Littlejohn Justification Manuscript

Chapter 1
Introduction
1
Chapter 2
The Orthodox Views
This chapter will build on two previous published papers. In the first, “The
Externalist’s Demon”, I explain how externalists can deal with the new evil de-
mon objection. In the second, “Defeating Phenomenal Conservatism”, I explain
why Huemer’s arguments for Phenomenal Conservatism are unsuccessful. In
addition, I shall explain why the standard arguments for reliabilism and proper-
functionalism should not persuade anyone to think of justification in externalist
terms.
2
Chapter 3
Epistemic Value
3.1 Introduction
In this chapter, we shall look at some of the ways in which considerations having
to do with epistemic value might help shape a theory of justification. Epistemol-
ogists have recently started to test competing accounts of knowledge by looking
at the implications these accounts have concerning the value of knowledge.1 Ev-
eryone thinks that a theory of knowledge has to be extensionally adequate (i.e.,
if we do not know that Tom stole the book, Dick knows that he is a member of
the executive branch, and Henry does not know that the building he saw was a
barn, this is what our theory should say). They say that extensional adequacy is
not enough. If by some miracle we were to hit upon an extensionally adequate
theory only to discover that the theory implied that knowledge was no more
valuable than a lucky guess, we should hope for a better theory. If we can rely
on our evaluative intutions in formulating and testing our theories of knowledge,
we should be able to do something similar when it comes to justification.
Let’s distinguish between two ways in which axiological considerations can
help shape a theory of justification. The first role is modest. If we have views
or intuitions concerning epistemic value, we should try to bring our theory of
justification into reflective equilibrium with these views or intuitions. Intuitively,
it is always good to believe with justification. In fact, it seems that it is better
to believe with justification than without it. These kinds of intuitions, I think,
are intuitions that many of us share and so I think many of us would agree
that it would be a mark against a view if it clashes with them. In appealing
to these kinds of intuitions, we are not assuming that the good is conceptually,
explanatorily, or metaphysically prior to the right. Appealing to these intuitions
should appeal to consequentialist and non-consequentialist alike.2
A far less modest role for axiological considerations to play is as the basis
of a theory of justification. An epistemic consequentialist could argue that the
1 See Kvanvig 2003, Pritchard 2007, and Riggs 2008.
2 Below, we shall look at Feldman’s 2000 value-driven argument for evidentialism.
3
CHAPTER 3. EPISTEMIC VALUE 4
good is, in some sense, prior to the right, and try to derive an account of right
believing from some prior conception of the epistemic good. Such a view has at
least this much to recommend it. It provides a neat explanation as to why it is
always better to believe with justification. It is by realizing epistemically good
states of affairs that beliefs are justified. Goodness makes for rightness. One
potential advantage of working from an account of the good to the right is that
it might free us from relying on intuitions about justification ascription we have
already seen cannot settle the debate between the internalists and externalists.
Although considerations having to do with epistemic value should play some
role in constructing a theory of justification, I doubt these evaluative considera-
tions can really advance the internalism/externalism debate. Someone could try
to argue, on broadly consequentialist grounds, for an externalist theory of justi-
fication.3 The move from epistemic consequentialism to externalism might make
sense, but the argument would fail because of its consequentialist underpinnings.
For reasons discussed below, we should be epistemic non-consequentialists and
deny that right or justified believing can be understood simply in terms of pro-
moting something of value. While there is nothing wrong with trying to bring a
theory of justification in line with the intuitions you have, these intuitions about
value will not tell us whether justification is an internalist or externalist notion.
At best, such intuitions suggest that there are internalist elements required for
justified belief. Since this is a view few externalists will deny, we will need to
look at other strategies for determining whether justification is an internalist or
externalist notion.
3.2 Epistemic Consequentialism

Epistemic consequentialism could take many forms, but in this section we shall
primarily focus on two, epistemic belief-consequentialism and epistemic rule-
consequentialism:
There is sufficient justification for you to believe p if there is no
alternative to believing p in which a greater amount of epistemic
value is realized (EBC).
There is sufficient justification for you to believe p if there is a set
of J-rules that permits believing p and there is no alternative set
of rules that prohibits believing p such that a greater amount of
epistemic value is realized by following this alternative set of rules
(ERC).4
EBC and ERC are intended to be the epistemic analogues of act- and rule-
consequentialism in ethics. One reason to focus on ERC is historical. Alvin
3 Goldman 1986 defends an externalist account of justification that is modeled on a rule-
consequentialist approach to right action.

4 These views are both impure. They are formulated in such a way as to allow the ad-
dition of a no-defeater clause. I doubt that the addition of such a clause is in the spirit of
consequentialism, but I wish to be generous to the epistemic consequentialists.
Goldman defends something similar to ERC in his work.5 To my knowledge, no

one defends EBC but I think it is worth discussing its merits. First, some have
tried to pin EBC on epistemic consequentialists on the grounds that indirect
consequentialism is not a genuine alternative to direct consequentialism. We
should ask whether EBC truly is a view we ought to avoid.6 Second, it is
easy to imagine how someone could try to argue from either ERC or EBC to
some form of externalism. One kind of internalist will say that the factors that
determine whether a belief can be justifiably held strongly supervene upon a
your non-factive mental states. Another might say that the reasons that bear
on whether to believe all have to be accessible to you. ERC and EBC seem
to be in tension with both views. The facts accessible to you do not seem to
be the sort of facts that determine whether it would be optimific for you to
believe a proposition at that time. The facts that determine whether it would
be epistemically optimific for you to believe p do not seem to strongly supervene
upon your non-factive mental states. The clash between standard versions of
internalism and EBC is rather obvious, but ERC is no friend to internalism,
either. Depending upon how J-rules are formulated, facts accessible to you
or facts that supervene upon your non-factive mental states might determine
which rule you conform to in forming a belief without determining whether that
rule is among the genuine epistemic rules or the rules that confer justification.
After all, what distinguishes J-rules from spurious epistemic rules has to do with
what results from following those rules and, depending upon how the rules are
formulated, it seems to be conceivable that two subjects in two different possible
worlds could follow the same rules with very different results.
Admittedly, the argument is sketchy and it ignores those versions of epis-
temic consequentialism that might seem more friendly to internalism.7 It is
an interesting question as to whether this an argument of this kind could be
developed into a plausible rationale for externalism. Below, I shall argue that
it cannot.
3.2.1 Epistemic Rule-Consequentialism

In Epistemology and Cognition, Goldman said that we should think of epistemic
justification in terms of epistemic rules:
I approach justification in terms of a rule framework. This is war-
ranted by purely semantic connotations. Calling a belief justified
implies that it is a proper doxastic attitude, one to which the cog-
nizer has an epistemic right or entitlement. These notions have a
strong deontic flavor . . . They are naturally captured in the lan-
guage of “permission” and “prohibition”, which readily invite the rule
formulation.8
5 See Goldman 1986 and 1999.
6 See Maitzen 1995. We shall discuss his objections to EBC and ERC below.
7 I have in mind more “subjective” objectivist views, such as the account found in Jackson
(1991).
8 Goldman 1986, pp. 59.
On his view, a belief is justified if it is formed in accordance with a right system

of epistemic rules, provided that the permission to believe is not undermined
by the believer’s cognitive state. He also defends the view that the justification
of a belief depends upon whether the processes that produce that belief are
sufficiently reliable. These claims are not obviously in tension, but it is also
not immediately obvious why we should combine them into a single account.
Why link rule conformity and reliability in this way? Goldman’s answer is
that he thinks we can model an account of epistemic justification on a rule-
consequentialist approach to right action.9 Perhaps the idea is this. Many of us
are already inclined to think that if we follow the right rules for belief formation
and revision, we will believe with justification. His account tells us how to
distinguish J-rules from spurious rules. The right rules are right because these
rules reliably lead to the truth. Spurious rules are spurious because nothing
good comes of following them.
Consequentialists say that the moral status of an action is determined by
the value of the consequences where those values are characterized in something
other than deontic terms. Act- and rule-consequentialism agree on this much
and so uphold the basic idea that goodness makes for rightness. Their accounts
differ in that the act-consequentialism has a one-level structure according to
which the moral status of a particular action is determined by comparing the
value realized by that action’s consequences to the values that could or would
have been realized by alternative actions.10 Rule-consequentialism has a two-
level structure. On this view, the moral status of a particular action is not
determined directly by its effects but the value of the effects of following or
internalizing some set of rules that permit the action. If the rules that would
be optimal to follow sanction the act, the act is permissible.11 Otherwise, it is
not, even if that action would bring about better results than its alternatives.
One of the apparent advantages of rule-consequentialism is that its ver-
dicts do not necessarily deviate from the verdicts of intuitive forms of non-
consequentialism as radically as the verdicts of act-consequentialism do. Non-
consequentialists and rule-consequentialists, for example, might agree that there
is a plurality of rules that determine whether an action is right and might agree
on which rules are genuine. Their disagreement is not always at the level of the
proper formulation of rules. It emerges when we look to see what justification
can be given for the rules. A non-consequentialist might say that the rules con-
fer justification but need no justification in turn. The rule-consequentialist will
likely say that rules only confer justification because they can be given a value-
theoretic justification. Rule-consequentialism thus might display the virtues
that attract some to rival non-consequentialist and consequentialist views. On
9 Goldman 1986, pp. 97.
10 The actualists think that the consequences that would have been realized are the conse-
quences that matter. The possibilists think we should focus on the consequences that could
have been realized. The difference between these formulations emerge when we think about
the ways in which present obligation depends upon future choices. The differences between
actualism and possibilism are interesting, but for our purposes it would be an interesting
distraction. For a discussion, see Zimmerman 1996.
11 See Hooker 2000 for a discussion of different formulations of rule-consequentialism.
the one hand, it has explanatory virtues that consequentialist views have that
non-consequentialist views seem to lack insofar as it supplies an independent
rationale for the rules it recognizes. On the other, it conforms to commonly
held intuitions that act-consequentialists have to try to explain away.
Of course, the critics of rule-consequentialism in ethics and epistemology will
say that this is all too good to be true. Here, I want to look at two kinds of
criticism. One kind of critic will say that we should adopt a non-consequentialist
view that recognizes a plurality of epistemic rules and denies that these rules can
be given a value-theoretic justification. Recently, a few authors have defended
accounts of epistemic justification modeled on Ross’ pluralist account of right
action and I want to see here whether it would be better to understand J-rules in
the way that Ross understood his principles of prima facie duty.12 Another kind
of critic will say that ERC is not a viable alternative to EBC. They say that the
consequentialists should follow the value-theoretic arguments where they lead
and, our critic says, they lead towards a direct form of consequentialism such
as EBC. Myself, I am not entirely convinced by these kinds of criticisms. In
the end, I think the critics of ERC are right to reject the view. After offering
a partial defense of the view, I shall try to explain where precisely ERC goes
wrong.
3.2.2 Can We Unify the J-Rules?

For Goldman, J-rules have conditions of application that make reference to an
individual’s cognitive states and say that there is a defeasible permission for the
subject to believe some proposition.13 So, he might recognize rules like these:
If it perceptually seems to S as if some object x is F (where F is a
perceptible property), and this causes or sustains in the normal way
S’s belief of x that it is F, than that confers prima facie justification
on S’s belief (PER).
If S seems to remember that p and this causes or sustains in the nor-
mal way S’s belief that p, then that confers prima facie justification
on S’s belief that p (MEM).
If it introspectively seems to S as if S is occurrently having a sensory
or perceptual experience such and such and this sustains in the nor-
mal way the belief that S is experiencing such and such, then that
confers prima facie justification on S’s belief (INT).
On ERC, a belief is justified (and not just prima facie justified) if it is formed
in conformity to some J-rule and the justification provided by the rule is not
defeated. These rules were lifted not from Goldman, but from Graham. He de-
fends an internalist and non-consequentialist approach to justification modeled
on Ross’ pluralist account of moral duty.14 According to pluralist epistemic
12 SeeGraham 2010 and Nelson 2002.
13 Goldman 1986, pp. 77.
14 Graham 2010. A similar view was defended by Nelson 2002.
non-consequentialism (PENC), a belief is jusitifed if it is formed by someone

who conforms to a J-rule and the justification provided by the rule is not de-
feated. The view is internalist because it says that the application conditions of
these rules make reference to internal states of the subject (e.g., the subject’s
non-factive mental states) and it denies that factors that do not supervene upon
those states can defeat the justification of the subject’s beliefs.
Assuming the rules listed are genuine J-rules, ERC says there is a value-
theoretic justification for these rules. These rules confer justification because
they reliably lead to truth. PENC denies this. Following these rules might reli-
ably lead to truth, but this is a contingent feature of these rules and inessential
to their status as J-rules. The version of PENC we will look at in this section
models its account of justification on Ross’ pluralist theory of justified action.
Ross thought that principles of prima facie duty identified features of the situa-
tion that go towards justifying an action and denied that these principles shared
anything in common apart from the fact that they specified features that go to-
wards making right acts right. He explicitly rejected the view that a principle’s
status as a principle depended upon whether it pointed to a feature that all
right acts shared in common, such as bringing about good states of affairs. In a
similar vein, the version of PENC we are interested in says that each of J-rules
identifies a different kind of justifying reason and denies that there is any feature
that all right beliefs share in common apart from the fact that they conform to
some J-rule or other.
If these views agreed on the rules, you might think that this gives us some
reason to prefer ERC. After all, it seems reasonable to prefer a theory that
explains the difference between J-rules and spurious rules to one that does not.
However, if ERC and PENC disagreed about whether some (putative) epistemic
rule was an actual J-rules, this might give us some reason for preferring one view
to the other. If ERC’s list of rules is wrong, it is either too long or too short. If
you say that it is too short, you have to say that there is a rule that is unreliable
that is nevertheless a genuine J-rule. If you say that it is too long, you have
to say that there is a reliable rule that does not confer justification. Since you
should not assert or believe without evidence, it seems that the defensibility of
your position depends upon having evidence that an unreliable rule justifies or
that a reliable rule fails to justify. But, it seems that if you have evidence that
a rule is unreliable, it cannot justify. If you have evidence that a rule is reliable,
it can justify. So, ERC avoids the objection.
Not quite. Rather than focus on actual rules and the actual consequences
of following them, we should focus on possible rules and possible situations in
which epistemic rules are mistakenly taken to be reliable by the denizens of
these worlds. Or, we could imagine situations in which rules that are actually
unreliable are reliable and relied on by the denizens of these possible worlds. If
ERC does not cohere with our intuitions about these possible but non-actual
circumstances, maybe this gives PENC an advantage over ERC. Recall our
earlier discussion of Norman, one of BonJour’s clairvoyants.15 Norman could
15 See his 1980, pp. 21.
reliably determine the locations of various individuals using clairvoyant powers

even though he had no evidence that he had this ability. While he also had
no evidence that he did not have this ability, BonJour thought that Norman’s
beliefs were not justified. Here, a proponent of PENC could say that ERC has to
classify Norman’s beliefs as justified because he forms his beliefs by conforming
to a rule as reliable as the rule we follow when we form perceptual beliefs, PER.
However, intuitively, his beliefs are not justified.
This objection has little bite. There has be a reason why Norman’s visions
do not justify his beliefs. ERC denies that it could be a brute fact that there is
a J-rule for perceptual experience and not for these visions and it seems right
for it to do so. If someone says that the reason that his apparent perceptual
experiences can justify his beliefs but his clairvoyant visions cannot is that he
has evidence that his apparent perceptual experiences reliably lead to truth,
then the problem is not really with the ERC approach to J-rules but that it
does not require evidence of reliability as one of the conditions necessary for
justification. If it simply added one in, the alleged difference between apparent
perceptual experience and clairvoyant vision would be lost. If the force of the
objection is supposed to be that it is intuitive to say that clairvoyance cannot
justify even in possible worlds where it is as reliable as perception and we have
the same evidence of its reliability we do for perception, the objection rests on
an intuition that is little more than an irrational prejudice against belief forming
processes that are not our own. The problem with this kind of objection to ERC
is perfectly general.
A proponent of PENC might instead try to argue that there is something
wrong with the consequentialist justification of the J-rules. On this front, ERC
faces two related objections. The first is a modal objection. PER says that if
you believe that some visible object has such and such perceptible quality on
the basis of an experience that represents that object as having that quality,
your belief is prima facie justified. Absent any defeaters, your belief just will
be justified. Perceptual experience is a reliable guide, but this is a contingent
fact about our experiences. We can easily imagine creatures with experiences
indistinguishable from ours having experiences that are an unreliable guide to
their surroundings. Just imagine that these creatures are deceived by a Carte-
sian demon. Intuitively, it seems these subjects are just as justified as we are
if they form their beliefs in accordance with PER, but ERC has to say that
their beliefs are not justified. So, ERC can say that there is a contingent con-
nection between forming justified beliefs and conforming to PER owing to the
contingent connection between following PER and reliably forming true beliefs.
The second objection is epistemic.16 We know apriori that even in epistem-
ically inhospitable environments, subjects that follow PER are justified in their
beliefs. We do not know apriori, however, that following PER will result in good
epistemic outcomes. Indeed, it seems we know apriori that following PER can
lead to epistemically disastrous outcomes and still confer justification. So, not
only does ERC say that J-rules have that status contingently when they seem
16 See Graham 2010 for a version of this objection.
to have their status necessarily, they say that we cannot know apriori whether
these rules confer justification when in fact it seems we do know apriori that
some J-rules have that status.
These objections are two sides of the new evil demon objection, and since
we have already discussed that objection, my response can be brief. Even if
it is necessarily true that subjects that conform to rules like PER are justified
in their beliefs, all that follows is that persons are justified in their beliefs,
not that their beliefs are adequately justified. Intuition does not support the
further claim that these subjects have justified beliefs and so intuition does not
support the further claim that forming beliefs in conformity to PER confers
doxastic justification even if it does not reliably lead to truth. And, since we
do not know apriori that PER confers justification even in epistemically hostile
environments, it is not obviously wrong to say that we cannot know apriori if
PER confers justification.
So far, ERC remains unscathed. At this point, defenders of ERC should
go on the offensive. One problem with PENC is that it seems unmotivated.
Intuition does not seem to support the modal or epistemological objections to
ERC, but intuition does seem to support the Rossian view that principles of
prima facie duty can confer justification upon actions that do not bring about
the best consequences. Intuitively, if you have to choose between two options
where one is not better than the other but one involves keeping a promise and
one involves breaking it, you ought to keep your promise rather than breaking it.
If the consequences of keeping a promise are dire enough, you should break it,
but you still regret that you had to break your promise. PENC is not supported
by similar intuitions. This is worrisome.
A second worry is that the specific reasons Ross had for adopting a pluralist
account of justified action do not support PENC. Why was Ross a pluralist?
Here, Ross explains why he is dissatisfied with the monism exemplified by util-
itarian theories:
. . . the theory of ‘ideal utilitarianism’ . . . seems to simplify unduly
our relations to our fellows. It says, in effect, that the only morally
significant relation in which my neighbors stand to me is that of
being possible beneficiaries by my actions. They do stand in this
relation to me, and this relation is morally significant. But they
may also stand to me in the relation of promisee to promiser, of
creditor to debtor, of wife to husband, of child to parent, of friend
to friend, of fellow countryman to fellow countryman, and the like;
and each of these relations is the foundation of a prima facie duty.17
A simple example should help illustrate Ross’ point. You could feed your hungry
child or you could feed an equally hungry stranger, but there is not enough bread
for both. In explaining why it would be right for you to help your child rather
than the stranger, we can say that there is a morally significant relation between
you and your child that justifies giving the child priority (You created that child
17 Ross 1930, pp. 19.
and so you put them in need of assistance. You did this without their consent.
So, you owe the child something you do not owe the stranger even if you owe a
stranger assistance when you are able to assist without failing to live up to some
other obligation.) In other situations, it might be wrong to use a second bit of
bread to feed your child rather than, say, save the lives of sixteen strangers.
Here a second kind of morally significant relation determines what you should
do. In the first case, the overriding reason is a reason is a reason you have
because you stand in some prior relation to someone. In the second, the reason
is a reason you can have to help a stranger. Helping the stranger is beneficent,
but it is no mere act of beneficence to feed your child. For evidence that the
reasons differ in kind, notice that morality does not require you to flip a coin
when the needs of your children and strangers are equally strong and can be
met equally well.
Modifying Ross’ rationale for pluralism in ethics does not provide us with
a suitable rationale for PENC. Maybe perceptual experience justifies because
it puts beliefs in the right relation to truth to be justified (whatever that is).
A belief that is not put in the right relation to truth (whatever that is) by
perception cannot be justified by memory or testimony unless memory or tes-
timony makes up for this deficiency by putting the belief in the right relation
to truth (whatever that is). No one really thinks that perception or memory
justifies without putting a belief in the right relation to truth (whatever that
is) or thinks that we can justify a belief that cannot be put in the right relation
to truth (whatever that is) by putting it in the right relation to something else
(e.g., fame, beauty, a pleasant illusion).
Finally, note that the evidence that seems to confirm Ross’ view seems to
disconfirm PENC. Remember that Ross thought that when duty compels us to
act against some defeated moral reason, that reason continues to exert a force
on us:
If, as almost all moralists except Kant are agreed, and as most plain
men think, it is sometimes right to tell a lie or to break a promise,
it must be maintained that there is a difference between prima facie
duty and actual or absolute duty. When we think ourselves justified
in breaking, and indeed morally obliged to break, a promise in order
to relieve some one’s distress, we do not for a moment cease to
recognize a prima facie duty to keep our promise, and this leads us
to feel, not indeed shame or repentance, but certainly compunction,
for behaving as we do; we recognize, further, that it is our duty to
make up somehow to the promisee for the breaking of the promise.18
According to Ross, one mark of genuine pluralism is the possibility of a kind of
rational regret. You can rationally regret a choice even if you know that this
choice is right. The regret is an indication that the agent faced a choice between
competing concerns where the goods contained in one of these options is not
18 Ross 1930, pp. 28.
contained in the other.19 You can rationally regret doing the right thing because
you can see in the neglected option something that is not contained in the option
chosen. To see this, notice that acting on the strongest reason and acting against
a defeated prima facie duty is importantly different from choosing between two
goods that do not differ in kind. Suppose someone offers you a choice between
a large and a small coffee on the house. There is nothing of value in the smaller
cup not contained in the larger cup. In this sort of situation where there is not a
plurality of goods, the right choice is not regrettable. In choice situations where
the right choice is regrettable, the regret registers that there was a kind of good
in the neglected option not contained in the right option. There is nothing like
the phenomenon of rational regret in the theoretical domain and this seems to
count against PENC. It posits a plurality of epistemic principles each of which
supplies a unique kind of justifying reason for belief while denying that there is
a common value these reasons all serve.
This is a point that should be credited to Williams. He observed that if
you are trying to decide what to believe and you weigh the evidence for and
against some proposition, you cannot judge that some proposition is the one
you should believe while also thinking of the evidence against that proposition
as anything but misleading evidence.20 The point is not that we cannot regret
losing a belief upon discovering (what we take to be non-misleading) evidence
against it. If you regret reasonably the loss of belief, this will always be due to
some desire of yours that is ill served by losing the belief (e.g., that you have
to abandon a pet theory, that you are caused pain by recognizing the facts that
falsify the belief, etc. . . ). The point is that there is nothing in the reasons that
seemed to support the belief that you think you lose out on by abandoning
that belief whereas there can be something in the reasons that you act against
that you see as attractive even when those reasons are pitted against stronger
reasons. A defeated practical reason to act might constitute a compelling reason
to act at another opportunity should it arise, but defeated reasons to believe p
that constitute misleading evidence do not later constitute compelling reasons
to believe p at another occasion.
Now, this point has been contested. Nelson has claimed that there are
epistemic residues:
[M]any of us will be familiar with such epistemic residues in the form
of perplexity. After searching fruitlessly through that article for the
passage I remember, I wonder, ‘I clearly recall reading that quote
from Flaubert in this article, but now I can’t find it anywhere. How
can this be!?’21
19 To be sure, if you had to choose between retriving the five dollar bill that fell from your
pocket and the ten dollar bill that fell from your pocket, this might be regrettable. What seems
regrettable, however, is not responding in the right way but that you had to respond to the
situation at all. For a useful discussion of the distinction between choice- and option-regret,
see D’Agostino 1988, pp. 32.
20 See Williams 1965, pp. 107.
21 Nelson 2002, pp. 273.
If epistemic residues were anything like moral residues, you could know that a
reason is defeated by an overriding reason and it would still continue to exert a
kind of rational force. For reasons sketched above, the phenomenon of rational
regret is supposed to require a plurality of kinds of justifying goods and so
would point in favor of PENC. I doubt perplexity involves epistemic residue.
I suspect that perplexity is due to the fact that the evidence you take to be
misleading is so robust. In taking the passage to be in those pages, you expect
not to find such strong evidence that it is not there. So, perplexity is due just
to the fact that the appearances you take to be misleading hold up so well
upon examination. To see that perplexity does not require pluralism and so
does not involve epistemic residue, notice that perplexity does not require a
plurality of kinds of sources of justification and so does not require a plurality
of J-rules. In cases of perceptual illusion, for example, there might be an object
that you visually attend to that appears to have properties it could not have
simultaneously. If the illusion is robust, you might be perplexed, but as the case
involves one source of justification only, we do not need to assume that there is
an irreducible plurality of J-rules to understand why this is.
To sum up, we have seen no reason to think that there is anything wrong
with offering a consequentialist justification for J-rules. Indeed, it seems to be
a mistake to deny that there could be a unifying explanation of these rules.
So far, ERC seems to favor reasonably well. Above I said that we should be
epistemic non-consequentialists. This is not because we should model epistemic
non-consequentialism on Ross’ view. In later chapters, I shall defend a non-
consequentialist account of justification. Before we get to that, I need to explain
why epistemic justification cannot be understood in consequentialist terms.
3.2.3 The Case Against Epistemic Consequentialism

It should come as no surprise ERC compares favorably to a pluralist view that
denies that J-rules share some common feature by virtue of which they can
confer justification. When the morally conscientious agents ask, “What should
I do?” there is not some common aim that they share in common apart from
responding appropriately to the diverse competing moral considerations they are
faced with. When the epistemically conscientious subjects ask, “What should
I believe?” there is something that they aim at in common. They aim to get
things right. There are many ways to aim to act rightly, but the only way to
believe rightly is to believe the truth.
ERC compares favorably to PENC, but I do think it fares quite poorly
overall. To see why, it will be useful to fill in some further details of Goldman’s
view. ERC tells us something about the relation between the right and the good,
but it does not tell us what the fundamental epistemic goods are. According to
Goldman, true beliefs are intrinsically good, false beliefs are intrinsically bad,
and these are the only values that matter in inquiry.22 This view, which he
dubs “veritism”, tells us what the bearers of epistemic value and disvalue are,
22 See Goldman and Olsson 2009, pp. 24.
but it does not tell us how to compare the value of true belief to the disvalue
of false belief. So, our account is still incomplete because it does not yet tell
you how to compare the value of the available alternatives. Goldman suggests
that the absolute value of true belief is the same as the absolute value of false
belief.23 Epistemically speaking, you are as well off believing nothing, as you
are forming beliefs on the basis of a process that gets things right half of the
time. This last feature of his view is not essential to veritism or ERC, so if the
need arises to adjust it, he can adjust it as need be.
Remember that the evaluative consequences of ERC and EBC might differ
because of ERC’s two-level structure. Initially, this seems to be a reason to
prefer ERC to EBC. EBC seems to say that no belief is justified unless that
particular belief promotes something of value. Suppose we were to say that true
beliefs are the bearers of epistemic value. Given this assumption, EBC seems
to imply that you cannot justifiably believe p unless in believing p you believe
something true. ERC avoids this implication. It denies that the justification of
a belief depends upon the value of the consequences of forming that particular
belief. So, while EBC suggests that there is too tight a connection between
justification and truth, ERC allows for a looser connection. You can justifiably
believe p even if in coming to believe p, you do not form a true belief.
Initially, it might seem that ERC enjoys an advantage over EBC because of
its two-level structure. In fact, it is a liability for ERC. To see why, consider
one of Rawls’ example:
In a game of baseball if a batter were to ask “Can I have four strikes?”
it would be assumed that he was asking what the rule was; and if,
when told what the rule was, he were to say that he meant that on
this occasion he thought it would be best on the whole for him to
have four strikes rather than three, this would be most kindly taken
as a joke.24
Suppose that when Bill Baseball drew up the rules for baseball, he did so in the
hopes of making the game as enjoyable for the fans as possible. He knew, never
mind how, that fans would enjoy the game more if batters were given three
strikes and umpires were not given any discretion in awarding more strikes
or depriving batters of a second or third strike. At one level, what justified
adopting the three-strike rule were facts about the fans and their preferences.
At a second level, particular calls are justified by the rule. The justification of
an umpire’s call depends upon whether it conforms to the rules and not upon
whether the fans would have been happier if the umpire made a different call.
On this model, if the value that justifies adopting the rule is not mentioned in
the specification of the rule’s application condition, the justification of the calls
governed by those rules do not depend upon how well any particular call serves
those values. This has nothing to do with the umpire’s epistemic predicament.
If everyone despises the Yankees, the umpire might know that calling one of
23 Goldman 1999, pp. 89.
24 Rawls 1955, pp. 26.
their batters out after two strikes would please everyone and still know that
there is no justification for making this call.
One lesson to draw from the example is this. If Goldman takes seriously the
parallels between ERC and rule-consequentialism, he should say that a particu-
lar belief that does not serve the values that justify the J-rules could nevertheless
be perfectly justified, provided the belief conforms to the J-rules. Indeed, this
is what he says. Think about the umpire’s call. Suppose the umpire’s call
conformed to the rulebook—he called the Red Sox batter out upon correctly
calling a third strike. Suppose this upsets everyone tremendously. Is there any
sense in which the umpire’s call was mistaken? It seems not. Or, to change
examples, suppose you have to choose between keeping a promise and breaking
it. You know that if you break the promise, you can produce a smidge more
happiness than you can if you keep it. This is perfectly consistent with the rule-
consequentialist saying that you are obliged to keep the promise. So, let us as-
sume that this is what the rule-consequentialist says. On rule-consequentialism,
there is no sense in which it is correct to say that it was a mistake to keep your
promise. There is no sense in which this view implies that you chose incorrectly
if you chose to keep your promise. The rule-consequentialist does not say that
the choice to perform the suboptimal action was justified if it conformed to the
rules but mistaken because it was suboptimal.
Now, let’s switch from action to belief. Suppose someone believes p in con-
formity with the J-rules but p is false. ERC implies that the belief is justified.
It also seems to imply that there is no sense in which forming the belief was a
mistake. Not if the analogy holds true. But, this is why ERC is mistaken. It
is essential to the state of belief that every false belief is mistaken and every
time you form a false belief you make a mistake. Just to be clear, the prob-
lem is not that ERC implies there are false, justified beliefs. Nearly everyone
agrees that there can be false, justified beliefs.25 The objection is that the
rule-consequentialist says that you can know that you ought to perform a sub-
optimal action if that action conforms to the right system of rules. When this
happens, you know that the correct way to deal with the situation is precisely
by doing what you know you should always do—follow the rules. The value-
theoretic considerations that determine which rules are genuine are not among
the considerations that determine whether you make the correct choice or make
a mistake in acting apart from selecting the rules. These values are screened
off, if you like. The worry is that in asserting that justification and truth are
connected only indirectly by means of J-rules, ERC asserts that if you follow
the rules you can form a false belief without making a mistake and denies that
there is any sense in which the right response to the rules is mistaken. This is
false to our concept of belief.
How might Goldman deal with this objection? Might he say that there is a
sense in which a false belief is a mistake and another sense in which it is not?
He might, but the objection does not assume that “mistake” is not ambiguous.
25 For dissent, see Sutton 2005 and Unger 1975. To come clean, I will argue that there are
no false, justified beliefs in later chapters.

