
mental reasoning with intentions. The same goes for the experience-to-belief
transitions.
So, the explanation based on the taking condition, even if successful in the
case of inferential justification, doesn’t carry over to explain how there can be
¹⁹ It’s seeming that a fact obtains or that an experience exists is not enough. Seemings do not ensure
the needed commitment of the subject. The point is familiar: if it seems to me that one rod being longer
than another, I need not be committed to its being longer.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
But this by itself doesn’t make the transition from experience E to believing
Q epistemically inappropriate. Whereas in the inferential case a reason to
doubt that P wouldn’t be true unless Q seems by itself to make the transition
from the belief that P to the belief that Q inappropriate, a reason to doubt that
having experience E supports Q at most makes it inappropriate to transition
from the belief that you have E to the belief that Q, but doesn’t make it
inappropriate to transition from the experience E itself to the belief that
Q. The Milk Taster case exemplifies this contrast well. If the milk taster
believed she was basing her belief on smell, it would be inappropriate for her
²⁰ I assume here that a doubt could appear in thought as an affirmation of a statement about chance.
One might question this. However, these issues could be bypassed by thinking of a case in which the
subject believes (justifiably) that it’s not true that she wouldn’t have E unless Q. We could then just load
this negation into the reasoning.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
to keep the belief; but if not, and especially if she believes (justifiably) that she
isn’t basing her belief on smell, it doesn’t seem inappropriate.
One might balk here, and indeed balk at Sturgeon’s (and my) judgments
about the Milk Taster case. Isn’t there something epistemically inappropriate
about basing a belief on a source while believing that there is a good chance
that that source is unreliable, even if you don’t have believe it’s your source?
But just what is inappropriate about doing this? It’s true that an ideal cognitive
agent would do no such thing, but that is because an ideal cognitive agent
would know its sources (Mellis 2014). It doesn’t follow that we lack justifica-
tion from the fact that we are non-ideal in this way. It might be claimed that
the real problem is that you’re conducting yourself in such a way as to
guarantee that your resulting belief fails to be knowledge if your belief about
your source’s unreliability were true and you still believed on the basis of that
source. If the milk taster’s belief that olfaction is unreliable in the circum-
stances were true, and if she still based her belief about the milk on olfaction,
it’s true that her belief wouldn’t be knowledge. But it’s hard to see what the
bearing of these facts is on whether the milk taster is in fact justified in
believing the milk is ok.
Thus, while there may be senses in which the milk taster’s belief is epistem-
ically non-ideal and senses in which the belief would be problematic if the
belief about the unreliability of olfaction were true, I don’t think these defects
are ones a theory of defeat needs to capture, because they do not eliminate or
reduce justification.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
The conclusions from the previous two sections imply that there is no such
thing as Pollock-style undercutting defeat of noninferential justification, only
of inferential justification. The nearest thing to genuine Pollockian attack-the-
connection defeat of a noninferential justification is what I’ve called higher-
order defeat: defeat in virtue of having reason to think that one’s belief is
unreliably based, a kind of defeat that requires beliefs about one’s bases or
sources and so about oneself and one’s mind. If these conclusions are correct,
we might use them to help us determine whether a justification in a certain
case is inferential or not. This section considers the implications for perceptual
and testimonial justification.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
It might be surprising that you start talking about yourself in the last line, but
it’s these facts about oneself that articulate the source of your justification for
not concluding it’s an eggplant. Contrast this final reply with a more
natural one:
: “Because if there are look-alike fakes around, then, even though this
looks like an eggplant, it could easily be one of the fakes.”
Similarly, in reply to a challenge about why you don’t conclude the wall is red
in a case of strange lighting, you might say:
: “Because if the lighting conditions are abnormal, then, even though the
wall looks red, it could easily be some other color.”
These more natural final replies don’t appeal to one’s own mental states. The
proposition that the thing looks like an eggplant does not entail anything
about my mind, or arguably about other people’s minds. It is a fact about
eggplants that can survive my demise and that of the rest of humanity.
Is there a way to block the “mentalization” of such defeaters for perceptual
justification? The classical foundationalist takes one’s justification in the
relevant cases to be inferential but still takes the justification to come from
equally mentalized premises, about which experiences one has. But there is
another sort of inferential approach available, suggested by the more natural
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
²¹ There would need to be a story of how the visual experience contributes to the justification of the
belief about the thing’s looking like an F. Are the latter beliefs noninferentially justified, or are they
themselves inferentially justified? Eventually, we must get down to beliefs that are not inferentially
justified. For them, undercutting defeat would be impossible. I am sympathetic to the view that the
noninferentially justified beliefs are beliefs about things having certain looks, and I suspect that such
beliefs aren’t subject to undercutting defeat, only higher-order and rebutting defeat of an object justifies
me in believing it looks like an F, and this, perhaps together with some background information or a
default assumption, inferentially justifies me in believing that it is an eggplant.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
²² For some development of this sort of account, see McGrath (2017, 2018).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
9.6. Conclusion
References
Press.
Balcerak Jackson, Magdalen and Brendan Balcerak Jackson. (2013). Reasoning as
a source of justification. Philosophical Studies, 164: 113–26.
Boghossian, Paul. (2014). What is inference? Philosophical Studies, 169: 1–18.
Broome, John. (Forthcoming). Linking beliefs. In: Balcerak Jackson, Magdalen
and Brendan Balcerak Jackson (eds.) Reasoning. Oxford: Oxford University
Press.
Burge, Tyler. (1993). Content preservation. Philosophical Review, 102(4): 457–88.
Casullo, Albert. (2018). Pollock and Sturgeon on defeaters. Synthese, 195:
2897–906.
Comesaña, Juan. (Forthcoming). Empirical justification and defeasibility. Synthese,
²³ I am grateful for comments from an audience at the University of Pavia, especially Tommaso
Piazza, James Pryor, and Crispin Wright. For comments on written drafts, I thank Jessica Brown, Al
Casullo, Juan Comesaña, and Jack Lyons.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
10
Defeaters as Indicators of Ignorance
Julien Dutant and Clayton Littlejohn
10.1. Introduction
1. Opposing Defeaters
2. Undercutting Defeaters
3. Reason-Defeating Defeaters
4. Collective Defeaters
5. Pragmatic Defeaters
6. Identification Defeaters
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
7. Higher-Order Defeaters
8. Negative Self-Appraisal Defeaters
We can think of the different items as standing for the different mechanisms
by which something threatens rationality. A good theory of defeat should tell
us which of the things on the list, if any, are genuine defeaters.¹
2. Monistic views are prima facie preferable to pluralistic views. Monists
and pluralists agree that there are different kinds of defeaters, but the monist
¹ In addition, there are discussions about whether defeaters are pieces of evidence, psychological
states or contents, things that we should have known about or believed, or facts. As interesting as these
discussions are, they don’t directly address questions about how defeaters defeat. And that is our focus.
For good discussions of normative defeat, see Goldberg (2017); Lackey (2006); Madison (MS). For
discussions of psychological defeat, see Bergmann (2006); Lackey (2006).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Here is our plan for what follows. In Section 10.1, we give a brief presen-
tation of some of Pollock’s seminal ideas about defeat and introduce an
alternative view that seems to satisfy our second and third desiderata. This
account characterizes rational belief in terms of confirmation and then char-
acterizes defeat in terms of things that have a negative impact on confirmation.
We shall argue that this approach cannot handle certain kinds of higher-order
defeat and that it mischaracterizes some epistemically beneficial factors as
defeaters. It doesn’t satisfy our first desiderata. In Section 10.2, we explain
where defeat should be located in a normative theory and argue that the first
view ran into trouble because it is a truth-centred view. In Section 10.3, we
propose a knowledge-centric view as an alternative, one that builds on the
strengths of the first truth-centred approach but improves upon it because it
handles the cases that cause trouble for the truth-centred approach.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
² Recall Spohn’s (2002) complaint that some theories (e.g. Pollock’s) were normatively deficient
because they offered laundry lists of (alleged) defeaters and no unified explanation as to why these
different defeat relations were toxic.
³ See Williams (1965: 107) for an interesting argument that the reasons we have to believe in
pluralism in the practical domain do not carry over to the epistemic domain. There might be reasons to
prefer a pluralistic view to a monistic one, but the existence of such reasons doesn’t change the facts
about what’s prima facie preferable to what.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
⁴ We should note that much of the literature on the defeat of rationality or justification is concerned
with the way in which defeaters defeat status or standing. On Pollock and Cruz’s characterization,
however, defeaters defeat reasons. The connection between defeating reasons and status is not obvious
without appeal to further assumptions about reasons and rationality. Because one of us doesn’t believe
that all beliefs need to be supported by reasons to be rational, it’s not clear to us how to connect the
defeat of reasons to the defeat of status. For arguments that beliefs needn’t be based on evidence or
reasons to be knowledge or to be rationally held, see Anscombe (1962); Littlejohn (2017); McGinn
(2012). For discussion of a more reason-centric approach, see Lord (2018).
⁵ Let’s assume that the relevant notion of probability is some kind of subjective probability or
evidential probability that captures the confirmation relation in question. Maria Lasonen-Aarnio
rightly pointed out that Pollock’s approach is not probabilistic. We are thinking of this as a way of
trying to build on some of his ideas in a probabilistic approach.
⁶ See Dorst (2019); Easwaran (2016); Sturgeon (2008) for defences of .
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
: the fundamental epistemic good is true belief and false belief is the
fundamental epistemic evil.⁸
Rational believers have the twin goals of acquiring true beliefs and avoiding
false beliefs. They’re often uncertain about whether they achieve these goals by
forming a belief and so they need to strike a balance between unreasonable
aversion to risk and recklessness. They rationally ought to believe in ways that
maximize expected epistemic value. This gives us the argument for .⁹
⁷ Chandler (2013) and Kotzen (2019) provide sophisticated formal characterizations of different
defeaters in the spirit of -.
⁸ See Goldman (1999); Lynch (2004); Sylvan (2018) for defences of veritism.
⁹ See Easwaran and Fitelson (2015); Dorst (2019) for details.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
some of our reasons for believing things like s are just the contents of justified
beliefs. Perhaps b is just such a reason in this sense. In that case, we might
imagine that there is some evidence e which supports b and which supports s.
(Perhaps e is a proposition about our perceptual experiences, but the precise
content doesn’t matter provided that e does not entail b or s.) Then we could
say that upon learning e, the probability of b and s increase and allow that it’s
possible that we learn something, d, where d is a reason-defeating defeater
because d lowers b’s probability without affecting the conditional probabilities
of s on b or on ~b.¹²
¹¹ Undercutting (or undermining) is sometimes understood very broadly (e.g., by Brown (2018)
who treats higher-order defeat as a kind of undercutting or undermining defeat), but we want to
understand it narrowly because we think that - cannot accommodate certain kinds of
higher-order defeat.
¹² Let P be your probability after e is acquired. For b to be a reason to believe s it must be that P(s|b) >
P(s|~b). Recall d doesn’t affect the conditional probabilities P(s|b) and P(s|~b), so P(s|b&d) = P(s|b) and
P(s|~b&d) = P(s|~b). Finally, d lowers the probability of b, so P(b|d) < P(b). From this, it follows that d
de-confirms s, namely, P(s|d) < P(s).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
¹³ See Leitgeb (2017); Pollock (1986); Smith (2016); Ryan (1991, 1996) for defences of views that
include this consistency requirement on justified belief.
¹⁴ See Christensen (2004); Easwaran and Fitelson (2015); Makinson (1965); Worsnip (2016) for
arguments that collective defeaters don’t invariably defeat the rational status of our beliefs. In Dutant
and Littlejohn (2020), we argue that views that posit collective defeaters are either overly sceptical or
too externalist (i.e. they deny us too much justification or they pick winners and losers on the basis of
considerations beyond the thinker’s ken).
¹⁵ For arguments that pragmatic considerations can help to determine whether a body of evidence
provides sufficient support for belief, see Fantl and McGrath (2009); Owens (2000); Schroeder (2012).
We think Schroeder is the first to describe these pragmatic considerations as defeaters.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
If readers believe that practical considerations (e.g. the stakes or the odds of a
salient bet) can create rational pressure to suspend without acquiring more
evidence than might normally be necessary, - can accom-
modate this kind of pragmatic defeater by positing that such defeaters move
the threshold.¹⁶
While - has a number of virtues, we struggle to see how it
handles further kinds of defeat that are discussed in the literature. Peacocke,
believes there are identification defeaters:
At first glance, this might look like a case of undercutting defeat. The expert
tells you that the proof is flawed so you might agree that the putative support
doesn’t actually support the conclusion. If this is a case of undercutting defeat,
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁶ For discussion of such cases, see Fantl and McGrath (2009); Owens (2000); Weatherson (2005).
For discussion of the ways in which moral considerations might provide pragmatic defeaters, see Basu
and Schroeder (2019); Moss (2018).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
¹⁷ See Dorst (2020); Dorst, Fitelson, and Husic (Forthcoming); and Williamson (2019) for discus-
sion of level-connecting principles.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
rational. Moreover, we think that the account will struggle with a kind of
higher-order defeat that we call negative self-appraisal defeat.
The third response is to our mind the most promising. The de-confirmers
might try to account for higher-order defeat by suggesting that it threatens
ex post rationality rather than ex ante rationality.¹⁸ The standard view is that
ex post rationality involves a basing requirement (i.e., that we competently
base our beliefs on the support that warrants them). We might account for
higher-order defeat by saying that it impacts ex post rationality on the grounds
that anyone who believes in spite of the presence of these defeaters cannot
meet the basing condition. Perhaps if Agnes believes (falsely but reasonably)
that she has been drugged, it is ex ante rational to believe the conclusion of a
proof but it fails to be ex-post rational because she cannot competently base
her beliefs on the good reasons that she has.
This strategy might help proponents of - accommodate
some higher-order defeat, but we don’t think this covers negative self-
appraisal defeat. A negative self-appraisal defeater is provided by consider-
ations about the epistemic status of some particular belief. Agnes might believe
something about, say, the afterlife. She might then be convinced that she
couldn’t possibly know anything about the afterlife. There seems to be a
rational tension between the belief that we’ll survive our own bodily deaths
and the belief that we couldn’t possibly know whether people can survive their
own bodily deaths.
Negative self-appraisal defeaters differ from the previous higher-order
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
defeaters in two ways. First, they concern the epistemic status of particular
beliefs (e.g. whether we know, whether we rationally believe, whether it is okay
to believe given the evidence we have, etc.) rather than pertaining directly to
the thinker’s belief forming processes or strength of evidence. Second, they
have a kind of epistemic content missing from more widely discussed cases of
higher-order defeat.
How can - handle cases like these? Suppose that a thinker
believes p and acquires evidence that confirms any of the following:
¹⁸ This idea was inspired by Di Paolo (2018); Silva (2017). They do not endorse -,
but we thought this work might help fend off one kind of objection. See also van Wietmarschen (2013).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
On its face, it seems that the evidence may raise the probability of (1)–(3)
without lowering that of p itself.¹⁹ So, if negative self-appraisal defeaters defeat,
they might show that there are defeaters that are not de-confirmers.²⁰
Think about lottery cases. Harman nicely captures part of what makes the
lottery case so puzzling:
Given the weak anti-sceptical assumption that our testimonial beliefs are
rational and Harman’s observation about the chance of error,
tells us that it is rational to believe lottery propositions. Suppose this is so and
suppose that we know a priori that we cannot know that we’ve lost the lottery.
If so, some Moorean absurdities cross the threshold. -
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁹ See Dorst (2020) and Williamson (2019) for arguments that show that once we allow that we can
be rationally uncertain to even a slight degree about what our evidence supports, we should expect that
there can be bodies of evidence that both make it very probable that p and very probable that it is not
sufficiently probable that p. Given , this would be evidence that would make it rational to
believe p and rational to believe that it is irrational to believe p. In truth-centric accounts, then, it seems
we can only account for the toxicity of (2) by denying or by accepting a kind of negative
introspection principle that says that where t is the relevant threshold for rational belief (which we
assume is less than 1), P(p)<t only if P(P(p)<t) =1.
²⁰ Bergmann (2006) claims that some beliefs provide negative self-appraisal defeaters even if those
beliefs are not adequately supported. We disagree. We think that he is making the kind of mistake that
people make when they conflate a wide- and narrow-scope requirement. We would agree that a thinker
should not believe p if she believes that she does not know p or if she believes that it’s irrational for her
to believe p, but we don’t think we can detach any interesting normative claim from this even if we add
the further (psychological) claim that the thinker does believe (1), (2), or (3).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Let’s step back and consider some foundational issues. We have already seen
two different rationales for . If we accept a truth-centred approach
to epistemic matters and either accept or , there is an
argument to be made that rational thinkers ought to conform to
because this represents the right way of handling certain risks (e.g. the risk of
²¹ See Smith (2016) and Weatherson (2016) for defence of the idea that non-sceptical accounts of
knowledge have to allow that some of the things we know have lower evidential probability than some
true propositions that we’re not in a position to know. Their defence uses lottery cases.
²² To see this, break down Agnes’s evidence into two components: that the file she thought to be
Inge’s isn’t, and that the file she thought to be Agatha’s is Inge’s. The first is a pure undercutting
defeater of Agnes’s initial reason to believe in Inge’s guilt (namely, the contents of the file she
mistakenly thought was hers). The second is a confirmer of Inge’s guilt, but not as strong a confirmer
as the previous was. Together, they are a mild undercutting defeater, on the de-confirmation account.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
believing in ways that are objectively bad or failing to believe in ways that are
objectively good or that of violating some objective norm). If were
correct, we could easily see how to give a monistic account of defeat and
explain why some familiar kinds of defeaters are rationally toxic without
having to add a clause to our theory of rational belief to handle defeat. That
would mean that - satisfies our second and third desiderata,
but doubt that it satisfies the first, extensional adequacy. Adding a clause to
handle tricky cases (e.g. higher-order defeat, puzzling intuitions about under-
mining defeat in lottery cases) might help with extensional adequacy, but then
the modified view would no longer satisfy our second and third desiderata.
Many epistemologists have raised concerns about along the
lines of those -previously discussed. Smith (2016), for example, thinks that the
problem has to do with this idea that justification is a matter of managing risk
and so he rejects the risk-minimization approaches to justification (of which
is an instance) in favour of an alternative view designed to allow
that we might be justified in believing in one case and unjustified in believing
in another even if the risk of forming a false belief is greater in the second kind
of case than the first. The view that we shall propose has a similar feature, but
we don’t want to be too hasty in rejecting the idea that much of rationality is a
matter of properly handling risk.
Before we reject the very idea that there is some connection between
rational belief and the management of risk, we should remember that there
are many different kinds of risk that we might need to manage. If a truth-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
centred approach were correct, we should worry about forming false beliefs
and failing to form true ones. If we were veritists, we would describe this as a
concern about managing the risks of forming bad beliefs and failing to form
good ones. If we believed in epistemic norms like truth norm, we would
describe this a concern about managing the risk of norm violation. It seems
strange to us to suggest that rationality shouldn’t be concerned with the
management of these kinds of risk. On its face, the idea that the rational
person’s responses are rational precisely because they manage these kinds of
risks well seems pretty plausible. Perhaps the lesson to draw from the diffi-
culties that arise for is not that we were mistaken to think that
rationality has a great deal to do with the management of risk but that we were
mistaken about the kinds of risks that need to be managed.
We’re not all veritists. And we don’t all believe that the fundamental
epistemic norms have to do with truth. Some of us prefer gnosticism to
veritism and prefer to think of epistemic norms as concerned with knowledge
and not merely with truth:
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
If we take either of these proposals as our starting point, the same arguments
that took us from veritism or truth norms to threshold take us from these
knowledge-centred claims about knowledge’s normative role to this account of
rational belief:
given what it would take to responsibly try to meet our objective epistemic
²³ See Hyman (1999); Littlejohn (2018); Williamson (2000); Unger (1975) on the value of knowledge
for motivations for .
²⁴ See Dutant and Fitelson (MS); and Littlejohn (2020) for discussion and defence of the idea that
rationality is understood in terms of the probability of knowledge. They are primarily concerned with
synchronic wide-scope requirements. We said earlier that we think of probability here as evidential
probability. Some might worry that if we combine this approach with the idea that everything known is
part of our evidence we will narrow the scope of the applicability of the account. (For example, if E=K,
our account says that in cases where you know that you know p, the probability that you know p is 1.
And this would mean that we couldn’t give an account of how belief in p could be defeated.) We aren’t
certain that the narrow scope is a problem and we are not taking any official stand on the relationship
between evidence and knowledge. In particular, we think that there are interesting independent reasons
to look for an account of evidential probability on which it’s possible to know p even if the probability
of p is less than 1. See Bacon (2014) and Brown (2018) for discussion of such reasons.
²⁵ In this way, we think we can follow Ghijsen, Kelp, and Simion (2016) in giving an account of the
subjective and objective dimensions of epistemic assessment without rejecting the idea that the
knowledge norm governs belief. Graham (2010) proposes a similar approach to reconciling an
objectivist approach to obligation with a more subjectivist account of rationality.
²⁶ We think a probabilistic approach to rationality that links rationality to the risk of norm violation
will help objectivists address some of Brown’s (2018) objections to Littlejohn and Williamson’s
proposals about excuses.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
obligations. Smith might have been right that rational belief isn’t just a matter
of believing in ways minimize the risk of error, but there are alternative risk
minimization approaches that we should consider. We have just offered one.
²⁷ We hadn’t realized that Gibbons (2013) had proposed that defeaters were evidence that we aren’t
in a position to know. Thanks to Errol Lord for pointing this out. One difference between his approach
and ours is that we introduced a rational standard that should help us decide whether some defeater is
full or merely partial. We also disagree about lottery cases, so he wouldn’t accept our points about
lotteries and undercutting defeat. Although we put Di Paolo’s (2018) work to use earlier in defence of
-, he said that he was sympathetic to a knowledge-centred view.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
be possible to believe each claim in the book whilst knowing that one of the
claims in the book is false. Our view differs from truth-centred views in that
they tend to treat lottery and preface cases in the same way. In our view, such
views deliver the wrong result in lottery cases or in preface cases. The
difference in the prospect of coming to know in such cases is the key to
understanding why we’re tempted to think of them as differing in a way that
should matter for our theory of rationality.
also provides a better treatment of negative self-
appraisal defeat. First, if the conjunction of some first-order belief and the
negative self-appraisal is not knowable (e.g. because the negative self-appraisal
represents the first-order belief as failing to fulfil some necessary condition on
knowledge), says that the conjunction cannot be
rationally believed because the probability that the conjunction is known is 0.