It rests on the intuition that there is no sense in which the umpire’s call is
mistaken if the umpire calls the batter out upon correctly calling a third strike.
So, to press the analogy, the worry is that there would be no sense in which ERC
can say that a belief is mistaken if it conforms to the J-rules. Could Goldman
instead deny that it is always a mistake to form a false belief? Of course he
could say this, but the response is both inherently implausible and damaging to
his veritism. If there can be false beliefs that are correct and true beliefs that are
incorrect, why would true beliefs be invariably intrinsically good and false beliefs
invariably intrinsically epistemically bad? What could be bad about believing
a false proposition if the false belief is both correct and justified? What could
be good about believing a true proposition if the belief is neither justified nor
correct?
The first problem arises precisely because of ERC’s two-level structure and
the way that truth is relegated to the role of selecting J-rules. The second prob-
lem has more to do with the way that epistemic consequentialists understand
the relation between the good and the right. In practical reasoning when op-
tions are tied, you cannot go wrong in picking between them. In theoretical
reasoning, if there is equally good evidence to believe both a proposition and
its negation, there is a conclusive reason to refrain from believing until you can
break the tie and find stronger evidence for believing one of these propositions
instead of the other.26
This is a familiar point and it points to a problem with ERC. Suppose that
the total epistemic value that results from following R1 is the same as the
total epistemic value that results from following R2. ERC should say that it
is permissible to follow R1 iff it is permissible to follow R2 even if following
these rules leads you to form different sets of beliefs. According to Goldman,
there is no obligation to form beliefs.27 J-rules give us permissions to believe
and determine which beliefs we are prohibited from having, but they do not
generate obligations to believe. So, suppose R1 forbids belief across a wide
range of cases (e.g., it forbids believing or disbelieving any proposition about
the past, the future, or the external world). In effect, if you follow R1, you
will believe nothing at all about the external world. Now, consider a rule that
permits you to form beliefs about the external world but only gets things right
half of the time. For example, suppose R2 says that for any p such that p
is a proposition about the external world, you are permitted to form a belief
about whether p by flipping a fair coin (i.e., believe p if heads or believe ∼p
if tails). In such cases, you know you have no better evidence to believe p
or its negation and that is why you know that you have conclusive reason to
believe neither. If you are permitted to follow R1 by believing nothing, you are
permitted to believe whatever you like about the external world by flipping a
fair coin. Consider another rule, R3. According to R3, you are permitted to
believe that the number of Fs is odd if you know that there is a finite number of
Fs and too many Fs to count. I am not in any position to judge that the number
26 Adler 2002, pp. 25.
27 Goldman 1986, pp. 77.
of koalas is odd, but it is just as likely to be odd as even given what I know
about koalas. We should reject any theory that says that you are permitted to
believe p when the evidence supports p and ∼p equally well. We should do the
same for a theory that says that you are permitted to believe p by following a
rule you know is as likely to lead you to form a true belief as it will lead you to
form a false one.28
What can Goldman say to deal with the 50/50 problem? As stated, the
objection assumes that the J-rules state permissions or prohibitions only, which
implies that you are not obliged to believe anything at all. This is a strange
position for the epistemic consequentialist to defend. As a rule, consequentialists
are not terribly keen on the doing/allowing distinction. Suppose we revise the
view so that its J-rules generate positive obligations to believe in ways that
promote epistemic value.29 Notice that this assumption is inessential to the
objection. Suppose the J-rules oblige you to do the best you can in promoting
verific consequences. If the number of Fs is finite and too large to count, you can
do no better than to either believe that the number of Fs is odd or the number
of Fs is even. If the aim is to maximize epistemic value, the rule that permits
believing the number of koalas is even should be expected to do just as well as
the rule that forbids this. Since there can be ties for first and consequentialism
denies that there is a normative difference between options of equal value, you
are permitted to believe both or permitted to believe neither. The problem
remains. Someone could say, plausibly, that you cannot justifiably believe on
the basis of a rule that is just as likely to get things right as it is to get things
wrong, but this misses the force of the objection. Why would ERC say that
J-rules only confer justification when you are ignorant of the chance of success?
In knowing that some particular J-rule leads you to form as many true beliefs
as false beliefs, you might know that following this rule is not suboptimal. So,
there is nothing available to the consequentialist to say that knowledge of the
low chance of success defeats the justification otherwise provided by the rule.
A different strategy would be to modify Goldman’s view concerning the
absolute values of true and false belief. Up to this point, we have assumed that
the magnitudes of the absolute value of true and false belief are the same (i.e.,
|VTB | = |VFB |). If you think that it is worse to believe a false proposition than
it is to fail to believe a true proposition, the way for the consequentialist to try
to capture this intuition is to say that |VTB | < |VFB |. If |VTB | < |VFB |, you
are better off not believing in accord with a rule that gets things right only half
of the time than you are by getting things right half of the time. Consider a
new rule. If you know that the number of Fs is finite but too large to count,
you have permission to believe that the number of Fs is not divisible by three.
It is likely that the number of koalas is not be divisible by three, but it also
seems that you are not in a position to justifiably believe that the number of
28 DePaul 2004 discusses a version of this problem.
29 Like Sutton 2005 and Nelson 2010, I think there are no positive epistemic duties to
believe. I cannot imagine any reasonable consequentialist rationale for denying that there are
any positive epistemic duties to believe. Add this to the list of complaints about epistemic
consequentialism if you are so inclined.
koalas is not divisible by three. Consider yet another rule. If you know that
the number of Fs is finite but too large to count, you have permission to believe
that the number of Fs is composite. It is likely that the number of tealeaves is
composite rather than prime, but you are not in a position to justifiably believe
this. Why not? On the present view, it has to be because the difference in
the absolute values of true and false belief are so great that you are worse off
believing on the rule in spite of the high probability of success. Any rule that
comes with a greater risk of leading to error cannot confer justification on your
beliefs. One worry about the present view, then, is that in dealing with the
50/50 problem, it implies that only very reliable rules can confer justification.
Consider yet another rule. For any lottery ticket you buy, believe that it is a
loser. Some have the intuition that you know you cannot know the ticket will
lose. It seems that if you know you cannot know p, your belief in p cannot be
justified.30 To accommodate these intuitions on the present view, there must
be a vast difference in the absolute values of true and false belief. The skeptical
implications of this view are straightforward. There are few things we believe
on the basis of rules that are more likely to get things right than the lottery
rule just described.
The third and final objection to ERC is the most serious. Unlike the previous
objection, it rests on no assumptions about the absolute values of true and
false belief. Unlike previous objections, it has implications for all version of
epistemic consequentialism. According to EBC, the justification of a particular
belief depends upon the value it promotes. According to ERC, the justification
of a belief does not depend upon the value the belief promotes but whether it
conforms to the right system of rules. By virtue of its two-level structure, it
seems that the evaluative consequences of ERC and EBC should differ. But,
some say that these views in the end really must come to the same thing.
Inspired by Lyon’s collapse objection to rule-consequentialism in ethics, Maitzen
argues that ERC faces a dilemma.31 Either it is extensionally equivalent to
EBC or it is incoherent. Why is that? Suppose you say that some rules are
J-rules only if there is not some alternative set of rules that you could follow
and thereby produce better outcomes. The best you could do is to follow the
rule EBC identifies as the fundamental epistemological rule. That rule simply
says we should believe in such a way as to bring about the best epistemic
consequences. According to EBC, in forming beliefs, you ought to believe p if
believing p is better than the alternatives and are permitted to believe p if there
is no better alternative in which you do not believe p. There are no beliefs that
ERC would classify as justified that EBC would classify as unjustified (or vice
versa) because ERC would say, in effect, that in cases where the views differed
it would defer to EBC.
Suppose this is right. What is wrong with EBC? Maitzen remarks:
30 Trivially, this is so if you cannot justifiably believe what you do not know. Even those
with internalist sympathies often defend the view that you cannot justifiably believe what you
know you are not in a position to know. See Huemer 2007.
31 See Maitzen 1995. For discussion of the collapse objection, see Lyons 1965 pp. 62-119.
If the nominal aim is the reason for having, or pursuing, justifica-

tion, then it ought to follow that beliefs are justified insofar as they
serve the nominal aim and unjustified insofar as they do not. But
this consequence gives rise to an obvious problem. If justification
is essentially a matter of serving the nominal aim [i.e., to maximize
true belief and minimize false belief], then it seems we would eval-
uate no true belief as unjustified and no false belief as justified . . .
The reason is straightforward. If one seeks, above all else, to maxi-
mize the number of true (and minimize the number of false) beliefs
in one’s (presumably large) stock of beliefs, then adding one more
true belief surely counts as serving that goal, while adding a false
belief surely counts as disserving it.32
We can sum up his argument as follows:
(1) Upon pain of incoherence, ERC must deliver the same verdicts
as EBC.
(2) EBC classifies each true belief as justified and each false belief
as unjustified.
(C1) Unless ERC is incoherent, it must classify each true belief as
justified and each false belief as unjustified.
(3) It is possible for true beliefs to be unjustified and false beliefs to
be justified.
(C2) ERC is either incoherent or delivers the wrong verdicts in just
the way EBC does.
One response to the argument is simply to deny (1). Rather than get bogged
down in a lengthy discussion about the coherence of rule-consequentialism, I
shall argue that (2) is false.33 Once we see why this is and and Maitzen’s
objection to ERC fails, we can see why Maitzen was right to reject both views.
Maitzen’s argument for (2) suggests that he thinks (2) is a straightforward
consequence of veritism. It is not, and this is why his argument fails. In ethics,
consequentialists do not think that the moral status of an action is determined
by the value of the action taken in isolation. Indeed, consequentialists will likely
be skeptical of the very idea of the value of the action taken in isolation. Con-
sequentialists endorse totalism, the thesis that total intrinsic value determines
normative status.34 Denying totalism is the equivalent of saying we should pre-
fer an acknowledged lesser good to a greater one. (To be sure, it might make
32 Maitzen 1995, pp. 870.
33 See Hooker 2000, pp. 93-108 for a response to Lyon’s collapse objection to rule-
consequentialism. While someone could say that Maitzen’s argument is outdated because
there are versions of rule-consequentialism that avoid the collapse objection, it should be
noted that Hooker’s specific reasons for denying that rule-consequentialism collapse into act-
consequentialism seem to have no obvious application to ERC and EBC. Also, Hooker’s rule-
consequentialism is not motivated by purely value-theoretic considerations and so differs from
the versions of consequentialism under consideration here.
34 Carlsson 1995, pp. 10.
sense to pick an acknoweldged lesser good over a greater good if you do so be-
cause of some non-consequentialist consideration that constrains choice, but we
are trying to use consequentialist considerations to determine what constraints
there are.)
The problem with Maitzen’s objection to EBC is that he neglects the fact
that on EBC it is the total epistemic intrinsic value that determines the justi-
ficatory status of a belief. If the belief is true, the true belief contributes some
positive value. If the belief is false, the false belief contributes some disvalue.
Beliefs also have consequences and come with opportunity costs. Epistemic con-
sequentialists should care about this. EBC implies that all and only true beliefs
are justified only on the assumption that it is impossible to form a true belief
without maximizing total epistemic value and impossible to form a false belief
while maximizing total epistemic value. Both assumptions are false, however.
It is possible that in forming the true belief that p the total value that results is
less than or equal to the value that would have resulted if you had instead either
formed the false belief that ∼p or believed neither p nor ∼p. It is possible that
in forming the false belief that p the total value that results is equal to or greater
than the total value that would have resulted if you had instead either formed
the true belief that ∼p or believed neither p nor ∼p. So, on straightforward
veristic grounds, it looks like (2) is false.
What might be true on EBC is that every time you justifiably believe a
proposition you also believe a true proposition, but these need not be the same
proposition.35 Believing p when p is true could trigger a chain of events that
causes the formation of large numbers of false beliefs. The result could be that
it would have been better not to believe p in the first place. Believing p when p
is true could set events in motion that prevent you from forming a large number
of true beliefs you would have formed otherwise. Given totalism, EBC says that
in these kinds of situations, p is true, but you should not believe it. If you
should not believe p, you cannot justifiably believe p. We can break the link
between false and unjustified beliefs constructing structurally similar cases.
Maitzen’s argument against epistemic consequentialism fails, but we can
now see why epistemic consequentialism is so deeply implausible. Our ordinary
practice of epistemic assessment is not just non-consequentialist. It is deeply
anti-consequentialist because we reject totalism. You can know that p is true
even if forming a belief in p brings about less overall epistemic value than
believing ∼p or suspending judgment. You can knowingly infer that p is true
even if you know that the result of forming this belief is that you will bring
about less overall epistemic value than believing ∼p or suspending judgment.
You cannot know p unless you justifiably believe p, so the justification of a
belief does not depend upon the total epistemic value of forming that belief.
35 Indeed, these need not even be believed by the same person. I do not think anyone believes
that the justification of your belief depends upon the value realized by the beliefs of others
as a result of your coming to believe what you do. My own intuitions strongly favor views
on which the justificatory status of my beliefs are not threatened by facts about what others
believe as a result of my beliefs. For a helpful discussion of the separateness of persons, see
Norcross 2008.
We should reject EBC because the totalizing assumption is at odds with the
anti-consequentialist nature of epistemic assessment. Even if ERC does not
collapse into EBC, it does not avoid the objection. There is nothing internal to
ERC that rules out the possibility of J-rules that classify beliefs as unjustified
by virtue of the value of the belief’s effects just as there is nothing internal to
rule-consequentialism that rules out the possibility of moral rules that classify
actions as unjustified by virtue of the value of the action’s effects. (Remember,
rule-consequentialists tend to agree with Ross that there are prima facie duties
of non-maleficence and beneficence.36 Their disagreement with Ross is about
whether there is an underlying rationale that explains why these rules are rules
and why there are more rules besides just these.) However, it is not just some
contingent fact about genuine epistemic principles or rules that they determine
the justificatory status of a belief apart from the belief’s consequences. So, while
we know it is impossible for a belief to count as unjustified simply in virtue of
the value of its effects, we know ERC does not rule this out and so ERC cannot
provide an adequate foundation for a theory of justification.37
Epistemic consequentialism can come in many forms. So far, we have as-
sumed the combination of consequentialism and veritism. Suppose true belief
is not intrinsically valuable. Suppose instead that knowledge is epistemically
valuable and any status short of knowledge is disvaluable. Or, suppose that
believing on the basis of evidence is intrinsically valuable and any irrationally
held belief is disvaluable. Modifying the value theory will not help. If the jus-
tification for believing p depends upon whether the total value of forming that
belief is at least as good as not forming that belief, we can construct examples
just like the examples described above to show that believing p on the evidence
or knowingly inferring that p is true prevents you from maximizing epistemic
value. So far we have focused on objective forms of consequentialism. The move
to a subjective form of consequentialism will not help. The examples do not
assume that the epistemically disvaluable effects of forming a belief are effects
the subject knows nothing about. Even if the subject is absolutely certain that
the cost of coming to believe the truth about p is a penalty of thousands of false
beliefs or irrationally formed beliefs, she knows this is irrelevant to the question
as to whether to believe p.
3.3 Evidentialism and Epistemic Value

In this section, I want to consider a value-driven argument for an internalist
version of evidentialism. On the view in question, the justification of your
beliefs is determined by relations between your beliefs and the evidence you
have, not on further relations between your beliefs and the states of affairs that
36 SeeHooker 2000.
37 Those with a taste for beating dead horses might also think about the separateness of
persons. Can epistemic consequentialists account for the fact that the justificatory standing
of yours beliefs does not depend upon how many “bad” beliefs others will form if you believe
the truth?
your beliefs concern.38 The view is an internalist view in that it says that your
evidence supervenes upon your non-factive mental states. Instead of working
from a theory of the epistemic good to a theory of right believing by means of
some consequentialist assumptions about the priority of the good to the right,
the argument we shall consider here appeals to intuitions about epistemic value
that many of us likely share.
The argument is intended to support an evidentialist view that upholds these
claims about epistemic wrongs and the right to believe:
It is wrong always, everywhere, and for anyone to believe without
sufficient evidence (EW).
It is right always, everywhere, and for anyone to believe upon suffi-
cient evidence (ER).
EW is perfectly harmless. This is the sort of claim that internalists and external-
ists alike can and should accept. ER might at first seem trivial, but evidentialists
do not deal in trivialities. If all ER said was that it is right to believe upon evi-
dence that is sufficient to provide a permission to believe, no one could sensibly
deny ER. So understood, if you did not have sufficient evidence for your beliefs,
you would not have whatever evidence is necessary, if any, to to permissibly
believe. Evidentialists defend something more contentious. They defend the
view that if you have just the same evidence as someone who justifiably believes
p, you also have sufficient justification for believing p.
The idea is that two individuals cannot have precisely the same evidence
but then differ in that on individual lacks sufficient evidence. This does not
tell us what sufficiency amounts to, but I do think that this is probably right
to say that if you know p, you have sufficient evidence for believing p is true.
Why think that? If you have knowledge, you have justified belief.39 You should
never believe without sufficient justification in the relevant sense. This is what
EW tells us. So, if you do justifiably believe, you have sufficient evidence in the
relevant sense (i.e., evidence sufficient to satisfy whatever requirement must be
satisfied so that your belief does not fail to be justified for a lack of evidence).
38 Not unless those states of affairs are themselves the sorts of things that obtain by virtue
of facts about your own non-factive mental states.
39 Does knowledge require justified belief? This is controversial, but I find Williamson’s
argument persuasive. Suppose S reasons competently from p and if p then q to q. If knowledge

does not require justified belief we should be able to suppose that S knows p without justifiably
believing it and justifiably believes the conditional without knowing that it is true. What
should we say about S’s belief in q? We cannot say that S’s belief in q constitutes knowledge
because an essential premise in the deduction was not known to be true. We cannot say that
S’s belief in q is justified because an essential premise in the deduction was not justifiably
believed. Surely, there is something going for believing q. It seems that it is reasonable for
S to believe q and it seems we would not say that S should suspend judgment rather than
have formed her belief in q. Would anyone ever say, “While she knew p and was not wrong to
think that p is true only if q, she should not have believed q until she had better reasons to
do so”? I think not, not with any plausibility, at any rate. If not, her belief was permissibly
formed. But that just is the mark of the justified belief. Knowledge is more than you need
for justified belief, so justified belief is required for knowledge. See Williamson 2007, pp. 111
Arguably, two individuals can have the same bodies of evidence but know dif-
ferent things.40 Given the conception of evidence our evidentialists are working
with, this is certainly so. If having the same evidence as someone in the know
is sufficient, ER says that if you have this evidence but fail to know for reasons
“external” to your evidence, you still can justifiably believe what you do. You
might have the same evidence as someone who could have known p while you
yourself fail to know p because p is false. You might have the same evidence
as someone who could have known p while you yourself do not know p because
your belief concerning p is Gettiered. Hopefully, the connection between EW,
ER, and this evidentialist supervenience thesis is clear:
If you and another individual have precisely the same evidence, you
have sufficient evidence for believing p iff this second individual
has sufficient evidence for believing p (EEJ).
Let us now turn to the argument for EEJ.
Feldman says we should accept evidentialism because, “following one’s evi-
dence is the proper way to achieve something of epistemic value”.41 Expanding
on this, he remarks:
While true beliefs may have considerable instrumental value, a per-
son who irrationally believes a lot of truths is not doing well epis-
temically. In contrast, a person who forms a lot of rational but false
beliefs is doing well epistemically . . . Consider a person who is con-
templating a particular proposition. To carry out the role of being
a believer in an epistemically good way, in a way that maximizes
epistemic value, the person must adopt a rational attitude towards
a proposition . . . To achieve epistemic value one must, in each case,
follow one’s evidence.42
We can restate his argument as follows:
(1) If you form the belief that is supported by the evidence, you form
the epistemically rational attitude towards a proposition.
(2) If you form the epistemically rational attitude towards a propo-
sition, you maximize something of epistemic value.
(3) In maximizing something of epistemic value, it is not wrong for
you to believe what you do.
(C1) Thus, it is right always and everywhere for you to believe upon
sufficient evidence.
40 This is controversial. We shall discuss Williamson’s account of evidence in the next two
chapters. For now, let us assume that two individuals in the same non-factive mental states
have the same evidence. This is Feldman’s conception of evidence and it is his argument for
an internalist conception of epistemic justification.
41 Feldman 2000, pp. 682.
42 Feldman 2000, pp. 685.
Feldman is probably right that some epistemic good is realized when you believe
in accordance with the evidence. He seems to assume that the value realized
either provides you with the right to believe or indicates that you have it. While
every value might call for some sort of response, it is not obvious that the value
realized by believing in accordance with the evidence calls for that response or
justifies it, come what may. Some values call for responses that do not involve
promoting the value in question (e.g., they might give us reasons to respect or
admire something without giving us reasons to try to bring about more of this
value).43 Some values call for promotion but provide only pro tanto reason to
bring these values about. So, even granting that it is always good to believe in
accordance with the evidence, it might also always be good in some sense to do
something that falls sort of being the right thing to do and something might be
good in some sense without giving any reason at all to bring that value about.
To see what the problem is, it might be useful to consider an argument
against ER and consider what Feldman could say in response:
(4) An agent can φ and her φ-ing can have moral worth even if it
was not right for the agent to φ.44
(5) If an agent’s φ-ing has moral worth, her normative judgment
that rationalizes her φ-ing has to have epistemic worth.
(6) If the normative judgment that justifies the agent’s φ-ing is jus-
tified, the agent’s φ-ing is justified if she φ’s accordingly.
(C2) Thus, a judgment can have epistemic worth without being jus-
tified.
(7) Any judgment that has epistemic worth has the value that at-
taches to reasonably held beliefs formed in response to the evi-
dence in a responsible way.
(C3) A judgment can have the value that attaches to reasonably
held beliefs formed in response to the evidence in a responsible
way without being justified.
Once we see what Feldman’s response to this argument is, we can start to see
how to blunt the force of his argument for EEJ.
Can an agent fail to do the right thing and still act in such a way that
it is creditable to her or laudable? I believe so, and Feldman seems to agree.
He thinks, for example, that an action can be wrong by virtue of its untoward
effects even if the agent happened to be non-culpably ignorant that the action
would have these effects. If the ignorance is “hard earned” in the sense that the
agent’s belief that the action would be for the best is based on sufficiently good
evidence, it is hard to deny that it is possible for the agent’s action to have just
as much moral worth as it would have had if only by some miracle those events
had not transpired and things really did turn out for the best. As Kant would
43 For a discussion of different modes of response to value, see Baron (1997) and Swanton
(2003).
44 This is controversial, but for defense, see Sverdlik 2001 and Zimmerman 1988, pp. 50.
stress, there is nothing in the production of those effects that shows that there
is anything at all defective in the agent’s will. The agent could have acted for an
appropriate moral motive and taken all the care in the world, but mother nature
likes bad things to happen because of good people. To be sure, Kant would say
that the permissibility of the action does not depend upon these effects, but on
this point, Feldman and I seem to be in agreement that Kant is wrong. I will
offer arguments against the sort of anti-consequentialist view associated with
Kant in the later discussion of norms of assertion and practical reason. Let us
bracket them for now and see if the rest of the argument holds up.
In (5), I introduce this notion of epistemic worth, a notion that is supposed
to parallel the notion of moral worth. According to Kant, an action has moral
worth only if it is done from a suitable motive. If that motive is taken to be
the motive of duty, many commentators take Kant’s view to be objectionable
insofar as it seems commendable for others to act from other kinds of motives
and it is not at all clear that there is anything admirable about the agent
who acts from a reverence for the law as opposed to the values that the moral
law is concerned with and directs us to respect, care for, produce, etc... For our
puposes, let us say that an action has moral worth iff it is creditable or laudable.
As Zimmerman stresses, it is probably best to acknowledge that while we speak
as if actions have moral worth, they do only because of the qualities of the
agents who perform them.45 What then is epistemic worth supposed to be? A
belief is supposed to have epistemic worth if the agent is motivated to hold it
because the agent both respects the right kind of reasons for holding a belief
and has exercised sufficient care in trying to respond to these reasons. Given
that the right kind of reasons for belief consist of truth-related considerations
(i.e., evidence), there is something plausible to the idea that epistemic worth
can be cashed out in broadly evidentialist terms. My objection to Feldman’s
argument is not so much that there is not something for the evidentialist to be
right about, it is that the thing that the evidentialist is right about is not the
standards that determine right or justified belief. The thought, then, behind (5)
is this. Suppose that the agent’s judgment about how to deal with a situation is
epistemically defective in some way that is a discredit to her as a person. This
would seem to indicate a lack of concern for considerations of moral importance
on the agent’s part. Surely, if I did not live up to my responsibilities in trying to
settle the question as to whether to believe I ought to φ, I cannot have lived up
to my responsibilities in trying to settle the question as to whether to φ since
the first question is transparent to the second.
I do not think that Feldman will deny (7), so by process of elimination, it
seems he must deny (6). In so doing, he needs to deny that there is a principled
connection between the normative standing of normative judgments and the
actions that these judgments rationalize. We, then, need to see if there are
counterexamples to the following principle:
If you judge that you must φ and your judgment is justified, you
cannot be obligated not to φ (EPJ).
45 Zimmerman 1988, pp. 38.
As we saw earlier in our discussion of phenomenal conservatism, there seem to

be a variety of counterexamples to EPJ. Suppose, for example, that you think
that if it seems to you that you ought to φ based on your own moral intuitions
and you have no reason to suspect that they are unreliable, you can justifiably
judge that you ought to φ. Given this understanding of justified judgment and
EPJ, we would have to sanction all manner of morally abhorrent behavior. (We
would, for example, have to say that the cannibals and terrorists discussed earlier
acted rightly if they followed their conscience and were isolated from the kinds
of defeating considerations that we take to show that their actions are morally
abhorrent). Or, suppose you think that you justifiably judge that you should φ
if you are the same on the inside as someone who knows she ought to φ. Such
a view might not imply that it is consistent with your moral obligations to eat
strangers or kill innocents in an attempt to live up to your religious obligations,
but such a view implies that the unknown and untoward effects of your actions
have no bearing on their justificatory status.46
Someone might say that she lacks the intuitions about these kinds of cases
to be moved by the argument, but this was not Feldman’s reaction. At one
point, he agreed that some of these cases constitute counterexamples to EPJ.
Because of this, he has to reject (6). He offers this rationale for adopting a view
on which EEJ is true but EPJ is false:
With regard to ethical justification, it is clear that there are cases in
which what is moderately subjectively justified differs from what is
objectively justified. Such cases occur when a person has reasonable
but incorrect beliefs about what action is morally best. One may
have reasonable but incorrect beliefs about what the consequences
of an action will be or about the values of correctly identified con-
sequences. In such cases, one may have a good reason to believe
that an action is best when it actually is not best. Thus, there can
be actions that are subjectively justified, but are not objectively
justified.47
46 Someone might object to this on the grounds that they reject the idea of an obligation an
agent non-culpably knew nothing about. Unknowable obligations, I take it, if there are such
things, are good candidates for obligations we know nothing about. There are good reasons
to believe in such obligations. Think of cases of conflicting pro tanto duties. Parents have to
decide when to let their children out to play. They might know that if they let them out too
often, they expose them to too much danger. If they let them out too little, they shelter them.
We can imagine a series of neighborhoods. At one end is the perfectly safe neighborhood and
at the other is an exceptionally dangerous neighborhood. Any parent who keeps their kids in
in the first is clearly sheltering them. Any parent who keeps their kinds in in the second is
probably letting them out to collect the insurance. There will be neighborhoods in the middle
where it is impossible to know whether it is too dangerous to let the kids play unsupervised.
This is not ignorance that can be remedied by giving the parents more crime statistics, more
reports of strangers in the area, a second look at the pollen count or the numbers of bees. I
am tempted to say that parents forced to decide will often make the wrong decision but be
nothing less than perfectly reasonable in their decision. Their acts, we might say, have just as
much moral worth as the permissible acts of parents in ever so slightly safer neighborhoods
who act from a similar sense of concern for their children. Both Jussi Suikkanen and Ralph
Wedgwood have made similar points in conversation. Also, see Sorensen (1995).
47 Feldman 1988, pp. 415.
On his view, when it comes to action, the evidence offers a kind of subjective jus-
tification. Objective justification or permissibility depends upon further factors
beyond the subject’s evidence. He says that this is not so for belief:
[A]ll subjectively justified beliefs are also objectively justified. When-
ever one is subjectively justified in believing p, then one is objectively
justified in believing that one’s reasons for believing p are good ones.
But then the evidence for this [second-order] belief together with the
reasons for thinking that those are good reasons constitute an objec-
tively good reason for believing p. Hence, . . . moderate subjective
justification implies objective justification.48
Let me register two concerns. His position seems to be that evidence constitutes
a reason for belief without constituting a reason for action. This might be so,
but someone who accepts EPJ does not need to say that evidence for forming
a belief provides reasons for action. Instead, someone who accepts EPJ can
say that the reasons there are not to act also constitute reasons not to believe.
If those reasons include reasons that the subject is unaware of, someone who
accepts EPJ can accept everything that Feldman says and simply add that the
reason EPJ is true is that when you ought not φ but reasonably believe that
you must φ the very same reasons that Feldman agrees prevent your action from
being justified (or, as he puts it, “objectively justified”) prevent your belief from
being justified. While some could object to the suggestion that a reason can be
decisive even if the subject is not aware of it, this is not just my view. This
is Feldman’s view. He thinks that there can be a decsive case against acting
even if the reasons not to act are inaccessible to the agent at the time of action.
Given that this is his view, it is reasonable to ask (again) why this is not so for
reasons that bear on whether to believe.
Dialectically, this puts him in an awkward spot. With very little honest toil,
we can construct an argument for EPJ that parallels his argument for EEJ:
(8) If you form the belief you must φ and the evidence supports this
belief, you form the epistemically rational belief and you perform
a practically rational action if you φ in accordance with your own
justified judgment.
(9) If you perform a practically rational action by φ-ing, you thereby
maximize something of practical value.
(10) If you maximize something of practical value, it is not wrong
for you to φ.
(C4) Thus, it is right always and everywhere for you to act on the
belief that is based on sufficient evidence.
Now, we can now get down to the theft. If Feldman thinks EPJ is false, he has to
reject one of the premises. (8) and (9) are perfectly sound. If it is epistemically
rational to believe that you must φ, it cannot be practically irrational for you
to act on that very judgment. Here, I agree with Foley who maintains:
48 Feldman 1988, pp. 416.
A plan (decision, action, strategy, belief, etc. . . ) is rational in sense

X for an individual if it is epistemically rational for the individual
to believe that it would acceptably contribute to his or her goals of
type X.49
The argument’s weakest link is (10). The value realized by acting reasonably,
rationally, responsibly, etc..., is a good, but either the good in question does not
offer a pro tanto justification for performing the act in question or it provides
merely a pro tanto justification for performing the act in question. Either way,
such a value does not indicate that you act rightly when you act reasonably.
Since this value is the practical analogue of the value Feldman says signifies that
the belief that has that value is justified, there is the very real possibility that
that value either does not provide a pro tanto justification for belief or provides
nothing more than a pro tanto justification. Since ER asserts that you have all
things considered justification in believing on the evidence, not just some pro
tanto justification.
While it is true that you should never believe without sufficient evidence,
arming yourself with sufficient evidence still leaves you exposed to moral luck.
Arguably, it also leaves you exposed to epistemic luck. Owing to bad luck, you
can believe on the evidence, act on your conscience, and just as Feldman thinks
that your practical conscience can lead you astray, your epistemic conscience can
lead you astray. By following the evidence and failing to live up to your practical
obligations, you did not live up to your epistemic obligations. We know why
Feldman rejects that last bit. He rejects that because it is reasonable to believe
on the evidence, there is something good about believing on the evidence, and
if something is so good, it cannot be wrong. It is not clear what is doing the
work for Feldman here. It cannot be the mere fact that there is something good
about forming the belief. There is something good about doing the wrong thing
if you reasonably thought it was right. Whatever that was, it was not a value
that called for you to act as you did and therein justified the action. In the
epistemic case, someone opposed to evidentialism can say that it is not true in
general that the rational is the mark of the permissible and so it is not true in
general that the value that attaches to the reasonable or the rational response
to the situation justifies that response. This has to be some special feature of
the epistemic realm. If Feldman concedes that this is a brute fact, this just is to
concede that the arguments for evidentialism fail. If Feldman is going to offer
evidence for ER, he has to explain what is special about epistemic rationality.
While I shall argue in later chapters that the evidentialist view of epistemic
rights and wrongs is wrong, at this stage I think it is safe to say that the value-
driven argument for evidentialism has been undermined. No justification has
been given for the argument’s crucial premises (2) and (3) and we are left with
an argument by analogy that suggests that they are false. One possibility is that
Feldman thinks the epistemic case differs from the practical case in that in the
practical case, there can be reasons not to act that do not supervene upon your
evidence but in the epistemic case, everything of epistemic value supervenes
49 Foley 2001, pp. 218.
upon the evidence. Stated crudely, the idea is that bad things can happen when
good people follow the evidence and act on the available intelligence. Iraq is a
bad example of this, but we knew what people were up to when they conceded
that mistakes were made. When people follow the evidence and believe on it,
however, nothing bad can come of it because facts that do not supervene upon
the evidence have no epistemic value whatsoever.
Developed in this way, the view is not that a justification does not aspire
to show that there are not decisive reasons not to believe or that a belief can
be justified even if a justification fails in its aspirations. Instead, the idea is
that there is nothing outside the head of epistemic value that could threaten
the normative standing of a belief whereas there are things outside the head
of practical value (e.g., facts about the needs of others and what would serve
our needs) that threaten a justification. Developed in this way, the evidentialist
would have to deny the veritist claims that false belief is epistemically bad and
true belief is epistemically good. The truth or falsity of most of what we believe
does not supervene upon the evidence. The evidentialist would have to say that
false beliefs have no epistemic disvalue, per se. And, they would have to say
that true beliefs have no epistemic value, per se. Evidentialists would need to
explain away the intuitions that seem to support the veritist’s claims to the
contrary. This would also create a problem internal to the evidentialist view.
Truth is an essential goal of doxastic deliberation. This is why following the
evidence concerning the truth value of a belief is the appropriate way to pursue
your epistemic goals and following the evidence that a belief has other desirable
features not related to truth is an inappropriate way to pursue your epistemic
goals. If no good can come of achieving your epistemic goals, why would it be
good to pursue them in the right way? If nothing bad can come of failing to
achieve your epistemic goals, why would it be bad to pursue them in the wrong
way?
The only way I can see to fix Feldman’s argument is to assert that the
rational is the mark of the permissible in both the practical and theoretical
domain. I do not explore this option in this chapter, but we shall return to it
later. Insofar as he denies that the rational is the mark of the permissible in
some domains, Feldman leaves his argument susceptible to the objection that
he is wrong about the proper mode of response to the value that attaches to
rational belief.
Anyone who agrees that the value of rational action does not justify the
rational action when there is a case to be made against acting that the agent is
ignorant of has to say that the value that attaches to rational action calls for a
response other than performing the action. Why? Because every value calls for
some response or other and the value in question is a value that impermissible
actions can have. When the morally conscientious agent tries but fails to live up
to her obligations because she is non-culpably ignorant of what it would have
taken to meet her obligations, we can say that her actions had moral worth. If
an agent’s actions have moral worth, the action might have only seemed to be
justified and so only seemed to have the values that justified it. That does not
mean that the action was valueless, only that the values the action had call for
a different response, such as honoring, commending, respecting, and forgiving.