What about the conjuncts, though? It’s one thing to say that nobody can
rationally believe certain conjunctions that we know cannot be known (e.g.
‘Dogs bark but I don’t know that they do’) and another to say that nobody can
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The difficult case is (2). If ex ante rationality were necessary for being in a
position to know, our treatment of (1) would cover this case.²⁸ Unfortunately
for us, we cannot say that rationality is necessary for knowledge. If we assume
probable knowledge, we should expect that there will be some ‘unreasonable
knowledge’ in the sense that there will be some things that we know that will
not be sufficiently likely to be known to also be rational to believe.²⁹ Still, while
we might want to allow for the possibility of unreasonable knowledge, we
might want to deny that it can be rational to both believe p and believe that this
belief is irrational.
²⁸ Let t be the threshold in . Suppose knowledge entails rational belief: by
, P(Kp) ≤ t entails ~Kp. Now suppose it is rational to believe that it is not
rational to believe p. By , P(P(Kp) ≤ t) > t. Since P(Kp) ≤ t entails ~Kp, P(~Kp) >
t. Assuming t ≥ ½, P(Kp) < t.
²⁹ See Lasonen-Aarnio (2010) for discussion and defence of unreasonable knowledge and Sylvan
(2018b) for defence of the idea that knowledge is a non-normative state that needn’t have a normative
dimension. We think that a non-normative account of knowledge can help address some concerns
about the very idea of defeat raised by Baker-Hytch and Benton (2015).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
For note that P(Kp)> t obtains just if P(Kp)≤ t doesn’t. Hence P(P(Kp)> t)≥ 1-t
is equivalent to P(P(Kp) ≤ t) ≤ t. Given the factivity of being in a position
to know, P(P(Kp)≤ t)≤ t only if P(KP(Kp) ≤ t) ≤ t. Hence given
t-, if P(Kp)>t then P(KP(Kp)≤ t)≤ t. This would mean that if the
probability that one knows p is above threshold, the probability that one
knows that it is not above threshold is itself below threshold, as desired.
Note that while t- is an inter-level constraint, it does not
force us to accept strong access constraints like the idea that it is rational to
believe p only if it is rational to believe that one knows that p. It still allows the
combination of P(Kp)>t and P(P(Kp)>t)≤ t and the combination of Kp and P
(Kp)≤ ½.³⁰ Note also that the higher t is, the milder the t-
requirement is. At t=.9, for instance, it merely requires that if it is highly (.9)
probable that one knows, it is at least somewhat (i.e. .1) likely that it is highly
probable.
While we don’t have a full defence of t-, we think that it
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
has some prima facie appeal and is worth exploring. Its initial appeal comes
from the idea that whatever makes it probable that one knows will typically
also make it probable that it is probable that one knows. What confirms to you
that you know p by memory or perception, for instance, is typically the fact
that your perception or memory is particularly clear or detailed, the fact that
you can give additional details pertaining to p, and so on. But those facts are
also evidence that you have evidence that you know, which is to say that they
contribute to make it probable that it is probable for you that you know.
The thought gets further support from an unlikely source: Williamson’s
(2014) unmarked clock models. While these models are meant to illustrate the
failure of various access constraints on knowledge, they turn out to validate
t-. We can illustrate how this works with an example and we
provide a proof in the Appendix. In Williamson’s models, a parameter (e.g.
time, temperature, etc.) takes different values along a line. Worlds correspond
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
to points on the line. What you know at a world is whatever holds at worlds
within a margin of error m to the left and to the right of the point on the line.
Your evidential probability is given by conditionalizing a uniform prior on
what you know. When you update in this way, every value outside of the
margins is ruled out and each of the values within the margins are equally
likely. Suppose the actual world (@) is at 50 and the margin for error (m) is 50.
What you know leaves open any world within [0:100]. Let p be some propos-
ition such that Kp holds throughout [30:90]. (For instance, p might be the
proposition that the value of the parameter is between -20 and 140.) Because
Kp occupies 60 percent of the worlds left open by what you know at @, the
probability of Kp at @ is.6. Note that any world between 40 and 80 will also
have within its margin the Kp-segment [30:90]: for 40 leaves open any world in
[-10:90] and 80 leaves open any world in [30:130]. So, at each of these worlds,
P(Kp)≥.6, too. This means that at least 40 per cent of worlds within @’s
margins are Kp-worlds. So, at @, we have an instance of 0.6-
, P(P(Kp)≥.6)≥.4. The reasoning generalizes to any Kp-proposition
that occupies a continuous stretch of 50 per cent of @’s margins.³¹
@±m
0 30 50 90 100
p
Kp
40±m
80±m
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
P(Kp)≥6
³¹ Note that propositions of the form Kp have to occupy continuous stretches within a world’s
margin. For if Kp holds at two points within @’s margins, those two points are at most 2m apart, which
means that their own margins connect and p holds throughout the segment between them, which in
turns means that Kp holds at any point between them.
What happens for t below ½? Suppose for instance that Kp holds throughout [80:100]. Then it
occupies 20 per cent of @’s margins, so at @ we have P(Kp) ≥.2 . Now all we can say is that every world
to the right of @ ‘sees’ the whole of [80:100]. So, we have P(P(Kp) ≥.2) ≥.5. More generally, the models
validate an alternative schema: for any t≤½, if P(Kp)> t, (P(Kp)> t)≥.5. This is good news as instances of
t- with low t are counterintuitive.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
seen how principles like - would fare if we worked with
different models (e.g. ones that do not assume a uniform prior, ones that do
not have a constant margin, etc.). At the very least, we think this gives us some
reason to explore principles like - further and reason to
hope that such principles will hold in a wide range of cases.
If it is impossible to rationally believe the propositions mentioned in (1)–
(3), our account handles cases of negative self-appraisal defeat quite nicely.
When it is rational for the thinker to believe, say, that she’s not in a position to
know or that it would be irrational for her to believe p, it would not be rational
for her to believe p and we can see why these negative self-appraisals can
function like defeaters. Alternative views that characterize rational belief in
terms of the probability of truth struggle to explain why it would be rationally
mandatory for a thinker who rationally believes that she doesn’t know, say, to
suspend judgment.
If our treatment of negative self-appraisal is correct, we can build on an
interesting proposal about suspension defended by Thomas Raleigh
(Forthcoming). He proposed that suspending is really a matter of believing.
We suspend, on his view, when we believe the evidence doesn’t support a
belief. We think that something very similar might be right. Provided that a
thinker is rational, her rational belief that she’s not in a position to know
whether p should ensure that she’s agnostic about whether p. (Even without a
belief, a sufficiently high degree of confidence should also work.) It is difficult
to explain why this should be in a truth-centric framework since, as we’ve seen,
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
it’s not clear why, if is correct, a rational thinker would suspend if
she were to judge that she’s not in a position to know or that her evidence
doesn’t support her belief. Our framework, however, explains why a rational
thinker suspends when she rationally believes that she’s not in a position to
know. We think that it’s a strength of our view that it helps us see what’s
attractive about Raleigh’s proposals about belief and suspension.
10.5. Conclusion
We have proposed a theory of rational full belief according to which it’s rational
to believe when it’s sufficiently probable that the belief would be knowledge.
From this account, we can derive a theory of defeat according to which defeaters
are indicators of ignorance, evidence that if you were to believe, your belief would
fail to constitute knowledge. We think that this account provides a better account
of every kind of (genuine) defeat there is than the leading alternative. If we think
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Appendix
We show that t- (for t≥.5) holds in Williamson’s (2014) unmarked clock
models. Worlds are points in the real number line, propositions are regions of the line;|p|is
the size of the region p; for every p, q, the proposition p&q is the intersection of p and q. At
each world w one knows every proposition entailed by the parameter’s value being within a
margin m>0 of w. We let k(w) = [w-m:w+m] and for any p, the proposition that one knows
p, Kp, is the set {w:k(w) ⊆ p}. The evidential probability at w, πw, is a uniform prior
conditionalized on k(w): πw(p)=|p&k(w)|/|k(w)|. Without loss of symmetry we set m=½
which gives us|k(w)|=1 and the simple relation: πw(p)=|p|. The proposition that the
evidential probability of p is above t, P(p)>t, is the set {w: πw (p)>t}. Now let t≥½, @
some world and p such that π@(Kp)>t.
Our aim is to show that π@(P(Kp)>t)≥1-t. Let a be greatest lower bound and b the least
upper bound of Kp&k(@). Since Kp holds at worlds (arbitrary close to) a and b, p holds
throughout the regions]a-m:a+m] and [b-m:b+m[. Since a and b are within k(@),|b-a|<2m
and the regions overlap and p holds throughout]a-m:b+m[. So Kp&k(@) holds throughout]
a:b[, and since π@(Kp)>t,|]a:b[|>t. Now let w be any world within]a+(t-½):b-(t-½)[. (This
interval has positive size since |]a:b[|>t≥½.) Since w>a+(t-½), w-m=w-½>a+t-1, which
given t≤1 entails w-m>a, and by parallel reasoning, w+m<b. Hence ]a:b[ ⊆ k(w). Since]a:b
[ ⊆ Kp and |]a:b[|> t, πw(Kp)>t. So P(Kp)>t holds at any world w within ]a+(t-½):b-(t-½)[.
Now since |]a:b[|>t≥½,|]a+(t-½):b-(t-½)[|>t-2(t-½)=1-t. Therefore π@(P(Kp)>t)>1-t.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
References
Adler, Jonathan. (2002). Belief ’s Own Ethics. Cambridge, MA: MIT University Press.
Anscombe, G. E. M. (1962). On sensations of position. Analysis, 22: 55–8.
Bacon, A. (2014). Giving your knowledge half a chance. Philosophical Studies, 171:
373–97.
Baker-Hytch, Max and Benton, Matthew. (2015). Defeatism defeated.
Philosophical Perspectives, 29: 40–66.
Basu, Rima and Schroeder, Mark. (2019). Doxastic wronging. In: B. Kim and
M. McGrath (eds.) Pragmatic Encroachment in Epistemology. New York:
Routledge: 158–78.
³² For feedback and discussion, we would like to thank Maria Lasonen-Aarnio, Josh Di Paolo,
Branden Fitelson, Errol Lord, Mona Simion, and particularly Kevin Dorst. Kevin was kind enough to
provide proofs that helped us think through the issues in Section 10.4.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
11
Competing Reasons
Justin Snedegar
One of the most important facts about the normative domain is that some
considerations are contributory, rather than decisive, when it comes to deter-
mining what we ought to, must, or may do.¹ They provide defeasible support
or defeasible opposition to an option. In epistemology, the theory of practical
reasoning, and ethics, it is standard to appeal to contributory or pro tanto
normative reasons to capture this fact. Most of our options have some reasons
in favor and some reasons against; to determine which option to take, we have
to determine how these reasons compare. In particular, on the standard view
reasons compete with one another, and the outcome of this competition
determines what you ought to, must, or may do.² That action A would be
mildly pleasant is a reason to do A, and that an incompatible action B would
prevent a great deal of suffering is a reason to do B. These reasons compete
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
with one another, since A and B are incompatible. If there are no other
relevant reasons, then what you ought to, must, or may do—A or B—will
depend on the outcome of this competition.