Intuitively, actions that call for these responses are good in some way. These
actions are better than overtly similar actions that were carried out recklessly,
maliciously, irresponsibly, etc. . . And, it seems that the value morally worthy
actions have is not purely instrumental. If the value were purely instrumental,
we would not say that it was better for an agent to φ responsibly than for an
agent to φ with indifference, or from the motive of profit, or from fear, etc. . . If
this much is right, perhaps we should say something similar for belief.
3.4 The Right and the Good

Epistemologists interested in epistemic value have focused primarily on the value
of knowledge and most of what has been said about the value of justification
has been said in the context of this discussion. The reader is probably familiar
with the basic contours of the discussion surrounding the Meno Problem.50
Knowledge is more valuable than mere true belief. It seems that way to many
of us, at least. We look for an explanation as to why this is. Could it be that
knowledge is tied down in such a way that it serves as a better basis for action
than true belief? Possibly, but if it is, this only shows that knowledge has an
instrumental and a practical value that true belief does not. This is the wrong
kind of value twice over.
A second explanation seems more promising. Knowledge is more valuable
than mere true belief because the former involves justification and the latter
need not. Unlike the previous attempt at explanation, this one at least involves
something of epistemic value. The explanation rests on three assumptions. The
first is that justification is not necessary for true belief but it is necessary for
knowledge. The second is that justification is always valuable from the epistemic
point of view. The third is that it is something we value for its own sake. So,
justification seems to be the thing we are after, the thing that explains why
knowledge is epistemically better than mere true belief. We know that the
discussion does not end here. Even if this explanation works, we still do not
know why knowledge is distinctively valuable (i.e., more valuable than any of its
proper parts). If we try to solve our first value problem by appeal to the value
of justification, a second problem crops up because there is more to knowledge
than just justified, true belief and (allegedly) more value in knowledge than
contained in justified, true belief. Even if we bracket this second value problem,
the assumptions we need to solve our first value problem are controversial.
Does justification have a non-instrumental epistemic value? It is hard to
say. Why do we value justification? The obvious reason is that justification is
supposed to be connected to truth and the truth is something we value. The
nature of this connection is hard to specify, but it is the connection to truth
that distinguishes epistemic justification from other kinds of justification.51 The
problem we now face is that if we cash out the value of justification in terms of
50 See Kvanvig 2003 for an introduction.
51 For discussion, see Cohen 1984.
the value of the truth-connection, we might spoil our explanation of the value
of knowledge.
To see what the worry is, suppose the value of knowledge, VK , is greater
than the value of true belief, VTB . We can appeal to the value of justified
belief, VJB , to explain this, but only if VJB is not itself contained in VTB . On
its face, it does seem plausible to maintain that VJB is not contained in VTB .
There can be unjustified, true beliefs, after all. It seems better to believe with
justification than without it. While these claims seem plausible, they can be
contested. Should we say that justified belief is valuable because of the value of
the truth-connection (i.e., the connection between a belief and the truth that
must be in place for the belief to be justified), VJTC , or should we say that
there is more to VJB than VJTC ? BonJour seems to think we value justification
precisely because we value the truth, and this suggests that VJB might just be
VJTC :
Why is such justification something to be sought and valued? Once
the question is posed this way, the following answer seems obviously
correct . . . We want our beliefs to correctly and accurately depict
the world . . . The basic role of justification is that of a means to
truth . . . We cannot, in most cases at least, bring it about directly
that our beliefs are true, but we can presumably bring it about di-
rectly . . . that they are epistemically justified. And, if our standards
of epistemic justification are appropriately chosen, bringing it about
that our beliefs are epistemically justified will tend to bring it about
. . . that they are true . . . It is only if we have some reason for
thinking that epistemic justification constitutes a path to truth that
we as cognitive beings have any motive for preferring epistemically
justified beliefs to epistemically unjustified ones. Epistemic justifi-
cation is therefore in the final analysis only an instrumental value,
not an intrinsic one.52
If he is right and VJB = VJTC , we face a problem that is structurally similar
to the swamping problem.53 VJTC appears to be an instrumental value. Unless
VJTC is also a non-instrumental value, we cannot explain how VJB could be a
non-instrumental value. If VJTC is merely instrumentally valuable (i.e., valuable
only because of the connection to truth), then when a belief is false, that belief
will not have some positive value by virtue of having VJTC . (Plausibly, if the
only value Fs ever have is instrumental value, on those occasions where nothing
of intrinsic value is realized, F has no value at all. To borrow an example, if
a horrible shot of espresso is made using the finest equipment, this shot is not
better for this reason than a qualitatively indistinguishable shot you could get
at the gas station.54 ) If VJTC is merely instrumentally valuable, then when a
belief is true, it should not have some additional value by virtue of possessing
VJTC , a value that unjustified, true beliefs lack. (If the only value the Fs ever
52 BonJour 1985, pp. 7-8.
53 SeeSwinburne 1999 and Pritchard Forthcoming for discussion.
54 Zagzebski 2003, pp. 13.
have is instrumental value, when something of intrinsic value is realized, the

presence of an F does not increase the total value. If a fantastic shot of espresso
is made using the poorest equipment, it is not less valuable than a qualitatively
indistinguishable shot of espresso made using the finest equipment.) So, if VJB
= VJTC , it seems we cannot explain VK > VTB in terms of VJB . So, it seems we
are under some pressure to explain how it could be that VJB > VJTC in order
to properly explain how it is that VK > VTB in terms of VJB and also under
some pressure to concede that VJB = VJTC .
It is always good to believe with justification. It is always better from the
epistemic point of view to believe with justification than to believe without
it. Both claims strike me as intuitive, but BonJour seems to suggest they are
mistaken. Perhaps he would argue as follows:
(1) The value of justification derives entirely from the connection
between justification and truth.
(2) On those occasions where a justified belief is false, the justified
belief is valueless.
(C) It is not always good to believe with justification.
For reasons that will emerge later, I reject (2). Justified beliefs do not lack
epistemic value when they are false because false beliefs are never justified.
So, even if (1) is true and the justified belief derives its value entirely from its
connection to truth, (1) does not compel us to accept the argument’s conclusion.
Having said that, if you accept (1) and reject (2) on these grounds, you face a
different problem. Does my view imply that unjustified, true beliefs have the
same value as justified, true beliefs? If so, the view implies that it is not better
to believe with justification than without it. In terms of their epistemic value,
true beliefs are on par whether they are justified or not.
In responding to BonJour’s argument, I want to expand on some remarks
from the previous section. Suppose that the conditions that justify believing
justify reasoning from that belief. In justifying a subject’s belief, these condi-
tions justify her treating what she believes as a reason for further beliefs and,
possibly, actions rationalized by these beliefs. Our agent’s action could have
had moral worth even if she did not live up to her obligations, provided that
she was suitably motivated in acting. To be suitably motivated in acting, the
agent had to exercise sufficient care in reasoning about what to do. To exercise
sufficient care, it had to be epistemically rational for the agent to believe that
she was dealing with her situation in the way she should. If the value that at-
tached to rational belief justified belief, it would justify acting on the belief, but
we are assuming that there is nothing that justified the agent’s actions, and so
this value did not provide sufficient justification. So, the value that attaches to
rational belief is an essential component of moral worth but it is not sufficient
for the moral permissibility of the action or the epistemic permissibility of the
beliefs that rationalize the action.
The argument suggests that the value of reasonable belief is an essential
component of moral worth. Combined, the value of reasonable belief and moral
worth do not constitute a justification for action or belief, but both intuitively
seem valuable in their own way even on occasions where the agent failed to live
up to her obligations, moral and epistemic. Because of this, it seems we do
not conceive of these values as purely instrumental. However, these values are
clearly connected to further ends. An action has moral worth only if an agent
is pursuing her moral ends responsibly and a belief has epistemic worth only
if the subject is pursuing her epistemic ends responsibly. Above, we saw that
BonJour’s remarks supported the view that VJB is really nothing over and above
VJTC . We defined VJTC in such a way that no justified belief can lack VJTC . It
is a value accrues to a belief if it stands in the connection to truth necessary for
the belief to be justified. While BonJour seems to think that this means that
VJTC has to be purely instrumental and that the same would hold true for VJB ,
the examples of moral worth and epistemic worth should give us pause. Could
epistemic worth be a component of VJTC ? It seems so. Epistemic worth is a
necessary condition on justified belief and any belief that lacks epistemic worth
is not related to truth in the way it must be to be justified. It also seems to be
but one component of VJB since beliefs can have epistemic worth even if they
are unjustified.
In making his case for the claim that VJTC is a merely instrumental value,
BonJour explained why you seek and value justification in instrumentalist terms.
You seek and value justification as a means to truth, he said. He is right. I
would add that from the perspective of first-person deliberation, your primary
concern seems to be the truth of your beliefs. However, it is one thing to say that
the aspect of justification you care about when trying to settle the question as
to whether p is whether the considerations you have in mind correctly settle the
question. It is another thing entirely to say this is the only aspect of justification
you care about. When you settle to your satisfaction the question as to whether
p only to later discover that ∼p, you might not value the justification you had
because you see know that it misled you. Contrast this case with the case
in which you settle to your satisfaction the question as whether p only later
to be told that while technically you were correct, you were unreasonable or
irresponsible in concluding that p is true given the reasons that were available
to you. On the instrumentalist picture, what concern would that be to you?
If the instrumentalist picture is right, it would be of no value at all, but this
does not ring true. In trying to settle the question as to whether p, you might
settle it to your satisfaction without shifting your focus from the truth to you
to consider how well you live up to your responsibilities, but in retrospect if
someone says that your beliefs lack epistemic worth, you care. At least, you
should. Notice, however, that you care about whether you pursued the truth
responsibly and reasonably because you care about the truth. So, even if there
is nothing more to VJB than VJTC , it does not follow that VJB is a purely
instrumental value. VJTC might be more than some mere instrumental value.
There might be many aspects of the truth-connection necessary for justification
and the subjective dimension that is connected to epistemic worth is part of it.
If VJTC accrues only to beliefs that have epistemic worth, we can explain why
VJB is not swamped by VTB . Not every true belief has epistemic worth and
not every belief with epistemic worth is true. Both truth and epistemic worth
confer value upon a belief, and so we might be able to explain VK > VTB by
appeal to a non-instrumental value contained in all justified beliefs.
One problem remains. In explaining why it is always good in some respect to
believe with justification, we really have not explained why it is always better
to believe with justification than without it. Epistemic worth is, arguably,
necessary for justified belief but not sufficient for justified belief. So, we have to
explain why it is better to believe with justification than to simply have beliefs
that have epistemic worth. It is better to believe with justification because it
is always better to live up to your obligations. Better even than failing to live
up to your obligations in admirable ways. You should never believe without
justification, but if you believe with justification, you have lived up to your
obligations. This is not because there is some independent notion of goodness
that determines what our obligations are. I want to suggest that this is because
our judgments about what is good or best often is a reflection of our independent
sense of what is right and what there is reason to do. This is a point that Foot
stressed in her critical discussions of consequentialist moral thought.55 She
would say that we do not have a grip on the idea good states of affairs that were
not good from particular points of view. I think she might have overstated her
case. Even if we have a grip on this idea, the fact is that our talk of what is
“good”, “better”, and “best” does not reflect our commitment to the idea that
the right is what is best from no particular point of view and our intuitions do
not suggest that moving to this mode of moral thinking would improve things.
Suppose that while backing out of your driveway you hit your neighbor’s
kid, Alice, with your car. She has a badly twisted ankle. You rush her to the
emergency room. Meanwhile, on the far side of town, I back my car into my
neighbor’s kid, Bernice. She has a badly twisted ankle and I take Bernice to the
emergency room. Alice and Bernice are both in pain, they respond equally well
to pain relievers, but there is only one pill left in the waiting room. You say
to yourself, “If there is only one pill to give, it would be better for me to give
it to Alice”. I say to myself, “It would be better for me to give it to Bernice”.
The nurse, meanwhile, says to herself, “It is not better to give it to Alice or
Bernice because they are both in equally bad shape, so it is best for me to flip
a coin”. I do not think that anyone is getting it wrong and I do not there is any
disagreement here. These judgments about what is better to do or best to do
simply reflect the agent’s recognition of the reasons there are for them to act
and not some measure of the value of the states of affairs they can bring about.
You know full well that it would be just as good for Alice and Bernice to get
the pill, but it is better for you to look after Alice first. It is better because
you have a reparative duty to Alice, not just some duty of beneficence. The
nurse knows full well that it would be just as good for Alice to get the pill as
Bernice, and since she has no prior relation to either of them (and an overriding
obligation to ignore such prior relations for professional reasons) this is why she
knows she should flip a coin if she decides who gets the pill.
55 See Foot 1985.
We can say the following about the case:

(3) You speak truthfully if you say, “I know it would be best to help
Alice. So, that is what I should do.”
(4) I speak truthfully if I say, “I know it would be best to help
Bernice. So, that is what I should do.”
(5) The nurse speaks truthfully if she says, “I know it would not be
better to help Alice or Bernice first, so it would be best to flip a
coin. That is what I should do.”
Owing to the factivity of “knows”, someone who thinks that judgments about
what is better or best are true in virtue of some facts about the good charac-
terized impersonally and without taking account of the speaker’s interpersonal
relationships will have to say that only the nurse speaks truthfully. This goes
against intuition. Not only do (3)-(5) seem true, it seems the speakers could
knowingly acknowledge that these are all true. This further reinforces the idea
that while judgments about what is best and what should be done might go
together, this is not because an impersonal notion of the best is doing all the
work in picking out our obligations.
If this is right and our judgments about rightness are often reflected in
judgments about what is better or best, this does a nice job explaining the
intuition that it is always better to believe with justification than without it.
But, it also suggests that this intuition is not the sort of intuition about value
that constrains a theory of justification. Here is where things stand. I have
tried to accommodate two intuitions about epistemic value. The first is that
it is always good to believe with justification. If this is true, I have suggested
that it is true because epistemic worth is an essential part of the justification
of a belief and it has a kind of non-instrumental value. While this notion of
epistemic worth is the sort of thing that should be cashed out in internalist
terms, this does not mean we should be internalists about justification. It only
means we should be internalists about one aspect of justification. Arguably,
epistemic worth is necessary for permissibility but not sufficient just as moral
worth is not sufficient for permissible action. I have not argued that there
is an additional externalist element to justified belief, but it seems the value-
theoretic considerations leave that possibility open. Whether we are internalists
who think there is nothing more to epistemic justification than epistemic worth
or we deny this, so long as epistemic worth is regarded as a necessary condition
for justification, this takes care of our first intuition. As for our second guiding
intuition, it is always better to believe with justification precisely because it is
better to live up to your obligations. This will not help us determine what our
epistemic obligations are, so this is why we need to turn to something else if we
are to make any headway.
Chapter 4
Evidence (I)
4.1 Introduction
According to one version of evidentialism, your beliefs are justified iff your beliefs
are supported by your evidence “on balance”.1 This view purports to tell us
something about the normative significance of evidence. Your evidence is what
gives you the right to believe. This right can be lost only if your evidence
changes and it no longer supports your belief. It tells you little else. It does not
say what your evidence consists of or what evidential support amounts to. Like
this evidentialist view, I have nothing to say about the nature of the evidential
support relation.2 What I shall do with this chapter is defend an externalist
conception of evidence. Like the evidentialists, I hope that the views defended
here can be combined with any plausible account of evidential support. With
this account of evidence in place, I can start to build a case against internalism
about the justification of belief.
If the acccount of evidence defended here is correct, I shall argue that some
important facts about the justification of your beliefs does not supervene upon
your mental states.3 The evidentialists think that two kinds of relation deter-
mine the justificatory standing of a belief. If a belief is not justifiably held, it is
either because it is not based on sufficiently good evidence or the believer is not
sufficiently sensitive to the evidence she has against her belief. Building on the
arguments of this chapter, I shall argue in the next chapter that this position
is a mistake. The relations between evidence and belief matter, but epistemic
appraisal does not focus on these relations to the exclusion of further relations
between your beliefs and the matters that your beliefs concern. This more am-
bitious argument will have to wait for the next chapter. In this chapter, I shall
lay down some foundations.
1 Conee and Feldman 2008, pp. 83.
2 To keep things simple, let us assume that p is evidence for q only if p raises the probability
of q: P(q p) > P(q).
3 In this chapter, our focus will be on non-factive mental states.
36
CHAPTER 4. EVIDENCE (I) 37
The externalist view I defend here is rather weak. I shall argue that this
supervenience thesis is false:
The conditions that determine what your beliefs are like justifica-
tionally strongly supervene upon your mental states (Supervenience
Internalism).4
Evidentialists such as Conee and Feldman are committed to Supervenience In-
ternalism because of their commitment to evidentialism and a certain conception
of evidence:
The conditions that determine (i) what your evidence consists of and
(ii) the support relations that hold between your beliefs and your
evidence strongly supervene upon your mental states (Mentalism).5
If two subjects’ beliefs are perfectly alike justificationally, these subjects have
the same justified beliefs, these beliefs are justified to the same degree, and
their beliefs stand in the same relations to the same justifying reasons. If two
subjects are perfectly alike evidentially, these subjects have to have precisely the
same evidence, their evidence must support the same beliefs for them, and the
degree of support provided by their evidence has to be the same. Supervenience
Internalism denies that it is possible for two internal duplicates to have different
evidence supporting their beliefs. This view is mistaken.
4.2 Getting a Fix on the Notion

The notion of evidence is notoriously slippery and so it is difficult to know how
to go about testing rival accounts. There are some general guidelines most
of us can live with. While each of these proposed guidelines is negotiable,
most of us would be skeptical of any view that violated them without good
reason. First, there are some general platitudes about evidence that many
of us recognize as platitudes. For example, your evidence is supposed to be
what you have to go on or what you can properly rely on in forming a view.6
This suggests that your evidence has to be relatively more accessible than the
unsettled matters you hope to resolve by following your evidence where it leads.
Second, there seems to be some agreement that your evidence plays certain roles
in inference, explanation, and the justification of belief. So, it would be a mark
against a theory, for example, if it denied that there was any sort of connection
between your evidence and the justification you have for your beliefs. Even the
externalists who deny that subjects who have the same evidence and believe
on the same evidence end up with the same justified beliefs agree that when
you have evidence against your beliefs, this threatens the justificatory status of
4 In addition to Conee and Feldman 2004, you can find defenses of Supervenience Internal-
ism in Audi 2001, Pollock and Cruz 1999, and Wedgwood 2002.
5 Conee and Feldman 2004.
6 See Conee and Feldman 2008, Kelly 2008, and Silins 2005.
those beliefs.7 Third, we have intuitions about what your evidence consists of
and what is consistent with your evidence. A conflict with intuition is a mark
against a view. Fourth and finally, there seems to be an important connection
between evidence and justifying reasons for belief. Your evidence does not (just)
explain why you believe. We look to your evidence to try to determine whether
you are reasonable or rational in your beliefs. This should be enough to get us
started.
One caveat is in order. The views defended here should be consistent with
plausible views about the nature of evidential support, but as I have nothing to
say about the relation of evidential support, I shall not be terribly concerned
with what it takes for something to be evidence for something else. Instead,
I shall be concerned with what evidence is and what it takes to have it. I
am not boldly suggesting that there is such a thing as evidence that is not
evidence for something or other. Nothing can be an ingredient without being an
ingredient for some possible dish.8 Just as we can speak of ingredients without
explicitly mentioning any dishes, we can speak of evidence without saying that
the evidence supports something or other.
4.3 The Refutation of Mentalism

Mentalists say that if two individuals do not differ mentally, these individuals do
not differ evidentially. As you might have guessed, the refutation of mentalism
consists of an argument that shows that it is possible for two subjects that do
not differ mentally to have different evidence to support their beliefs.
Mentalism does not tell us what evidence is, it only tells us when two indi-
viduals have the same evidence:
Necessarily, if two individuals do not differ mentally, p is part of the
first subject’s body of evidence and plays such and such a rule in her
epistemological economy iff p is part of the second subject’s body
of evidence and plays the same role in her epistemological economy
(Mentalism+).
The basic idea behind Mentalism+ is that by holding fixed the subject’s psycho-
logical states, we hold fixed facts about which propositions are elements of this
subject’s body of evidence and the role that this evidence plays in determining
the justificatory standing of this subject’s beliefs. So formulated, Mentalism+
assumes:
Evidence consists of propositions (Propositionality).9 Propositionality is
controversial. Pollock remarks:
7 See Goldman 1986.
8 Hyman 2006, pp. 892.
9 The two most prominent defenders of Mentalism say that they are agnostic as to whether
Propositionality is true. See Conee and Feldman 2008, pp. 101. In criticizing rival views, they
never do so on the grounds that these views assume Propositionality. I would be surprised
if they thought that the only reason my attempted refutation of their view fails is that it
assumes Propositionality.
What is it that justifies a belief? ... For example, consider the case
of a person who believes there is a sheep in the field because he sees
a dog that looks very much like a sheep–so much like a sheep that
anyone would be justified in taking it to be a sheep until he examined
it quite closely. One is apt to say that it is the fact that the dog
looks like a sheep that justifies the person in thinking that there is
a sheep in the field. But this is misleading. What is important in
deciding whether the person is justified in his belief is not the fact
itself but rather the person’s belief that it is a fact. After all, if
the person did not believe that the dog looked like a sheep, then
his belief that there was a sheep in the field would not be justified,
although it would of course still be a fact that the dog looked like a
sheep. Thus we must say that what justifies a belief is always another
belief. It is a person’s “doxastic state” which determines which of his
beliefs are justified. Of course, we can still talk about facts, states of
affairs, etc., justifying beliefs, but this must be understood in terms
of beliefs justifying beliefs.10
It seems that Pollock denies that evidence or a reason to believe could consist
be a fact or a true proposition because if you subtracted a belief, that fact
or true proposition could not justify anything. This is true, but Pollock’s test
for determining whether something could be a reason is too crude insofar as it
does not distinguish between reasons and enabling conditions. Suppose there
is a difference between the reason and that by virtue of which the reason is a
reason. Subtract conditions that fall into that second category and there would
be no reason, but we have stipulated that the conditions in this category are
not reasons.11
Even if Pollock’s argument against Propositionality is not persuasive, it is
fair to ask for reasons to believe that evidence consists of facts or propositions
rather than mental states or events. Evidence seems to be a kind of normative
10 Pollock 1974, pp. 25.
11 For a useful discussion of the difference between reasons and enabling conditions, see
Dancy 2004. In a recent paper, Turri argues that mental states are not enablers that enable
us to have facts or true propositions as reasons. He suggests that if someone asks why some
event transpired, it is strange to then ask why the enabling conditions obtained (e.g., it is
strange to ask, after being told that someone’s belongings were destroyed by a fire, why there
was oxygen in the apartment). Then, he observes that it is not strange to ask someone who
has just explained why something has occurred by citing the fact that p why they believe p.
This, he suggests, shows that mental states are not enabling conditions. See his 2009, pp. 504.
Myself, I do not think it is always strange to ask this (e.g., we might have expected that the
oxygen had been evacuated from the apartment). He says that if a speaker tries to explain
p in terms of q and we ask about X, if it is natural to ask about X, X is a reason rather
than an enabling condition. Suppose someone says that they believe Tiger will win because
he excels at putting. Turri takes the it that since it is perfectly natural to ask the speaker
why he believes Tiger excels at putting, this belief is a reason why (for?) the speaker to hold
the further belief about Tiger’s chances of success rather than an enabling condition. I do not
find this convincing. Suppose we modify the example slightly. The speaker says that Tiger
won because he excels at putting. Someone might ask why someone thinks that Tiger excels
at putting and the question might be natural, but nobody thinks that this fact about the
speaker’s attitudes has anything to do with explaining anything about Tiger’s performance.
reason, a reason to believe rather than something as opposed to (just) a reason

why we believe something. Normative reasons for action are, by their very
nature, relational beasts. It is hard to imagine a world in which there are
reasons that are not reasons for such and such an agent. How does something
become a reason for an agent to do or avoid such and such a thing in such and
such circumstances? There might be many paths to reasonhood, but the most
obvious way something gets to be a reason is by counting in favor or counting
against.12 While some reasons might not count in favor of anything at all, most
of the reasons I can think of are reasons precisely because they count in favor of
doing something or count against the doing of it. From here, it is a short step
to the rejection of the rival view that treats normative reasons as psychological
states of the agent. Unless we all harbor systematic and massive confusions
about what counts in favor of acting, the things that count in favor of, say,
lending a hand, are facts having to do with who needs help and how they can
be helped.
Once this point is granted, it is hard to deny Weak Propositionality:
Some of our evidence consists of propositions (Weak Propositional-
ity).
The reason is that the things we take to bear on whether to act are very often
the very same things we take to bear on whether to believe. As Fantl and
McGrath observe, “Whether it is an action, an intention, or plan, or even an
‘ought’ judgment, still, we draw practical conclusions from the same premises
from which we draw theoretical conclusions.” 13 If you know that I always vote
for candidates from the same political party and ask what possible reason I
could have to do that, the reasons I will offer are the very same reasons I will
offer if you then ask why I believe I should vote that way. In trying to settle
a question about what to do and a question about what to believe, when the
belief that settles for me the question about what to do, I will draw upon the
same reasons in practical and theoretical deliberation. If practical reasons are
picked out by that-clauses, a great number of epistemic reasons will be picked
out in just the same way. To deny that reasons for action or belief consist of
the facts that figure in deliberation is to charge us with a kind of error it would
be implausible to say we make. Either we would have to be wrong in taking the
favorers to consitute an important class of reasons or wrong in taking facts to
be what count in favor.
Someone could try to argue from Weak Propositionality to Propositionality
by arguing against possible pluralist views that recognize both propositional
and non-propositional evidence. In the context of arguing against Mentalism,
this would be a distraction. A sound argument against Mentalism+ constitutes
a sound argument against Mentalism given the assumption of Weak Proposi-
tionality. The pluralist who recognizes both propositional and non-propositional
evidence has to concede that if Mentalism+ is false, some facts about an individ-
ual’s evidence do not supervene upon the facts about this subject’s mental life.
12 See Dancy 2000.
13 Fantl and McGrath 2009, pp. 73, fn. 16. See also Gibbons Forthcoming.
If two subjects differ only in that one subject has less propositional evidence for
her beliefs than the other, we should expect that her beliefs are supported to
a lesser degree even if non-propositional evidence can provide some evidential
support.
With these preliminary points out of the way, here is the refutation of Men-
talism+:
(1) My evidence includes the proposition that I have hands.
(2) If my evidence includes the proposition that I have hands, I have
hands.
(3) The fact that I have hands does not supervene upon facts about
my mental states.
(C) Thus, it is possible for two individuals to differ evidentially with-
out differing mentally–the proposition that I have hands is not
evidence that my handless mental duplicates have.
There are not many proofs in philosophy, but we can hope for the occassional
valid argument with premises supported by argument. Anyone who accepts
Supervenience Internalism, Mentalism, or Mentalism+ has to reject one of the
premises. On the account of evidence defended below, each of the premises
turns out to be true.
4.3.1 FactivityE
In this section, I shall argue that evidence consists of true propositions:
If p is part of your evidence, p is true (FactivityE ).
Evidence consists of propositions, but only true propositions constitute evi-
dence. The metaphysics of this seems to makes sense. Saying that a proposi-
tion constitutes evidence only if it stands in a relation to something else (e.g.,
an individual’s attitudes, things that ensure that the proposition is true, etc...)
is kind of like saying that a hunk of matter makes for a statue, but only if it
stands in some relation to a sculptor. If the reader prefers to think of evidence
as consisting of facts, I should say now that I do not have a view concerning the
relation between facts and true propositions. The arguments here can be taken
as supporting the disjunctive proposition that pieces of evidence are either true
propositions or facts. My main concern is to argue that ascriptions of evidence
that say that p is part of someone’s evidence are true only if p is true.
Mentalists do not need to deny FactivityE . There are truths that supervene
upon your mental states, after all. For them, such true propositions are certainly
fair game. Remember that nothing I argue for in this section directly contradicts
Mentalism, Mentalism+, or Supervenience Internalism. The arguments that
support FactivityE do not rest on any particularly contentious claims about our
knowledge of the external world or the nature of mind. You can construct a
compelling argument for FactivityE on purely linguistic grounds. This is where
we will start.
4.3.1.1 The Linguistic Case for FactivityE

When he was younger, Unger argued that nothing you ever do, think, or feel
could be rational. Nothing could be rational, he thought, because nothing can
be rational without normative reasons. You have no normative reasons, he
thought, because something cannot be a reason for you to act, to believe, or
to feel unless you know that it is true. You could only know p if there is not
something else you could be more certain of, so the chances are good that you
do not know any of the propositions you take yourself to know. Buried in this
argument for universal ignorance and his argument for the irrationality of all
feeling, thought, and deed, was an interesting observation about the relation
between knowledge and reasons ascriptions:
If p is your reason for V-ing, you know p (Known Reasons).14
What was his evidence for Known Reasons? Consider:
(1) I’m going to the store for the reason that I’m out of candlesticks,
but I don’t know I’m out of candlesticks.
(2) I believe I should go to the store for the reason that I’m out of
candlesticks, but I don’t know I’m out of candlesticks.
Both (1) and (2) are clearly defective. They seem contradictory. Is this evidence
for Known Reasons? Not good evidence, you might think, because this seems
contradictory, too:
(3) I’m out of candlesticks, but I don’t know that I am.
This is just a Moorean absurdity. It seems contradictory (in some sense), but
the proposition it expresses could be true. This is evidenced by the fact that if
we restate (3) in the third-person it is perfectly felicitous:
(3’) Plum is out of candlesticks, but she does not know she is.
The appearance of contradiction, you might think is incredibly weak evidence
for Known Reasons.
On the hypothesis that (1) and (2) are defective for the reason that (3) is,
(1) and (2) should be felicitous if restated in the third-person:
(1’) Plum is going to the store for the reason that she’s out of can-
dlesticks, but she doesn’t know that she’s out of candlesticks.
(2’) Plum believes she should go to the store for the reason that
she’s out of candlesticks, but she doesn’t know she’s out of can-
dlesticks.
Even if we restate them in the third-person, they seem defective. So, we do have
some evidence for Known Reasons and FactivityE . The obvious explanation as
14 Unger 1975.
to why these seem contradictory is that they are. If p is Plum’s reason for φ-ing
only if she knows p, p is among Plum’s reasons for φ-ing only if p is true.
While it would make things easier for me if there was a convincing argument
for Known Reasons, but there are reasons to be skeptical. Known Reasons also
predicts that these could be a successful challenges to (1) and (2):
(1c) Plum believed that she was out of candlesticks, but her belief
was not justified. There was tons of misleading evidence that
we planted in the hopes of tricking her into thinking that she
had candlesticks. She knew of the evidence, but she completely
ignored it. So, her reason for going wasn’t that she was out of
candlesticks.
(2c) Plum was out of candlesticks and she also reasonably believed
that she was. But, she believed this on the basis of a note she
reasonably but mistakenly thought was addressed to her saying
that there were no candlesticks left. So, her belief was only ac-
cidentally true and so her reason for going wasn’t that she was
out of candlesticks.
While the considerations offered might show that Plum’s belief did not satisfy
the conditions necessary for knowledge, these considerations do not threaten (1)
or (2). One explanation as to why this is is that there seem to be contexts in
which we use “knows” to pick out something like a true belief. Think about
when unfaithful lovers say that others “know” of their affair even if the lovers
know that others have little evidence to support their beliefs.15 In such contexts,
the behavior of “knows” is not reliable evidence for claims about knowledge and
its ascription. So, someone could reasonably doubt that the evidence on offer
provides good support for Known Reasons. Of course, one reaction to this is
to say that if the argument for Known Reasons fails because “knows” picks out
true beliefs rather than knowledge, the linguistic evidence thought to support
Known Reasons would support FactivityE . While someone could say this, I
would rather look for further evidence. This evidence seems tainted.
This seems to be better evidence for FactivityE :
Scarlet: Do they have solid evidence against Mustard?
Green: The prosecution thinks it does. Here’s the evidence they
have: that he was the last one to see the victim alive, that he lied
about his whereabouts on the night of the crime, that his fingerprints
were on the murder weapon, and that he wrote a letter containing
details the police think only the killer could have known.
Scarlet: But, didn’t you say that he wasn’t the last person to see
him alive and his fingerprints couldn’t have been on the weapon?
Green: That’s right. He also didn’t lie about his whereabouts and
wasn’t the last one to see him alive.
15 Goldman 2002, pp. 183.
Green’s remarks seem contradictory. In stating the facts of the case as he takes
them to be, it appears that he contradicts his claims about the prosecution’s
evidence. If evidence ascriptions were non-factive, Green’s remarks should be
perfectly coherent. Ascribing evidence and denying that the propositions as-
cribed as evidence were true would be akin to ascribing false beliefs.
His remarks here seem defective as well:
Scarlet: Do they have solid evidence against Mustard?
Green: People seem to think they do. Here’s the evidence they have:
that he was the last one to see the victim alive, that he lied about
his whereabouts on the night of the crime, that his fingerprints were
on the murder weapon, and that he wrote a letter containing details
the police think only the killer could have known. That being said, I
don’t know if he’s the last one who saw the victim alive and I don’t
know if he lied.
There is nothing at all strange about assertions such as:
(4) The prosecution believes on reasonably solid evidence that Mus-
tard was the killer, but I don’t know if they are right. I want to
hear Mustard’s side of things.
But this seems defective:
(5) The prosecution knows that Mustard was the killer, but I don’t
know if they are right. I want to hear Mustard’s side of things.
This is further evidence that evidence ascriptions are factive in the way knowl-
edge ascriptions are and belief ascriptions are not.
4.3.1.2 The Objection that Proves the View

Briefly, I want to consider an objection to FactivityE . Versions of the objection
have appeared in a number of places as an objection to E=K. One version of
the objection alleges that FactivityE is incompatible with an intuitively plausi-
ble closure principle.16 Another says that cases of mistaken belief force those
who accept FactivityE to say implausible things about the evidential bases of
mistaken beliefs.17 These objections all assume something in the neighborhood
of this principle:
If you justifiably believe p, p is a justifying reason of yours that can
justify further beliefs (Justified Basis).
Let’s start with an example:
16 Comesaña and Kantin 2010.
17 See Conee and Feldman 2008, Goldman 2009, and Rizzieri Forthcoming.
I believe that nobody can enter my office (O for now) because I be-
lieve that I have just locked the door (LD for now). Let us stipulate
that I have inferred (O) from (LD). I pushed the lock in and gave it
a quick twist to the left, which usually does the trick; however, my
lock is damaged and does not work. Hence, (LD) is false.18
Rizzieri says this about his example:
If Williamson’s proposal that (E=K) is correct then (LD) cannot
serve as an evidential ground for (O). This generates problems for
(E=K). The first difficulty is that it is very plausible that (LD) does
partially constitute my evidence for (O). After all, I am justified in
believing (LD), (LD) supports (O), and an explicit inference from
(LD) is my most immediate basis or ground for (O).19
Given the features of the case, he says it is difficult to deny that LD is evidence
for O because LD renders O more probable than it would have been otherwise.20
Adding to the difficulties facing E=K and FactivityE , Comesaña and Kantin
allege that these theses are incompatible with an attractive closure principle:
If your belief in p is justified, you have sufficient justification for
believing the obvious consequences of p and can justifiably believe
these consequences if these beliefs are arrived at by means of com-
petent deduction (J-Closure).
They say that if we assume that the proposition that p can justify you in
believing something only if it is part of your evidence, E=K implies that the
following is true:
The proposition that p justifies you in believing that q only if you
knows that p (E=K1 ).
They then argue against E=K1 and E=K as follows:
[S]uppose that Terry is a recently envatted human. On the basis
of an experience very much like the one that you have when you
are facing a dog in your neighborhood, Terry believes that there
is a dog in her neighborhood. Of course, Terry doesn’t know that
there is a dog in her neighborhood (if only because it is false, let
us suppose, but not only because of that), but she is still justified
in believing it. She then deduces from that belief that there is a
nonhuman animal in her neighborhood. Isn’t she thereby justified
in believing that there is a non-human animal in her neighborhood?
J-Closure (and intuition) say ‘‘Yes,’’ E=K1 says ‘‘No.’’ But the
proposition that there is a non-human animal in the neighborhood
is a lightweight implication of the proposition that there is a dog in
18 Rizzieri
Forthcoming, pp. 2.
19 Forthcoming, pp. 3.
20 Forthcoming, pp. 3.
the neighborhood. Therefore, again, if E=K1 is true then closure

fails miserably.21
While there are reasons to doubt E=K, it is not because it commits you to
FactivityE .
For the first case to constitute a counterexample to FactivityE , we have to
assume:
(1) That I have just locked my door is evidence that nobody can
enter my office.
Assume that if p is evidence for q, the probability of q has to be higher on p
than it would have been otherwise. Given this assumption, (1) entails:
(2) Because I just locked my door, it is more probable than it would
have been otherwise that nobody could get into my office.
The problem is that (2) entails:
(3) I just locked my door.
The case is only a potential counterexample to FactivityE if (3) is false. The
argument just sketched shows that (∼3) entails (∼1). So, it seems that the
objection to FactivityE comes to this: if you accept FactivityE , you have to
deny something false.
This response rests on two assumptions. The first is that if p is evidence for
q, there is some explanatory connection between p and q’s evidential probabil-
ity. The evidential probability of q is higher than it would have been otherwise
because p. The second is that “because” is factive in the explanans and ex-
planandum positions. These assumptions have both been called into question.
With regards to the first, someone could say that it is possible for p to be evi-
dence for q even if the probability of q on the total evidence is not increased by
the addition of p.22 This worry is easily addressed. The objection to FactivityE
assumed that evidence is evidence for something only if it raises its probability.
Suppose that this is not the only way for something to serve as evidence for
something else. There has to be some necessary condition on evidential support
for p to be evidence for q. It might be a highly disjunctive condition, but what-
ever that condition is, C is that (possibly disjunctive) condition. Unless “p is
evidence for q” entails “q satisfies C”, p is not evidence for q. To argue against
Factivity, you have to start with a putative counterexample in which (i) is true:
(i) p is evidence for q.
This entails:
(iia) Because p, q satisfies C.
(iib) q satisfies C because p.
21 Comesaña and Kantin 2010, pp. 453.
22 Thanks to [omit].
But, these entail:

(iii) p is true.
If (iii) is true, the example is not a potential counterexample to Factivity. If (iii)
is false, (i) is false, and so the example is not a potential counterexample to any
claims about what it takes for p to be evidence for q. Notice that the crucial
move in the argument is the move from the claim that something is evidence to
the further claim that the thing that constitutes evidence explains something
about what is supported by the evidence. If this works for evidence, it should
work for other kinds of normative reasons as well. Any attempt to show that,
say, normative reasons or justifying reasons consists of false propositions will fail
because if something has the status of a reason to believe or act, it has to explain
something about the normative properties of the belief or act in question. It
might be something trivial (e.g., it might explain why it is that the belief or act
in question has something going for it), but even trivial explanations require
facts rather than false propositions.
My response also assumes “because” is factive. This too has been called
into question, but the linguistic evidence for the factivity of “because” is solid.
Consider:
(4) The colonists protested because the tea was taxed. Not only
that, the tea was taxed.
(5) He knows that they are angry and confused protestors. Indeed,
they are angry protestors.
(6) I have two tea members of the tea party living in my building.
Indeed, I have precisely two. I’m lucky not to have more.
You cannot reinforce entailments (e.g., in (4) and (5)). If you try, you end
up with redundant conjunctions. You can reinforce pragmatically imparted
information (e.g., in (6)).23
This seems contradictory and this seems to provide further support for the
factivity of “because”:
(7) The bolt snapped because there were too many people on the
bridge, but nobody was on the bridge.
It would not be surprising that (7) is contradictory if these were equivalent:
(8) The bolt snapped because there were too many people on the
bridge.
(9) There were too many people on the bridge. That’s why the bolt
snapped.
These do seem to be equivalent. It seems contradictory to assert (8) and (∼9)
or to assert (∼8) and (9). Also, you cannot reinforce (8) with (9) or vice-versa.
23 I owe this point to Stanley 2008 who credits it to Sadock 1978.
Finally, suppose you were under the impression that LD was true and so
thought you had evidence that O was true. Suppose you said as you were
leaving the office:
(10) It is likely that nobody will get into the office because the door
is locked. Later, we discover that the door had not locked.
Looking back, we cannot say that you knew (10) was true when you uttered
it. Why not? You could have had excellent evidence for (10). You believed
(10). If ‘because’ is not factive, (10) could be true. My guess is that the reason
you cannot say you knew (10) was true is that (10) is false if the explanans
proposition is false.
Comesaña and Kantin’s objection fails for essentially the same reasons.
Their case is a counterexample to FactivityE only if we assume that Terry’s
belief that there is an animal in the neighborhood is based on the justified belief
that there is a dog in the neighborhood and that the proposition that there is
a dog in the neighborhood is part of Terry’s evidence. If this were so, then this
would have to be true:
(11) That there is a dog in the neighborhood is a reason for Terry
to believe that there is an animal in the neighborhood.
This entails:
(12) Because there is a dog in the neighborhood, Terry has a reason
to believe there is an animal in the neighborhood.
In turn, this entails:
(13) There is a dog in the neighborhood.
But, we were told to stipulate that (13) is false. If it were true, the case would
not threaten FactivityE . If (13) is false, so is (11).
The linguistic evidence discussed in the previous section is suggestive, but
the linguistic evidence for FactivityE does not focus on any of the essential
functional features of evidence. The discussion in this section does focus on one
of the roles essential to evidence. Whatever constitutes evidence has to figure
in explanations. If something constitutes a reason to act, it often does so by
explaining why it is that there is something good or attractive about so acting.
If something constitutes a reason to believe a proposition, it has to explain
something about the kind of rational support there is for believing it. It might
do this by explaining why something believed is more likely than it would have
been otherwise or it might do it in some entirely different way, but it has to
explain some fact about the support available for a belief. False propositions
explain nothing, so false propositions explain nothing about normative standing
or normative properties of a belief. So, false propositions do not constitute
normative reasons for action or for belief.
4.3.2 Having Evidence