An important task for normative theory, then, is to explain how the compe-
tition between reasons works. A large part of this task will involve investigating
relationships of defeat between reasons, since when reasons compete, at least in
many cases, some reasons will defeat others—that is how they win the compe-
tition. Though the concept of defeat is perhaps more familiar in epistemology,
I will be concerned mostly with the practical domain.
¹ See Ross (1930) for an early and very influential development of this idea; for a recent overview, see
Lord and Maguire (2016).
² I’m using this slightly long-winded phrase, ‘what you ought to, must, or may do,’ because the
differences between these different normative statuses will be important later. To note the difference
between ‘ought’ and ‘must’ here, consider the following, which may be said about some supererogatory
action: “Well, no, I’m not saying you must do it, but you really ought to.”
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
05.003.0011
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
³ In fact, this is too simple. There are cases of partial undercutting in which a reason is not fully
undercut, but just weakened. So, a bit more precisely: when a reason is merely rebutted, it retains its full
strength, but when it is undercut, it loses at least some—and perhaps all—of its strength. See Schroeder
(2007): Ch. 5; compare with Dancy (2004): Ch. 3, on attenuators.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
something I should take into account in deciding where to send her. It may
seem that my promise is an undercutting defeater of the financial reason to
send our daughter to Good Public School. But in fact there is an important
difference between undercutters, as in the coerced promise previous case, and
exclusionary reasons.⁴ In the undercutting case, the initial reason is no longer
in force at all—it is just not a reason anymore. But, in the exclusionary case,
the reason is in force, it’s just that I have a reason not to take it into account in
my deliberation.⁵ In the undercutting case, there is no reason to be taken into
account. Nevertheless, even though the reason in the exclusionary case is in
force, it will lose the competition—in fact, since it is taken out of consider-
ation, it doesn’t even get to compete.⁶
So, we have seen three different ways in which a reason may be defeated, or
in which reasons can compete. The reason may be rebutted, undercut, or
excluded. Arguably, the most straightforward of these is a rebutting defeat, in
which a reason is defeated by a stronger reason for a conflicting option. One
goal for this paper is to suggest that rebutting defeat is actually not as
straightforward as it seems. In particular, I will argue that we should recognize
two different varieties of rebutting defeat. First, there are cases in which a
reason for one option is defeated by a stronger reason for an incompatible
option. Second, there are cases in which a reason for one option is defeated by
a stronger reason against that same option (or vice versa). Many philosophers
have not recognized this distinction between two kinds of rebutting defeat,
I believe, because they have not recognized a distinction between reasons for
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
⁴ See Horty (2012) and Horty and Nair (2018) for the claim that exclusionary reasons are just
undercutters. See Raz (1999): 184; and Whiting (2017): 400, for a rejection of this identification.
⁵ Whiting (2017) makes essentially the same point, but he understands exclusionary reasons as
second-order reasons not to base one’s actions or attitudes (including decisions) on certain ordinary,
first-order reasons. Raz (1999), especially in the postscript, often seems to understand them in this way
as well.
⁶ I am glossing over some complications here. For example, there may be cases in which the
excluded reason is so weighty that it should not be excluded. A full theory of the competition between
reasons, and in particular of how exclusionary reasons function, will need to explain this.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
First, I’ll introduce what I take to be the most common way of thinking
about the competition between reasons, which I call balance accounts. Then,
I’ll argue this picture—as so far developed—is inadequate, primarily because it
does not recognize the importance of reasons against. I then consider a very
different way of thinking about the competition between reasons, which I call
criticism-based accounts. This kind of view does recognize the importance of
reasons against but misses out on the importance of the competition between
reasons bearing on competing options. Finally, I sketch a positive view that
incorporates elements of both kinds of view. Open questions remain, but
I argue that it gives us the beginnings of an attractive account of how the
competition between reasons determines the overall status of our options.
⁷ A different metaphor comes from Ross (1930), who thinks of the competition between pro tanto
considerations as analogous to the competition between different forces acting on an object. I take the
two physical metaphors to suggest the very similar conceptions of the competition between reasons.
⁸ See, e.g., Broome (2004). ⁹ On accrual, see Nair (2016).
¹⁰ Often authors use ‘ought’ instead of ‘required’ here. I think that most of them have in mind what
I’ll mean by ‘required,’ including that you’d be acting impermissibly by not doing what you “ought” (in
their sense) to do. In any case, this won’t matter much for my discussion of the Balance Account,
though the difference between these overall normative statuses will become important later.
¹¹ Compare Parfit (2011): 32; Schroeder (2015); and Lord (2018): 10–11.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Balance: you are required to A when the reasons for A outweigh the reasons
for any of the alternatives.
The intuitive thought here is that if you have more reason for A than for
anything else, then you are required to A. On this view, we imagine the
deliberating agent facing a range of alternatives. Each alternative will have
some reasons in its favor. We weigh these up and find out where the balance of
reasons lies; that is the option the agent is required to take. This picture relies
heavily on rebutting defeat, of course: the reasons for incompatible options
compete, and the weightier ones rebut the less weighty ones.¹²
This is a very natural picture, but as so far described, it faces problems.
Suppose that pressing either button 1 or button 2 will guarantee that I get a
particular $100 bill. But while pressing button 1 has no cost, pressing button 2
will deliver a painful electrical shock. Assuming these are the only relevant
factors in this case, it is clear that I am required to press button 1 and that I am
required not to press button 2. But note that in this case, my reason for
pressing button 1 is the same as my reason for pressing button 2—doing so
will get me $100. So, I do not have more reason for pressing button 1 than for
pressing button 2, and so, according to Balance, it isn’t true that I am required
to press button 1. What we need, clearly, is to add reasons against: there is a
reason against pressing button 2—the electrical shock—that is not a reason
against pressing button 1, and this explains why pressing 1 is required and why
pressing 2 is impermissible.
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
This problem may seem fairly obvious, and so you may worry that I’ve been
unfair to defenders of balance accounts. The reason this problem hasn’t been
apparent, I believe, is that defenders have (in most cases, implicitly) made the
assumption mentioned in the introduction, that reasons against an option are
just reasons for alternatives to that option. If this were true, then Balance
would take reasons against into account, since reasons against just are reasons
for alternatives. The obvious thought in this case is that if we understand the
fact that pressing button 2 would give you a shock as a reason for the
alternative to pressing 2—that is, pressing 1—then we get an extra reason
for pressing 1, and so that is what you are required to do.
An alternative, defended by Schroeder (2007), can also avoid this problem.
This is because Schroeder understands the relevant kind of competition to
be the competition between reasons for an option and reasons against that
¹² The picture will be more sophisticated than this, to accommodate undercutting defeat, as well as
what Dancy (2004) calls enablers, disablers, intensifiers, and attenuators, at least.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
know how to think about reasons for on this picture: they’re like marbles
that go into the relevant pan. It isn’t immediately obvious how to build in
reasons against, on the other hand. But, if these are just reasons for alterna-
tives, then we know how to accommodate them. Less metaphorically, this
assumption will simplify our theory of the competition between reasons, since
we only have to understand one kind of rebutting defeat: the kind that holds
between reasons for incompatible options. But, as I argue in the next section,
the assumption that reasons against an option are simply reasons for alterna-
tives to that option is false. So, we cannot appeal to this assumption to defend
the balancing account.
¹³ See Schroeder (2015) for relevant discussion, and Nagel (1970): 47, for the same assumption.
¹⁴ See Pendlebury (ms), who also makes this observation.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
the blue shirt is just to say that this fact is a reason for not wearing the blue
shirt. I believe that this answer is plausible when we think about the choice
the agent is facing as a choice about whether to wear the blue shirt—that is, a
choice between wearing the blue shirt and not wearing the blue shirt. But, in
this case, the choice is more fine-grained than this: it is a choice between
wearing the blue shirt, wearing the red shirt, and wearing the green shirt.
Reasons against wearing the blue shirt, like the fact that it is boring, are
clearly relevant for this choice, and not only the more coarse-grained choice
between wearing the blue shirt and not wearing it, since they count against
wearing the blue shirt. But, as so far stated, the view in question doesn’t
really tell us how it bears on this choice. It must count against wearing the
blue shirt by counting in favor of some alternative option(s). But not
wearing the blue shirt is not one of the options. The same kind of objection
would apply to the view that a reason against wearing the blue shirt is just a
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
reason for the disjunction of the alternatives, since the disjunction of the
alternatives is not one of the options, either.¹⁶
A second initially plausible answer is that the fact that the blue shirt is
boring is a reason for each of the alternatives. Just as a reason for an option is
or indicates something positive (in some sense) about that option, a reason
against an option is or indicates something negative about that option—
something we’d like to avoid if possible. Wearing the red shirt would let me
avoid what’s bad about wearing the blue shirt, namely its boringness. So,
would wearing the green shirt. But this view is false. Suppose that the green
shirt is even more boring than the blue shirt. Then the fact that the blue shirt is
boring—though it is a reason against wearing the blue shirt—is not a reason
for wearing the green shirt. So, reasons against an option are not simply
reasons for each of the alternatives. Moreover, this case shows that even the
weaker claim that each reason against an option is also a reason for each of the
alternatives (though reasons against may not be reducible to reasons for
alternatives) is also false.¹⁷
A third answer is that what it is to be a reason against an option is to be a
reason for some alternative to that option. The fact that the blue shirt is boring
is a reason against wearing the blue shirt. On this third view, this means that
this fact is a reason for some alternative to wearing the blue shirt. The two
alternatives are wearing the red shirt and wearing the green shirt. Plausibly a
reason for wearing the red shirt—that it’s very comfortable—is a reason
against wearing the blue shirt, if the blue shirt is not as comfortable. But the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
fact that the blue shirt is boring doesn’t seem to be a reason for any particular
alternative to wearing the blue shirt. If we suppose that the green shirt is even
more boring than the blue one, as discussed, then it is clearly not a reason for
wearing the green one. Maybe it is a reason for wearing the red one, but the
view in question seems to get the direction of explanation backwards: it isn’t
that it’s a reason against wearing the blue shirt because it’s a reason for wearing
the red one, or that its being a reason against wearing the blue one just
amounts to its being a reason for wearing the red one. Rather, it’s a reason
¹⁶ Appealing to a kind of transmission principle, according to which a reason for not doing A, or for
doing the disjunction of alternatives to A, bears on the choice at hand by being a reason for each way of
not doing A, or of doing the disjunction, makes the view equivalent to the view I discuss in the next
paragraph.
¹⁷ This brings out further objection to the view that reasons against A are reasons for the disjunction
of alternatives to A. Some of the alternatives to A may be even worse than A in the relevant respect (i.e.
the respect which explains the reason against). If so, then doing that alternative is not plausibly a way of
complying with the reason against doing A, but the view that reasons against are just reasons for the
disjunction of alternatives seems to imply that it is, since it is a way of doing that disjunction.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
for wearing the red one—if it is—because it’s a reason against wearing the
blue one.¹⁸
So, none of these three natural versions of the idea that reasons against an
option just are reasons for alternatives to that option seem promising. We
could offer more complicated answers, for example, that reasons against an
option are reasons for disjunctions of only some of the alternatives (e.g.,
leaving out the alternatives that are even worse along the relevant dimension).
But answers like this will not be very systematic, and if reasons against an
option are to be identified with or reduced to reasons for alternatives, we might
have hoped for a more systematic account. I think we do better to allow for the
existence of reasons against options that are not just reasons for any of the
alternatives to that option. These reasons against will bear on choices involv-
ing the option. There can also be reasons against an option that are not reasons
for alternatives that are even worse along the relevant dimension. Finally, we
do not need to find any particular alternative that the reasons against one
option are reasons for, in order for them to factor into the competition.
It is also good to step back and think about why we were trying to do
without a distinct category of reasons against in the first place. It is indeed hard
to see how a reason could count against an option other than by counting in
favor of alternatives to that option if we remain wedded to the balancing
metaphor. But it’s well known that this metaphor is oversimplified. Once we
move past that, there seems to be no special problem with reasons against: if
we can understand the idea that a reason for an option counts in favor of it,
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
I call the kind of view I consider in this section a criticism-based account. Both
Greenspan (2005, 2007) and Gert (2003, 2004, 2007, 2016) defend views of this
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
sort. As I’ll explain briefly, the version I describe here is closer to Greenspan’s
version, but I take the most important part of the view (for my purposes) to be
common between Greenspan and Gert.
According to criticism-based accounts, the conception of reasons and
rationality lying behind the balancing account of stacking up reasons for
various options and weighing them against each other is mistaken. Instead,
as Greenspan says, reasons are “invoked primarily to offer or answer criticism
of action,” and rationality is not about doing what your reasons most strongly
support, but rather “avoiding being subject to significant criticism or having a
response to it available.”¹⁹
Note that there are two importantly distinct functions of reasons here, and
correspondingly two distinct ways of acting rationally. First, reasons might
offer, or ground, criticism of an option. Second, reasons might answer such
criticism. You can act rationally either by not acting in ways for which you
could be criticized (or “significantly” criticized), or by having an adequate
response to such criticism. This contrasts with the balanced account, which
understands reasons to have a single underlying function, that of counting in
favor of options. Since one part of the challenge for balance accounts was to
explain two different ways in which reasons could compete, the criticism-
based account seems to be an improvement. In particular, the criticism-based
account includes a distinct role for reasons against—that of grounding (or
perhaps just serving as) criticisms.
The competition between reasons, on this view, works as follows. Reasons
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
¹⁹ Greenspan (2005): 389. ²⁰ I take this example from Gert (2007): 538.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
namely that there are cars hurtling past, that can morally justify you in staying
on the sidewalk.
So far this seems compatible with the balance account, since we could just
understand all of this in terms of the weights of the reasons involved.²¹ What is
distinctive of this view, for the purposes of this paper, is the insistence that
reasons against an option cannot be understood as reasons for alternatives to
the option (including reasons for not performing the option). The central
notion of criticism gives us a basis for a substantive distinction by understand-
ing reasons against an option as grounding criticisms of the option, and
reasons for as answering those criticisms. To criticize one option is not
necessarily to answer a criticism of another option. That the blue shirt is
boring does not necessarily answer any criticism of wearing the red shirt. In
fact, it’s plausible that criticizing one option does not even necessarily provide
any support for alternatives. At the very least, we can understand the argu-
ments against identifying reasons against with reasons for as showing that it is
far from straightforward to identify which alternative a criticism of a given
option would support, in a way that both gives plausible results (by not falsely
claiming that a criticism of A supports B when B is subject to the same kind of
criticism, for example) and explains how reasons for and against can play the
appropriate role in deliberation (by explaining how they actually bear on the
options under consideration). In addition, answering a criticism of one option
is not the same as criticizing an alternative option, so reasons for alternatives
to A cannot just be reasons against A.²² For example, that I like the way the
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
blue shirt fits may answer the criticism that it’s boring, but it does not
necessarily amount to any criticism of the alternatives. This means that the
opposite kind of reduction to that suggested on behalf of the balancing
account—reducing reasons for alternatives to A to reasons against A, instead
of reducing reasons against A to reasons for alternatives to A—won’t work,
either. Thus, on this view, we have a sharp distinction between reasons for and
reasons against.
Gert (2004, 2007) focuses on a distinction between the requiring and
justifying strengths of reasons. The mechanics of the competition between
²¹ In fact, one of both Greenspan’s and Gert’s main points is that we cannot properly handle these
kinds of cases just by appealing to some single notion of weight. In particular, we can’t explain why you
are permitted to run into the street but not required to do so.
²² Though of course sometimes it can, in a sense. Kurt Sylvan (p.c.) suggests that some criticisms of
one option, A, might be so serious that they suffice to answer or at least justify ignoring criticisms of the
alternative, B. But here it is more natural to say that the reason for B, or the answer to the criticisms B,
are that you can avoid A, or maybe just that the (only) alternative is so terrible. If so then the reason for
B is not identical with but rather derivative of the reason against A.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
reasons are similar to that previously sketched: justifying reasons (or reasons
with justifying strength) are used to justify you in performing options that you
would otherwise be required not to perform, due to the requiring reasons
against them. Gert’s examples often involve reasons against options with
requiring strength and reasons for those options with justifying strength. For
example, he holds that in the case involving the child stranded in the street, the
fact that the child is stranded in the street has significant justifying strength in
favor of running into the street—enough to justify you in doing so even in the
face of the strong requiring reason against doing so, but no requiring strength,
since it cannot generate a rational requirement to run into the street (whether
it can generate moral requirements is a different question). But in fact, the
distinction between requiring and justifying and that between reasons for and
against are cross-cutting. Nevertheless, on Gert’s view the primary competition
between reasons is always between reasons for an option and reasons against that
same option, where the two kinds of reasons are playing importantly different
roles. It’s just that sometimes, for Gert, the reasons for play the requiring role
while reasons against play the justifying role, and sometimes the reverse is true.
Thus, in focusing on the competition between reasons for an option and reasons
against it, Gert’s view is relevantly similar to the one sketched in this section,
which is most directly based on Greenspan (2005, 2007).
The criticism-based view, understood in terms of reasons for and reasons
against, gives a plausible account of the competition between reasons for an
option and reasons against it. When the reasons against an option win out, you
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
are required not to perform that option. When the reasons for an option win
out, you are permitted to perform it. Both Gert and Greenspan emphasize that
this is the most we can say here: we cannot go on to say that you are required
to perform the option, just because the reasons for the option win out over the
reasons against it. This is because both reject a maximizing conception of
rationality. All that rationality requires is not acting as you have overriding
reason against acting or avoiding unanswered criticism. So, your reasons will
generate a requirement for a particular action only by ruling out all of the
alternatives: an option is required if and only if all of the alternatives are
impermissible, while it is permissible.²³
Even granting this non-maximizing conception of rationality, this leaves
out an important element for a normative theory, namely an account of what
you ought to do. Though what you ought to do is not always properly
²³ This last clause is important if we accept the possibility of dilemmas, in which all of our options
are impermissible, and none are required.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
distinguished from what you are required to do in moral philosophy, these are
two importantly distinct overall normative statuses. What you are required to
do is plausibly very closely tied to what you can be criticized for not doing. But
what you ought to do is not necessarily tied to criticism in this way.²⁴ Think of
the way we might describe the normative status of some supererogatory
action: “You really ought to help out, though of course you don’t have to.”
On some demanding moral theories, there may not be a substantive distinc-
tion between what you ought to do and what you are required to do, but
I think it is clear that there is at least a conceptual distinction.²⁵
The criticism-based view has trouble accounting for what you ought to do,
in this sense.²⁶ This is because ought is closely connected to what’s best, in the
sense of being the most strongly supported by reasons.²⁷ ‘Best’ and ‘most’ are
comparative terms: you ought to do something when it’s better or more
strongly supported by reasons than the alternatives. The account of the com-
petition between reasons sketched so far on behalf of the criticism-based view
is focused on the competition between the reasons for an option and the
reasons against it. We don’t yet have any way of comparing different alterna-
tives in terms of how well supported they are by the reasons.²⁸
The balance account focuses on the competition between reasons for
competing options but does not have a natural way to account for the
competition between reasons for an option and reasons against it. The
criticism-based account focuses on the latter kind of competition but does
not have a natural way to account for the former. The latter’s trouble account-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
ing for what we ought to do, as opposed to what we’re required to do, is a
symptom of this problem.
Suppose that reasons for one option are also reasons against alternatives,
even if we don’t identify reasons against with reasons for alternatives. Then we
have some sort of relationship between the reasons bearing on different
alternatives. But even if this is true, such that reasons for alternatives to an
²⁴ Perhaps some sort of criticism is appropriate in some cases in which you don’t do what you ought
to do, even if you aren’t required to do it; see, e.g., Driver (1992); Macnamara (2013). But the criticism
will at least typically be weaker and not call for apologies, reparations, punishment, and so on.
²⁵ For a sampling of work on this issue, see McNamara (1996a, b); Bedke (2011); Snedegar (2016).
²⁶ Bedke (2011) also argues that Gert’s view fails to explain ought as opposed to requirement, though
in a different way. See also Snedegar (2016).
²⁷ For early work on the connection between ‘ought’ and ‘best,’ see Sloman (1970); Finlay (2010,
2014) also emphasizes the connection between ‘ought’ and ‘best.’
²⁸ Perhaps using ‘ought’ to describe the best action is sometimes too strong, or at least can sound
wrong, for example when the best action is only best by some trivial amount. The objection to the
criticism-based view is essentially just that it leaves us no way to make comparisons between the
options in terms of how strongly supported those options are. Focusing on the overall status expressed
by ‘ought’ is a natural way to bring this out. Thanks to Kurt Sylvan here.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
option are factored into the competition between reasons for that option and
reasons against it, this still doesn’t yet solve the main issue raised previously.
The problem is to explain how to compare competing alternatives in terms of
how well they are supported by the reasons. Without this, we won’t be able to
explain how your reasons determine what you ought to do, in addition to what
you must and may do. The reasons bearing on alternatives get factored into
the competition between reasons for and reasons against a given option; but
the outcome of this competition determines what you must or may do. So, we
still don’t have an answer to the question of how your reasons determine what
you ought to do. In the next section, I sketch a positive account that draws on
aspects of both the balance account and the criticism-based account, and argue
that it gives us a more complete picture of how the competition between
reasons determines the overall normative status of our options. But I’ll also
show that open questions and problems remain.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
only reasons for A, then the combined weight of the reasons for A is simply the
weight of R1 plus the weight of R2, or that if R3 is a reason for A and R4 is a
reason against A, and these are the only reasons in play, then the overall
weight of reasons for A is simply the weight of R3 minus the weight of R4.
Nevertheless, for convenience I will use ‘+’ for the combining weights oper-
ation and ‘ ’ for the difference in weights operation. Issues involving the more
complicated ways that reasons actually seem to combine and weigh against
each other will have to be left aside here.³¹
The first kind of competition is that between reasons for an option and
reasons against it. What we need for this is the combined weight of all the
reasons for A, W(RFA), and the combined weight of all the reasons against A,
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
will get you a particular $100 bill but pressing 2 will also deliver a mild electric
shock. You would be willing to endure this shock to get $100, so it seems that
the reason for pressing 2, that it will get you the $100 bill, is weightier than the
reason against, that you will be shocked. So according to the account, pressing
button 2 is permissible. But this is incorrect: pressing button 2 in this case
would be foolish.
I have tried to explain permissibility (and, shortly, requirement) just in
terms of the reasons for and against the particular option, whereas what you
ought to do will be explained in terms of comparisons among the alternatives.