In the previous section, I argued that if my evidence includes the proposition that
I have hands, I have hands. Since hands are items to be met in space, mentalists
have to deny that propositions like this could be part of our evidence. In this
section, I shall complete the argument against Mentalism+.
It is tempting to argue as follows:
(1) I know I have hands.
(2) If I know that I have hands, I justifiably believe that I have
hands.
(3) If I justifiably believe that I have hands, my evidence includes
the proposition that I have hands.
(5) My evidence includes the proposition that I have hands only if I
have hands.
The argument rests on the relatively mild anti-skepticial assumption that we
have knowledge of the external world. I defended (2) earlier. (3) is a consequence
of Justified Basis. (4) follows from (2) and (3). I defended (5) in the previous
section.
As stated, the argument suffers from two apparent defects. The first is that
the the premises entail that it is not possible for false beliefs to be justified. The
second is that it does not take care to distinguish between evidence and justifying
reasons. Under pressure from the arguments for FactivityE , the mentalists might
be tempted to reject (3) and reject Justified Basis and offer this in its place:
If you justifiably believe p and p is a justifying reason, p is a jus-
tifying reason of yours that can justify further beliefs (Weakened
Justified Basis).24
According to Weakened Justified Basis, it is possible to justifiably believe p even
if p is not itself a justifying reason. The conditions that determine whether
your beliefs are justified might determine which justifying reasons, if any, sup-
port your beliefs, but they do not determine whether your justifying reasons
include what you believe. Justifying reasons might not supervene upon your
mental states, but the conditions that determine whether your beliefs are justi-
fied might.
The second problem with the argument is that it neglects a distinction Conee
and Feldman draw between “ultimate” and “intermediate” evidence.25 Every
piece of evidence is a justifying reason, but some justifying reasons might not
24 Fantl and Mcgrath defend only this weakened version of Justified Basis.
25 Conee and Feldman 2008, pp. 87.
constitute evidence in the strict sense. If you justifiably believe p on inductive

grounds and deduce that some obvious consequence of p is true, there is some
sense in which p is a justifying reason of yours but it seems to be a derivative
reason. So, in response to the argument above, Conee and Feldman can say
that Mentalism+ is a thesis about ultimate evidence and the problem with the
argument above is that it assumes that our ultimate evidence includes contingent
worldly propositions. Sure, they might say, you can justifiably treat contingent
worldly propositions as reasons for further beliefs, but that does not mean that
these propositions are part of your evidence. So, the mentalists can reject (4)
on the grounds that they reject (3) if taken as a claim about ultimate evidence.
Let us say, then, that justifying reasons are any true propositions justifiably
believed and reserve the term “evidence” for what Conee and Feldman describe
as “ultimate evidence”. We need to know what evidence is, what distinguishes
a bit of evidence from a justifying reason. They offer this characterization:
Some philosophers have argued that only believed propositions can
be part of the evidence one has. Their typical ground for this claim is
that only believed propositions can serve as premises of arguments.
Our view differs radically from this one. We hold that experiences
can be evidence, and beliefs are only derivatively evidence ... Experi-
ence is our point of intersection with the world–conscious awareness
is how we gain whatever evidence we have.
Furthermore, all ultimate evidence is experiential. Believing a propo-
sition, all by itself, is not evidence for its truth. Something at the
interface of your mind and the world –your experiences –serves to
justify belief in a proposition, if anything does. What we are calling
your “ultimate evidence” does this without needing any justification
in order to provide it.26
In another passage they say that a person’s ultimate evidence is, “evidence one
has for which one need not have evidence.” 27
It seems they want to defend two claims about evidence:
If p is part of your ultimate evidence, p is experiential (Experien-
tiality).
If p is part of your evidence, p is a justifying reason that you do not
need evidence for in order for p to have that status (Basicality).
Given that the mentalists are neutral on Propositionality, we have to be careful
in how we interpret Experientiality. Surely your ultimate evidence will include
propositions that you know about experience on the basis of introspection, but
the question is whether mentalists can also say that the propositions that are
the contents of experience are themselves part of your evidence. Let’s consider
two versions of Mentalism+. The first understands Experientiality as follows:
26 Conee and Feldman 2008, pp. 87-88.
27 Conee and Feldman, pp. 88.
p is part of your ultimate evidence only if p is the content of some

introspective state and not a contingent worldly proposition (Narrow
Experientiality).
The second version of Mentalism+ allows for a more liberal conception of evi-
dence:
If p is part of your ultimate evidence, p is either the content of an
introspective state or the content of an experience (Broad Experien-
tiality).
Problems arise for both versions of Mentalism+.
The problems with the argument against Mentalism+ just sketched above
are easily surmounted. Suppose Basicality and Broad Experientiality are true.
We can the anti-mentalist argument as follows:
(1’) I know non-inferentially that I have hands.
(2’) If I know I have hands non-inferentially, I believe that I have
hands, this belief is non-inferentially justified, and this belief is
true.
(3’) If I believe that I have hands, this belief is non-inferentially
justified, and this belief is true, then my evidence includes the
proposition that I have hands.
(5) My evidence includes the proposition that I have hands only if I
have hands.
The argument assumes only Weakened Justified Basis, not Justified Basis. (1’)
is a relatively weak anti-skeptical assumption. We might quibble a bit about
whether I can know non-inferentially that I have hands. Maybe all I know non-
inferentially is that the facing surface of an object with the shape of a human
hand exists in space. This fact does not supervene upon anyone’s mental states,
so we can modify the argument accordingly and it still poses a problem for
Mentalism+. (2’) is no more controversial than (2). Suppose Broad Experien-
tiality is true. The content of perceptual experience includes contingent worldly
propositions as evidenced by the fact that the veridicality conditions of such
experiences make reference to facts external to us. I am assuming that (1’) is
true because the proposition that I have hands (or, better, that these are hands)
is part of the content of an experience. The only way to block the argument, it
seems, is to deny Broad Experientiality.28
28 There is an interesting question as to whether the scope of non-inferential perceptual
knowledge is limited to propositions that are part of the representational content of experience.
Mentalists are free to deny Broad Experientiality, but doing so comes at a

cost. Consider an argument inspired by an argument of Pryor’s.29 Imagine we
discovered that there were two kinds of people. Some people formed their be-
liefs about the external world by taking perceptual experience at face value and
some only formed beliefs about present experience and arrived at their external
world beliefs by means of inference. This seems to be the the sort of thing that
cognitive scientists discover, so let us imagine that they did discover this. It
seems intuitive to say that this psychological difference is normatively insignifi-
cant, these persons all know the same range of propositions about the external
world provided that they end up with the same beliefs and end up reasoning in
similar ways (exclusive, of course, of the early transitions in thought). If this is
right and we assume Basicality, some of these subjects would have contingent
worldly propositions as part of their ultimate evidence. This would land us right
back where we were. So, if we want to say that Basicality is correct we could as-
sert that Narrow Experientiality is correct and thus deny that the subjects that
base their beliefs directly on the evidence of the senses have knowledge. But, the
guiding intuition seemed to be that when it came to knowledge, the difference
in these subjects’ wiring is normatively insignificant. So, it would seem that
the mentalists either need to deny Basicality as well as Broad Experientiality
or embrace skepticism.
Mentalists are of course free to deny Basicality, embrace external world skep-
ticism, or do both. Suppose they deny Basicality and reject skepticism as well.
The mentalists would have to say that p can fail to be part of your evidence
even if p is a justifying reason of yours that you can treat as a reason for further
beliefs without needing further evidence to justifiably believe p. This would
rob the notion of evidence of its interest. The difference between perception
and introspection is a psychological difference, not a normative one. If they
agree that claims about which psychological faculties (if any) deliver ultimate
evidence should be determined by normative considerations about what can be
treated as a reason without the need for prior reasons, then they could accept
Basicality if they accept skepticism. The obvious problem with this option is
that it concedes everything to the external world skeptic. The more damning
problem is that the mentalists sold us out to the skeptics at so cheap a price.
They have declared that the external world skeptics the victors on the grounds
that the beliefs we form in direct response to experience could be false. Remem-
ber that the problem with my initial argument against Mentalism+ was that the
premises of that argument implied that there cannot be false, justified beliefs.
There are also interesting questions as to how to go about determining how “broad” the
representational content of experience is. For an argument that the scope of perceptual
knowledge is not limited to propositions that are themselves the contents of our perceptual
experiences, see Brewer 1999 and Millar 2000. As for the content of experience, I am not
entirely sure that perceptual experience has a content, but I am not persuaded that the
content of perceptual experience is quite so broad that it would include kind concepts. For
discussion, see Siegel 2005.
29 See Pryor 2000. Feldman 2004 endorses the argument discussed here. There is an extended
line of argument in Greco 2000 that seems to show that you give the game to the skeptic if
you deny that perception can provide direct knowledge of the external world.
Now it seems that the mentalist turned skeptic is attacking us on the grounds
that we are fallibilists about justification. Which is worse, the view that denies
that there are false, justified beliefs or the view that denies that there is suffi-
cient justification in the absence of entailing evidence? The evidentialists who
say that all a belief needs to be justified is ’on balance’ support should reject
both views. This is one reason why evidentialists should not be mentalists.
It should be clear now why Mentalism+ and Supervenience Internalism are
not viable options. Mentalists can tie themselves in knots to avoid the argu-
ment just sketched above. Given the arguments for FactivityE , they can only
do this by embracing an implausible form of skepticism. The argument against
Mentalism+ shows that there is this flaw in Supervenience Internalism. Accord-
ing to Supervenience Internalism, you and your systematically deceived mental
duplicate do not differ justificationally. But, the only way for you and your
systematically deceived mental duplicate to be perfectly alike justificationally
is you and your duplicate to have the same support relations between your evi-
dence and your beliefs. But, the only way for these relations to be the same is
for you to have just the same evidence your systematically deceived counterpart
to have. Given Broad Experientiality and FactivityE , these relations between
belief and evidence are the same in both cases only if you are systematically
deceived as well. Assuming you are not systematically deceived, Supervenience
Internalism is false.
For the purposes of the arguments against Mentalism+ and Supervenience
Internalism, I did not have to take a position on whether Experientiality is true.
There are probably different ways of interpreting Experientiality and I think
there are some readings on which it is perfectly harmless. There is one reading
that strikes me as not being perfectly harmless, so I thought I should comment
on that. Someone could interpret Experientiality as saying that p is evidence
you have if p is part of the content of some perceptual experience. On this view,
having evidence can be understood in non-normative terms. Understood in this
way, I have two concerns about Experientiality. The first is that the view offers
us a non-normative account of what it is to have evidence. The second is that
it implies that it is possible for you to have p as part of your evidence even if
you do not suffer from any sort of self-deception or failure of any failure of self
knowledge and either do not believe p or disbelieve p.
It seems intuitive to me to say that if you have p as part of your evidence,
you have the right to treat p as a reason for further beliefs. If I think Plum
did it and tell you that I think this for the reason that her prints were on the
murder weapon, it would be odd to say that I should not have assumed that her
prints were on the weapon if you concede that this is part of my evidence. If,
however, having p as part of the content of an experience is enough for having
p as part of your evidence, it seems you cannot say that this is so. Having an
experience as of p being the case is compatible with having strong reasons to
think p is not the case or having strong reasons to think that there is something
amiss with your experience. In either case, I think you might not have the
right to treat p as a reason for forming further beliefs, so the non-normative
account seems to cut the link between having evidence and having the right to
use it. It also seems strange to think that you can believe that something is
part of your evidence without believing that it is true or while believing that it
is false. It is possible, however, to believe that p is part of the content of some
experience while believing either that p is false and the experience is delusory
or while suspending judgment as to whether p is true because you have some
reason to think that something is amiss. To me, these points suggest that we
should prefer a doxastic account of having evidence to a non-doxastic account
and a normative understanding of the relevant notion of having to one that is
non-normative. It may well be that experience is one of the primary ways by
which we acquire evidence, but that is not to endorse the further claim that you
have as your evidence whatever the content of your experience is.
4.4 Evidence and Knowledge

FactivityE tells you something about the constitution of evidence, but nothing
about its possession. In the course of arguing against Mentalism+, I suggested
that your evidence will include what you know non-inferentially:
If you know p non-inferentially, p is part of your evidence (IKSE).
Because I defend both IKSE and FactivityE , my view is similar in many respects
to Williamson’s. Given E=K, he is committed to two claims about the relation
between knowledge and evidence:
If you know p, p is part of your evidence (ESK).

If p is part of your evidence, you know p (KSE).
KSE entails IKSE and ESK entails FactivityE . E=K gives you necessary and
sufficient conditions. So far, I have only defended a necessary and a sufficient
condition. In this section, I shall argue that both halves of E=K are mistaken.
The literature is now filled with purported counterexamples to E=K and my
impression is that a fair number of epistemologists think there is something
seriously wrong with Williamson’s equation. Myself, I think this is an overreac-
tion. Most of the objections to E=K are not convincing. Those that are require
only minor modifications to his view. Still, the view needs modification. It is
worth taking the time to make the necessary changes to his view. The resulting
view, if even approximately correct, seems to show that the orthodox accounts
of doxastic justification are all seriously mistaken.
Is it possible to have knowledge without evidence? According to IKSE, it is
not possible to have non-inferential knowledge without evidence. If you know p
non-inferentially, p is true. If you know p non-inferentially, your belief in p is
non-inferentially justified. That it is justified entails that it is not wrong to treat
it as if it is a reason. That it is true removes the main objection from taking p to
be the right sort of beast to be a bit of evidence or a justifying reason. Suppose
you know p inferentially. If you know p on the basis of entailing evidence, is p
an extra bit of evidence you acquire by means of deduction? If you know p on
the basis of non-entailing evidence, is p an extra bit of evidence you acquire by

means of inductive inference?
It seems strange to suggest that deductive inference is a way of acquiring
new evidence if it is a way of extending your knowledge, but it also is hard to
imagine what harm could come of describing the things you come to know in this
way as evidence. An example suggests that KSE might be too liberal. Suppose
you read in a magazine that biologists believe only female foxes eat berries on
the grounds that after observing thousands of hungry foxes, only the females
would eat berries. The male foxes preferred root vegetables and chicken. You
look out the window and see that there is a fox in the yard. You see that it is
eating berries off of a berry bush, so you judge:
(1) There is a female fox in the yard.
You then deduce:
(2) There is a vixen in the yard.
You then remember that (2) entails (1), so you think to yourself (correctly)
that (∼2) is inconsistent with what you know iff (∼1) is inconsistent with what
you know. You then note that if KSE is correct, you know (1) only if (∼2) is
inconsistent with your evidence. Intuitively, it seems that (∼2) was consistent
with the evidence you initially had for believing (1) and I do not think this
changes simply because you added (1) to the set of propositions you knew or
noted that (2) and (1) are mutually entailing. According to KSE, however, you
cannot know (1) unless (∼2) is inconsistent with your evidence, and as that
evidence consisted of past observations of other foxes, it is hard to see how (∼2)
could be inconsistent with your evidence.
This sort of problem does not arise for (IKSE). It denies that your evidence
includes the propositions you know via deduction and so denies that your ev-
idence will include the propositions that you know on inductive grounds and
denies that your evidence will include all the subsequent propositions you come
to know via further deductive inferences. There is a perfectly good sense in
which (1) and (2) are justifying reasons. Consistent with FactivityE , we can say
they are the right sorts of things to be reasons if (1) and (2) are true. Accord-
ing to IKSE and IJSE, they can be justifiably treated as if they are reasons if
justifiably believed. The picture that seems intuitive to me says that there is a
division of labor. Some processes provide the evidence. Reasoning allows you to
extend knowledge and justify new beliefs on the basis of old evidence without
adding evidence in the process. In the grand scheme of things, I think nothing
particularly deep or important turns on whether KSE or IKSE is true given
that we can easily distinguish between derivative and non-derivative justifying
reasons or ultimate and non-ultimate evidence. To accomodate intuitions about
the case described above, I think we should say that while (∼1) and (∼2) are
inconsistent with what you justifiably believe, neither is inconsistent with your
evidence. Your evidence constitutes the foundations and fixes the evidential
probabilities of propositions like (1) and (2). As it seems implausible to say
that the evidential probability of (2) is 1 if you know (1), I hope the reader is
willing to make a minor modification of Williamson’s view.
Given the reasons for thinking that there can be knowledge without evidence,
I do not think the problems that arise for KSE are particularly interesting and
do not think the move to IKSE is particularly illuminating. The claim that your
evidence consists of what you know directly or immediately is consistent with
the knowledge-first program, after all. In this section, I shall argue that it is
possible to have evidence without knowledge. If the arguments offered in this
section are sound, the concept of knowledge plays a much more diminished role
in the account of evidence than Williamson maintains.
An initial worry you might have about E=K is that it seems to conflict with
the apparent platitude:
You have unproblematic access to your evidence (Unproblematic Ac-
cess).
Silins suggests that we can think of this kind of access in terms of Armchair
Access:
If p is part of your evidence, you can know that p is part of your
evidence from the armchair alone (Armchair Access).
He objects to E=K on the following grounds.30 Consider:
(3) You know that E=K is true from the armchair.
(4) You know p on the basis of observation.
(5) p is part of your evidence. [(3), (4)]
(6) If p is part of your evidence, you could know that p is part of
your evidence from the armchair alone. [Armchair Access]
(7) You knows that p is part of your evidence from the armchair
alone. [(5), (6)]
(8) You know from the armchair that if you know that p is part of
your evidence from the armchair alone, you can know p from the
armchair alone. [(3)]
(9) You can know p from the armchair alone. [(7), (8), and K-
Closure]
To avoid the conclusion, we either have to deny that we could know from the
armchair that E=K is true or say that anything we know we can know from
the armchair. The argument assumes that knowledge is closed under known
entailment:
If you know p and know q is an obvious consequence, you can come
to know q by means of competent deduction (K-Closure).
30 See Silins 2005.
Although the principle is controversial, it is unlikely that the argument fails

because of counterexamples to K-Closure.
Note that the objection applies with equal force to any non-skeptical view
that incorporates IKSE and FactivityE . Replace (3) with:
(3’) You know IKSE and FactivityE from the armchair.
You get the same result. If there is anything we can know non-inferentially
about the external world, we can know that this proposition is true from the
armchair. Either empirical investigation is needed to know what your evidence
is or it is too easy to conduct an empirical investigation as we can know the
external world from the comfort of our armchairs.
The problem that we face is structurally similar to the problem McKinsey
thought arose for thought content externalism.31 If there are necessary connec-
tions between the external world and the conditions that determine the contents
of your mental states, he thought that you could not know the contents of your
thoughts until you knew what the external environment is like. Suppose that the
thought a speaker expresses in uttering “Water is wet” depends upon whether
the speaker is in a world containing H2 O or XYZ. In a world containing H2 O
(but not XYZ), the speaker expresses the proposition that water is wet. In a
world containing XYZ (but not H2 O), the speaker expresses the proposition that
twater is wet. If the first speaker says, “I believe water is wet”, she expresses
a proposition that is true only in a world that contains H2 O, so how could she
know that the proposition expressed is true unless she knows that H2 O exists?
If she cannot know that the proposition expressed is true unless she also knows
that H2 O exists, either we are wrong about how easy it is to know the mind or
wrong in thinking that it is relatively harder to know the world.
While these problems are structurally similar, they do not admit of struc-
turally similar solutions. Objections to content externalism based on the McK-
insey worry rest on a mistake about the content externalist’s commitments.32
The content externalist says that we know apriori that our thought contents
are world-dependent, but not world-involving. We know from the armchair, for
example, that if two hypothetical communities are in different environments, dif-
ferences in their environments can be responsible for differences in their thoughts
about those environments even if the thinkers are the same from the skin in.
Imagine people just like us whose ancestors used the term “unicorn” to speak
about a kind of horse. We know from the armchair that if our ancestors never
used that term successfully, our ancestors meant something different by their
use of “unicorn” than the ancestors of this counterfactual community. We do
not know from the armchair whether our concept of “water” is world-dependent
in the way that “unicorn” is for our ancestors or world-dependent and world-
involving in the way that it is for the ancestors of the members of this coun-
terfactual community. So, you do not know from the armchair that if you have
31 For a defense of content externalism, see Burge 1979. See McKinsey 1991 for a discussion
of the problem of reconciling content externalism with privileged access.

32 See Brueckner 1992.
water-thoughts water exist just as you do not know from the armchair that if
you have unicorn-thoughts unicorns exist. What you know from the armchair
is that if the external world differed in various ways, you would grasp different
thoughts. Unfortunately, the objection to E=K does not have the commitments
of that view wrong. Some other response is needed.
There are problems with E=K, but the objection has little force as it is
stated. Armchair Access seems to lead to skeptical problems of its own:
(10) You can know from the armchair that your evidence is limited
to propositions you can know from the armchair belong to your
evidence. [Armchair Access]
(11) If p is part of your evidence, p is true. [FactivityE ]
(12) If you know non-inferentially that p is true, p is part of your
evidence. [IKSE]
(13) If your evidence includes the proposition that you have hands,
then you have hands and you know from armchair that you have
hands. [(10), (11), (12)]
(14) But, it is absurd to think you could know from the armchair
that you have hands.
(15) If your evidence cannot include the proposition that you have
hands, either you are handless or you cannot know non-inferentially
that you have hands. [IKSE, FactivityE ]
(16) Your evidence cannot include the proposition that you have
hands. [(13), (14)]
(17) Either you are handless or you cannot know non-inferentially
that you have hands. [(15), (16)]
(18) If you have hands, you cannot know non-inferentially that you
do. [(17)]
(19) If you do not have hands, you cannot know non-inferentially
that you do.
(20) So, whether you have hands or not, you cannot know non-
inferentially that you have hands. [(18), (19)]
The argument assumes IKSE and FactivityE . To block the argument from
Armchair Access to the skeptical result, you would either have to deny IKSE
or deny FactivityE . Suppose, then, that we relax our assumptions a bit. We do
not need FactivityE to show that Armchair Access leads to the same sorts of
unpalatable skeptical results that Mentalism+ did earlier:
(21) If you know from the armchair that p could be part of your
evidence only if you had empirical justification to believe p, then
you cannot know from the armchair that p is part of your evi-
dence.
(22) You know from the armchair that if the proposition that you
have hands is part of your evidence, you would have to have
empirical justification to believe that you have hands.
(23) You know that cannot know from the armchair that you have
empirical justification to believe that you have hands.
(24) You know that you cannot know from the armchair that the
proposition that you have hands is part of your evidence. [(21),
(22), (23)]
(25) Your evidence cannot include the proposition that you have
hands. [(24), Armchair Access]
(26) If you could know non-inferentially that you have hands, your
evidence could include the proposition that you have hands. [IKSE]
(27) You know you cannot know non-inferentially that you have
hands. [(25), (26)]
As the notion was introduced, armchair knowledge is knowledge that does not
depend constitutively upon experience and the justification it provides. So, (21)
and (23) are harmless assumptions. As for (22), the thought is that you cannot
properly treat contingent worldly propositions such as the proposition that you
have hands as reasons for further belief unless they received some sort of support
from experience. This seems to be rather obvious, something you could know
upon reflection alone. So, it is a good candidate for armchair knowledge. Now, if
you recall earlier our discussion of Pryor’s argument for liberal foundationalism,
he argued (plausibly) that it is a contingent matter whether the scope of non-
inferential knowledge includes contingent worldly propositions. The argument
just sketched here shows that if Armchair Access is true, no such proposition
could be part of anyone’s evidence regardless of whether they were wired in
such a way that their beliefs about such contingent worldly propositions were
based on propositions about the subject’s own mental life or not. If the lesson
you took from his argument is that any argument that purports to show that
we cannot in principle have non-inferential knowledge of contingent worldly
propositions would show that we cannot have knowledge of such propositions,
the argument just sketched here shows that Armchair Access commits us to the
very unpleasant skeptical view that you cannot know that you have hands at
all.
If the Armchair Access objection to E=K has any bite, we have to assume
that the intuition behind Armchair Access is intuition that underwrites Un-
problematic Access. If it is, then perhaps the thought is that facts that you
know about the external world non-inferentially cannot belong to our evidence
because if they did, you would have only epistemically problematic access to
the evidence. While there is something strange to the suggestion that you have
only problematic access to your own evidence, there is also something strange
to the suggestion that you can know p non-inferentially even if your access to
the truth of p is epistemically problematic. If it were problematic, it seems you
could not justifiably believe p without independent evidence. But, if you could
not justifiably believe p without independent evidence, your belief in p would

not constitute non-inferential knowledge. Suppose that Unproblematic Access
does not commit you to Armchair Access. Suppose the latter principle is more
restrictive and more demanding. If so, then the problem is not that we have
only problematic access to facts about the external world. Rather, it is conceded
that we have epistemically unproblematic access when we have non-inferential
knowledge. Then it seems to make little sense to complain about IKSE and
FactivityE . After all, it has been conceded that you have unproblematic ac-
cess to certain facts and can justifiably treat them as reasons without needing
independent evidence to do so. That seems to suggest that what you know
non-inferentially bears all the marks of evidence.
These problems can be mitigated if we revise Armchair Access as follows:
If p is part of your evidence, it is possible for you to know that p
is part of your evidence without needing any empirical justification
for believing that p is part of your evidence beyond the justification
needed for p to be part of your evidence (Revised Armchair Access).
Unlike Armchair Access, Revised Armchair Access does not imply that anything
you cannot know from the armchair is epistemically problematic. So, Revised
Armchair Access does not seem to give us any reason to deny IKSE or FactivityE .
Whereas the combination of Armchair Access and FactivityE commits you to
saying that your evidence cannot include any propositions you cannot know
on the basis of introspection alone, Revised Armchair Access does not force
you to choose between FactivityE and the plausible claim that the evidence
provided by perception is not limited to the evidence you could have acquired
via introspection alone. Revised Armchair Access still causes problems for E=K.
Consider:
(28) You know on the basis of observation that p is true.
(29) p is part of your evidence. [(28), (E=K)]
(30) You know that p is part of your evidence without further em-
pirical investigation. [(29), (Revised Armchair Access)]
(31) You know from the armchair that if p is part of your evidence,
you know p. [(E=K)]
(32) You know that if you know p, your belief in p is not Gettiered.
(33) You know from the armchair that if p is part of your evidence,
your belief in p is not Gettiered. [(31), (32)]
(34) You know without further empirical investigation that your
belief in p is not Gettiered. [(32), (33)]
If only fake barn detection were so easy. Given how weak Revised Armchair
Access is, I think this is a serious objection to E=K. Given that you can have
Gettier cases for non-inferentially justified beliefs, the troubles for E=K would
also arise for the weaker thesis that identifies your evidence with what you know
non-inferentially:
Your evidence includes p iff you know p non-inferentially (E=IK).

Someone might say that even Revised Armchair Access is problematic, but
there is a more straightforward objection to E=K and E=IK. Suppose a non-
factive mental duplicate of yours sees a real barn, a barn that is qualitatively
identical to the barn you know you saw on your drive through real barn country.
Suppose this subject’s barn is surrounded by a sufficient number of fakes so
that this subject doesn’t know the building that she saw was a barn. We can
stipulate that her belief is true and reasonably held. It seems counterintuitive
to say that she has less evidence than you because of the fake barns. I can’t
think of any principled reason to think that your evidence couldn’t include
the proposition that the building you saw was a barn, so it is tempting to
say that your counterpart’s evidence includes a proposition she doesn’t know is
true—that the building she saw was a barn. Also, it seems that the difference in
what you two know is due to extra-evidential factors (e.g., the presence of fakes
in her case and the absence of fakes in yours). This extra-evidential explanation
is ruled out by ESK. ESK implies that if your counterpart’s evidence includes
the proposition that the building she saw was a barn, the subject would know
that the building was a barn and so would not be in a Gettier case. So, ESK
fails to accommodate intuition and rules out a plausible explanation of that
intuition.
There is a view that seems to fall squarely within the knowledge-first camp
that can accomodate intuitions about Gettier cases and access worth consider-
ing:
Your evidence includes p iff you know p or fail to know p for reasons
external to your mental states (E≈K).
This view takes its inspiration from Bird’s account of justification on which a
justified belief is either knowledge or a belief that fails to constitute knowledge
for reasons that, in some sense, you cannot be held responsible for.33 Such a view
seems to accomodate intuitions about access and Gettier cases, but there are
worries about how it can handle cases of false propositions. If the argument for
FactivityE are sound and the Gettier objection to E=K is also sound, it would
be strange for someone who bought into the knowledge-first program to say
that the truth of the proposition believed is required for it to be evidence while
conceding that some of the external conditions necessary for knowledge had no
bearing on whether the proposition was part of your evidence. Both truth and
the absence of Gettier conditions are necessary for knowledge, so it would seem
arbitrary to say that one of these conditions is involved in determining what
evidence you have while the other is not.
These objections to E=K do not threaten a weaker view that incorporates
IKSE and FactivityE . There is no reason to think that IKSE and FactivityE
imply that you do not have unproblematic access to your evidence because
FactivityE says that your evidence consists of true propositions and IKSE rules
33 Bird 2007, pp. 84. There is no indication that he would accept such a view of evidence.
out facts that you have only epistemically problematic access to. You can con-
sistently maintain that IKSE and FactivityE are true while denying (31).
While I think IKSE and FactivityE are true, tacked together, they do not
constitute a view. Let me start to sketch an alternative to E=K and E=IK. In
motivating IKSE, I noted that IKSE is implied by IJSE. If your belief in p is
non-inferentially justified, it is proper to treat p as if it is a justifying reason for
forming further beliefs. So, if your belief in p is non-inferentially justified, it is
proper to treat p as if it is a piece of evidence. If, however, your belief in p is
not justified, it is not proper to treat p as a reason for further beliefs. Finally,
if p is not true, p is not itself a piece of evidence. So, consider this equation:
Your evidence includes p iff p is true and your belief in p is non-
inferentially justified (E=IJTB).
E=IJTB does not deny that evidence is factive, it implies that IKSE is true, and
it provides us with necessary and sufficient conditions for evidence ascriptions.
The worries about access and Gettier cases that arose for E=K and variants of
that view do not arise for E=IJTB.
We should pause to consider an objection. Comesaña and Kantin have ar-
gued in a recent paper that FactivityE implies that there are no Gettier cases.34
If this is right, it is not fair for me to use Gettier cases to criticize E=K as
FactivityE commits me to saying that there are no Gettier cases. They ask us
to consider one of Gettier’s examples, Coins:
Suppose that Smith and Jones have applied for a certain job. And
suppose that Smith has strong evidence for the following conjunctive
proposition:
d. Jones is the man who will get the job, and Jones has ten coins in
his pocket.
Smith’s evidence for (d) might be that the president of the company
assured him that Jones would in the end be selected, and that he,
Smith, had counted the coins in Jones’s pocket ten minutes ago.
Proposition (d) entails:
e. The man who will get the job has ten coins in his pocket.
Let us suppose that Smith sees the entailment from (d) to (e), and
accepts (e) on the grounds of (d), for which he has strong evidence.
In this case, Smith is clearly justified in believing that (e) is true.
But imagine, further, that unknown to Smith, he himself, not Jones,
will get the job. And, also, unknown to Smith, he himself has ten
coins in his pocket. Proposition (e) is then true, though proposition
(d), from which Smith inferred (e), is false. In our example, then,
all of the following are true: (i) (e) is true, (ii) Smith believes that
(e) is true, and (iii) Smith is justified in believing that (e) is true.
But it is equally clear that Smith does not know that (e) is true;
34 Comesaña and Kantin 2010.
for (e) is true in virtue of the number of coins in Smith’s pocket,

while Smith does not know how many coins are in Smith’s pocket,
and bases his belief in (e) on a count of the coins in Jones’s pocket,
whom he falsely believes to be the man who will get the job.35
Coins is one of Gettier’s cases, but is it a Gettier case? It is if it is a case where
(e) is true, Smith is justified in believing that (e) is true, but Smith does not
know that (e) is true.
Comesaña and Kantin argue that anyone who accepts FactivityE has to deny
that Coins is a Gettier case as follows:
(1) According to FactivityE , no false propositions can constitute ev-
idence.
(2) If no false propositions can constitute evidence, Coins is not a
genuine Gettier case.
(3) Coins is a Gettier case.
(C1) False propositions can constitute evidence.
The argument’s crucial premise is (2). They say this in its defense:
Now, everyone should agree that the proposition that Jones has ten
coins in his pocket is something Smith knows, and that is part of
what justifies Smith in believing that whoever got the job has ten
coins in his pocket. Everyone should also agree that the proposition
that the secretary said that Jones got the job is something that you
know . . . and it certainly plays some role in justifying Smith in
believing that whoever got the job has ten coins in his pocket. But
for this strategy to work, it should be the case that everything that
justifies you in believing that whoever got the job has ten coins in
his pocket is a proposition that you know. [E=K implies] ... that
a proposition p cannot be part of your justification for believing
something unless you know that p . . . And there is no argument
that we can think of to the effect that Smith’s belief that Jones got
the job plays no part whatsoever in justifying Smith in thinking that
whoever got the job has ten coins in his pocket.36
Their objection assumes that Smith’s evidence includes (d) because Smith justi-
fiably believes that (d) is true. The objection is dialectically ineffective because
it assumes:
If you justifiably believe p, p is part of your evidence (E=J).
E=J appears to commit you to:
You cannot justifiably believe p if the evidential probability of p for
you is less than 1 (InfallibilityJ ).
35 Gettier 1963, pp. 122.
36 Comesaña and Kantin 2010, pp. 450.
InfallibilityJ commits you to the skeptical thesis that you cannot have justi-
fied beliefs based on inductive inference. This kind of skepticism is inherently
unattractive, but that is not the main problem with InfallibilityJ . Coins is
a genuine Gettier case only if Smith justifiably believes (d). Smith’s belief is
based on non-entailing evidence. (It must be based on non-entailing evidence
because (d) is false and it is based on propositions that Gettier says are true
when we describe the case.) So, InfallibilityJ seems to imply that Coins is not a
genuine Gettier case because it seems to imply that Smith does not justifiably
believe (d). Since E=J implies InfallibilityJ , E=J implies that Coins is not a
genuine Gettier case. Note that the argument from E=J to (∼3) did not assume
FactivityE . So, I think it is safe to say that the objection to FactivityE failed.
There might be ways of reformulating their objection so as to avoid this
worry, but we need not worry too much about revised versions of this objec-
tion. First, their argument does not show that FactivityE commits us to the
impossibility of Gettier cases because there are Gettier cases that do not involve
reasoning from any false beliefs. At best, the objection suggests that FactivityE
forces us to deny that Coins is a Gettier case. Second, their objection rests on a
questionable description of Gettier cases insofar as Gettier’s own description of
the case involves forming a false belief without forming it on the basis of any false
evidential propositions and then inferring a true belief from that. Finally, the
objection neglects the distinction between personal and doxastic justification.
Remember that early in the post-Gettier literature, some authors thought that
Gettier cases were not genuine. The worry was that Smith’s beliefs could not
be justified because they were beliefs in false propositions, false propositions do
not constitute evidence, but a belief is justified only if the proposition believed
constitutes evidence or a justifying reason. In response, Lowy argued that these
objections all missed their mark. As she noted, Getter was interested in the
conditions that determined whether a believer was justified in holding a belief,
not in the conditions that determined whether the believer’s belief is justified.37
Consistent with the standard intuitions about Gettier’s cases is the claim that
Gettier cases are one of the cases where personal and doxastic justification come
apart. Smith is justified, sure, but Smith’s beliefs are not justifiably held. So
even if Smith’s beliefs in (d) in (e) cannot be justified, Smith can be justified in
holding these beliefs. By all accounts, Smith was fully rational and responsible
in believing (d) and (e). If all a Gettier case requires is a case in which Smith
is (personally) justified in believing p, believes p, and p is true where Smith
does not know p, Coins is the case we need. It does its job even if FactivityE
is true. (Indeed, it does its job even if InfallibilityJ and E=J are true.) Thus,
E=IJTB does not imply that Gettier cases are impossible and so it is fair for me
to appeal to intuitions about such cases in arguing that E=IJTB is preferable
to E=K.
37 For a discussion, see Lowy 1978.
4.5 Evidence and Epistemic Rationality

The aim of this chapter is to defend an account of evidence that I think is
both correct and a useful corrective. In the course of arguing for an externalist
conception of evidence, I did not address one of the standard arguments for the
internalist conception of evidence favored by the mentalists. The argument is
rather simple. It starts with this rather unfortunately plausible assumption:
The conditions that determine whether you are epistemically ra-
tional strongly supervene upon your mental states (Supervenience
InternalismR )
The epistemically rational believer respects the evidence. That is pretty much
agreed upon by all sides. If we think of epistemic rationality as a matter of
respecting the evidence, we can argue from Supervenience InternalismR to Men-
talism+ as follows. Suppose White and Plum are in precisely the same mental
states. Both believe that Mustard is the killer. White saw Mustard kill his vic-
tim. Plum underwent an indistinguishable hallucinatory experience and seemed
to have seen Mustard kill. Suppose Mentalism+ is false and some externalist
view such as E = IJTB is correct. White’s evidence includes everything that
Mustard’s evidence includes, but it includes more besides. Now, if Plum knew
that her evidence included only the evidence someone would have in the bad
case (i.e., the case in which her beliefs are mistaken but she is in just the same
non-factive mental states she is in now because of a hallucinatory experience),
she ought to be significantly less confident in her beliefs than White is. Perhaps
if she knew this she ought to suspend judgment as to whether some proposi-
tions White knew to be true really were true. If she knew that her evidence
included just the evidence someone had in the bad case and did not adjust her
attitudes accordingly, she would not be as reasonable or rational as White is.
But, if E = IJTB is true she is not in a position to know that her evidence is
less than White’s and her ignorance seems necessary for our saying that she is
no less reasonable or rational than White. But, then we have to say that she
is rational or reasonable only because she fails to know what her evidence truly
consists of. How can we say both that she is nothing less than fully reasonable
or rational and that epistemic rationality is a matter of respecting the evidence
when she is ignorant of what her evidence truly consisted of? This combination
of views is puzzling. However, it seems that this combination of views is what
any externalist is saddled with unless they are willing to deny that epistemic
rationality is a matter of respecting the evidence. So, it seems there is a plau-
sible line of argument from Supervenience InternalismR to Mentalism+. Given
the intuitive plausibility of Supervenience InternalismR , we have a good case for
Mentalism+.
Williamson has offered two points in response to this kind of argument for
Mentalism+. First, he maintains that the demands of rationality cannot be lu-
minous because no non-trivial conditions are.38 While I agree that the demands
38 Williamson 2000b, pp. 624.
of rationality are not luminous and agree that subjects in the bad case are typi-
cally mistaken about what their evidence is, this line seems more promising if you
are trying to undercut the intuitive support for Supervenience InternalismR than
if you are trying to undercut the argument from Supervenience InternalismR to
Mentalism+. As most parties seem to agree that subjects are perfectly rational
in forming beliefs in response to hallucination, the challenge is to explain how
this could be if subjects in the bad case have less evidence than similarly situ-
ated subjects in the good case. I do not think that we can explain why these
subjects are reasonable or rational in terms of their ignorance of the demands
of epistemic rationality.
Second, Williamson suggests that even if subjects in the bad case have less
evidence for their worldly beliefs than subjects in the good case do, that does
not mean they lack sufficient evidence to justifiably believe what they do.39 So,
there is no good reason to think that it follows from the fact that subjects in
the bad case have less evidence by virtue of being in the bad case that they are
thereby anything less than perfectly rational or reasonable.
This second point needs to be handled with some care. I do think that
someone can have less evidence a subject has in some good case and still have
sufficient evidence for her beliefs. I want to bracket the question as to whether
someone has sufficient evidence for her beliefs in the case of hallucination until
the next chapter. What worries me here is that while Williamson is right that the
externalist about evidence can say that subjects in the bad case are reasonable
or rational, it is not entirely clear that they can say this while also saying that
subjects in the good case are also perfectly reasonable and rational. For, suppose
subjects in the good and bad case are equally confident in believing what they
do on the basis of their respective experiences. It seems that given the extra
evidence subjects have in the good case, they ought to be more confident than
subjects are in the bad case. So, some subject is either too confident or not
confident enough. So, some subject is not perfectly rational and the externalist
about evidence has to deny Supervenience InternalismR .40
There are two points to make in response to these worries. First, for reasons
discussed earlier, it is important to be careful about the link between deontic
judgments and judgments about rationality. The rational, the reasonable, and
the responsible are not the mark of the permissible or the proper. If this point is
granted, then it must also be granted that two subjects can be equally rational
or reasonable in how they respond to the reasons that apply to them even if there
is a difference in the reasons that apply to them and even if the right response to
these reasons differ. In the case of excusable wrongdoing, for example, the agent
counts as being equally reasonable or responsible as the agent who acts rightly
even though she acts against an undefeated reason. (If the reason had been
defeated, this would have been a wrongful act that was justified by overriding
reasons.) So, suppose we just bite the bullet and say that if subjects in the case
of hallucination and perception are equally confident, one of these subjects is
39 Williamson 2000a, pp. 197.
40 This version of the objection is inspired by some of Silins’ 2005 remarks.
not as confident as she should be. If they ought to have different degrees of
confidence, there is an undefeated reason for them to adjust their degrees of
confidence accordingly. It does not follow that one of these subjects is less than
perfectly rational because failing to respond to the reasons there are is not a
failure of rationality, per se. There has to be an argument that these reasons
are somehow special insofar as failing to respond to them is always a failure of
rationality.
On the bullet biting response, White and Plum are equally confident in their
perceptual beliefs. Both are sufficiently confident to believe on the basis of their
respective experiences, but at most one of these subjects ought to be as confident
as they are. So, one of them ought either to be more confident or less so. Here
is an argument for the bullet biting response. At most, one of these subjects
is permitted to believe on the basis of her experiences. If a subject ought not
believe p but she believes p, she ought to be less confident in p. How much less
confident? She should lower her level of confidence below whatever threshold
she must to avoid having the false belief. Clearly, Plum is too confident because
she believes White is the killer but, I shall argue later, she ought not believe
this. She ought not believe this because her belief is based on hallucination. It
does not follow that she is less than fully rational or reasonable, mind you, but
she ought to be less confident.
My argument for the bullet biting response assumes that if you believe p
but ought not believe p, you ought to be less confident that p. Someone could
deny this, but then claims about the justificatory standing of a belief seem to
have little to do with claims about the proper degree of confidence and so the
argument for Mentalism+ simply fails at an earlier stage. This talk of degrees
of confidence was a distraction. So, even if the degree of confidence you ought
to have matched perfectly the degree of confidence it is reasonable or rational
to have, the (purported) fact that subjects in the good and bad case ought to
be equally confident does not cut against the claim that subjects ought to form
different beliefs in the good and the bad case. So, if what you ought to believe
is fixed by the evidence, we cannot determine what evidence someone has by
determining what degree of confidence they ought to have.
What does it mean to say that someone is epistemically rational? Someone
is rational in φ-ing if she φ’s in light of what she takes to be reasons where
should not have expected that there would be a decisive case against φ-ing.
There is a division of labor here. If the perceptual faculties do not do what they
are supposed to, the subject can believe without seeing that there is a decisive
case against believing what she thought she saw. What if the subject believes
for no reason at all? Can she be rational in believing for no reason at all or for
prudential reasons? Even if this were possible, the account does not imply that
the subject is rational in forming beliefs in this way unless the subject should
not have expected that there would be a decisive case against forming beliefs
in this way. Later, I shall argue that there is always a decisive case against
believing without evidence, so this should help to mitigate some of the worries
behind this objection. I think it is false to say of normal subjects that they
should not have expected that there would be a decisive case against believing
if they did not live up to the ordinary intellectual standards that we regularly
apply to one another but I cannot rule out the possibility of someone who, for
theoretical reasons, has good reason to think that it is permissible to believe
without evidence or on practical grounds. Happily, I think I am not alone in
this. My guess is that anyone who dismisses this possibility is employing an
externalist conception of rationality that many of us would find objectionable.
The basic idea behind judgments of rationality is that we can defend the subject
from the charge that she has failed in her responsibilities as someone who must
respond to the reasons that apply to her. On its face, the natural way to do this
is to argue that the subject should not have expected that she would fail to live
up to her responsibilities. If she should have expected this, she is unreasonable.
If she should not have expected this and you fault her, you are unreasonable.
The subject that fails to live up to the standards of rationality is either pitied
or the proper object of the reactive attitudes. If the subject did not have the
capacities to determine whether there is a case against her beliefs or the ability
to exercise those capacities, she ought to be exempted from criticism. If the
subject had the capacities and the power to use them but believed when she
should have expected that there would be a case against so believing, her beliefs
are irrational or unreasonable.
4.6 The Refutation of Supervenience Internalism

In arguing against Mentalism+, I argued that it is possible for two mental du-
plicates to differ in terms of the evidence that supports their beliefs. It follows
from this that the support relations that hold between two subjects beliefs can
differ even if their mental states do not. The evidence that supports the sub-
ject’s belief in the good case might include contingent, worldly propositions but
her mental duplicate in the bad case has beliefs that are either supported by
different evidence or no evidence at all. This might not seem to be a particu-
larly interesting result on its own. This result is consistent with the internalist
thought that the justificatory status of a belief does not depend upon the fac-
tors that distinguish the good case from the bad. So, the argument constitutes
a refutation of Supervenience Internalism, but only because it shows that the
internalist is wrong about something that seems not to matter so much in the
grand scheme of things. Still, it is a wedge. In the next chapter, I will show
that it is a useful wedge.
Chapter 5
Evidence (II)
5.1 Introduction
In the previous chapter, I defended an inelegant, externalist account of evidence
(E=IJTB) and argued against an internalist supervenience thesis on the grounds
that it denied that our beliefs can be justified, in part, by things we know
directly upon the basis of our perceptual awareness of our own surroundings. A
critic might say that the account of evidence is too gerrymandered to have any
plausibility. Another might say that even if the account defended in the previous
section is correct, it is of little interest to the larger internalism/externalism
debate. There are internalist views that say that only facts that supervene
upon your mental states can have anything to do with the justification of your
beliefs, but there are internalist views that do not say this.
Let me address the first worry first. E=K is simple and elegant in precisely
the way E=IJTB is not. Other things equal, simpler views are preferable.
Now, the arguments from the previous chapter were supposed to show that
other things are not equal. Still, it might be fair to say, as Williamson does,
that E=IJTB is, “a rather unnatural hybrid: the truth-condition is an ad hoc
afterthought, not an organic consequence”.1
Williamson is right that the right view cannot be as complicated as E=IJTB.
To tidy things up, I shall argue that the truth-condition is strictly speaking
redundant. If your belief in p is non-inferentially justified, p is part of your
evidence. If p is part of your evidence, p must be true. The truth-condition has
to be satisfied for your belief to satisfy the justification condition:
You cannot justifiably believe p unless p (FactivityJ ).
The truth-condition is an organic consequence of a view that says that p is part
of your evidence if your belief in p is non-inferentially justified. You cannot con-
sistently maintain that normative reasons for belief or action consist of facts or
1 Williamson 2009, pp. 311.
69
CHAPTER 5. EVIDENCE (II) 70
true propositions and also maintain that the satisfaction of the truth-condition
has nothing to do with whether a belief is justified.
Below, I shall argue FactivityJ follows from these theses about the ontology
of reasons for belief and action:
If p is a justifying reason of yours to believe, p is true (FactivityT ).
If p is a justifying reason of yours to act, p is true (FactivityP ).
The argument for FactivityJ rests, in part, upon the idea that if you justifi-
ably believe something, what you believe can justify further beliefs or justify
actions that the belief (partially) rationalizes. The argument should show that
an account of the ontology of reasons has a direct bearing on an account of
justified belief and so could potentially help us decide whether justification is
an internalist or externalist notion. It is not entirely clear whether internalists
have to deny FactivityJ , for they can say that justification is unattainable in
a wide range of cases. If, for example, we can have no justified beliefs about
contingent worldly facts, we cannot argue from FactivityJ against an internalist
view that says that individuals in the same non-factive mental states will not
differ justificationally. So, one of the things I need to do in this chapter is try
to show that FactivityJ is at least consistent with the commonsense view that
says that we have extensive knowledge of the external world. If, as it seems
plausible to say, that we have extensive knowledge of the external world, then
the case for FactivityJ does show that justification cannot be understood along
internalist lines.
5.2 Reasons for Belief

In this section, I shall offer an argument for Justified Basis, an argument that
shows that FactivityT commits you to FactivityJ . According to Justified Basis, if
you justifiably believe p, p is a justifying reason. Those who agree that evidence
and justifying reasons for belief consist of true propositions but deny FactivityJ
have to distinguish between the conditions under which you can permissibly
treat something as if it is a reason for belief and the conditions under which
it is a reason for belief. The distinction between the reasons there are and the
reasons that you have because you are aware of them is a familiar distinction
and perfectly harmless, but this is not the distinction that you need if you reject
FactivityJ but agree that FactivityT is true. The reasons you have and aware
of are reasons. What you need is a view on which it can be permissible to treat
something that is not a reason as a reason. Against this view, I shall argue that
since the right to believe comes with the right to treat what you believe as a
reason, the right to believe depends upon whether what you believe is a reason.
Why think there cannot be false, justified beliefs? The following thesis enjoys
rather widespread acceptance:
If you justifiably believe p, you have some justifying reason and your
belief in p is based on it (Proper Basis).
Those who deny Proper Basis have to say that it is possible for a belief to
be justified even if it is not based on evidence or a genuine justifying reason.2
Anyone who denies Proper Basis faces a dilemma. Either they have to say you
do not need evidence for p to justifiably believe it or they have to say that you
have to have evidence but don’t have to base your beliefs on it. If Williamson,
for example, says we do not need evidence that supports our beliefs to justifiably
hold these beliefs, he cannot then say that knowledge is what justifies belief. But
this is central to his account of evidence. Should we say instead that justified
belief doesn’t have to be based on evidence? I think not. We want to capture
the intuition that someone who believes on the evidence is epistemically better
off than if they just happen to have evidence that supports what they would
have believed anyway. As Pollock and Cruz put it:
One could have a good reason at one’s disposal but never make the
connection. Suppose, for instance, that you are giving a mathemat-
ical proof. At a certain point you get stuck. You want to derive a
particular intermediate conclusion, but you cannot see how to do it.
In despair, you just write it down and think to youself, “That’s got
to be true.” In fact, the conclusion follows from two earlier lines by
modus ponens, but you have overlooked that. Surely, you are not
justified in believing the conclusion, despite tha fact that you have
impeccable reasons for it at your disposal. What is lacking is that
you do not believe the conclusion on the basis of those reasons.3
It makes little sense to endorse the standard view that doxastic justification en-
tails propositional justification while allowing that that the propositional justifi-
cation you have to have to have justified beliefs might play no role in supporting
your beliefs. If you are not going to use it, why would you have to have it on
hand?
Someone might be skeptical of the reasons offered in support of Proper Basis.
They might think that if reasons are facts or true propositions, the justification
of a belief does not depend upon whether the belief is based on a genuine
reason, but only whether the subject can justifiably take it that the reasons for
which she believes are genuine reasons. To address this sort of skepticism, let
me remind the reader that we also saw earlier that there is linguistic evidence
that supports the orthodox account of the logic of justification ascriptions. On
the orthodox view, doxastic justification requires propositional justification. If
someone’s belief is justified, there is a reason or a justification for that belief
and that is the believer’s reason for believing. This seems to best explain why
it seems contradictory to say, “She has no reason to believe that it’s raining
outside” having just conceded that her belief that it is raining outside is perfectly
justified.
2 Fantl and McGrath 2009, pp. 104 say that this is a viable view for someone who agrees
that only true propositions can justify but wants to allow for the possibility of false, justified
beliefs.
3 Pollock and Cruz 1999, pp. 35.
Suppose Proper Basis is true. For a large class of beliefs, it will be impossible
to justifiably have such beliefs if they are mistaken. If p is non-inferentially
justified, maybe your belief in p is based directly on the fact or some factive
mental state (e.g., seeing that p). Given the argument for FactivityT , it is
obvious that such a belief can only be justified if true. The content of the belief
and the justifying reason that serves as the belief’s basis are the same. If p
is inferentially justified and based on entailing evidence, you cannot justifiably
believe p if ∼p. Not if the arguments for FactivityT are sound.
If there are false, justified beliefs, they have to be inferentially justified beliefs
based on non-entailing grounds. So, could these be cases where you justifiably
believe a false proposition on the basis of true propositions? Not if Same Basis
is true:
If you and another subject both believe p on the basis of a justifying
reason, these will only be different justifying reasons if your justifying
reasons for believing p differ or there is some difference in your non-
factive mental states (Same Basis).
The thought behind Same Basis is that your justifying reasons for believing
something are not just justifying reasons (i.e., facts), they are the things you
treat as if they are reasons and that depends upon your mental states rather
than the facts. What you treat as a reason for your beliefs is determined by
your psychological states.
Suppose that you believe p on the basis of non-entailing evidence, r. You are
the non-factive mental duplicate of someone who believes p on the basis of r in
a p-world. Are you also in a p-world? Yes. You both deduce q from p because
you both know that q is an obvious consequence of p. According to J-Closure,
you both justifiably believe q. According to Same Basis, you both believe q for
the reason that p. According to Proper Basis, you justifiably believe q only if p
is a justifying reason. But, Factivity says, this is true only if p is true. So, yes,
you are in a p-world. If p is non-inferentially justified, Proper Basis says that
p is the justifying reason for believing p. FactivityT says that p must be true.
If p is inferentially justified and based on entailing evidence, FactivityT implies
that p is true. If p is inferentially justified and based on non-entailing evidence,
p still turns out to be true. So, there are no false, justified beliefs based on
non-entailing evidence. The argument for FactivityJ is now complete.
How might someone who accepts FactivityT try to block the argument? In
his discussion of perceptual error, Williamson says this:
In unfavorable circumstances, one fails to gain perceptual knowl-
edge, perhaps because things are not the way they appear to be.
One does not know that things are that way, and E=K excludes the
proposition that they are as evidence. Nevertheless, one still has
perceptual evidence, even if the propositions it supports are false.
True propositions can make a false proposition probable, as when
someone is skillfully framed for a crime of which she is innocent. If
perceptual evidence in the case of illusions consists of true proposi-
tions, what are they? The obvious answer is: the proposition that
things appear to be that way. The mountain appears to be that

shape.4
Williamson can say that your evidence in the case of illusion consists of propo-
sitions about appearances and say that this is the evidence the belief is based
on. If he says this and also accepts Same Basis, he has to say that our beliefs
cannot be based on evidence that consists of propositions about the external
world. Either, this means that our knowledge of the external world cannot jus-
tify our beliefs or we cannot have knowledge of the external world. He wouldn’t
want to say such things. So, should he deny Same Basis? To say that someone
based her belief on such and such reasons is to say, in part, that the reasons
for which they believe are such and such. The form such a reason explanation
takes should not depend upon whether the agent’s beliefs are true or false.5
A subject’s reasons for believing are limited to what she takes to support her
beliefs, and it seems impossible for two subjects to differ in terms of what they
take to support their beliefs if these subjects are non-factive mental duplicates.
We know why Williamson thinks there can be false, justified beliefs. He says,
“Knowledge figures in the account primarily as what justifies, not as what gets
justified. Knowledge can justify a belief which is not itself knowledge, for the
justification relation is not deductive.” 6 He is right. The justification relation
is not deductive. You can justifiably believe p on the basis of non-entailing
evidence. This does not force us to deny FactivityJ . The justification of a belief
might be locked up with evidence that rules out the possibility of error, but it
does not need such evidence. The justification of a belief depends both on what
it is based on and what it can do for you. A belief is not justified if it cannot
provide reasons for further beliefs. True beliefs based on sufficiently strong but
non-entailing evidence can do that, but false beliefs based on the same evidence
cannot. This is why there cannot be false, justified beliefs.
The mistake Williamson makes is in thinking that the justificatory standing
of a belief is fixed by what the belief stands on, its basis or the evidence that
supports it. Williamson’s critics often make the same mistake in the course of
attacking E=K, as we saw in the previous chapter. The justificatory standing
of a belief depends, in part, upon whether it stands on a proper basis, but
also upon whether it can shoulder its burden in providing support for further
beliefs. Given the arguments for Factivity, only true beliefs can do that. Given
the arguments for FactivityT , there is no reason to think that only beliefs based
on entailing evidence can do that. So, Williamson is right that the justification
relation is not deductive, the remark is misleading.
5.3 Externalism and Epistemological Disjunctivism

Even if the arguments for FactivityJ are sound, the case for externalism is not
closed. Internalists could simply deny that it is possible to have justified beliefs
4 2000, pp. 197.
5 RizzieriForthcoming stresses this point in his discussion and he is right to do so.
6 2000, pp. 9.
concerning contingent matters of fact that do not supervene upon the internal
facts. It will take some work to defend the idea that someone can consistently
say that there can be no false, justified beliefs while at the same time say that
many of our beliefs about the external world are justified. One view to consider
is a kind of epistemological disjunctivism that says we have justified beliefs
about our surroundings because our beliefs are based on factive reasons, mental
states that embrace worldly facts.7 This is one way to go, but I would prefer
not to enter into any entangling alliances with any disjunctivist view. It might
seem, however, that I need epistemological disjunctivism to avoid skepticism.
To see why, consider Cohen’s objection to FactivityJ :
The strongest view one could take regarding the truth connection
is that taken by Descartes. The Cartesian view is that justification
logically entails truth. To put it schematically: It is a conceptual
truth that, if conditions C justify belief B for subject S, then C
logically entails that B is true.
He says that this Cartesian view faces a decisive objection:

The legacy of the Cartesian view is scepticism. Descartes demon-
strated this in the first meditation that no such connection is forth-
coming . . . Given any plausible specification of C for any S, it will
always be logically consistent to suppose that not B. That is what
the evil demon argument shows. Where, e.g., C comprises facts
about sensory data, and where B is a belief about the truth of some
empirical proposition, it is always logically possible that the evil de-
mon has arranged for C to obtain where B is false. Not wishing to be
saddled with this sceptical result, most contemporary philosophers
have rejected the Cartesian view and have opted instead for a falli-
bilist theory of justification. A fallibilist theory allows that where C
makes B justified for S, it is still possible that B is false.8
The worry is that Factivity entails skepticism because it entails a kind of infal-
libilism.
As stated, it is easy to deal with this objection. According to the knowledge
account of justified belief, S’s belief that p is justified iff that belief constitutes
knowledge.9 Because knowledge is factive, the knowledge account is committed
to FactivityJ . To show that the knowledge account is committed to skepticism,
you have to show that we cannot have knowledge. If you could somehow show
that we do not have knowledge of the external world, we should embrace the
(alleged) skeptical consequences of FactivityJ . If we do have knowledge of the
external world (we do), the knowledge account implies that FactivityJ is true
and carries no skeptical consequences. Of course, the knowledge account is
7 See Brewer 1999, Gibbons Forthcoming, Neta and Pritchard 2007, McDowell 1998, and
Williamson 2000.
8 Cohen 1984, pp. 281.
9 See Sutton 2005.
wrong. Knowledge suffices for justified belief but justified belief is not sufficient
for knowledge. If less is required for justified belief than knowledge, then it
will be harder to show that we do not have justified beliefs than it will be to
show that knowledge is unattainable. Since the skeptic is wrong and no one can
“show” that we lack this knowledge, we have nothing to fear from the skeptic
simply because we think that justified belief is like knowledge in that it requires
truth.
Cohen’s argument seems to say, in effect, that you should not combine ex-
ternalism about the conditions that determine whether a belief is justified with
internalism about the conditions that can justify belief. So long as the exter-
nalist denies that the factors common to you and your systematically deceived
counterpart exhaust the conditions that determine whether your beliefs are jus-
tified, the objection seems to have no force. But, perhaps this brusque dismissal
is too brusque. Conee agrees that Cohen’s objection is flawed, but he thinks its
flaws are easily remedied:
Suppose you have the belief that someone is speaking. You infer
this from your justified belief that Mr. Jones is speaking. Thus,
your external world belief that someone is speaking is a belief for
which you have an entailing justification, your justified belief that
Jones is speaking. However, it is quite plausible that your belief
that Jones is speaking must itself be justified in order to justify
any other belief. In general, it is quite plausible that a belief can
contribute epistemic justification only if the belief is justified. When
we consider candidate justifications for entailing justifiers like the
belief that Jones is speaking, it becomes plain that at some point
there is always a proposition that is justified without being entailed
by its justification. In the present instance, the nonentailing justifier
may well be your justification for the belief that Jones is speaking.
This belief may be justified by the experience of its seeming to you
that you hear what you seem to recall to be the sound of Jones’
voice. This experience does not necessitate that Jones, or anyone
else, is speaking. But it may be all that you have, and all that
you need, in favor of the belief that Jones is speaking. Exactly
how this justification works is another matter ... [I]n any plausible
view, at some point in the justification of each external world belief
that is justified, there is justification without entailment. When this
further assumption is added to the assumption that the entailment
account is correct, we have a valid argument for the conclusion that
no external world belief is well enough justified to be known ... The
entailment claim is the argument’s least plausible assumption. So, if
the skeptical conclusion is to be avoided, then the entailment account
of the truth connection is the best candidate for rejection.10
Some now deny Conee’s claim that we lack a sufficient stock of entailing jus-
10 Conee 2004, pp. 245.
tifying reasons for our worldly beliefs, and would reject his argument on the
grounds that he assumes we do not. I confess that I find myself sympathetic to
some of what Conee says, so helping myself to this stock of factive reasons is
not the sort of thing I want to do here.
To address this worry, we need to make a brief detour into the issues having
to do with the nature of experience and the reasons provided by experience.
Let’s start with a tempting, popular, but flawed line of reasoning:
If there is a cat in the corner and it looks to you as if there is, you have
good reason to believe there is a cat in the corner. Indeed, you might
have good enough reason to believe this. Since it can look to you as if
there is a cat there even if the nearest cat is miles away, experience
can provide you with a sufficiently good reason for belief even if
there is no cat. The reasons provided by veridical experience give
you the right to believe. The same is true for the reasons subjectively
indistinguishable hallucination provide. If so, the justificatory work
is done by the elements common to hallucination and perception.
These elements do their justificatory work just as well in cases of
perception and hallucination. After all, you have the same evidence
either way.
For their part, the mentalists say that there is nothing wrong with this sort of
reasoning. On their view if two individuals have the same evidence, the same
reasons bear on their beliefs and it is impossible for two individuals to have
different evidence if they happen to be non-factive mental duplicates. As they
see it, the conditions that determine whether your experience is veridical or not
do not determine the nature of your experience, they do not determine what
evidence you have, and so they have nothing to do with the proper description
of your reasons for believing any of the worldly propositions you believe. So,
they endorse:
Veridical experience and subjectively indistinguishable hallucination
provide you with the same evidence for your worldly beliefs (Same
Reasons).11
McDowell agrees that this line of reasoning is defective. He thinks the mistake
is in thinking that since it can look as if there is a cat in the corner whether
there is or not if it looks as if there is a cat there you have the same reason
to believe there is, cat or no cat. The conditions that distinguish veridical
experience from hallucination are essential to perceptual knowledge. Everyone
agrees to that. Knowledge, he says, is a standing in the space of reasons.12 So,
the difference between perceptual knowledge and ignorance requires a difference
in the reasons there are to believe worldly propositions in the case of veridical
perceptual experience and subjectively indistinguishable hallucination. As he
sees it, Same Reasons leads to skepticism. To avoid skepticism, he thinks we
should say:
11 This is a popular view. See also Huemer 2006 and Silins 2005.
12 McDowell 1995, pp. 877.
The evidence veridical experience provides is better than the evi-

dence provided by subjectively indistinguishable hallucination in the
sense that veridical experience provides evidence that hallucination
does not (Better Reasons).13
Of course, those who accept Same Reasons typically reject skepticism, but he
thinks they have no right to do so.
Suppose Same Reasons does lead to skepticism. Does Better Reasons save
you from the skeptic? No, he says, not on its own. If you were to say that
the nature of the psychological states and events by virtue of which it looks
to you as if such and such is the case are the same in the case of perception
and hallucination, the view leads right back to skepticism. On such a view,
the qualities by virtue of which your reasons are thought to be better would be
blankly external to you. For McDowell, this is verboten:
The root idea is that one’s epistemic standing . . . cannot intelligibly
be constituted, even in part, by matters blankly external to how it
is with one subjectively. For how could such matters be other than
beyond one’s ken? And how could matters beyond one’s ken make
any difference to one’s epistemic standing? . . . But the disjunctive
conception of appearances shows a way to detach this “internalist”
intuition from the requirement of a non-question begging demonstra-
tion. When someone has a fact made manifest to him, the obtaining
of this fact contributes to his epistemic standing on the question.
But the obtaining of the fact is precisely not blankly external to his
subjectivity, as it would be if the truth about that were exhausted
by the highest common factor.14
The one point on which McDowell and the mentalists seem to agree is that noth-
ing can confer any justificatory benefit or constitute some superior epistemic
position unless it corresponds to some mental difference that distinguishes you
from those who do not enjoy the benefit. Because he thinks that experience
can embrace worldly facts, McDowell is happy to say that the veridicality of an
experience can provide a justificatory benefit by virtue of which you enjoy a su-
perior epistemic standing when compared to the epistemic standing of someone
undergoing an indistinguishable hallucination. The mentalists either deny that
there are factive mental states of the kind McDowell thinks marks the difference
between the case of perceptual knowledge and hallucination or deny that such
states can confer any benefit upon you.
As the passage indicates, the problems that arise for the mentalists arise for
anyone who denies:
An appearance can either be a mere appearance, as with halluci-
nation, or a fact made perceptually manifest. The nature of the
psychological states and events by virtue of it looks to you as if p
13 This is also the view defended by Williamson 2000.
14 1998, pp. 390.
depends upon whether you are hallucinating or your experience is

veridical (Disjunctivism).
Thus, McDowell’s target seems to include most of the orthodox accounts of
epistemic justification in that these accounts deny that the veridicality of a
particular experience is part of what determines what justifies beliefs formed in
response to taking such experiences at face value.15
We can summarize McDowell’s epistemological argument for Disjunctivism
as follows. Given the internalist intuition that epistemic standing cannot be con-
stituted by factors blankly external to you or beyond your ken, Same Reasons
leads to skepticism. Knowledge is an epistemic standing and Same Reasons as-
serts that the conditions essential to that standing are blankly external to you in
the case of veridical experience. If you endorse Better Reasons but deny Disjunc-
tivism, you do not avoid the skeptical consequences of Same Reasons because
your view commits you to saying that the conditions essential to knowledge are
beyond your ken even in cases of veridical experience. The only alternative to
skepticism is a view that combines Better Reasons with Disjunctivism. So, on
the plausible assumption that we have perceptual knowledge, we have to reject
the traditional conception of experience.16
Those who take a dim view of the argument for Disjunctivism might say that
McDowell tries to derive an implausible claim about the nature of experience
from implausible claims about the justification of perceptual belief. Not only is
he wrong to think that Same Reasons leads to skepticism and wrong to endorse
Better Reasons, he is wrong to think Disjunctivism could explain Better Rea-
sons. I also have some reservations about his argument, but the problem with
his argument is not that it assumes Better Reasons. Not only is Better Reasons
true, so is this stronger thesis:
Only in the case of veridical perception do you have good enough
reason for your worldly beliefs. If you believe on the basis of hallu-
cination, you cannot believe with justification. You can believe with
sufficient justification if your experience is veridical (Good Enough).
The questionable step in McDowell’s argument is the step where he says that
claims like Better Reasons and Good Enough commit us to Disjunctivism. Once
we see why Better Reasons and Good Enough are true, we can see why we do
not need to take a stand on whether Disjunctivism is true.
15 Thus, McDowell’s target is broader than Conee and Feldman’s mentalist view or Huemer’s
phenomenal conservatism. Externalist views such as Goldman’s 1986 reliabilism, Bergmann’s

2006 proper-functionalism, and Comesaña’s 2010 evidentialist reliabilism are all targets. It
is not at all clear that those who defend these views defend Better Reasons, but even if
they thought that they had the resources to do so, they cannot point to causal differences
between cases of hallucination and perception as the feature that explains why the reasons
experience provides in the good case are “better” than those provided by an indisitnguishable
hallucinatory experience.
16 Remember that McDowell’s ambitions are relatively modest. He hopes to describe the
conditions under which a kind of skeptical argument fails, not provide premises for refuting
the skeptic. See his 2008, pp. 378.
5.3.1 Preliminary Objections