But the lesson of this case is that even for permissibility and requirement, the
alternatives matter: it is only because of the availability of pressing button 1
that pressing button 2 is impermissible. The task for now is to say exactly how
it is that the alternatives matter for permissibility and requirement.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
One way in which the alternatives can matter is by the reasons bearing on
those alternatives transmitting to the option in question. Sometimes reasons
for one alternative are also reasons against incompatible alternatives. This is
one way to explain the notion of an opportunity cost: among the costs of an
option are the good features of the alternatives that you’ll miss out on. This is
consistent with denying that all reasons against options are nothing over and
above reasons for alternatives. But in this case, given that you will get the $100
either way, the reason for pressing button 1—that it will get you $100—is not a
reason against pressing button 2.
Rather, I think that in cases like this, the alternatives matter by helping to
determine which considerations are or are not reasons for the option in
question. In particular, since you will get the $100 either way, the fact that
you’ll get the $100 by pressing button 2 is not a reason for pressing button 2.
Hence, there is only a reason against pressing button 2 (the shock), and so it is
impermissible.
The surprising consequence of this view is that the $100 also does not give
you reason to press button 1. So, if the money and the shock are the only
relevant considerations, there won’t be any reason to press button 1.
Nevertheless, pressing button 1 will be permissible, since there’s at least as
much reason for it as against it (namely, no reason either way). In fact it will be
required, since the only alternative is impermissible.³⁴ Though it may seem
initially surprising to deny that the fact that you can get $100 by pressing
button 1 (or button 2, for that matter) is a reason for pressing button 1, I think
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
this is plausible. The crucial point is that you’ll get the $100 no matter what
you do (assuming pressing 1 and pressing 2 are the only options). So, you
cannot, for example, justify yourself in pressing button 1 (instead of button 2)
by pointing out that you’ll get $100. Also, it would be odd to base your decision
to press button 1 (instead of button 2) on the fact that you will get $100 by
pressing button 1. Being able to serve as justifiers or at least sensible bases for
decision are commonly thought to be earmarks, if not conceptual require-
ments, for some consideration’s being a reason.³⁵ So it should not seem so
³⁴ In fact, things will be a bit trickier since it seems that we can make true reasons ascriptions like,
“That you will get shocked by pressing button 2 is a reason to press button 1, instead.” I think that this
is true, but that the reason ascribed is a derivative reason—derivative on the reason against pressing 2.
A full theory here will need to explain when we get these derivative reasons and how their weights do or
do not contribute to the overall status of the options in question. It would seem to be a mistake, for
example, to treat this derivative reason for pressing button 1 as having weight that is independent of
and in addition to the weight of the reason against pressing button 2 on which it is derivative. Compare
Maguire and Snedegar (forthcoming).
³⁵ See, e.g., Schroeder (2007): Ch. 2; Setiya (2014); Gregory (2016); Silverstein (2016); Way (2017);
Snedegar (2019).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
surprising that in this case, that you can get $100 by pressing button 1 will not
be a reason to press button 1, since you’ll get the $100 no matter what you do.
If this kind of solution is on the right track, and can be generalized, then we
have the start of an account that tells us when an option is permissible in terms
of the reasons for and against it. We also get an account of when an option is
required: an option is required when all of the other options are
impermissible—that is, when the reasons against each of the other options
outweigh the reasons for them. If more than one option is permissible, then no
single option is required.³⁶ There are clear similarities between this under-
standing of permissibility and requirement and the one we get from the
criticism-based account.
But, as I argued in the previous section, this leaves out an important part of
the normative domain—namely, what we ought to do, as opposed to what
we’re required to do. To quickly review the main points from the previous
section: we ought to do is very plausibly what’s best, in some sense.³⁷ But, to
say that some option is the best is to say that it’s better than the alternatives.
So, to account for what we ought to do, we need to introduce a comparison
among the alternatives.
The overall weight of the reasons bearing on an option, A, which I called W
(RA), is determined by the difference between the weight of the set of reasons
for A and the set of reasons against A. So, we can compare these weights of
overall reason for each of the options. What we ought to do is the option that
ranks at the top of this comparison: you ought to do A when W(RA) > W(RB),
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
for all the alternatives, B. Whereas the competition between reasons for an
option and reasons against that option are most directly relevant for deter-
mining which options are permissible and required, the competition between
the reasons bearing on competing alternatives is most directly relevant for
determining what you ought to do.³⁸
This picture has some attractive features. First, it explains important overall
or verdictive normative statuses—permissibility, impermissibility, requirement,
³⁶ An open question at this point is whether it is possible for the reasons against to win out for every
option, making every option impermissible. Many people think that such tragic dilemmas are impos-
sible. We could, for example, argue that the fact that an option is the “least bad” is always a decisive
reason to do it.
³⁷ Cf. Sloman (1970); Finlay (2010); von Fintel and Iatridou (2008).
³⁸ As noted in the Introduction, I am focusing on the moral case. In the epistemic domain, it is much
less clear that there is a substantial difference between ought and requirement. One possible explan-
ation for this, which I cannot explore at any length here, is that in the epistemic domain, unlike in the
moral domain, reasons for believing p, or evidence for p, are always reasons against believing, or
evidence against, propositions that are incompatible with p. If so, then it would plausibly turn out that
any time the reasons for believing p outweighed the reasons against believing p, the reasons against
believing any incompatible alternative q would outweigh the reasons for believing q.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
and ought—in terms of pro tanto reasons. This is a popular and attractive idea,
but it has proved challenging to provide satisfactory accounts of all of these
notions in terms of reasons.³⁹ Second, the picture captures the correct logical
relationships between permissibility, requirement, and ought. To focus on the
relationship between requirement and ought: if some option A is required, then
it is also what you ought to do. Since A is required, the reasons against each
alternative B outweigh the reasons for B; thus W(RB) will be negative for all the
alternatives, B. Moreover, since A is required, it is the only permissible option, in
which case it is the only option for which W(RA) is positive. So, W(RA) > W(RB)
for each alternative B, and thus A is what you ought to do. But A might be what
you ought to do without being required, since it is compatible with W(RA) > W
(RB) that W(RB) is nevertheless positive—that is, B is permissible. This is just
what we want, because it has been widely noted in both linguistics and philoso-
phy (the former focusing on the modals that express these normative statuses
and the latter focusing on the statuses themselves) that there is a one-way
entailment from requirement to ought.⁴⁰ This captures the commonsense idea
that there are things that you really ought to do, though you are not required to
do so, and are not acting impermissibly if you don’t—think of paradigm cases of
supererogation, for example.
This is just a sketch, and open questions remain, but I do think it gives us
the beginnings of an attractive account. As I have already noted, I don’t take
this view to be any radical departure from an orthodox way of thinking about
how reasons compete. Rather, what is important about it is the incorporation
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
of the two different kinds of competition between reasons. Both the standard
balance account and the criticism-based account are incomplete because they
focus on just one kind of competition.
References
Bader, R. (2016). Conditions, modifiers, and holism. In: Lord, E. and Maguire, B.
(eds.) Weighing Reasons. Oxford: Oxford University Press: 27–55.
Bedke, M. (2011). Passing the deontic buck. In: Shafer-Landau, R. (ed.) Oxford
Studies in Metaethics. Vol. 6. Oxford: Oxford University Press: 128–52.
Berker, S. (2007). Particular reasons. Ethics, 118(1): 109–39.
³⁹ See Snedegar (2016) for arguments against some recent attempts, including Bedke (2011).
⁴⁰ See, for example, Wertheimer (1972); McNamara (1996b); von Fintel and Iatridou (2008);
Portner (2009); Snedegar (2012, 2016).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Broome, J. (2004). Reasons. In: Wallace, Scheffler, and Smith (eds.) Reason and
Value: Themes from the Moral Philosophy of Joseph Raz. Oxford: Oxford
University Press: 28–55.
Dancy, J. (2004). Ethics Without Principles. Oxford: Oxford University Press.
Driver, J. (1992). The suberogatory. Australasian Journal of Philosophy, 70(3):
286–95.
Finlay, S. (2010). What ‘ought’ probably means, and why you can’t detach it.
Synthese, 177: 67–89.
Finlay, S. (2014). Confusion of Tongues. Oxford: Oxford University Press.
von Fintel, K. and Iatridou, S. (2008). How to say ‘ought’ in foreign: The com-
position of weak necessity modals. In: Guéron and Lecarme, (eds.) Time and
Modality. Dordrecht: Springer: 115–41.
Gert, J. (2003). Requiring and justifying: Two dimensions of normative strength.
Erkenntis, 59: 5–36.
Gert, J. (2004). Brute Rationality: Normativity and Human Action. Cambridge:
Cambridge University Press.
Gert, J. (2007). Normative strength and the balance of reasons. Philosophical
Review, 116(4).
Gert, J. (2016). The distinction between requiring and justifying: nothing to fear.
In: Lord, E. and Maguire, B. (eds.) Weighing Reasons. Oxford: Oxford
University Press: 157–72.
Greenspan, P. (2005). Asymmetrical practical reasons. In: Reicher, M. and Marek,
J. (eds.) Experience and Analysis: Proceedings of the 27th International
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Raz, J. (1999). Practical Reason and Norms. Oxford: Oxford University Press.
Ross, W. D. (1930). The Right and the Good. Oxford: Oxford University Press.
Schroeder, M. (2007). Slaves of the Passions. Oxford: Oxford University Press.
Schroeder, M. (2015). What makes reasons sufficient? American Philosophical
Quarterly, 52(2): 159–70.
Setiya, K. (2014). What makes reasons sufficient? Philosophical Studies, 167:
221–35.
Sher, I. (2019). Comparative value and the weight of reasons. Economics and
Philosophy, 35(1): 103–58.
Silverstein, M. (2016). Reducing reasons. Journal of Ethics and Social Philosophy,
10(1): 1–22.
Sloman, A. (1970). “Ought” and “better.” Mind, 79(315): 385–94.
Snedegar, J. (2012). Contrastive semantics for deontic modals. In: Blaauw, M. (ed.)
Contrastivism in Philosophy. New York: Routledge: 166–33.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
12
Perceptual Reasons and Defeat
Mark Schroeder
as reasons to believe, and conversely. ‘Reason’ and ‘evidence’ talk also exhibit
parallel distinctions between objective talk about reasons or evidence that are
out there, to be discovered, and subjective talk about the reasons or evidence
that someone has, or from her perspective. It is also possible to do or to believe
things for reasons, and similarly, it is possible to believe things on the basis of
evidence. These facts, which could be enumerated at much greater length, all
constitute strong evidence that reasons and evidence are closely related.
From an ontological perspective, many kinds of thing can be cited appro-
priately as reasons. For example, the height of the Empire State Building is a
reason not to jump off, Robin Jeshion is a reason to study philosophy of
language at the University of Southern California (USC), and the bruise on
Alyssa’s leg is reason to believe that she has gotten hurt. Heights, people, and
bruises all belong, I take it, to different categories. But it is also true that one
reason not to jump off of the Empire State Building—and not, intuitively, a
¹ Schroeder (2021).