Some object to McDowell’s view on the grounds that it is committed to a kind
of infallibilism and that this infallibilist view committed to an unpalatable form
of skepticism.17 If knowledge is a standing in the space of reasons and the
difference between knowledge and ignorance cannot be “blankly external” to
you or beyond your ken, the difference between knowledge and ignorance can
never simply be a difference in whether the relevant belief is true.18 In fact,
the difference between knowledge and ignorance can never be due to factors
that do not supervene upon a full description of the reasons you have to believe.
Thus, it seems that the difference between knowledge and ignorance must reflect
a further difference in the reasons you have for your beliefs, so it seems that
McDowell is committed to this thesis:
If you know p, you believe p on a different basis than anyone who
believes p but does not know p (Different Basis).
Different Basis entails that if you know p, you believe p on a different basis than
anyone who fails to know p. So, in ∼p-worlds, subjects in very similar epistemic
situations believe p on a different basis than you do. But, that just means that
if you know p, you believe on a basis that is incompatible with ∼p. This just is
the infallibilist view:
If you know p, your belief must be based on something incompatible
with ∼p (Infallibilism).
The argument from Infallibilism to inductive skepticism is straightforward. In
cases of inductive inference, the basis for your belief is a basis you could have
even if your belief is mistaken. If I believe correctly that the n+1st draw from my
bag will be black on the basis of n observations of black draws and you believe
incorrectly that the n+1st draw from your bag will be black on the basis of n
observations of black draws, there is a perfectly good sense in which we believe
what we do on the same basis. I get things right, but you do not. According to
Infallibilism, I cannot know unless everyone who believes on your basis knows.
But, you did not know the next marble would be black. You pulled the first
white ball.
To block the objection, we have to deny Different Basis. Instead we should
say that it is possible to know p even if you believe p on the same basis as
someone who mistakenly believes p. The lesson is supposed to be that if you
reject Different Basis, you have to also reject Better Reasons and Good Enough.
Unless Different Basis is true, your (allegedly) better reasons cannot make you
better off, epistemically, because the qualities by virtue of which your reasons
are better are blankly external to you or beyond your ken. If you deny Better
Reasons, you also have to deny Good Enough. How could you have the same
reasons as someone else and only one of you have reasons that are good enough?
17 See Comesana 2005.
18 A point noted by both Comesana 2005 and van Cleve 2004.
If Same Reasons is true and the reasons in the case of hallucination are not good
enough to justify belief, those reasons cannot be good enough to justify belief
in the case of veridical experience. My guess is that McDowell and McDowell’s
critics might agree that Better Reasons and Good Enough require Different
Basis. If they agree on this point, then I think they are both making a mistake.
We shall see that externalists can deny Different Basis even if they accept Good
Enough and Better Reasons, but this is something we shall return to below.
Let us consider a second objection. In explaining how it is possible to have
the kind of knowledge the skeptic denies we could have, McDowell rejects Same
Reasons and argues that Disjunctivism is needed to explain Better Reasons.
Nothing could be a reason that contributes to the justificatory standing of your
belief unless that reason is part of your basis for believing. For reasons we
have touched on, having such a reason requires having a kind of unmediated,
unbroken mental contact with the facts you come to know via perceptual experi-
ence. Conee objects to this on the grounds that Disjunctivism could not explain
Better Reasons because such an explanation would run afoul of the following
principle:
A subject’s justification for a belief is not stronger than a second
subject’s justification for the same belief, if their respective justifi-
cations are prone to being equally well defeated by the same defeaters
(Defeat).19
If Defeat says that two reasons defeated by the same defeater cannot differ in
strength, the principle is not very plausible. A full house is stronger than a pair
even if four aces would beat both hands.
On a more charitable reading, Defeat says that the justification provided
by two conscious experiences is equally strong if these justifications are liable
to defeat by all the same defeaters. This is more plausible, but still hardly
self-evident. It is not obvious that the strength of a reason can be measured in
terms of what can defeat it. Forget about reasons for a moment and think about
boxers. Nobody can defeat Mustard in a boxing match. Apart from Mustard,
nobody can defeat White or Plum. White and Plum cannot box against each
other because they share gloves. Plum and Green cannot box each other because
they share trunks. No one can box without both gloves and trunks. Suppose
you have debts that you can only repay if you come into some quick money.
The only way to come into some quick money is to set up a boxing match
for tomorrow night. You have to bet on the boxer you send to the ring and
you manage White and Plum. You do not know whether the opponent will be
Green, Mustard, or someone else. You know the fight will not take place if you
try to send Plum up against Green, so there is stronger reason to send in White.
While White and Plum would lose to the same boxers, you have stronger reason
to send White in. One lesson to take from this is that if reasons are like boxers,
strength cannot simply be measured in terms of who could defeat the reason or
19 Conee 2007, pp. 19.
boxer you have. Surely some reasons are like boxers. Reasons to pick between
boxers are reasons and they behave a bit like boxers.
Not only do I think we have reason to doubt Defeat, we have pretty good
reason to think that Conee’s objection fails. He thinks veridical perceptual
experience and subjectively indistinguishable hallucination are equally well de-
feated by the same defeaters because they are subjectively indistinguishable. If
his objection is sound, it shows that if two conscious experiences are indistin-
guishable, the reasons they provide for your beliefs are equally strong and these
experiences will justify the same beliefs to the same degree. Consider two theses
about indiscriminability and justification:
TransitivityI : (x)(y)(z)[(Ixy & Iyz) → Ixz)].
TransitivityJ : (x)(y)(z)[(Jxy & Jyz) → Jxz)].
According to TransitivityI , a and c must be indiscriminable or indistinguishable
for you if you cannot distinguish a from b and cannot distinguish b from c.
According to TransitivityJ , if a and b justify the same (i.e., justify the same
beliefs to the same degree) and b and c justify the same, a and c must justify
the same as well.
Arguably, TransitivityI is false. Suppose a, b, and c are perceptual expe-
riences you have while looking at three different paint chips in good viewing
conditions. It seems possible for a and b to be indiscriminable, b and c to
be indiscriminable, even if you can discriminate a from c. If these chips differ
only slightly, you might be unable to distinguish the first from the second and
the second from the third even if you can discriminate the first from the third
by sight.20 What goes for the chips goes for the perceptual experiences of the
chips. Although it seems that TransitivityI is false, TransitivityJ is true. For
TransitivityJ to be false, there would have to be some proposition, p, such that
the degrees to which a and c justified belief in p differed even though both a
and c justified belief in p to the same degree that b does. This is impossible.
With this in mind, I shall argue that Conee cannot use Defeat to show that
Disjunctivism cannot explain Better Reasons. His objection assumes:
(1) (x)(y)(Ixy → Jxy).
Let me introduce a further assumption:
(2) (x)(y)(∼ Ixy →∼ Jxy).
The justification for (2) is that in discriminating between two things, you can
know that these two things are distinct.21 If you can discriminate between a
and c, you will have stronger reasons for believing that you are undergoing a
while undergoing a than you will have for believing that you are undergoing
some experience you can knowingly discriminate from a (e.g., c).
If TransitivityI is false, we can coherently suppose that a is indiscriminable
from b, b is indiscriminable from c, but you can discriminate between a and
20 For discussion, see Williamson 1994, pp. 237-44.
21 See Williamson 1990.
c. (1) entails that a and b justify the same beliefs to the same degree. It also
entails that b and c justify the same beliefs to the same degree. It follows by
TransitivityJ that a and c justify the same beliefs to the same degree. But, if (2)
is correct, this contradicts the further assumption that a and c are experiences
that you can discriminate between. The most obvious way to avoid this con-
tradiction is to deny (1). If (1) is false, Conee’s Defeat principle is no threat to
Better Reasons. His objection was that McDowell’s view implied that it is pos-
sible for indistinguishable states to provide different reasons for belief, reasons
that differed in strength. His objection assumed that indistinguishable states
can be defeated by precisely the same considerations and that states that can be
defeated by precisely the same considerations cannot offer reasons that differ in
strength. We know now that these assumptions cannot both be correct. Either
the reasons provided by two indistinguishable states are not defeated by the
very same considerations or the reasons provided by two states can be defeated
by the same considerations even if these states provide different reasons.
There is a deeper problem with Conee’s objection. It is tempting to think
that claims like Better Reasons and Good Enough are only true if the reasons
we have in the case of perceptual knowledge are stronger than the reasons we
have in cases of hallucination. While we do have stronger reasons in the case of
veridical perception than hallucination, it is also important to remember that
strength of epistemic position is not simply a function of the strength of reasons
to believe. Normative standing is a function of both reasons for and reasons
against. This is something to keep in mind as we try to sort out the connections
between Better Reasons, Good Enough, Different Basis, and Infallibilism.
5.3.2 The Epistemological Argument for Disjunctivism

The argument for Better Reasons and Good Enough builds on the account of
evidence defended earlier. Why does veridical experience provide better rea-
sons than subjectively indistinguishable hallucination? Better Reasons follows
from IKSE and FactivityT . Because the scope of things that you know non-
inferentially in the case of veridical perception is greater than the scope of
things you can know non-inferentially when you undergo a subjectively indis-
tinguishable hallucination. If it looks as if there is a cat in the corner, you can
know non-inferentially that a cat is there if there is a cat. You cannot if there
is no cat. That there is a cat in the corner is a better reason to believe there
is a cat in the room than that it looks as if there is a cat in the corner. In the
case of veridical perception, you have both reasons. In the case of hallucination,
you have only one of these reasons. The argument for FactivityJ gives us an
argument for Good Enough. If you take experience at face value in the case of
veridical perception, there seems to be no reason not to hold such beliefs. If
you take experience at face value in the case of hallucination, there seems to be
a reason not to hold such beliefs–such beliefs lack a proper basis.
McDowell might agree with some of this, but insist that this does not go far
enough. Nothing in the arguments for Good Enough or Better Reasons told us
anything about the nature of perceptual experience. If the traditional view of
experience is left in place, all is lost. Why is that? Remember that McDowell
wanted to hold onto the internalist thought that your epistemic standing cannot
be constituted even partially by matters blankly external to you. Why not?
Because, he says, such matters are beyond your ken and what is beyond your
ken cannot make any difference to your epistemic standing.
If I understand the reasoning right, it goes something like this:
(1) If q is blankly external to your subjectivity, q is beyond your
ken.
(2) If q is beyond your ken, q cannot make a difference to your
epistemic standing.
(3) Thus, if q is blankly external to your subjectivity, q cannot make
a difference to your epistemic standing.
What does it actually mean to say that something is blankly external to your
subjectivity? One interpretation that seems plausible is given by van Cleve–
q is blankly external to your subjectivity iff a complete description of your
psychological states neither entails q nor ∼q.22 What does it mean to say that
q is beyond your ken? Whatever it means, we know that McDowell’s conclusion
is that whether you know something cannot depend upon q if q is beyond your
ken. And, so, let us say that if q is beyond your ken, you are not in a position
to know q non-inferentially. If, however, you are in a position to know q non-
inferentially, q is not beyond your ken.
We can now restate the argument as follows:
(4) If a full description of your psychological states entails neither q
nor ∼q, you cannot know whether q know non-inferentially.
(5) If you cannot know whether q non-inferentially, q cannot make
a difference to the justificatory status of your beliefs.
(6) Thus, if a full description of your psychological states entails
neither q nor ∼q, q cannot make a difference to the justificatory
status of your beliefs.
Does this compel us to accept Disjunctivism?
McDowell is probably right that if something is beyond your ken, it cannot
confer any epistemic benefit upon you. However, I think it is a mistake to
say that your epistemic standing cannot be determined, in part, by features
that are beyond your ken. In fact, McDowell should say as much. On his
view, there can be matters beyond your ken that can partially determine the
justificatory standing of your beliefs–that you are in the bad case, for example,
is not blankly external to your subjectivity but it is, nevertheless, something
that partially determines your epistemic standing. It does if Disjunctivism is
true and either Better Reasons or Good Enough is true. Better Reasons and
Good Enough say that there is a justificatory difference between the good and
22 van Cleve 2004, pp. 486.
bad cases and Disjunctivism says that this corresponds to a difference in the
psychological states and events by virtue of which it looks to you the way it
does in these cases. Thus, we should restate (5) and (6) as follows:
(5’) If you cannot know whether q non-inferentially, q cannot make a
difference to the justificatory status of your beliefs by conferring
any sort of epistemic benefit upon you.
(6’) If a full description of your psychological states entails neither
q nor ∼q, q cannot make a difference to the justificatory status
of your beliefs by conferring any sort of epistemic benefit upon
you.
With this fix in place, we have our argument for Disjunctivism.
McDowell is right to deny that something inaccessible to you can confer
upon you an epistemic benefit. Consider some examples. Suppose someone
does something there is reason not to do. Suppose that there happens also to
be reason to do it. Bernie shoots a kid carrying a weapon (that is something
there is a pro tanto reason not to do), but doesn’t know that the kid is carrying a
weapon. Maybe the kid was going to use that weapon to attack a bunch of people
(perhaps that’s a pro tanto reason to shoot the kid). Since this has nothing to do
with Bernie’s reasons for shooting, it is hard to see how facts about what the kid
was carrying and what the kid planned to do with his weapon could be cited to
justify his deeds. Even if Bernie were made aware of the kid’s weapon, if Bernie is
shooting the kid just because he hates kids it is hard to see how these facts could
justify his conduct. To justifiably act against a reason, it seems that it is not
enough that there is overriding reason that happens to be out there somewhere.
It seems that this reason to act has to be the reason for which the subject acts
if that reason is going to be the reason in virtue of which some other agent’s
deeds are going to have a moral standing superior to the standing of Bernie’s
deeds. The reasons that count in favor of acting seem to contribute positively
to moral standing only if they play some motivational role. They cannot play
that motivational role, however, if they are beyond the subject’s ken. Indeed,
one argument for the claim that considerations beyond your ken cannot confer
any justification is predicated on the assumption that considerations can only
justify when they play some motivational role. If Bernie’s reasons for shooting
were not the reasons for which he shot, those reasons seem to do nothing to
justify his action even if he is aware of them but is motivated instead wholly by
malice. We do not need practical examples to make the point. One lesson you
might take from BonJour’s clairvoyant examples is precisely that considerations
that are inaccessible to you cannot be reasons that justify forming beliefs.
This much seems right. It seems to be the sort of thing that might lead
McDowell to say that there is something a subject in the good case is cog-
nizant of that explains why a subject in this case ends up with beliefs better
justified than beliefs formed in the bad case. The reasons that count against
acting, however, can contribute negatively to the normative standing of an ac-
tion without playing any motivational role. Moreover, the reasons that count
against acting can contribute to normative standing of an action even if the

agent is non-culpably ignorant of them. Think about cases where someone is
imprisoned for a crime that we later discover that they did not commit. In the
wake of this discovery, we discover that we have a duty of reparation and must
compensate the victim. Such reparative duties are, however, not mere duties of
beneficence. Such reparative duties should leave the victim better off than they
were, but unlike duties of beneficence the duty is one that arises between the
victim and the subject(s) that harmed the victim. These duties can exist when
the parties responsible for imprisoning the victim were non-culpably ignorant of
the fact that the accused was innocent. (Just think about cases where reliable
eyewitnesses came forward to suggest that the victim was guilty and it was only
later developments in forensic science that exonerated the person imprisoned.)
These duties only exist when the agent acted against some genuine reason that
contributed negatively to the normative standing of the original act. (Other-
wise, helping the wrongly accused would not be a response to some past wrong
and would be a mere duty of beneficence.) If this is right, the act of putting
the innocent victim away and forcing them to suffer the hardships of prison
was wrongful and wrongful for reasons that all relevant parties could have been
non-culpably ignorant of.
Examples like these suggest that there is an important asymmetry between
reasons for belief or action and reasons against.23 Even if reasons for believing or
acting cannot contribute to normative standing unless the subject is cognizant
of them, reasons against can contribute negatively to normative standing when
the subject is not cognizant of them. McDowell himself seems to concede this
much if he accepts Better Reasons and accepts that subjects in the bad case
are in no position to realize that their reasons are defective. Since comparative
normative standing is a function of both the reasons for and reasons against,
there is a serious lacuna in McDowell’s argument for Disjunctivism. Why? Well,
suppose there are reasons not to believe p on the basis of how things look when
its looking as if p is due to hallucination. It could be that beliefs in the good
case are comparatively better off even if there is not something internal to the
subject’s experience that is distinctive of the good case. The disparity is due
entirely to reasons not to believe that are present only in the bad case that make
beliefs formed in that case defective.
Notice that there is a way of accomodating the internalist point about rea-
sons to believe. None justify if they are beyond your ken. However, if he must
concede that reasons not to believe can do their work by making it wrongful to
believe even if they are beyond the subject’s ken, we can explain the difference
in epistemic standing between the good case and bad in terms of this difference
in the reasons not to believe. We could say, if we wanted, that there were the
same reasons to believe in these cases. Thus, it seems that the right to believe
does not depend upon the possession of reasons that entail that the belief in
question is true. One might have such reasons on hand, say, in the case of
non-inferentially justified belief, but there is no necessary connection between
23 For further discussion of this asymmetry, see Gardner 2007.
rightly held belief and entailing evidence.

In this way, I think we can block the argument for Disjunctivism without
necessarily denying (5’) or (6’) and so without denying what seems right about
the internalist thought. What about (4’)? Suppose we were to deny it and
adopt some picture of experience on which there is no psychological difference
that distinguishes the good case from the bad. We might say this to McDowell.
When, say, I know non-inferentially that the cat is in the corner because it looks
as if it is, the fact that the cat is in the corner is not beyond my ken because I
know it non-inferentially. It may well be that the fact is not one entailed by a
full description of my psychological states, but it is nevertheless not beyond my
ken. And, for this reason, it matters little whether it is blankly external to my
subjectivity.
McDowell might say that this just begs the question against him, but in my
defense I would say that I am basically engaged in the same project that he
is. He is supposing we have a kind of perceptual knowledge and is trying to
describe the conditions under which it is possible to have it.24 This is what I
have done, and I have described conditions under which we have such knowledge
without assuming that we recieve any epistemic benefits by virtue of conditions
that are beyond our ken. Nothing we have seen thus far compels us to say
that the conditions under which we could have such knowledge is that we are
in mental states that entail that the beliefs in question are true. So, until we
see an argument for (4’) that goes beyond the support it can receive from the
internalist thought that nothing beyond your ken can contribute positively to
the justificatory standing of your beliefs, we need not worry so much about
denying (4’).
If there is no reason to endorse (4’) and we can consistently maintain that
Better Reasons and Good Enough are true, we avoid the worry that arose above
insofar as nothing in the argument for Better Reasons or Good Enough supports
Different Basis. Since nothing we have said thus far commits us to Different Ba-
sis, it seems nothing we have said thus far commits us to the kind of infallibilist
view pinned on McDowell that seems to generate inductive skepticism. So, we
can now see our way through the dilemma. On the one horn was the disjunc-
tivist view that seemed committed to Infallibilism and so committed to a kind
of skepticism many of us want to avoid. On the other was the fallibilist sort
of mentalist view that I have argued earlier leads to skepticism. If, as I have
argued, the epistemic standing of a belief is determined both by reasons to hold
it and reasons not to hold it and the latter need not be reasons that are accessi-
ble to you, the total set of conditions that determine the justificatory standing
of your belief need not supervene upon the basis of your belief or the basis of
your belief taken together with all of your evidence. So, asserting that truth is
required for justified belief does not commit you to the view that the right to
believe p depends, inter alia, upon possessing some antecedent reason to believe
p that is inconsistent with ∼p. And, asserting that Infallibilism is false does
not commit you to the view that you can have the right to believe p simply by
24 See his 2008, pp. 379.
having the right sort of basis for your belief no matter how things external to
you turn out.
5.4 Reasons for Action

In this section, I shall argue that FactivityJ is a consequence of FactivityP . The
justification of a belief depends upon the facts external to us because reasons
for action consist of facts external to us and the justification of a belief depends
upon whether it provides us with these reasons.
A widely held view is that belief aims at truth.25 No one thinks this could
literally be true, so I suppose that the widely held view is that this talk of
belief’s aim is a useful metaphor. A natural way to unpack this metaphor is in
functionalist terms. True beliefs can do what beliefs are supposed to do. False
beliefs cannot. Why are false beliefs constitutionally incapable of doing what
beliefs are supposed to do? Here is a hypothesis. Beliefs are supposed to provide
us with reasons from which we can then reason to conclusions about what to do
or the way the world is. Since reasons are facts, false beliefs provide us with no
reason at all.26
This way of understanding the aim of belief works only if we work from the
assumption that reasons are facts. This is controversial if the kind of reason
at issue is a motivating or explanatory reason. This is not the kind of reason I
have in mind. The reasons beliefs are supposed to provide are normative reasons,
reasons to act or believe. A simple argument suggests that there cannot be false,
justified beliefs if reasons for action are facts:
(1) The belief that p is true can only contribute a normative reason
to practical deliberation if p is true. (If p is false, the belief
that p could only pass off a non-reason as a reason if included in
deliberation.)
(2) There is no normative reason to include the belief that p in
practical deliberation if the belief would merely pass off a non-
reason as a normative reason if included in deliberation.
(3) There is, however, a normative reason to exclude the belief that
p from practical and theoretical deliberation if the belief would
pass off a non-reason as if it were a genuine normative reason if
included in deliberation.
(4) If there are reasons not to believe or act, the belief or act can
be justified only if there are equally strong reasons to believe or
act.
(C1) Thus, only true beliefs can be justifiably included in practical
deliberation.
25 See Velleman (2000), Wedgwood (2002), and Williams (1973).
26 See Collins (1997), Dancy (2000), Hyman (1999), Scanlon (1999), and Williams (1981).
(5) A belief cannot be justified if it cannot justifiably be included in

the process of practical deliberation.
(C2) Thus, only true beliefs can be justified.
Since the argument’s first premise is an obvious consequence of FactivityP , we
should focus on the argument’s other premises. (2) strikes me as being rather
intuitive, but if someone did not accept (2), they could try to offer a coun-
terexample. The only kind of counterexample I could imagine would have this
structure. We do not know if the number of stars is (a) odd or (b) infinite or
even. Suppose someone offers us a large sum of money to assert sincerely that
the number is odd. You cannot do this unless you believe that the number is
odd, so the money is a reason to believe that the number is odd. It is also a
reason to reason from the belief that the number is odd to the conclusion that
you should say believe and say that the number is odd. If you have reason to
believe what you say is true, you have reason to get yourself to believe that the
number is odd by any means that would be sufficient. If banging your head
against the wall would do the trick, go bang.
Examples like this are unhelpful for two reasons. First, it would be too easy
to rewrite the argument and say that in the absence of such reasons, you cannot
justifiably believe what is false. Since those who deny FactivityJ do not do so
on the grounds that there can be practical reasons to cause yourself to form
false beliefs, denying (2) on these kinds grounds would do nothing to spare the
orthodox accounts of justification from the argument. Second, the reason in
question is the wrong kind of reason to be a reason to believe. It is merely a
reason to cause yourself to believe, and so it is not a reason that bears on the
normative standing of the belief. A standard test for distinguishing the right
kind of reasons from the wrong kind of reasons is to ask whether in accepting
the reason you will thereby settle a question about what to do or believe by
forming an intention or a belief.27 The (alleged) reason fails the test.
According to (3), there a normative reason to exclude non-reasons from
reasoning? If there were no such reasons, then treating non-reasons as if they
were genuine reasons would be reason enough to do so. It is often said that
ought implies reason, so if there is no reason not to φ, it must not be that you
ought not φ. So, φ-ing would be permitted. In other words, treating something
as a reason would not be the sort of thing that called for a justification. Does
treating something as a reason call for justification? It seems so. We criticize
people for reasoning from assumptions that they had no good reason to think
were true. Assuming this practice is not badly misguided, it seems that (3)
must be true.
(4) is a plausible claim about justification and conflicting reasons. There
can be some justification or some reason for φ-ing even when there are stronger
reasons not to φ but the question is not whether there is some justification
for φ-ing, but what it takes for there to be sufficient justification for φ-ing. I
doubt there can be sufficient justification under the very same conditions in
27 See Hieronymi 2005.
which there are reasons that defeat the case for φ-ing. The point seems rather
obvious. If all it took to justify an action was that there was some reason to
do it, all sorts of actions there are strong cases against could be justified by the
trivial reasons that count in favor of them. It would be excruciatingly painful
to stick a fork into a live outlet, but it would be interesting to know what it is
like to receive that sort of jolt. If those are the operative reasons, I think I have
just explained why you should not stick a fork into the outlet.
As for (5), the thought behind this premise is that if you have the right
to believe, you have the right to treat what you believe as reasons for further
beliefs. Essentially, this is an application of J-Closure. Does that right end
when that belief is included in deliberation about what to believe about what
to do? J-Closure says that it does not. What is the relation between practical
deliberation and deliberation about what to believe about what to do? Don’t
we do the one by doing the other. All that (5) says is that there will not be
a decisive epistemic case against treating p as if it is a reason for action when
there is not a decisive case against believing p in the first place. It is difficult for
me to imagine a case in which there is a decisive purely epistemic case against
reasoning from p where that does not constitute a case against believing p that
threatens the justificatory status of belief in p.
Let’s take stock. We have seen two lines of argument from claims about
the ontology of reasons to a conclusion about the justification of belief. We
can summarize the argument as follows. The justification of a belief depends,
inter alia, upon whether what you believe can properly be treated as reasons for
further beliefs. Whether you can properly treat what you believe as a reason
depends upon whether it is a reason. Since reasons are facts, beliefs have to fit
the facts.
5.5 Reasons and Motivation

What happens when someone acts for a reason? Someone acts and she acts for
a reason. This seems like the obvious answer, but it is not at all obvious how
it could be right. If the agent were to act for a reason, there would have to be
a reason and she would act for it. Some powerful arguments suggest that such
a thing could never happen. So, our question might rest on a mistake. What
happens when someone acts for a reason? What happens when you divide the
number of horses and unicorns by the number of elves?
Some who defend Psychologism deny that motivating and normative reasons
belong to the same ontological category. It seems they deny that it is possible
that the reasons we act for are in the right sort of category to be good reasons.28
Normative reasons are the reasons that apply to us, make demands on us, and
count in favor of an action. These are facts about the situation or worldly
states of affair that an agent has in mind when deciding what to do. Motivating
reasons are states of the agent or the contents of those states. They help explain
28 See Dancy 1995 and 2000
why the agent behaves as she does. They say that there is nothing in us or in
the world that plays both roles:
When we have such a reason, and we act for that reason, it becomes
our motivating reason. But we can have either kind of reason without
having the other. Thus, if I jump into the canal, my motivating
reason was provided by my belief; but I had no normative reason
to jump. I merely thought I did. And, if I failed to notice that the
canal was frozen, I had a reason not to jump that, because it was
unknown to me, did not motivate me.29
The problem with such a view is that it denies what seems obvious to many of
us. When someone acts for a reason, there is a reason that is at least potentially
a valid reason and the agent acts for it. If all goes well, the agent’s reason for
acting was a good reason.
Dancy has argued that if this is so, we ought to think of both normative
and motivating reasons as constituted by the worldly facts or states of affairs
we have in mind when acting rather than the states of mind or the contents
of those states.30 An acceptable theory of reasons, he says, should accomodate
these two constraints:
Any normative reason is capable of contributing to the explanation
of an action done for that reason. (Explanatory Constraint)
Any motivating reason must be capable of being among the reasons
that count in favor of acting. (Normative Constraint)
The problem with his view is that it too seems to deny what is obvious to many
of us. When the agent acts on a mistaken belief, the agent’s reason for acting
cannot be a worldly state of affair or fact because the facts do not fit the agent’s
beliefs.31 So, what happens when that happens? Dancy says that it sounds
“too harsh” to say that such an agent acts for no reason at all.32 So, it seems
that the reason for which the agent acts must be an attitude or the content of
an attitude. If this is right, there must be something wrong with the argument
against Psychologism.
Someone could respond to Dancy’s argument against Psychologism by deny-
ing that Explanatory and Normative Constraints. They could deny that the
reasons we act for and the reasons there are to act belong to the same onto-
logical category. Instead, they could argue that Dancy was wrong to say that
Psychologism violates the Explanatory and Normative Constraints.33 Turning
29 Parfit 1997, pp. 99. Smith 1987 is also often saddled with this sort of view.
30 See Dancy 2000.
31 Gibbons 2009, Hornsby 2007, Lord 2008, Miller 2008, Turri 2009, and Wiland 2002 argue
that cases of error cause trouble for Dancy’s view.

32 Personal communication.
33 Miller 2008, Schroeder 2008, and Gibbons Forthcoming all defend views that are supposed
to accomodate both the Normative and Explanatory Constraints. They all reject the view that
motivating reasons are worldly facts or states of affairs. For Miller, motivating and normative
reasons are Fregean propositions. For Schroeder, both kinds of reasons are propositions, but
Dancy’s argument on its head, some now defend views on which both norma-
tive and motivating reasons are either our attitudes or the contents of these
attitudes.
Arguments from error make it hard to give up Psychologism about motivat-
ing reasons. In this section, I want to do two things. First, I want to argue that
Dancy was right to reject Psychologism in both of its forms. Second, I want
to offer a response to the argument from error that saves what is right with
Dancy’s view. Is it possible to act for good reasons? The argument from error
does not force us to deny that it is even if we insist that the good reasons are
typically facts about the situation or states of the world rather than states of
mind.
5.5.1 Motivational and Normative Psychologism

Motivational Psychologism, as the name suggests, is a view concerning the on-
tology of motivating reasons, the reasons for which we act. Normative Psy-
chologism is a view concerning the ontology of normative reasons, the reasons
there are to act that make demands on us, apply to us, or count in favor of an
action. On the assumption that the Explanatory and Normative Constraints
are correct, Motivational and Normative Psychologism go hand in hand. It will
be helpful to distinguish between two versions of Psychologism:
Normative and motivating reasons are constituted by your mental
states. (PsychologismS)
Normative and motivating reasons are constituted by the proposi-
tions that are the contents of your mental states. (PsychologismP )
On the first view, reasons are attitudes.34 On the second, reasons are provided
by your attitudes because they are the contents of those attitudes.35
The case against Motivational Psychologism builds on the case against Nor-
mative Psychologism and it might be useful to remind ourselves why that view
strikes many as being so implausible. Normative reasons by their very nature
seem like relational beasts. It is hard to imagine a world in which there are
reasons that are not reasons for such and such an agent and even if we add the
agents in, I think it is extremely difficult to imagine these reasons matching up
with their agents without being reasons-for the agents to do or avoid various
things. How does something become a reason-for, a reason for an agent to do
or avoid such and such a thing in such and such circumstances? There might
normative reason ascriptions are factive because the thing that is a reason is only a normative
reason if it corresponds to a fact. For Gibbons, both kinds of reasons are psychological states
of the agent. These states need not be non-factive mental states, mind you. He thinks that
knowledge is a state of mind.
34 Gibbons Forthcoming argues that normative and motivating reasons are states of mind
because states of mind make things reasonable and that is what reasons are in the business of
doing. Turri 2009 defends the view that motivating reasons are states of mind, but does not
endorse the further claim that normative reasons are also states of mind.
35 See Fantl and McGrath 2009, Lord 2008, Miller 2008, and Schroeder 2008.
be many paths to reasonhood, but the most obvious way something gets to be
a reason is by counting in favor or counting against. So, while some reasons
might not count in favor of anything at all, most of the reasons I can think of
are reasons precisely because they count in favor of doing something or count
against the doing of it. From here, it is a short step to the rejection of Normative
Psychologism. Unless we all harbor systematic and massive confusions about
what counts in favor of acting, the things that count in favor of, say, lending
a hand, are facts having to do with the external situation or worldly states of
affairs. We need not be too terribly picky about which of these options to set-
tle for because Normative Psychologism rejects both. It asserts that normative
reasons are the sorts of things that supervene upon our mental states, so they
are either states of mind or the contents of those states with their veridicality
or accuracy bleached out.
This first argument against Normative Psychologism is the implausible error
argument. Ordinary agents may well be mistaken about the facts on the ground
and so the actions they think will turn out favorable might not. That kind of
error is often unfortunate, but often understandable. It is implausible to accuse
ordinary agents of failing to know what it would take for actions to turn out
to be favorable in some respect or other on the grounds that it is facts about
the agent’s beliefs rather than facts the agent has beliefs about that determines
whether things turned out favorably for them. If I drink a tonic in the belief
that it will help my headache and it only makes the pain more intense, it would
be implausible to say that things turned out favorably for me. If counting in
favor is what confers reasonhood upon a reason, it is facts about the efficacy
of the tonic rather than my beliefs about its efficacy that determines whether
there was the reasons to drink I took there to be. If counting in favor cannot
confer reasonhood upon a reason, this just seems like one more implausible error
to impute to ordinary agents. If Moore had asked, “I know that such and such
counts strongly in favor of doing it, but what reason is there to do it?” we never
would have been so fascinated by the open question argument.
This is one objection to Normative Psychologism, but it is not the only one.
Myself, I think Normative Psychologism cannot do justice to our intuitions
about right action. In some recent defenses of Normative Psychologism, some
have argued for their view on the grounds that it preserves the link between
the right and the reasonable. Reasons, they say, are things that make things
reasonable and so the reasonable judgment of the morally conscientious agent
is the mark of the permissible.36 If this is right and the reasons demanded that
the agents acted against their own reasonable judgments about what to do,
the reasons would make unreasonable demands. But, reasons are, if anything,
reasonable things. And, if the reasons accede and do not demand that you do
not φ when it would not be reasonable from your point of view to do something
other than φ, φ-ing just is permissible for you. After all, if you ought not φ,
there is a reason not to φ and that reason is the winning reason. Remove that
reason, and the obligation not to φ goes away.
36 See Gibbons Forthcoming.
To see why this view is problematic, consider two plausible claims about
what it is reasonable to judge about what you should do:
It is reasonable for you to judge that you should φ if you are the
mental duplicate of someone who knows she should φ.37
It is reasonable for you to judge that you should φ if it seems intuitive
that φ-ing is the thing to do, these intuitions are robust, you have
no available reason to distrust these intuitions, you have no reason
available to think that φ-ing is not the thing to do, or you reasonably
judge that ψ-ing is necessary for some further end, ψ-ing, where you
reasonably judge that ψ-ing is the thing to do and that judgment is
not threatened by any available defeaters.38
It seems unreasonable to reasonably judge that you should φ and refrain from
φ-ing, so these claims tell us something about what is reasonable to do. The
problem with Normative Psychologism is this. Given the second account of
what is reasonable to judge and do, we end up denying that facts that the agent
is non-culpably ignorant of can bear on whether φ-ing is the thing for the agent
to do. Whether these are non-normative facts (e.g., facts about the effects of
action, the historical features of the situation) or normative facts (e.g., facts
about which normative principles are genuine, facts about which of the relevant
reasons are stronger), since these facts are not fixed by facts that Normative
Psychologism says determines which reasons apply to you, these facts do not
determine which reasons apply to you. It should not be terribly difficult to
construct any number of counterexamples to this view. Non-culpable factual
ignorance excuses.39 It does not obviate the need to justify an action that
results in an overall bad state of affairs. Less controversially, we can make
reasonable mistakes about which normative principles are genuine or which of
the reasons we are considering is overriding.
It is more difficult to counterexample the first view about what is reasonable
to judge. It seems to be something of a contingent fact about human psychology
that no actual person has the sorts of moral intuitions that would make acting
like a vampire, cannibal, or Neo-Con reasonable, but since it is a mere contingent
fact about human psychology that this is so, this fact counts against the second
view of reasonable judgment and action. The first view escapes this because
someone who is the same on the inside as a vampire, cannibal, or Neo-Con may
well not be the same on the inside as someone who knows what to do. These
horrible creatures fail to do what they ought because they act against necessarily
true principles and while these principles might not be inviolable, the reasons
these creatures have for acting against them do not justify the violations. The
problem with this view, it seems, is that it avoids counterexamples but abuses
the notion of the reasonable. Someone can make a reasonable mistake about
37 See Gibbons Forthcoming.
38 See Huemer 2006.
39 We shall come back to this later. In Chapter Y, I shall argue that non-culpably held
mistaken beliefs might excuse, but they do not obviate the need to justify an action that
brings about some bad state of affairs.
whether some reason is stronger than another and in so doing might judge
that φ-ing is the thing to do even though no one could knowingly judge that
that is so. If the hope is to link the reasonable to the right, I worry the first
view avoids counterexamples by means of a technical trick. We know what it
would take for it to avoid all the counterexamples, it would have to deny that a
conscientious and careful moral reasoner can reason to a reasonable judgment
about what to do if given the wrong intuitions as inputs. But, the thought
that someone can reason carefully and correctly from the firm intuitions she
has to a judgment about what to do and fail to be reasonable precisely because
she has the wrong inputs smacks of a strange kind of externalism. It is akin
to saying that someone who hallucinates cannot have reasonable beliefs about
the external world because the inputs were defective. The reasonable, it seems,
is more intimately connected to the agent’s perspective on things and the first
view avoids the counterexamples that arise for the second only by denying this.
So, here is a second argument against Normative Psychologism. It is possible
for two equally reasonable subjects to judge that they should φ and act accord-
ingly where one subject is permitted to φ but the other is obliged not to φ. Such
a difference in obligations requires a difference in the reasons that apply to them
because ought implies reason. In such cases, the reasons are typically grounded
in features external to the subject (e.g., facts that the subject is non-culpably
ignorant of, facts about the comparative weight of reasons that the agent is
non-culpably ignorant of, or facts about which principles are genuine that the
subject is non-culpably ignorant of). So, some reasons are neither attitudes of
the agent nor the propositions that the agent has in mind. So, some reasons are
constituted by facts external to the agent.
While defenders of Psychologism can try to accomodate intuitions about the
right by cutting the connection between the agent’s perspective and the reason-
able, they do violence to our intuitions about reasonable judgment. Instead,
they can try to accomodate intuitions about the reasonable by upholding the
link between the reasonable and the agent’s perspective, but then they do vi-
olence to our intuitions about right action. Of course, they can deny that the
reasonable judgment is the mark of the permissible, but then they undercut the
argument offered for Normative Psychologism. It seems that the last option is
the best option for Psychologism. If the defenders of Psychologism deny that
the reasonable judgment is the mark of the permissible, this undercuts one ar-
gument for Normative Psychologism but leaves Psychologism untouched. In the
next section, we shall consider another argument for Psychologism, the argu-
ment from error. I hope to show below that Psychologism cannot respect the
Explanatory and Normative Constraints if it is motivated by the argument from
error. If we treat these assumptions as axiomatic, there might be difficulties that
arise for Dancy’s view, but Psychologism is not a tenable alternative.
5.5.2 The Argument from Error