Mark Schroeder, Perceptual Reasons and Defeat In: Reasons, Justification, and Defeat. Edited by: Jessica Brown and Mona
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
98847205.003.0012
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
In Objective Defeat, Roberta cannot know that there is something red in front
of her unless she has some independent source of evidence that that is so. Even
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
if she does not realize that she is wearing rose-colored glasses, since they make
everything look red, they are incompatible with Roberta having the discrim-
inatory capacity to know that there really is something red in front of her, even
if there is. In Subjective Defeat, it seems on the face of it that Roberta cannot
reasonably believe that there is something red in front of her, since she
rationally believes herself to be in Objective Defeat, and in that case she cannot
know. But it seems that knowledge requires reasonable belief, so if it is not
reasonable for her to believe it, then it seems that she cannot know it, either—
again, unless she has some independent source of evidence, a qualification
which I will henceforth ignore.
There are many pairs of cases that stand to one another as Objective Defeat
and Subjective Defeat stand to one another, in the first of which some worldly
fact is what stands in the way of a subject knowing, and in the second of which
the subject’s rational belief in that same worldly fact is what stands in the way
of her knowing. Least interestingly, there is the pair in which there is in fact
nothing red in front of Roberta, and in which she rationally believes that there
is nothing red in front of her—exhibiting the plausibility of the truth and
rational belief conditions on knowledge. Some pairs consist in the availability
of compelling counterevidence and Roberta’s rational belief in the compelling
counterevidence. According to pragmatic encroachers, there are also pairs in
which Roberta faces high stakes and in which she rationally believes herself to
face high stakes. And many pairs are like the one that I have exhibited—cases
in which the force of Roberta’s evidence appear to be undercut by some further
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
fact. Though all of these pairs are possible, the views with which I will be
concerned in this paper are all accounts of the paradigmatic kind of defeas-
ibility typically referred to as undercutting defeat—cases in which the strength
or force of one’s evidence has been called into question.
I want to be a little bit careful about how I describe the data about these two
cases, since some of the views that I go on to discuss will disagree about how to
describe them. My own view is that both cases can and should be taken at face
value—that there are two dimensions along which knowledge can be defeated,
both by the facts, independently of what the subject believes about them, and
by the subject’s other beliefs. I will call these two dimensions objective defeat and
subjective defeat, respectively. I also believe that these two dimensions are
independent—cases of objective defeat without subjective defeat are (typically²)
recipes for Gettier cases, and cases of subjective defeat without objective defeat
undermine the justification or, as I have been putting it in this paper, the
² Failures of the truth condition are not, of course, recipes for Gettier cases.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
reasonability condition on knowledge. But as we will see later, some theorists are
going to struggle to accommodate subjective defeat. They may still accept that
there is an appearance of defeat of knowledge in cases of subjective defeat, but
explain this appearance in some other way, because something else nearby,
but somewhat different, is true.
According to what I will call the classical theory of the defeasibility of percep-
tual evidence, it can be properly subsumed to the non-monotonicity of
inference. Non-monotonic inferences are cases where R is a reasonable con-
clusion to draw from P, but not a reasonable conclusion to draw from
P&Q. For example, if you know that Tweety is a bird, it is reasonable to
draw the conclusion that Tweety probably flies, but if you also know that
Tweety is a penguin, then this is not a reasonable conclusion to draw.
Non-monotonicity in inference is possible whenever inferences are rea-
sonable without being entailed. So according to the classical theory, the
evidence about the external world that you get when you have a visual
experience as of something red in front of you does not entail anything
about the external world. It is, for example, that it appears to you that there is
something red in front of you, or that you have a visual experience as of
something red in front of you, that you are appeared to redly, that there is a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
red sense-datum, or the like. There are many ways of developing the classical
theory by filling in a more detailed account of the nature of perceptual
evidence, according to the classical theory, but for purposes of this paper,
they share a package of virtues and vices.
The classical theory is, in the first instance, a theory about subjective defeat.
This is because non-monotonicity is a property of reasonable inference, and
only subjective defeat concerns reasonable inference, at least directly. The key
idea of the classical theory is that it is reasonable to infer that there is
something red in front of you from the evidence that it appears that there is
something red in front of you, but not from the evidence that you are wearing
rose-colored glasses and it appears that there is something red in front of you.
But the classical theory can also be extended to offer a treatment of objective
defeat. This was essentially the project of defeasibility analyses of knowledge,
such as those offered by Klein (1970); Lehrer and Paxson (1969); Annis (1973);
Ackerman (1974); Olin (1976); Levy (1977). What all of these theorists noted
was that there seems to be a general pattern whereby knowledge has to be not
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
only reasonable belief, but belief whose reasonability “stands up” to the facts,
in some way.³ The project explored in their papers was how to say more
precisely and in a way free from counterexamples exactly how knowledge must
be belief whose reasonability “stands up” to the facts, but all of their accounts
agree that this is what goes wrong in cases of objective defeat like Roberta’s
that are not also cases of subjective defeat. In such cases, Roberta may
reasonably believe that there is something red in front of her, and this belief
may be reasonable because she does not realize that she wearing rose-colored
glasses, but it is only reasonable because she is not aware of this. If you add this
proposition to her stock of beliefs, it would no longer be reasonable for her to
believe that there is something red in front of her, and that is why, on this view,
its truth defeats her knowledge.
The strength of the classical theory is the elegance with respect to which it
subsumes the defeasibility of basic perceptual evidence under the defeasibility
of evidence elsewhere, by incorporating it into a general account of non-
monotonic inference. Indeed, the literature on nonmonotonic inference in
general is replete with examples that take for granted that the classical theory is
correct about the defeasibility of perceptual evidence. Its home case is the case
of subjective defeat, and although defeasibility theorists in the 1970s famously
discovered that it was hard to pin down exactly what the “standing up to the
facts” condition on reasonable belief is that is required for knowledge, so it is
intelligible to worry that they will not actually be able to adequately capture all
and only the correct cases of objective defeat, classical theorists at least have a
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
However, the weakness of the classical theory has been regarded by many to be
the key assumption about the nature of perceptual evidence that is required in
order for it to subsume perceptual defeasibility to the monotonicity of infer-
ence more generally. And this is the assumption that perceptual evidence does
not entail anything about the external world. This assumption, in essence, set
all of the central problems of epistemology in the twentieth century, and
shaped the space of possible options.
It is the assumption that perceptual evidence does not entail anything about
the external world from which it follows that there are skeptical scenarios that
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
tual evidence to beliefs about the external world, even though these conclusions
are not entailed. This is what dogmatists claim. But in order to avoid skepticism,
dogmatists must claim that inferences from perceptual evidence to conclusions
about the external world are more reasonable than inferences to conclusions
about the intentions of the evil demon, or about the locations of ones and zeros
on the hard-drive of the matrix. Since nothing in the content of the perceptual
evidence breaks this symmetry, something else must be appealed to in order to
explain the break in symmetry, and this is where epistemological externalism
originally came in, in the late 1960s and early 1970s—it looked like every
non-skeptical, non-coherentist, non-rationalist view had to give some causal,
subjunctive, reliabilist, or otherwise fundamentally externalist answer to what
justifies moving from basic perceptual evidence to external world beliefs over
alternative conclusions. This was, as Armstrong (1973) argued for these reasons,
inevitable. And so, externalists argued, everyone needed to accept the adequacy
of such externalist explanations, which obviated the need to explain knowledge
or the reasonability of belief in terms of evidence at all.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Since the classical view both requires and is enabled by the assumption that
perceptual evidence does not entail anything about the external world, its
alternatives adopt the assumption that perceptual evidence does entail things
about the external world. They assume, as I will put it, that perceptual evidence
is world-implicating. But world-implicating views disagree amongst them-
selves, as we will see, about exactly what sort of content perceptual evidence
is, and also about what sort of condition a subject must satisfy, in order to
possess this sort of evidence.
Recall that classical theorists believe that perceptual evidence does not entail
anything about the external world. But classical theorists also believe that being in
possession of some particular piece of perceptual evidence does not entail anything
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
about the external world. If your perceptual evidence that there is something red in
front of you is that it appears to you that there is something red in front of you, then
you count as having this perceptual evidence in virtue of it’s being true that it
appears to you that there is something red in front of you. Or perhaps you count as
having this evidence in virtue of believing that it appears to you that there is
something red in front of you. Either way, not only does the content of your
evidence not entail anything about the external world, the condition of your having
that evidence does not entail anything about the external world, either.
The first class of world-implicating views that I want to consider disagrees
with the classical theory on both of these fronts. These views claim not only
that the content of your perceptual evidence is world-implicating, but that the
relationship that you must stand in to that evidence in order for it to be
evidence that can make beliefs reasonable for you is world-implicating as well.
Let us call these views doubly world-implicating.
One example of a doubly world-implicating view of perceptual evidence
comes from Williamson (2000). Williamson holds that your evidence is what
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
you know (E = K), that knowledge is the most general factive stative attitude,
and that seeing that there is something red in front of you is a factive stative
attitude. So it follows that Williamson holds that when you see that there is
something red in front of you, that makes available to you as part of your
evidence the fact that there is something red in front of you. The content of
your evidence is a proposition about the world outside of your head, so this
view is world-implicating along the content dimension. But the condition of
this being your evidence is also world-implicating—it is that you see that this is
the case. And seeing that something the case is a factive relationship to the
world—a relationship that only holds if there really is something red in front
of you. Let us call this view the factive content view.
Another example of a doubly world-implicating view comes from an
interpretation of some of John McDowell’s views offered by Comesaña and
McGrath (2015). According to this view, which I’ll call the factive attitude
view, in normal cases your evidence that there is something red in front of you
is that you see that there is something red. And you come to have this reason
because it is true—you really do see that there is something red in front of you.
The factive attitude view shares with the factive content view a theory about
the conditions under which you have perceptual evidence about the world but
disagrees about what that evidence is. But like the factive content view, it holds
that both the content of your evidence and the condition of your having that
evidence entail something about the external world. Both entail something
about the external world because seeing that is a factive relation, and so you
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
cannot see that there is something red in front of you unless there really is
something red in front of you. So, it is doubly world-implicating.
Doubly world-implicating views avoid the consequences of the classical
theory explored in Section 12.2.2. But they do so by undermining the classical
theory’s elegant subsumption of the defeasibility of basic perceptual evidence
to the nonmonotonicity of inference more generally. The inference from ‘there
is something red in front of me’ to ‘there is something red in front of me’ is not
nonmonotonic, and neither is the inference from ‘I see that there is something
red in front of me’ to ‘there is something red in front of me.’ This non-
monotonicity is precisely what avoids the gap between perceptual evidence and
conclusions about the external world that the skeptic exploits and that the
coherentist, rationalist, and externalist all try to close, so there is no avoiding
the conclusion that avoiding the consequences of the classical theory also means
giving up on its elegant treatment of the defeasibility of perceptual evidence.
But doubly world-implicating views—at least, both of the ones that I’ve
described here—come with an alternative treatment of perceptual defeasibility.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
And that is because someone who is wearing rose-colored glasses cannot see
that something is red. Seeing that something is red does not only entail that it
really is red—it is also incompatible with having insufficient perceptual
discriminability to discriminate red from other colors. Since rose-colored
glasses interfere with perceptual discriminability, wearing them is incompat-
ible with seeing that something is red. So, if you are wearing rose-colored
glasses, you do not know that something is red, just by having a visual
experience as of it being red.⁴
Recall that the classical theory is most at home in its treatment of subjective
defeat. That is because what it explains, in the first instance, is why certain
inferences are unreasonable, in the presence of further background beliefs. In
contrast, we can now see that doubly world-implicating views are most at
home in their treatment of objective defeat. That is because these views impose
an objective condition on the subject’s relationship to the world, in order for
her to have a reason at all—that she sees that something is the case. And seeing
that something is red is undermined by wearing rose-colored glasses.
But subjective defeat involves background beliefs that might (at least, when
subjective defeat does not overlap with objective defeat) be false. Having a
background belief that you are wearing rose-colored glasses does not prevent
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
your vision from actually working properly. So, it does not prevent you from
seeing that there is something red in front of you. So, whereas the doubly
world-implicating views get an immediate explanation of why cases of object-
ive defeat are cases in which the subject lacks reasons for her belief (and hence
don’t know), they have no such obvious explanation for why cases of subject-
ive defeat are cases in which agents lack reasons for their beliefs.
⁴ It is worth considering how this strategy might be extended to fake barn cases, which bear many
similarities to paradigmatic cases of undercutting defeat but are slightly more contentious. Fake barn
cases are more contentious because some hold that visual knowledge of a barn is possible even in fake
barn country. But the strategy of doubly world-implicating views can be extended to fake barn cases so
long as either fake barn country undermines both knowing that and seeing that, or it undermines
neither. Fake barn country cases could be an obstacle to this strategy if fake barn country undermines
knowing that without undermining seeing that—a possibility that is opened if we reject, as I do
(Schroeder (Manuscript, ch. 4)), the claim that seeing that entails knowing that. But even if we deny
that seeing that entails knowing that I still believe that we should either accept fake barn country cases
as undermining both, or as undermining neither. The reasons that seeing that does not entail knowing
that are that seeing that something is the case does not entail either belief or reasonable belief that it is
the case, but knowledge does, and this cross-cuts considerations about fake barn cases. Compare
Prichard (2012).
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
whether the subject knows, but also with respect to whether, and why, the
subject’s belief is reasonable.
Now, it is open to doubly world-relative accounts to give an alternative
account of the evidence that is available to a subject who suffers from objective
defeat, in order to explain why her belief is reasonable after all, even though
she lacks the evidence that is possessed by someone in the good case.
A theorist who goes this way is an epistemological disjunctivist—that is, she
tells different stories about the reasonability of perceptual beliefs in the good
and bad cases. Another way of being an epistemological disjunctivist is to deny
that perceptual beliefs are reasonable in the bad case of objective defeat at all
and explain why they seem reasonable by pointing out that the subject does not
know that they are not reasonable.
So doubly world-relative views of the defeasibility of perceptual evidence
have twin vices—they have to explain why knowledge seems to be defeated in
subjective defeat cases, and they have to explain why reasonability seems not to
be defeated in objective defeat cases, and if they do try to explain why beliefs
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
The classical theory and doubly world-implicating views disagree about two
things. They disagree about whether the content of perceptual evidence can be
world-implicating, and they disagree about whether the condition that a
subject must satisfy, in order to have perceptual evidence, can be world-
implicating. But the features of their view that doubly world-implicating
views cite as their advantage over the classical theory derive from their view
about the content of perceptual evidence, and the features that cast their
treatment of defeasibility into doubt derive from their view about the condition
requires to possess that evidence. The idea behind singly world-implicating
views is to split this difference by adopting the view that the contents of
perceptual evidence are world-implicating but denying that the condition
required in order to possess this evidence is world-implicating.
Let’s take these points one at a time. To see that the advantages of the
doubly world-implicating views turn on what they say about the content of
perceptual evidence, it suffices to recall that the problem with the classical view
was that it required the assumption that skeptical hypotheses can be compat-
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
ible with the totality of your perceptual evidence. It is necessary and sufficient
to face this problem that the content of perceptual evidence—the proposition
such that your perceptual evidence can be accurately reported as being that
p—entails something about the external world.
In contrast, the problems for the doubly world-implicating views turn on
what they say about the condition of having perceptual evidence. Since they
say that this condition entails something about the external world, it follows
that this condition cannot be shared across good and bad cases, and hence that
the story about why belief is reasonable must be different across the good and
bad cases, since the same evidence is not available to justify it. In contrast, if
the condition under which a thinker counts as having a certain piece of
evidence entails nothing about the external world, then it will cross-cut the
good case/bad case distinction, and so we will be able to tell a non-disjunctive
story about why thinkers are reasonable to believe in the way that they do.
So that is what motivates singly world-implicating views. But there is one major
reason why philosophers have not explored singly world-relative views. And that
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
is that it is common to assume that evidence must be true. If evidence must be true,
then singly world-implicating views are impossible. For, if evidence entails some-
thing about the external world, and having that evidence entails that it is true, then
the condition of having the evidence must entail something about the external
world as well. So, proponents of singly world-implicating views must reject the
view that evidence (or for that matter, reasons) must be true.
This can sound bizarre, but it is not such an unnatural view. We may
distinguish between objective and subjective reasons, and correspondingly
between objective and subjective evidence. Objective reasons count in favor
of what it is advisable to believe, but subjective reasons count only in favor of
what it is reasonable to believe. Objective reasons must be true, because what it
is advisable to believe (or do) depends on what is true, but subjective reasons
need only be believed (perhaps rationally), because what it is reasonable to
believe (or do) depends only on what you (rationally) believe.
the condition under which this is your evidence is that you see that there is
something red in front of you. If we relax the condition that the perceptual
condition for having this evidence must be a factive perceptual relation, we are
led to the non-factive content view, according to which your perceptual
evidence is the same—namely, that there is something red in front of you—
but the condition under which this counts as your evidence is that you have a
visual experience as of something red in front of you.
Similarly, recall that according to the factive attitude view, your perceptual
evidence when you see that there is something red in front of you is that you
see that there is something red in front of you, and the condition under which
this is your evidence is that it is true. Again, we can relax the world-implicating
condition of truth that this view requires in order for this to be part of your
evidence, and replace it with some other, more subjective condition, such as
that it seems true to you. Let us call this the non-factive attitude view.
Of all of the views that have been discussed here, the non-factive content
view offers by far the least promising account of the defeasibility of perceptual
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
evidence. Let us take the case of objective defeat first. In this case, you have a
visual experience as of something red in front of you, but unbeknownst to you,
you are wearing rose-colored glasses. According to the non-factive content
view, you have available as evidence the proposition that there is something
red in front of you. But this evidence is both available to you, and true. So, it is
hard to see why it can’t ground knowledge that there is something red in front
of you, especially because you have no contrary evidence. But it shouldn’t be
possible to know, on the basis of visual evidence, that there is something red in
front of you, if you are wearing rose-colored glasses. The non-factive content
view has no obvious answer to this problem. It loses the elegant answer that the
doubly world-relative views gave to this problem, because it relaxes the
constraint on having evidence too far.
The non-factive content view also fails to adequately treat cases of subjective
defeat. In this case, you have a visual experience as of something red in front of
you, but you also realize that you are wearing rose-colored glasses. According
to the non-factive content view, as with the doubly world-implicating views,
this latter belief does not interfere in any way with your satisfying the
condition required to have world-implicating evidence—in this case, your
evidence being that there is something red in front of you. The doubly
world-implicating views, as we saw, might attempt to try to explain why it is
nevertheless not rational to believe that there is something red in front of you
in a way that was derivative from their treatment of objective defeasibility. But
as we’ve just seen, the non-factive content view does not have a successful
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
treatment of objective defeasibility. And so, it can’t borrow this strategy either.
In contrast, the non-factive attitude view offers elegant, parallel, treatments
of both objective defeasibilty and subjective defeasibility. It is able to do this, in
effect, because it borrows more from the doubly world-implicating views—it
takes on board their idea that factive perceptual relationships to the environ-
ment play an important role in basic perceptual evidence. The fact that you are
wearing rose-colored glasses defeats your visual evidence that there is some-
thing red in front of you, on this view, because it cannot be true that you see
that there is something red in front of you if you are wearing rose-colored
glasses. This suffices to defeat your knowledge, because knowledge cannot be
based on false lemmas. But it doesn’t defeat the reasonability of your belief,
because reasonability of belief is not defeated by false lemmas. So, this is
exactly the right kind of explanation of the objective defeat to avoid leading
to epistemological disjunctivism.
The non-factive attitude view also offers an elegant treatment of subjective
defeat. In this case, you believe that you are wearing rose-colored glasses, but
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
part of your evidence about the world is that you see that there is something
red in front of you. But as we’ve been reminded again and again throughout
this paper, these claims cannot both be true. So, either you should give up your
belief that you are wearing rose-colored glasses, or you shouldn’t infer any-
thing from the proposition that you see that there is something red in front of
you. So either your belief that you are wearing rose-colored glasses is not
rational, or, if it is, it is not reasonable to infer anything from this particular
piece of perceptual evidence, and hence it is defeated as a sort of reasonable
belief about the world.
The non-factive attitude view carries other, substantive, commitments,
however. Of the views that I have surveyed, here, it is the only view which is
committed to the possibility that creatures who can have perceptual evidence
about their environments must also be able to represent themselves as seeing
that something is the case. This can seem like a particularly stringent kind of
cognitive demand. But perhaps it is not so implausible. Imagine seeming to see
your friend Bob in front of you, and then learning that what you are in fact
seeing is a hologram of Bob that is triggered by Bob standing in front of you. If
this sounds like a case of an illusion, then we must say what about your
experience makes it illusory. What makes it illusory, a proponent of the
non-factive attitude view may claim, is that it represents itself as a factive
relation to your environment.⁵
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
12.5. Summing-Up
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
References
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 22/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Index
For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on
occasion, appear on only one of those pages.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
286
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
287
inquiry 35, 112–13, 133n.28, 136n.29, 223n.1, 225n.5, 237n.28, 247n.1, 250n.11
137–9, 141 Lord and Maguire 247n.1
instrumentalism 110–13 Lord and Sylvan 118, 125n.16
intention 13–14, chapter 8 passim, 215, 274 Lynch 226n.9
internalism 1, 9–11, 56, 58, 74n.3, 78–9, Lyons 3–4, 11–12, 39, 45n.15, 57n.35,
89–90, 94–5, 162–3, 173, 208 60n.40, 62, 146n.1, 149n.9, 150,
151n.10, 174n.36
James 138
Johnson, Hashtroudi, and Lindsay 76–7 Macnamara 259n.24
Joyce 154n.13 Maguire 261n.30
justification Maguire and Snedegar 261n.30
doxastic 42, 59n.38, 61–2, 64, 203 Makinson 229n.15
inferential versus noninferential 13, 202, Markovits 118
210–21 Matheson 39n.1, 48n.26
perceptual 204–5, 214n.18, 218–20 Matilal 72, 89–90
propositional 42, 51, 53, 58, 59n.38, 61–4 McNamara 259n.25, 265n.39
testimonial 13, 202, 217, 220–1 McGinn 225n.5
McGrath 201, 220n.22
Kelly 5–7, 121n.6, 122n.7 McKinnon 21n.8
Kelp and Simion 1 Mele and Moser 179, 186, 188
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
288
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
289
anti-reductionism 2, 20
reductionism 2, 20 Ye 26nn.14,15, 85n.8
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.
OUP CORRECTED AUTOPAGE PROOFS – FINAL, 19/2/2021, SPi
Copyright © 2021. Oxford University Press USA - OSO. All rights reserved.
Reasons, Justification, and Defeat, edited by Jessica Brown, and Mona Simion, Oxford University Press USA - OSO, 2021.