This argument from error is intended to be an argument for some version of
Psychologism. Suppose Plum and White are running down two very similar
halls in two very similar houses. There is a killer chasing Plum and she knows
it. White believes that there is a killer chasing her, but there is no one after
her. Keep Plum and White as psychologically similar as you can in keeping
with what I have just said. To introduce some jargon, Plum is in the good case,
White is in the bad. (Obviously, goodness and badness is measured in epistemic
terms rather than practical terms. Most of us would think our case is not made
better by putting a killer into it much less one that gives us good reason to
run screaming down a hall.) Given anti-Psychologism about motivating and
normative reasons, it is tempting to say Plum’s case is a case where there are
good reasons to run and Plum runs for those reasons. So, we might say:
(1) Plum’s reason for running down the hall was that the killer was
after her.
(2) Plum’s reason for running down the hall was that the killer was
after her and this was a good reason for her to run.
(3) Plum ran down the hall for the reason that the killer after her.
(4) Plum ran down the hall for the reason that the killer after her
and this was indeed a good reason for her to run.40
What should we say about White? According to Dancy, “The distinction be-
tween true and false beliefs on the agent’s part cannot affect theform of the
explanation which will be appropriateto his actions.” 41 Why think this? I
would defend the idea this way. Think about the implausible error objection. If
the form the explanation took depended upon whether the agent’s beliefs were
correct, in the case of error we would need to describe the agent’s reason for act-
ing as something that the agent is right about. While the agent is wrong about
the facts on the ground, the agent is, presumably, right about the facts in her
head. So, we would have to describe the agent as acting for the sort of reason
that only a muddled agent would think counts in favor of acting. If we are not
trying to explain the behavior of muddled agents, we should not describe the
agent’s reason for acting in psychologized terms.42 So, it seems that if (1)-(4)
are correct, these should be correct as well:
(5) White’s reason for running down the hall was that the killer was
after her.
40 If “reason” in (1)-(4) meant different things depending upon whether it was the kind of
reason that could be good or the kind of reason for which somene φ’s, (2) and (4) would be
zeugmatic (e.g., “She saw a crack and the killer in the mirror”). They both seem perfectly fine.
So, there is at least a tiny bit of evidence that the Explanatory and Normative Constraints
are correct. The “this” in (2) and (4) pretty clearly refer to whatever it is that was Plum’s
reason for running and (2) and (4) are correct only if what “this” picks out is a good reason.
41 Dancy 1995, pp. 13.
42 Someone could defend the idea in this way. The explanations we are after are causal
explanations and the cause of behavior does not depend upon the correctness of the agent’s
attitudes. Some of the relevant attitudes are about the future and the present and past do
not depend causally upon future facts. The reason I did not offer this kind of justification is
that it is controversial as to whether the explanations we are interested in are purely causal.
Others might try to justify Dancy’s point in these ways, but Dancy would not. He thinks that
these reasons explanations are not causal.
(6) White’s reason for running down the hall was that the killer was
after her and this was a good reason for her to run.
(7) White ran down the hall for the reason that the killer after her.
(8) White ran down the hall for the reason that the killer after her
and this was indeed a good reason for her to run.
Since there was no killer much less a killer chasing White, it seems that White’s
reason for running could not have been that the killer was after her. So, it seems
that (5)-(8) should be false. If (5)-(8) are false, (1)-(4) must be false as well. So,
neither Plum nor White ran for the reason that there was a killer after them.
It was, in some sense, the thought that was their reason for running.
Dancy responds by saying that there can be correct non-factive explanations
(e.g., (7)).43 While he would not describe White’s case by means of (5)-(8), that
is not because he thinks (5)-(8) are false. He would prefer to describe the case
this way:
(5d) White’s reason for running down the hall was that, as she sup-
posed, there was a killer after her. However, there was no one
after her.
This is supposed to be a correct explanation because the explanation depicts the
light in which the agent acted. It is supposed to be non-factive, however, because
the truth of (5d) does not turn on the truth or falsity of the agent’s relevant
beliefs (i.e., (5d) is supposed to be true even if there is no killer after her). He
thinks it is “too harsh” to deny that White acted for a reason, but I think (5d)
sounds too harsh in a different way. To my ears (5d) is a contradiction. Here, I
have to side with Hornsby who remarks:
... [I]t is a very strange idea that explanations are ever non-factive.
To many ears, “He φ-ed because, as he supposed, p” is true only if
it is true that p. (One plausible account of “as X supposes” used
parenthetically within a sentence s will treat it [as a sentence adverb
such as “luckily” should arguably be treated] as conveying something
about what is said in s without affecting its truth-conditions. If so,
then, given that “p because q” requires the truth of p and of q,
introducing a parenthetic “as X supposes” within it will not produce
anything non-factive.44
On this point, she and I are in perfect agreement.
There is further evidence against Dancy’s proposal. Consider:
(9) White ran down the hall because the killer was after her.
Dancy agrees that in the circumstances described, (9) is false.45 He agrees that
(9) is false because he agrees that it is obviously factive, but if (∼9) is false.
43 Dancy 2000, pp. 131.
44 Hornsby pp. 292
45 On this point, he and Schroeder both said in personal correspondence that they agree.
This is why he denies that it is a consequence of (7). It had better not be a

consequence of (7), for Dancy’s response to the argument for error is supposed
to show how (7) could be correct. Dancy would probably agree that there
is some connection between (7) and (9). In many conversational contexts I
imagine that we would at least take it that someone who asserted (7) would
assent to (9) if asked. Someone could say that the connection between (7) and
(9) is weaker than entailment, but there are ways of testing this. For example,
if (7) merely conversationally implied that (9), then this implication should be
cancellable. I don’t think the implication is cancellable. Moreover, you cannot
properly reinforce entailment, but you can properly reinforce things that are
conversationally implied.46 So, consider:
(10) Plum knew that the killer was in the kitchen. Indeed, the killer
was in the kitchen.
(11) Mustard has put a killer behind bars. Indeed, he has put many
killers away.
(12) Mustard has put a killer behind bars, but only one. After the
stress of that, he retired.
Compare these with this:
(13) White ran down the hall for the reason that a killer was after
her. Indeed, she ran down the hall because the killer was after
her.
To my ear, (13) is a redundant conjunction much in the way that (10) is. So,
we have further linguistic evidence for the hypothesis that (7) entails (9). This
is a problem because, as we have seen, “p because q” is factive. Consider:
(14) White ran down the hall because the killer was after her. In-
deed, the killer was after her.
(15) The killer was after White. That is why she ran down the hall.
It seems that (14) is a redundant conjunction, so there is some evidence that (9)
is true only if there is a killer coming after White. Also, note that (15) seems
to be equivalent to (9). (15) entails:
(16) The killer was after White.
If (16) is not factive, nothing is.
Hornsby accepts some of this, possibly all of it. She offers a disjunctivist
account of acting for a reason according to which you can be influenced by the
facts you know to be true. On her account, since agents in the good and bad
case know different things, the reasons for which they act (can) differ accordingly
even if these subjects are in the same non-factive mental states:
46 A point I owe to Stanley 2008 who credits it to Sadock 1978.
(17) Plum ran down the hall for the reason that a killer was after
her and White ran down the hall for the reason that she believed
a killer was after her.47
There are concerns, of course, whether (17) really properly describes the light in
which White acted. As she sees it, this does what we want a reasons explanation
to do because we manage to express that both Plum and White treated some
consideration as they would have if they knew it to be true. If either knew that
there was a killer after them, they would run. That is what (17) conveys and
that does a pretty good job depicting the light in which they acted. Neither
tried to be a hero, both tried to get to safety.
One of the difficulties I have in accepting this view is that it clashes with
the thought that the from the explanation takes depends upon the accuracy of
the agent’s beliefs. So, for example, if we did not know whether it was Plum or
White who correctly believed that the killer was after them but knew that one
of them had correct beliefs, we could not say whether it was (17) or (18) that
was correct:
(18) White ran down the hall for the reason that a killer was after
her and Plum ran down the hall for the reason that she believed
a killer was after her.
What is it that we are not supposed to know if we do not know whether it is
(17) or (18) that is correct? Whatever it is, it is something we do not know if
all we know is this:
(19) White and Plum both ran down the hall because they both
believed that a killer was after them.
If that contains all we need to know to explain their action, what is wrong with
a conjunctive account that denies both (17) and (18) and simply offers (19) in
its place?
Perhaps she would reply by saying that (19) does not tell us the reasons for
which White and Plum acted, but we can easily enrich (19) as follows:
(20) White and Plum both ran down the hall because they both
believed that a killer was after them and both thought that the
fact that there was a killer after them was a good reason to run.
Both knew that if they knew that there was a killer there, running
was the way to respond.
This does not tell us whether (17) or (18) is true, but it seems to tell us every-
thing we need to know about White and Plum. On the disjunctivist account,
full understanding requires knowing whether it is (17) or (18) is true, and I just
do not see what the disjunctivist thinks is gained if we gain this extra bit of
knowledge beyond what is contained in (20). We do learn that there was a killer
47 Hornsby 2007, pp. 300.
after one of our agents, but it is not at all clear what this has to do with reasons
or understanding the agent’s action.
There is a further strange feature of the view. The that-clauses we use to pick
out motivating reasons often employ propositions that are true only if certain
future events transpire. Indeed, these events might take place after we offer the
explanation of the agent’s action. White put all of her money into a hedge fund
that Mustard was running. It is strange to say that her reason for investing her
money with Mustard is one thing if it pans out and something else if it does
not. But, on the view that says the explanans will depend upon whether White
knows or merely believes that she will make a good return on her investment,
this is precisely how things are.
If neither of these views seem satisfactory, it is tempting to embrace Moti-
vational Psychologism. Like Dancy’s view, it denies that the form the reasons
explanation takes depends upon events that will transpire only after the action
occurs and asserts that the form that the explanation takes does not depend
upon the accuracy of the agent’s mental states. Like Hornsby’s view, it does
not respond to the argument from error by saying things that are contradictory.
Miller says this on behalf of PsychologismP:
[U]nless we are infallible about what facts there are, there will be
plenty of instances in which we invoke motivating reasons in our
practical deliberation and yet at the same time are quite mistaken
about the existence of the facts to which they make putative refer-
ence.48
Think about White. She is mistaken about the facts. Is she mistaken about
the reason for which she runs? According to Miller, if you were to ask either
Plum or White why they were running down the hall so quickly, both would be
disposed to say, “I am running down the hall for the reason that the killer is after
me”. They would then politely excuse themselves and continue running. On his
view, motivating reasons are propositions, not propositional attitudes, so these
remarks are not elliptical for a longer statement that makes explicit reference
to attitudes. Suppose, then, that this is a case where White is mistaken about
the facts but not thereby mistaken about her reasons. White would thus speak
the truth if she said:
(21) I am running down the hall for the reason that the killer is after
me.
But the problem here is obvious. The proposition her utterance expresses is
false. (21) entails (5) and (7), which entail (9). But (9) is false. He is not wrong
in saying that we are fallible about the facts. Obviously, we are. He is wrong in
denying that this fallibility extends straightforwardly to judgments about our
reasons for acting or the reasons others acted for. White cannot correctly assert
(21) if there is no killer after her and we cannot correctly assert (7) if there is
no killer after her.
48 Miller 2009, pp. 229.
In the debate between Dancy and the defenders of PsychologismP, both

parties agree that our ascriptions of motivating reasons do not typically make
reference to the agent’s attitudes even in cases of error. Dancy says that in the
good case, a reason is a fact. Miller says that it is a Fregean proposition that
might correspond to some fact. The problem with using the argument from error
as an argument for PsychologismP is that the Dancy and PsychologismP seem
to agree on which sentences correctly describe Plum and White’s reasons. If
sentences of the form “Her reason for φ-ing was that p” entail that p is the case,
the argument from error applies to both views with equal force. Miller rejects
PsychologismS as does Dancy. It is hard to see how PsychologismS can avoid the
implausible error objection. It is also hard to see how Hornsby’s disjunctivist
proposal avoids the implausible error objection. So, where does this leave us?
It leaves us looking for a new view.
5.5.3 Acting for a Reason as an Achievement

If the argument from error constitutes a decisive refutation of Dancy’s view, it
constitutes a decisive refutation of PsychologismP as well. It seems to me that
there is one further view worth considering. It is a compromise of sorts, but one
that is designed to please no one. It at least has the mark of a good compromise.
It seems that the parties to this debate have been assuming that acting for
a reason is not a kind of achievement. Sure, responding to real reasons is an
achievement, but acting for a reason is something you can successfully pull off
even if you do not manage to respond to real reasons. This seems to me to be
a pretty promising argument to the contrary:
(1) The reasons for which an agent φ’s when the agent φ’s for a reason
are picked out by means of that-clauses that deploy propositions
that are the contents of the beliefs that figure in the agent’s
deliberation and so are typically propositions about the situation
rather than propositions about their own propositional attitudes.
(2) When the agent φ’s for a reason, the form the explanation takes
of the agent’s φ-ing does not depend upon the accuracy or veridi-
cality of the agent’s propositional attitudes.
(3) The ascriptions that report the reasons for which the agent φ’d
are factive.
(C) So, if ∼p, “She φ’d for the reason that p” is false and if the agent
is the non-factive psychological duplicate of someone who acted
for a reason (i.e., a subject it would be true to say of, “She φ’d
for the reason that p”), she herself did not succeed in acting for
a reason.
The support for (1) comes from the implausible error argument and the argu-
ments for the Explanatory and Normative Constraints. The thought behind (2)
is that the form an explanation takes does not depend upon whether the agent’s
attitudes are veridical or not. If we have two subjects that are non-factive men-
tal duplicates who both φ and we want to say the reasons for which they φ, we
cannot then say that hte reasons for which they φ differ (i.e., it is a fact about
an agent’s mind in one case and a fact the agent has in mind in the other). As
for (3), that seems pretty well supported by the linguistic evidence. If you think
the idea of a correct but non-factive explanation makes little sense, you should
accept (C).
How does this view differ from disjunctivism? It flips disjunctivism on its
head. According to the disjunctivist:
(4) In the good case, she ran down the hall for the reason that the
killer was after her.
(5) In the bad case, she ran down the hall for the reason that she
believed the killer was after her.
The disjunctivist thinks that the propositions that specify the agent’s motivating
reasons in the good and bad case provide the explanans that correctly explain
the same explanandum proposition in the good and bad case. This contradicts
(2) because it says that there are two cases (i.e., the good and the bad) with
the same explanandum where agents are in the same non-factive mental states
and the reasons that explain their behaviors differ. On the present view:
(7) In the bad case, she ran down the hall, but she did not run for
any reason at all. At best, she took it that there was a reason to
run.
The view is consistent with (2). I say that the explanandum proposition that
we explain by describing the agent’s motivating reasons in the good case is a
proposition that is false in the bad case. So, the question, “What was the agent’s
reason for acting?” rests on a mistake if the agent is in the bad case, but not in
the good. If acting for a reason is something that happens only in the good case
and not in the bad, we can accept the principle that states that there will not
be different correct explanations of the same phenomenon in both cases. Why
hold this view? Given (1) and (3), all the candidate explanans propositions that
we use in the good case are excluded if we try to explain how the agent in the
bad case managed to act for a reason. She acted in the bad case, but failed to
act for a reason.
Does that mean that the present view makes it impossible to explain the
agent’s behavior in the bad case? Not at all. The view agrees with disjunctivism
in saying the following:
(8) In the good case, she ran down the hall because the killer was
after her.
(9) In the good case, she ran down the hall because she believed that
the killer was after her.
(10) In the bad case, she ran down the hall because she believed that
the killer was after her.
Not only do the disjunctivists seem to agree that these all come out to be true,
all parties seem to agree that this comes out false:
(11) In the bad case, she ran down the hall because the killer was
after her.
Suppose Dancy’s view and Motivational Psychologism accept (8)-(10) but reject
(11). Consider:
(12) She ran down the hall because the killer was after her.
(13) She ran down the hall because she believed that the killer was
after her.
If they say that the truth of (12) or (13) depends upon whether the agent is in
the good case or bad, the only position for them to take that is consistent with
(2) is the position I am advocating, which is that the thing you try to explain
by citing the reasons for which an agent V’s is a feature unique to the good
case. If to deal with this point you accept (9) and (10) but deny (8), it seems
you also have to deny:
On its face, (14) entails (8). But, to deny (8) is simply to deny (1). It is to assert
that the reasons for which we act really are correctly picked out by propositions
that report our attitudes instead of the propositions that are the contents of the
attitudes that figure in deliberation.
The most significant obstacle the present view faces is that in asserting (7),
it seems the view suffers from an explanatory deficit that other views do not.
In response, notice that those who defend PsychologismP or Dancy’s view have
to deny “She φ’d for the reason that p” entail “She φ’d because p”. They agree
that “She φ’d because she believed p” is true whenever “She φ’d for the reason
that p” is true. In asserting (4), I am committed to saying that in the bad case,
it is false that White ran down the hall because the killer was after her. On this
point, all the views are in agreement. I am not committed to denying that she
ran down the hall because the killer was after her. This causal explanation is
one that all parties seem to agree is correct. I say that the explanans proposition
in “She φ’d because she believed p” does not ascribe the reason for which the
agent acts. All parties seem to agree on that point as well. So, I do not think
my view suffers from any explanatory deficit. My view accepts all the “because”
claims that the other views accept and offers the same causal explanation of the
agent’s behavior in terms of the agent’s attitudes that alternative views do.
The difference is that on the view defended here, there are more true “be-
cause” claims than on the rival view. In the good case where the agent correctly
believes p, you can correctly say, “She φ’d because p”. So, maybe the problem
with the view is not that it suffers from an explanatory deficit. Does the view
offer too many explanations? What is the extra thing that motivating reasons
explain? It has to be something that distinguishes the good case from the bad.
Here is what it is. We know what something has to be to be a motivating
reason–it has to be something that could “turn out” to be a good (normative)
reason if the facts fit the attitudes. We know that each instance of acting for a
reason involves a motivating reason and each behavior that can be understood
in terms of motivating reasons is acting for a reason. So, acting for a reason is
an achievement. When you act for a reason, there is a reason and you act for
it. You saw something in the situation and have responded to it rather than re-
sponded in a predictable way given psychological inputs that might misrepresent
the circumstances in which you acted.
Why have two notions? Why have causal explanations of behavior that cite
the agent’s psychological states and reasons explanations that cite facts that fit
those states? One idea is this. In saying that someone acted for a reason, we
impart two pieces of information. Part of it has to do with specifying the agent’s
reasons to say what the agent’s intentions were in acting. (This is something
we can do in the good case and the bad by describing the agent’s psychological
states.) Part of it has to do with reporting facts that the agent confronted
because those facts are facts that we all potentially might have to deal with.
Causal explanations of behavior that cite the agent’s psychological states do not
convey this extra bit of information, which is that the agent saw something in
the circumstance that she took to be something that called for her to act in the
way that she did.
Properly understood (8) and (9) are complementary, not competing. They
do not compete because facts and beliefs explain different things. In some
contexts, we want to know what it is that the agent got out of acting in the way
that she did. We want to know what she accomplished or what she achieved.
These are the contexts in which we say what the reasons for which an agent
acted. If it turns out that the agent’s attitudes were mistaken, our question
rested on a mistake. We had thought that the agent got something out of
acting in the way that she did that she had hoped for, but she did not. In
contexts where we do not want to know what the agent got out of acting, we
are looking for a psychological explanation of the agent’s behavior. So, if we
know that the agent’s attitudes are false or do not know whether the agent’s
attitudes are false, our interest is in what would make the agent’s behavior
intelligible. Here, psychological states of the agent are useful. If the agent is in
the good case, we can ask both sorts of question and that is why both (8) and
(9) turn out true. If the agent is in the bad case, we can ask only one sort of
queston and that is why (10) turns out true rather than (11).
Acting for a reason is thus similar to knowing and different from believing
insofar as knowing and acting for a reason are both achievements. In contexts
where we can safely assume that the agent’s attitudes were correct, we can ask,
“Why did she do that?” and get in return an account of what she achieved by
acting in the way she did. This is akin to contexts where someone asks, in a
non-skeptical or non-challenging way, “How does she know that?” Just as we can
ask “How does she know that?” to learn something either about how she learned
that p or why she came to believe p, we can ask, “Why did she do that?” either
to learn what she gained or what she had hoped to gain and receive different
answers. If taken in the first way, the answer cites facts. If taken in the second,
the answer can cite attitudes. In contexts where we cannot safely assume that
the agent’s attidues were correct, we can ask, “Why did she do that?” and get
a correct answer that describes the agent’s attitudes, but that does not tell us
the reasons for which she acted. We are not interested in the reasons for which
she acted, we want a psychological story that makes sense of her behavior that
remains neutral on the question as to whether she had any reasons to act as she
did.
In this way, we can allow that psychological states of the agent do have a
role to play in explaining behavior. In asking why some event occurred, if we
want to know the causes, we can cite the psychological states as causes. These
psychological states are the reasons why someone’s body moved in such and
such a way. The reasons why an event occurred are not reasons for the event to
occur and they are not the reasons for which an agent acted.49 The thesis that
psychological states are reasons why events take place is extremely plausible
and I think that this is all that Psychologism can be right about. And this, is
just to say, that Motivational Psychologism is false.
Should this be upsetting to those who defend Psychologism? I do not think
it should be at all upsetting to Smith, for as I understand his view, he does not
really deny (1). Rather, for him, “motivational reason” is a term of art that has
more to do with explaining why actions occurred. I do not think Smith assumes
something the parties to this debate have assumed, which is that motivating
reasons are whatever we pick out using that-clauses when we describe the reasons
for which an agent acted. Rather, he assumed that motivating reasons are
whatever psychological states we pick out that can figure in causal explanations
of events that involve agents acting purposively. So, in the end, nothing I have
said against Motivational Psychologism speaks against the Humean view that
he defends. Whether that view is correct depends pretty much on what he
thought it depended on, not on whether the things we describe as the reasons
for which an agent acts are facts, beliefs, the contents of beliefs rather than
desires, but on whether the correct causal story of an event that is an action
involves psychological states with differing directions of fit. If the reader prefers
to think of motivating reasons as reasons why people behave in certain ways,
nothing I have argued here has called that view into question.
In the previous chapter and this one, we have covered considerable ground.
I have argued for an externalist conception of justification by arguing for an
49 That reasons why are not the same thing as motivating reasons is something that Dancy
2000 and some of his critics agree on. It is less clear that this is something that all defenders of
Psychologism agree to. Some might say that motivating reasons are kinds of reasons why and
they might deny that they are reasons why because they are considerations in light of which
the agent acted. Instead, they are states by virtue of which there seemed to be something
in the situation that called for a response. My sense is that this is closer to the view that
someone like Smith prefers.
externalist account of reasons for belief and action. Such an account causes
trouble for the orthodox internalist and externalist views insofar as it asserts
that the justification of a belief depends upon whether it is true. Justification
depends upon truth because normative reasons are constituted by truths and
the justification of a belief depends upon whether it can provide us with reasons
from which we can reason to a conclusion about what to do or what to believe.
This view faces a number of objections. Some say that the view either
motivates a kind of skepticism or requires an implausible account of the nature
of experience. I have tried to show that this is not so and the arguments to the
contrary rest on an assumption that many take for granted that I think we must
reject. The objection rested on the assumption that the normative properties
that determine whether a belief is justified have to be properties of the basis
that the subject has for holding that belief. If we reject this assumption, we
can say that while having a justified belief requires that the belief is true for it
to be justified, the justification offered in support of that belief need not entail
that the belief is true.
There was a further worry that taking reasons to consist of facts or true
propositions clashes with the thought that we can explain an agent’s behavior
and an agent’s beliefs by citing psychological states of the agent. Since mo-
tivating reasons have to be mental states or the contents of those states and
normative reasons have to belong to the same ontological category as moti-
vating reasons, normative reasons cannot be facts or true propositions. This
objection does not hold up. If the objector takes motivating reasons to be the
reasons in light of which someone believes or acts, motivating reasons are facts.
If, however, the objector takes motivating reasons to be elements of a causal ex-
planation that explain why someone believes or acts and denies that these need
to be the reasons for which the agent believes or acts, the objector’s mistake is
in thinking that my arguments call this view into question.
Chapter 6
Assertion
6.1 Introduction
There has been considerable discussion recently of epistemic norms governing
assertion. Assertions can be evaluated along a number of dimensions, but can
they be assessed epistemically? It seems so. Some locals tell you that the water
is not safe to drink and so you buy some bottles. Then they tell you that the
reason the water is not safe is that it has been fluoridated. Now, we might
assume that the water is unsafe for reasons our local knows nothing about. In
some sense, then, it was a good thing the local said what he did and you believed
it. If there is something wrong with the speaker’s assertion, it is not that it was
not true and it was of no benefit to you. Still, you can resent the speaker for
saying what he did. He should not have thought that what he told you was
true because he should not have taken his grounds to be grounds. The grounds
for your resentment have to do with the credentials of the speaker’s belief. So,
there is work here for epistemologists to do. We can try to work out what
the proper standards are for determining whether an assertion is epistemically
defective and whether there are norms that can help sort between warranted
and defective assertion.
If this is right and we could work out an account of warranted assertion,
this might help us in trying to work out an account of justification. There are
two reasons to think so. First, the concepts of warrant and justification are
both normative. A belief is justified only if it is permissibly held and “warrant”
is a technical term that stands for permissible. Some of the arguments that
shape the internalism/externalism debate are intended to show that justifica-
tion is an internalist notion on the grounds that it is a deontological notion.
Warrant is also a deontological notion. If warrant is understood along exter-
nalist lines, these arguments that purport to show that justification must be
understood along internalist lines because it is a deontological notion are un-
persuasive. Second, an increasing number of contributors to the literature on
warranted assertion are convinced that assertion and belief are governed by
106
CHAPTER 6. ASSERTION 107
common standards.1 So, arguments that purport to show that warrant should
be understood in internalist or externalist terms might show tha the same is
true for justification.
Assertions are correct only if they are true. In this respect, assertion is like
belief. If we go so far as to say that an assertion is correct iff it is true, someone
could say that warranted assertions just are true assertions:
Your assertion that p is true is warranted iff p (TA).2
Williamson thinks this will not do and we know why it will not. In the example
sketched above, our local’s assertion is true but unwarranted. He and I both
agree that there is a truth norm that governs assertion, but he thinks that this
norm governs assertion because knowledge is the fundamental norm of assertion:
Your assertion that p is true is not warranted unless p is true (TNW).
Your assertion that p is true is warranted iff you know that p is true
(KA).
TNW governs assertion on his view because knowledge requires truth. I shall
start by reviewing the case for the knowledge account of warranted assertion.We
shall see that the arguments offered in its support only support weaker views
on which the internal components of knowledge are required for warranted as-
sertion:
Your assertion that p is true is warranted iff you reasonably believe
p (RA).3
Having said that, I do not here defend the view that you have sufficient warrant
to assert anything you happen to reasonably believe. Sometimes you do, but
sometimes you ought to be excused for asserting something you should not have
for the reason that you were reasonable in thinking what you said was true.
What I have not seen in the literature on warranted assertion is a convincing
argument for TNW and that is what I hope to offer here in the course of arguing
for a justification account of warranted assertion:
Your assertion that p is true is warranted iff you justifiably believe
that p is true (JA).
It might seem that this account is in tension with TNW. Combined, JA and
TNW entail that there are no false, justified beliefs. Although many do hold
to the view that there cannot be false, warranted assertions, they almost all
insist that there can be false, justified beliefs. This combination of views, I shall
argue, is untenable. We should think of justification in deontological terms. If
you justifiably believe p, you cannot be under any epistemic obligation to refrain
from so believing. Among the norms governing assertion is a norm that enjoins
1 See Adler 2002, Sutton 2005, and Kvanvig 2009.
2 Weiner 2005 defends this view.
3 This view is close to views defended by Douven 2006, Lackey 2007, and Kvanvig 2009.
you to refrain from asserting false propositions. Among the norms governing
belief is a norm that enjoins you to refrain from believing p if you do not have
sufficient warrant to assert p. So, you cannot have it both ways. If warrant
requires truth, so does justification.
6.2 Warrant and Normativity

It seems plausible that knowledge warrants assertion. If it did not, you could
assert p, know that p is true, but fail to live up to your epistemic obligations
because you were not in a good enough position to assert that p is true. It
is hard to imagine how this could happen. It is probably not hard to imagine
cases where you should not assert that p is true unless you are absolutely certain
that p is true. Perhaps the heavens would fall if you spoke falsely. In such a
case, knowledge might not warrant speaking, but it seems the reason this is
so has to do with the high practical stakes. To show that knowledge does not
warrant assertion, we need cases where it is wrong for purely epistemic reasons
to assert that p is so even if you know that p is so. If somone says that you are
in no position to say that p is true and adds that you merely know that you
are right, it is hard to say exactly what the criticism is supposed to be. The
knowledge account is controversial, I take it, because it claims that you cannot
have sufficient warrant to assert p unless you know that p is true. Is that too
demanding?
If it is too demanding, we need to be put something less demanding in
its place. For various reasons, Williamson thinks that weaker rules than his
knowledge rule cannot accomodate the data we shall discuss below:
You must: assert p only if you know that p is true (KR).
Thomson says that this is too strong.4 If we read it as I think she reads it, she
is probably right. As she reads it Williamson is Kant, but worse. To say that
you “must” not assert what you do not know is supposed to be stronger than
just saying you “ought” not assert what you do not know. Surely, however, Kant
was wrong and if the thing you must do to hide someone in your house from the
Nazis is lie, lie. Since lies do not constitute knowledge, it is surely too strong to
say that you must never assert what you do not know.
I cannot imagine Williamson ever meant to deny this. He is neither Kant nor
worse. Buried in this picky semantic point about “must” is an important point
about warrant and normativity. Among Williamson’s reasons for accepting KR
is that he thinks this rule is a constitutive rule, a rule that is essential to the
speech act of assertion. Just as there are rules in games that tell us which moves
are correct, Williamson says that there are rules that govern speech acts that
determine whether something is a correct asserting. Even if this is so, it is hard
to see how this would lead us to accept KR even if we were to weaken it a bit
and say simply that you ought not assert what you do not know. To modify
4 Thomson 2008, pp. 88.
one of Thomson’s examples, there are rules that tell us the correct way to move
a rook in chess.5 If the only way to save the lives of some hostages is to move
the rook incorrectly, this is what you should do. The rules that distinguish the
correct from incorrect ways of moving the castle do not really tell us that we
must move our castles in such and such a way. (They also do not tell us that we
must move them certain ways if we are to move them at all.) If they did, the
rules of chess would seem to conflict with the rules of any sensible moral code.
Chess is not immoral. Do the rules of chess even really tell us what we ought
(in some sense) to do? It would do no good to say that these rules tell us only
what we oughtchess to do. Maybe the moral rules tell us what we oughtmoral to
do, but once we see that the rules of chess and morality require us to move the
rook in different ways, we have to decide what we ought (or oughtreally , if you
like) to do. Surely, morality is concerned with what you oughtreally to do. If
chess is also concerned with what you oughtreally to do, which it must be if it is
giving us advice, the rules of chess would conflict with morality. But, they do
no such thing. So, they really do not tell us anything about what we oughtreally
to do. I suspect that the rules governing assertion are like this, but then they
do not so much as tell us that we oughtreally not assert something if we do not
know it to be true.
If the rules of assertion are anything like the rules of chess, they do not tell us
what we ought not say or must not say. How damaging is this to Williamson’s
project? Not very, I think. As Thomson herself says, Williamson is surely right
that there are cases where you should not assert something that you do not
know to be true and she seems tempted by the thought that in some of these
cases you should not assert something because you do not know that it is true.6
To deal with this mess, we should probably say this. First, if there is some sort
of knowledge rule for assertion, the rule does not constrain our intentions in
such a way that we intend to conform to the rule in order to count as asserting
that something is so. Parties to this debate tend to agree that lies are assertions
and clearly liars do not at all intend to say what they know to be true. Second,
we have to take some care in saying what sort of obligation we are under to
avoid saying implausibly that it is not permissible on purely epistemic grounds
to lie. If the Nazis are at your door and silence would give everything away, the
right thing to do is to lie. If such cases are not counterexamples to the accounts
of warranted assertion considered below, they cannot be taken to be accounts
of what we must not say. I am all for distinguishing between epistemic and
non-epistemic obligations, but it does no good to say that while morally you are
under an obligation to lie to the Nazis you are under an epistemic obligation to
either tell them the truth or say nothing. Suppose the epistemic norms said that
you must not assert p unless you know it. Insofar as the right thing to do is to
lie to the Nazis, what epistemology said was false. Suppose epistemology only
says that you oughtepistemic not assert p unless you know it. We then would
have to work out how this obligation relates to other obligations such as the
5 Thomson 2008, pp. 166.
6 Thomson 2008, pp. 94.
obligation we have to lie to the Nazis. Since we are under an obligation to lie,
whatever reason associated with this epistemic obligation is defeated. If there
is such an obligation, then, to refrain from asserting without warrant, it is a
pro tanto obligation or duty only. There can be principles of pro tanto duty
that bend without breaking, but when they bend, there is supposed to be some
sort of residue that rationalizes regret. Do you really have to regret lying to the
Nazis at your door? No. Is it really regrettable for epistemic reasons? Certainly
not.
Let’s switch cases. Your friend recently told his class that cows have no
difficulty walking down stairs because of the way their legs are jointed. He
knew this was false, but he also knew that if he did not tell the class this,
horrible things would happen to them. If he knew he could never correct this
and explain to the students the reasons he had for deceiving them, I can imagine
that he would regret having had to deceive them. We can imagine variants on
the case. The teacher is forced at gun point to assert something he had no
reason to believe is true. Again, I could imagine he would regret having to do
this even though he knows that it is best for him to do this.
What distinguishes this case from the case with the Nazis at the door? It is
not that in one of these cases the speaker knows that his overriding obligation is
to say what he knows or takes himself to know. In both cases, the speaker has
overriding reason to assert something she knows she does not know to be true.
What distinguishes these cases might be this. In the second case, the speaker
has assumed a certain role and assumed the responsibility that comes with
occupying this role but realizes his role obligation is overridden by a weighty
moral obligation. In the first, the speaker never assumed this sort of role and so
had no reason to assume any responsibility for the veracity of what she said. In
the second, the speaker was under a standing obligation to assume responsibility
for her remarks and was not able to cancel that obligation. While the speaker
rightly decided competing reasons having to do with the welfare of the students
were overriding, there was still the pull of the defeated reason that made it
regrettable that the students had to be deceived.
Perhaps we can now formulate our issue this way. There are some conversa-
tional exchanges in which a speaker has a responsibility to another as a testi-
monial source and the speaker assumes this responsibility in choosing to assert
that something is so. In some such cases, the obligation might be overridden by
practical obligations (e.g., the teacher who is compelled to deceive his students
to save them from harm), but there is another set of cases in which a party to the
conversational exchange has no responsibility to another as a testimonial source
(e.g., the speaker who is forced to deal with the Nazis at the door conducting
searches). Our focus is on those cases where the subject’s responsibility as a
testimonial source is her overriding responsibility. What standards determine
whether she has lived up to her obligations in such a case? If the speaker lives
up to her responsibilities in such a case, her assertions are warranted. If not,
they are not.
This much seems obvious. In asserting that something is so, the speaker tells
her audience that it is so and claims that something is true. Whether this means
that TNW is correct is a delicate issue. It depends, in part, upon whether there
is an epistemic reason not to assert false propositions and also upon what such
a reason demands. For the purposes of the discussion in this chapter, let us
assume that if there is a norm that enjoins you not to φ unless some condition
C obtains, this norm says that there is a reason not to φ unless C obtains
and this reason demands conformity.7 Further, let us assume that if there is
reason not to φ and this reason is undefeated, you cannot be warranted in φ-ing
against such a reason without an equally strong reason that demands that you
φ. Given these assumptions, someone who denies TNW either has to say that
there are epistemic reasons to assert propositions that are not true that are just
as strong as whatever reasons there are not to assert false propositions or deny
that there is any epistemic reason not to assert false propositions. Similarly, to
argue against KA, you either need to argue that there is no epistemic reason
to refrain from asserting what you do not know or when you are warranted
in asserting things you do not know, there are epistemic reasons as strong as
whatever reasons there are not to assert that enjoin you to assert. It is an
interesting question as to whether the assumptions just introduced are true. I
shall later argue that they are, but doing so requires a careful discussion of what
the reasons associated with norms demand. With these preliminaries out of the
way, let us look at the case that has been offered in support of the knowledge
account of assertion.
6.3 The Knowledge Account

Should we refrain from asserting what we do not know? I think not. The argu-
ments offered in support of KA seem to support only weaker views of warranted
assertion. The view also sometimes delivers the wrong verdicts. We start by
looking at some arguments having to do with the aim of belief and assertion
and then look at some arguments having to do with Moore’s Paradox. It seems
to me that there are ways of accomodating the data thought to support the
knowledge account given the resources of an account that takes truth to be the
fundamental norm of assertion.
6.3.1 Truth as the Aim

The first argument for the knowledge account works from the assumption that
belief and assertion have a common aim and that this aim has a kind of norma-
tive significance. In saying that belief or assertion aims at something and that
you ought not assert or believe unless this aim is achieved, it is not at all clear
which of these claims is explanatorily prior to the other. It might be that in
saying that belief aims at truth or knowledge, I am expressing the thought that
beliefs that are not true or do not constitute knowledge are not as they ought
to be, but that might be because I think this talk of aims is a useful metaphor
best cashed out in normative terms.
7 If there is a reason to φ, you conform to that reason iff you φ.
Suppose, as seems plausible, that belief does aim at the truth and that
belief has no further independent aim such that a false belief might fail to hit
its first target but succeed in hitting some second equally significant target.
Given this, it seems natural to say that even if some good comes of believing a
false proposition, such a good has little significance when it comes to justifying
the belief in question. Our concern is with assertion, not belief but I think
we can make some progress in understanding the norms governing assertion by
thinking about the norms that govern belief. As Williamson notes, “assertion is
the exterior analogue of judgment, which stands to belief as act to state”.8 Given
that this is so, it seems a reasonable default assumption would be that assertion
and belief share common aims and are governed by common standards. If belief
aims at the truth, so does assertion. If truth is normative for assertion, it is
probably normative for belief. If the assumption that truth is the fundamental
norm for belief is too weak to account for all of the normative demands that we
are under as believers, we should expect the same to be true for assertion and
the demands that we are under if we assume the responsibilities we have to our
audience if we decide to tell them that something is so.
Williamson suggests that if the norms of assertion can be derived from a
proper description of the norms and aims of belief, you can derive the knowledge
account of assertion from the assumption that truth is among the aims of belief
and assertion. How is this supposed to work? If there is such a derivation,
it is hardly straightforward. Utilitarians do not think that the justification of
an action depends upon whether the agent knew that it was optimific at the
time of action, only that it was optimific. I think no one has ever thought of
faulting them for this omission. It is not hard to see that you are courting
disaster if you advance a view on which there are positive duties alongside the
further claim that all duties must be knowingly discharged. Such a combination
of views would seem to lead rather quickly to the untenable view that says that
there are unknowable obligations that are only obligations insofar as they are
knowable.
It helps that here it seems natural to say that our duties are all negative. We
do not fail to live up to our responsibilities if we simply choose to say nothing.
If there is a truth norm that governs assertion, it enjoins us to say only what is
true, not to say that things are true:
You should not assert p unless p is true (TNW).
Williamson seems to think that you cannot say that this norm governs assertion
while denying that there is an evidential norm that also governs assertion:
You should not assert p unless you have sufficient evidence for be-
lieving that p is true (ENW).
If this is right, he thinks we cannot then deny that knowledge is the norm of
assertion:
You should not assert p unless you know that p is true (KNW).
The idea seems to be that the evidential norm is derivative from the truth norm
and that the view that combines these two norms but does not incorporate the
knowledge norm will only be able to handle some of the data.
If TNW is among the norms that governs assertion, we have to reject the
following claim about epistemic wrongs and fault:
All epistemic wrongs are fault-implying wrongs (i.e., any condition
that makes believing or asserting p wrongful is a condition that the
believer can be faulted for failing to take account of if she asserts or
believes p when that condition obtains) (Fault1 ).
You can consistently reject Fault1 and accept this claim about fault and epis-
temic wrongs:
Any condition that grounds the charge of epistemic fault is a condi-
tion that makes assertion and belief wrongful (Fault2 ).
It seems that Williamson’s reason for thinking ENA is a consequence of TNA
might be something along the lines of Fault2 . He says:
[I]f one must not bury people when they are not dead, then one
should not bury them when one lacks evidence that they are dead.
It is at best negligent to bury someone without evidence that he is
dead, even if he is in fact dead.9
Let’s suppose this is right. I do not think these considerations support the
knowledge account.
Williamson says that we do not satisfy the evidential norm governing as-
sertion unless we have evidence that puts us in a position to know that the
proposition we assert is true.10 If we think about lottery propositions, it seems
we do not have adequate evidence to believe or assert such propositions. It
seems the best explanation as to why this is rests on the observation that the
evidence we have for believing lottery propositions without insider’s information
does not put us in a position to know that these claims are true.
There are two ways of reading Williamson’s lottery argument. On the first
reading, his remarks concerning lottery propositions gives us a clue as to what he
thinks it takes to satisfy Fault2 . If you believe or assert without first gathering
evidence that puts you in a position to know that p, you seem to be at fault
even if your belief turns out to be true. After all, you could have weakened your
commitment by simply believing that p is likely or probable. On the second,
we appeal directly to ENA and let intuition serve as our guide in determining
what it takes to satisfy ENA rather than appeal to assumptions linking fault
and justification.
On the first reading, the argument amounts to this.
9 Williamson 2000, pp. 245
(1) You should not assert p unless p is true.

(2) If you do not know whether you would violate a strict prohibition
by φ-ing but φ anyway, you are at fault for φ-ing.
(C) It follows that you should not hold assert p if you do not know
that p is true.
So formulated, the argument rests on this assumption about fault and knowl-
edge:
You are irresponsible to φ if you do not first know whether φ-ing is
permissible (Fault3 ).
Without the assumption, you cannot derive anything stronger than the claim
that you ought not φ unless you have conformed to TNA and reasonably assume
that you have given your evidence.
There are two reasons to reject Fault3 . First, suppose we say that knowledge
is a condition necessary for permissible assertion. Suppose Fault3 is true. In
saying this, we would have to say either that mere knowledge of p’s truth is
insufficient for permissibly believing p or we would have to endorse a KK thesis
according to which you cannot know p unless you are in a position to know
that you know that p is true. There are persuasive objections to the KK thesis.
Knowledge of p’s truth requires that the means by which you arrived at the
belief that p could not have easily led you to be mistaken about p. According to
the KK thesis, you do not satisfy the conditions for first-order knowledge unless
you are in a position to know that you satisfy these conditions. Second-order
knowledge also requires that you would not easily be mistaken in the second-
order belief (i.e., the belief that your first-order belief constitutes knowledge).
Our ordinary knowledge ascriptions suggest that knowledge does not require
being in a position to know that you know. We readily ascribe knowledge of
p’s truth to someone knowing that she could have easily been mistaken in her
second-order belief that she knew that p. We might know that the margin of
error for second-order knowledge is slim but the margin of error for first-order
knowledge is sufficiently wide so they could not have easily been mistaken about
whether p but could have easily been mistaken about whether they knew p.
Concerning such cases, not only does it seem we readily ascribe you knowledge,
we do not think that it is wrong for you to believe p. This seems to disconfirm
both the weak KK thesis and the thought that permissible belief involves more
than just knowledge. If we reject both, however, we have to reject Fault3 .11
There is the second reason to deny Fault3 . If combined with the knowledge
account, it commits us to the JTB analysis of knowledge and an infallibilist
conception of justification. Epistemic irresponsibility can make an otherwise
justifiable belief unjustified. According to Fault3 , if you fail to know for any
reason, you can be charged with epistemic irresponsibility. Thus, if you cannot
be charged with epistemic irresponsibility because you are justified in believing
p, the fact that you are justified in holding your belief is logically incompatible
11 See Williamson 2000 for discussion of these points.
with (a) your belief being mistaken or (b) your belief being Gettiered. One
consequence of this is that you cannot satisfy the justification condition if it
is possible that someone should have just your reasons but be mistaken about
whether p. Thus, your reasons must entail p if your belief that p is justified.
But, no one seems to think that you must have entailing grounds to permissibly
believe p. Second, it seems that Fault3 has the consequence that if someone
does not know that p, they are not justified in believing p, in which case Gettier
cases are impossible.
In light of these problems, I think an alternative reading of the argument
might be more charitable.12 On this reading, the argumentative burden is shoul-
dered not solely by assumptions about fault and epistemic responsibility, but
also by intuitions concerning cases involving lottery propositions. The assump-
tions about fault are supposed to support the idea that some sort of evidential
norm governs belief. Our intuitions about lottery propositions are supposed to
help us see what it takes to satisfy this evidential norm. We start from the
assumption that you should not assert or believe lottery propositions. The nat-
ural explanation for this is that you do not have evidence for believing these
propositions that would put you in a position to know that they are true. From
here, the argument might go in one of two directions. If someone said that it
followed from this alone that you ought not assert what you do not know, this
would repeat the mistakes we have just discussed. All that follows is that you
ought not believe p if you are in a position to appreciate that your evidence
does not put you in a position to know p. In Gettier cases and in cases where
you do not know that p is a lottery proposition, it seems p is not known, you
do not know that you are not in a position to know p, but it is not obvious
that you have violated ENA. If you think you do violate ENA in such cases, it
seems you will once again be forced to accept the JTB analysis of knowledge
and an infallibilist account of justification. To avoid these difficulties, someone
should instead argue as follows. Intuition tells us that evidence is needed for
permissible belief. Intuitions about lottery cases tell us that beliefs in lottery
propositions are defective because there is not adequate evidence for believing
them outright. The best explanation of the observation that you do not satisfy
ENA unless your evidence puts you in a position to know that the relevant
proposition is true is that KA is the fundamental norm of assertion.
This argument rests on a pair of assumptions. First, the argument assumes
that beliefs in lottery propositions do not constitute knowledge. Second, it
assumes that you ought not believe lottery propositions. If we deny the first
assumption, we cannot appeal to intuitive verdicts about lottery cases to moti-
vate the knowledge account. If we were to reject the second assumption while
accepting the first, lottery cases would provide counterexamples to the claim
that you should not believe what you know you do not know.
In response to this argument, I want to say two things. The knowledge ac-
count cannot give the best explanation if independent considerations show that
KA is not a norm that governs assertion. The verdicts the knowledge account
12 Jonathan Sutton suggested that this was the more plausible way of reading the argument.
delivers for covert lottery beliefs and for Gettier cases are counterintuitive. In
addition, cases involving covert lottery beliefs suggest that while the knowledge
account delivers the right verdict about some familiar lottery cases, it gives the
wrong reason for thinking this is the right verdict. Intuitions concerning cases of
covert lottery beliefs suggest that the reason we ought not believe lottery propo-
sitions is not that they cannot constitute knowledge per se, but that subjects
that believe lottery propositions are wrong to do so in light of considerations
accessible to them (i.e., considerations about the kinds of grounds they have for
believing lottery propositions). While such grounds might not put the subject in
a position to know, the normative significance of this is not what the knowledge
account takes it to be.
The distinction between covert and overt lottery beliefs is a familiar one,
but the terminology is not. Let us say that a covert lottery belief is a belief
whose truth or falsity depends on the outcome of a lottery when the believer is
not in a position to appreciate that this is so. Let us say that an overt lottery
belief is a belief in a lottery proposition held by someone who has no insider’s
information. If you look at your bank statement and see that you are down to
your last few dollars, you might reasonably believe that you will not be able to
go on safari. If your mother has just purchased you a ticket for a lottery drawing
being held later this afternoon without telling you, that belief is a covert lottery
belief. Were you to believe that the ticket that your mother bought you will
lose, that would be an overt lottery belief. (We are assuming that you know that
you would be able to afford to go on safari if only you were to win the lottery
drawing being held this afternoon.) It seems that overt and covert lottery beliefs
will either both constitute knowledge or neither will. If you think that safety is
necessary for knowledge, it will be just as easy for a covert lottery belief to turn
out to be false as an overt one to turn out to be false. If you think that some
suitably formulated closure principle holds true, someone will be in a position
to know that a covert lottery belief is true only if this subject is in a position to
know that an overt lottery belief is true. Assuming, as we are, that overt lottery
beliefs fail to constitute knowledge, it seems we have two reasons for thinking
that covert lottery beliefs similarly fail to constitute knowledge.
If this much is correct, the knowledge account commits us to saying that
you should not hold or form covert lottery beliefs. I think this is bad news for
the knowledge account. First, in defences of the knowledge account, the focus
has been on the judgments that overt lottery beliefs should not be held and
cannot constitute knowledge. No intuitive support has been offered to back the
claim that neither type of lottery belief ought to be held. In fact, you might
think that one of the reasons that the lottery paradox is so interesting is that
we are not naturally disposed to think of covert lottery beliefs held by others
as beliefs they should not continue to hold for reasons of which only we are
aware (i.e., that unbeknownst to them the truth of their beliefs is contingent on
the outcome of a lottery). Second, not only is the knowledge account’s verdict
about covert lottery cases not intuitive, it seems positively counterintuitive. To
see this, consider a modified version of one of Hawthorne’s examples. A friend
writes you an email on Monday before a lottery is held, but you only read it
Tuesday after the results of that lottery are known to you. It contains the
following line of reasoning:
The ticket for tomorrow’s lottery is a loser. So if I keep the ticket I
will get nothing. But if I sell the ticket I will get a penny. So, I’d
better sell the ticket.13
You know now that the first premise was not known to be true because of the
grounds the subject had for that belief and know that the belief turned out to
be true. Retrospectively, it seems you would agree with Hawthorne that this
reasoning is unacceptable and would likely further agree that its unacceptability
is due to the speaker’s belief in the argument’s first premise. Assuming that you
should not hold beliefs that should not be trusted for the purposes of practical
deliberation, we would arrive at the view that the speaker should not have held
the first belief. Even without that assumption, you might agree that the subject
should not have held the first belief regardless of whether it was fit to figure in
practical deliberation.
Now, suppose a different friend writes you an email on Monday before a
lottery is held, but you only read it Tuesday after the results of that lottery are
known to you. You had purchased this friend a ticket for this lottery without
telling them, but now know that the ticket was a loser. They had written:
I want nothing more than to go on safari. If I were to go on safari, I
would want nothing more than to buy a new elephant gun. The gun
will be useless, however, since I cannot afford to go on safari. So I
guess I will use that money instead to do some repairs around the
house.
The subject’s belief in the first premise is known to you to be a covert lottery
belief. The lottery was held and the ticket lost. You know this, so you know that
the speaker’s belief in the first premise was not known by the speaker to be true
and that the speaker was in no position to appreciate this fact (i.e., it was an
“unknown unknown” in Sutton’s terminology). I think you would not take this
reasoning to be unacceptable. However, the knowledge account regards both
instances of reasoning as unacceptable and takes them to be unacceptable for
the very same reason. It says neither piece of reasoning is acceptable because
both bits of reasoning involve crucially beliefs not known to be true.
That the knowledge account delivers the wrong verdict in the case of covert
lottery beliefs suggests that knowledge is not the norm of assertion or belief.
Additionally, it suggests that the knowledge account gives the wrong explanation
for the unacceptability of the first bit of reasoning. The knowledge account seeks
to explain the unacceptability of this reasoning in terms of a fact that is not
accessible to the individual engaged in this bit of reasoning (i.e., that one of
the beliefs involved in the reasoning is not known to be true). However, if overt
and covert lottery beliefs have different normative statuses (i.e., one ought never
hold overt lottery beliefs but may permissibly hold some covert lottery beliefs),
13 Hawthorne 2004, pp. 29.
it seems that the proper explanation as to why you should not reason from
overt lottery beliefs should be given in terms of features distinctive of overt
lottery beliefs (e.g., the kinds of ground available for overt lottery belief) rather
than ignorance, per se. We have not found a route from the truth norm or
the thesis that belief aims at the truth to the knowledge norm. It is not for
a lack of trying. Williamson is right that anyone who thinks there is a truth
norm should think there is an additional evidential norm governing belief, but
we know from Gettier that there is more to conforming to the knowledge norm
than conforming to these two.
6.3.2 Knowledge as the Aim

Rather than try to derive the knowledge account from the truth norm, we might
try a different approach. Someone could argue as follows. Belief does not aim
at just the truth. Belief aims at knowledge. Any belief that fails to constitute
knowledge is wrongful precisely because there is no distinct aim a belief serves
that could potentially provide a justification for believing without knowing.
Since the argument assumes nothing about justification and fault, it should not
face the problems the previous argument did. The argument assumes that the
aim of belief is knowledge.14 The norms that govern belief also govern assertion.
So, since you should not believe what you do not know, you should not assert
what you do not know.
The problem with the argument is simple. Knowledge is not the aim of belief.
To test proposals about aims, we should consult our intuitions to determine
what an external observer would say if she knew that another’s belief fails to
constitute knowledge. We know that belief aims at the truth, for example,
because we know that if someone knows that someone else’s belief about p is
not true, this outside observer has sufficient warrant for asserting that this belief
is incorrect or mistaken. On the hypothesis that belief also aims at knowledge,
we should expect that those who know we don’t know that p for any reason will
be disposed to say we have made a mistake, we were wrong to believe what we
did, or that we should suspend judgment. This is not what we find. If this is
how we evaluate claims about the epistemic aim and the epistemic ought, we
not only fail to find support for the knowledge account, we find evidence for
denying that belief is governed by the knowledge norm. If beliefs that fail to
conform to no norms are justified, we find evidence for denying that knowledge
of p’s truth is necessary for the justification of the belief that p. Suppose you
think you saw a barn. You did, but you did not realize that you were in the
land of fake barns. Because the hills were filled with convincing fakes, we do not
think your belief constitutes knowledge. Knowing this, however, I do not think
that your belief failed to fulfil its aim. Knowing that you do not know and why
your belief is not knowledge, I would not be inclined to tell you are wrong to
believe what you do or that you have made a mistake by believing that you see
a barn. If a belief such as this does not miss its mark, nothing is left of the view
14 See Bird 2007, pp. 93, Sutton 2005, and Williamson 2000, pp. 48.
that belief aims at knowledge. The fakes prevent your belief from fulfilling its
aim only when they fool you into believing a fake barn is genuine.
Now, it should be noted that some of these authors have argued that knowl-
edge is the aim of belief, and it is worth taking a moment to address the argu-
ment. It is clear, I think, that truth is at least among the aims of belief. A false
belief is mistaken, there is no sense in denying that. What is wrong with saying
that truth is all that belief aims at? Sutton remarks:
Another assumption is almost as common as the assumption that
truth maximization/falsity minimization is a primary epistemic goal–
the assumption that a central fact about belief is that it aims at
truth. Known unknown beliefs again suggest that this is not so. If
belief aims at truth, then the belief that one will lose the lottery ...
will, in almost all cases, succeed in fulfilling that central aim, and so
should be impeccably formed, that is, justified. If, as I will argue,
the known unknown beliefs are not justified, and are not justified be-
cause they do not constitute knowledge, we should rather say that
belief aims at knowledge.15
The argument assumes that if something is an aim or, perhaps, a legitimate aim,
of φ-ing, φ-ing is justified if it fulfills that aim. The thought here is that if there
are beliefs that are true, they should be justified whether or not they constitute
knowledge if belief’s aim is truth but not knowledge. The problem with this
argument is that it ignores the possibility of side-constraints that distinguish
between legitimate and illegitimate ways of pursuing legitimate aims. One could
have a view on which the justification of φ-ing turns entirely upon whether the
aims of φ-ing were pursued by legitimate means or also upon whether φ-ing
fulfilled it aims, but Sutton’s argument suggests that neither of these views are
possible. This, I think, is why his argument fails.
If we set aside the question about aims and focus on the normative question,
it seems that if someone said that you should not believe it is a barn knowing that
your belief fails to constitute knowledge simply because the belief is Gettiered,
it seems that they have made the mistake, not you. If that is right, there is
nothing left of the view that knowledge is what is necessary for permissible
belief. In saying that it is not epistemically wrong to believe p if that belief
has been Gettiered, it might seem I am denying something Reynolds says in
his discussion of Gettier cases and warranted assertion.16 He says the locals
who know that you have been driving through fake barn country would not say
that you should believe you saw a barn. They know that you were reasonable in
holding this belief, but they know that given the grounds on which your belief is
based, you did not have the power to distinguish fake from genuine barns. This
may be true, but the knowledge account is not necessary for explaining why the
locals would not (and ought not) say it is permissible for you to believe you saw
a barn. Reynolds says the locals do not know whether the particular belief you
15 Sutton 2007, pp. 23.
16 Reynolds 2002, pp. 150.
have formed is true. They only know that your grounds are not effective for
determining whether your belief is true. Because of this, it would be wrong for
them to assert that you should believe what you do because they do not have
a true and reasonable belief that your belief satisfies the truth norm. Once we
give the locals the additional piece of information that your belief is correct and
they know that the sole reason you fail to know has to do with factors beyond
those that determine whether you are justified or you are right, speakers are
not disposed to think you should revise your beliefs and suspend judgment.
Without this information, however, we cannot use their responses to evaluate
the respective merits of the knowledge account or the weaker truth account.
6.3.3 Moore’s Paradox

We shall now look at arguments that appeal to observations about Moore’s Para-
dox in the hopes of motivating the knowledge account of warranted assertion.
Consider the statement ‘Custer died at Little Big Horn, but I believe he did not’.
It seems contradictory to assert this, but it easily could have been true. (I know
next to nothing about the history of the United States.) What accounts for the
appearance of contradiction in the absence of contradiction? One suggestion is
that anyone who holds the beliefs associated with Moorean absurd statements
holds beliefs that conflict with the rational commitments that come with those
very beliefs.17 For example, it is thought that belief has as its aim the truth,
and someone who holds the beliefs associated with ‘Custer died at Little Big
Horn, but I believe he did not’ would be committed to denying the accuracy
of the belief expressed by the first conjunct. That is akin to a contradiction.
The appearance of contradiction is explained in terms of the conscious conflict
between the beliefs of which the subject is aware and the rational commitments
that come with the beliefs associated with the Moorean absurd statements (e.g.,
such as the fact that, by her lights, she has misrepresented how things stand by
believing Custer died at Little Big Horn).
The arguments we are about to consider all purport to show that knowledge
is the norm of belief and this, in turn, is supposed to show that knowledge is
a norm that governs assertion as well. Those who deny either or both thesis
are supposed to have trouble explaining why Moorean absurd thoughts and
assertions are defective. They all rest on a methodological assumption that I
shall grant for the purposes of this discussion:
Anyone who holds the beliefs associated with a Moorean absurd
statement holds beliefs that conflict with the rational commitments
that come with those very beliefs (MA).
Our first argument tries to establish that knowledge is the norm of belief on
the basis of two assumptions, one of which is undeniable and one of which is
thought to follow from MA. According to the second, our intuitions will lead
us to classify types of statements as Moorean absurdities that could only have
17 See Adler 2002, de Almeida 2001, and Huemer 2007.
that status if knowledge is in fact the norm of belief. According to the third,
alternatives to the knowledge account are too weak to explain why certain kinds
of Moorean absurdities have that status.
In making the case for the knowledge norm of belief, Huemer helps himself
to two assumptions:
Consciously believing p rationally commits you, upon reflection,
to comprehensively, epistemically endorsing your belief that p
(MCP).
Knowledge attribution is the most comprehensive epistemic endorse-
ment (ETK).18
Given these assumptions, if you believed p it would be wrong to have that belief
without endorsing it as knowledge. But, you should not endorse that belief as
knowledge unless it is knowledge. Therefore, you ought not believe p unless you
know p.
In defence of MCP, remember that MA tells us that whenever someone
utters a Moorean absurdity, they have uttered something absurd because the
beliefs associated with that statement conflict with the rational commitments
that come with those beliefs and so those beliefs cannot be comprehensively
endorsed. As for ETK, there seems to be no more comprehensive epistemic
endorsement of a belief than one that says the belief constitutes knowledge.
Suppose we were to grant ETK. Here is an initial worry about Huemer’s
strategy. His argument implicitly assumes that each of the conditions that fig-
ure in a comprehensive epistemic evaluation pertains to the permissibility of
belief. This does not seem right. Consider moral evaluation. A moral eval-
uation that focused on just the permissibility of some action would not be a
comprehensive evaluation. A comprehensive evaluation of an action should not
only tell us whether the act was permissible, but also whether the act had moral
worth. That my act lacks moral worth does not show that I have acted imper-
missibly and so does not show that my act lacked justification. Similarly, that
my action was not supererogatory similarly does not show that my action calls
for a justification. Huemer offers no reason to think that a comprehensive eval-
uation of belief would concern only a belief’s deontic status. Thus, even if we
assume ETK, it would not be surprising if some of the conditions necessary for
knowledge were not necessary for permissible belief.
We should be given some reason for thinking that a comprehensive epistemic
endorsement is only concerned with properties that are of deontic significance,
and he gives us no such reason. We have already seen some reason to think that
the permissibility of believing p does not turn on that belief’s being properly
endorsed along all lines of epistemic evaluation. Suppose you thought that if
a belief constitutes knowledge, that belief is more valuable from the epistemic
point of view than a belief that fails to so constitute knowledge. Along one line
of evaluation, a true belief someone is justified in holding might be less valuable
than an item of knowledge. (Think of Gettier cases and covert lottery beliefs.)
18 Huemer 2007.
The additional value that attaches to beliefs that constitute knowledge would
only be necessary for permissible belief if we were to say that you should never
harbour covert lottery beliefs or hold beliefs in Gettier cases. So, we either
have to say that a comprehensive epistemic endorsement concerns more than
just that which bears on the permissibility of belief or deny the evaluative claim
that items of knowledge are epistemically more valuable than beliefs that fail to
do so. If a comprehensive epistemic endorsement concerns more than just those
properties of a belief that are of deontic relevance, a belief might not be one we
can endorse and not amount to knowledge even if it is permissibly held.
Maybe this worry is relatively minor. This next worry is much more serious.
While the knowledge account does seem to follow from MCP and ETK, you
also get the result that you should not believe p unless you know that you
know that you know (etc. . . ) that p. Clearly, the argument needs revision. It
is not hard to find the needed fix. We simply have to rewrite MCP and say
that consciously believing p requires that you should not believe yourself to fail
to satisfy the standards of a comprehensive epistemic endorsement while also
consciously believing p. This revision is independently motivated. It is simply
not true that if you permissibly believe p, you ought to believe that your belief
satisfies a comprehensive epistemic endorsement. Failing to have any belief
about whether you know p when you happen to believe p (and happen not to
be wrong to do so) is no sin at all. What rationality requires is that you revise
your beliefs if you believe that you cannot comprehensively endorse them (i.e.,
by judging that they are false, that they don’t amount to knowledge, etc. . . ).
We should replace MCP with:
Consciously believing p rationally commits you, upon reflection, to
refrain from believing both p and that p cannot satisfy the standards
of a comprehensive epistemic endorsement (MCP2 ).
If we replace MCP with MCP2 , we undermine the argument for the knowledge
account. According to ETK and MCP2 , all that permissibly believing p requires
is that you do not both believe that you do not know that p while holding the
belief that p, and that is a requirement you could easily satisfy even if you did
not know that p. All it takes to satisfy this requirement is not forming the belief
that you do not know p.
Adler offers a similar argument for the knowledge account, but it seems not
to suffer from the same difficulties.19 He thinks we can use our intuitive sense of
which combinations of attitudes would constitute Moorean absurd combinations
to tell us something about the norms governing those attitudes and is as follows.
According to MA, if any judgment that expresses the belief that p is coupled
with the acknowledgement that some condition C does not obtain, you are
cognizant that, by holding these beliefs, you violate rational requirements on
holding those beliefs. The judgment expressed by ‘p but I do not know it’ is
incoherent in the way Moorean absurdities are. Therefore, in judging that you
19 See Adler 2002.
do not know p, you know there is something wrongful in believing p—namely,

that you do not know p.
This is an advance because the argument does not assume the Metacoherence
Principle. Instead, it relies on the incoherence test:
If it is incoherent in the way Moorean absurd statements are to φ
while acknowledging that C does not obtain, in acknowledging that
C does not obtain, you are cognizant of something that makes it
wrong to φ—namely, that C does not obtain (IT).
Given our methodological assumption concerning the proper resolution of Moore’s
Paradox, it seems we can test proposals concerning the norms of belief as follows.
If it is incoherent to simultaneously believe p while believing C does not obtain
in the way a Moorean absurd thought is (i.e., apparently contradictory without
being a belief in a contradiction), there is a norm that enjoins us to refrain from
believing p when C does not obtain. For example, it seems incoherent to believe
the following:
(1) I believe Custer died at Little Big Horn, but he did not.
In representing the belief about Custer as being false, it seems we have an
incoherent combination of attitudes without a contradiction. Perhaps this is
due to the fact that belief is governed by the truth norm. It seems similarly
incoherent to believe:
(2) I believe Custer died at Little Big Horn, but there is no reason
for me to think that.
On the assumption that the only thing that could be a reason for me to believe
is a piece of evidence, it seems we can infer from the fact that (2) is a Moorean
absurd thought that the belief about Custer is governed by an evidential norm.
It has been observed that it is incoherent to believe both that p is true and that
this belief fails to satisfy one of the conditions necessary for knowledge. It is
also incoherent to believe that p but that p is not known to be true. So, by
similar reasoning, it seems that we ought to accept that knowledge is the norm
of belief.
Unfortunately, this test is insufficiently discriminating and might only be
useful for uncovering normative requirements governing combinations of belief
rather than useful for uncovering the norms that govern those beliefs individu-
ally. To appreciate the first problem, consider an example:
(3) God hates my atheism and it is raining outside.

This is a Moorean absurd thought.20 However, there is no norm that enjoins
us to refrain from believing that it is raining outside unless God forgives the
non-believer. It is no mystery as to why (3) is incoherent. It is incoherent
because the belief that God hates my atheism is a Moorean absurdity in its own
20 This is a modified version of an example from Sorensen 1988.
right. At the very least, IT needs to be reformulated to avoid these sorts of

example. Even if we grant that for every incoherent pair of attitudes there is
something you are cognizant of that makes one of the attitudes you are conscious
of wrongful, we could say what makes it wrongful is precisely that it is held in
combination with the belief that C does not obtain. It might be that this belief
alone is absurd, irrational, or contravenes an epistemic norm. We might say, as
it were, that whenever believing p while believing C does not obtain constitutes
a Moorean absurdity, all that follows is that:
(4) You should not believe: p and that C does not obtain.
That is different from:
(5) If C does not obtain you should not believe p.
The former is a normative requirement and the ‘ought’ takes wide scope. The
latter is a norm in which the ‘ought’ takes narrow scope. The former tells us
what combinations of attitudes we ought to avoid. The latter tells us what sorts
of conditions bear on whether to hold the belief in question. As we are trying
to derive norms such as the knowledge norm (i.e., if you do not know p you
must not believe p) from judgments about rational combinations of attitudes
(i.e., it is irrational to believe both that p is true and not known to be true), we
need some reason to think that we can proceed from intuitive judgments about
irrational combinations of attitude to judgments about attitudes we have reason
to refrain from holding when certain non-mental conditions obtain (i.e., that we
have reason to refrain from believing falsehoods or those beliefs not known to
be true even when we have no clue that our beliefs are false or fail to constitute
knowledge).
Maybe these problems are not insuperable. To deal with the first, we can
revise IT as follows:
IT2 : If it is incoherent in the way Moorean absurd statements are
to φ while acknowledging that C does not obtain and the belief that
C does not obtain is not itself incoherent, in acknowledging that C
does not obtain, you are cognizant of something that makes it wrong
to φ—namely, that C does not obtain.
To deal with the second and more fundamental problem, we might say this. The
reason that it is irrational to believe both that p is true and that C does not
obtain is that in representing your present situation as one in which C does not
obtain you thereby appreciate that if that belief is correct, you should expect
there to be reason not to believe p. Moreover, if that belief is incorrect it is
still by your lights a situation where there is reason not to believe p. To believe
against what you take to be good reasons is itself a kind of epistemic wrong.
Perhaps this suffices to address the difficulties that arose for IT. We might
have saved the test, but the revised version of the test does not support the
knowledge account. To see this, note that the test only applies when the belief
that C does not obtain is a belief that is not incoherent taken on its own. While
we might grant that if C is a condition necessary for knowledge it is incoherent to

believe both that p is true and that C does not obtain, we only find confirmation
of the knowledge account if we assume also that for any condition C such that
C is a condition necessary for knowledge it is coherent to believe on its own that
C does not obtain. This is not what we find.
It is not incoherent to believe that you do not know p because p is false.
So, according to IT2 there is a norm that enjoins us to refrain from believing
the false. It is not incoherent to believe that you do not know p because your
evidence does not put you in a position to know whether p. So, according to
IT2 , there is a norm that enjoins us to refrain from believing without evidence.
What of the other conditions necessary for knowledge? So far, we have
only confirmed that you should not believe the false and not believe without
evidence. The belief that my belief about p is Gettiered, like the belief that
God will not forgive my atheism, is incoherent taken on its own. There are
many ways to Gettier a belief, but I shall focus on two. In the first sort of case,
your evidence for believing p is undermined thanks to true propositions of which
you are unaware. In the second sort of case, your evidence for believing p only
accidentally leads you to the correct judgment concerning p. It is reasonably
clear why you cannot coherently and correctly believe yourself to be in the first
sort of case. To believe yourself to be in such a case, you have to believe of
some piece of evidence (i) that you are unaware of it and (ii) that it would,
if combined with your present evidence, undermine the justification you have
for believing p. However, you would have this evidence in mind if you were to
believe this. The case is not possible. If you were aware of the evidence you
would no longer be justified in believing what you did. This means this would
not be a Gettier case.
What about the second sort of Gettier case? In this case, you fail to know p is
true due to the accidental connection between your grounds and the truth. You
cannot coherently believe that you fail to know that p simply because you are
only accidentally related to the truth about p. To believe that your connection
to the truth is accidental is (roughly) to believe that if p turns out to be true,
this is not to be expected. The judgment that you are in this sort of Gettier
case amounts to the complex judgment that p is true but that you are not in
an epistemic position to expect that it is true. This is a Moorean absurdity in
its own right. We can account for its absurdity by noting that someone with
such attitudes would take their belief that p is true to fail to fulfil the evidential
norm since what is to be expected is a function of the evidence available. The
subject’s taking it to be the case that the truth of the relevant belief is not to be
expected is acknowledging that by her lights she cannot expect to be right given
her evidence. IT2 fails to confirm that this condition is necessary for permissible
belief.
The challenging case for those who do not think knowledge is the norm of
belief is this one:
(6) p but I do not know p.
If you take yourself not to know that p because you take yourself not to believe
p, say, then we have a situation where your second-order belief is transparently

falsified by a fact about your own mind. If you take yourself not to know that
p because you take it to be that p is false, we can explain the incoherence by
appeal to the truth norm. If you take yourself not to know that p because you
have insufficient evidence by your own lights, we can explain this in terms of the
evidential norm. That norm in turn can be derived from the truth norm in just
the way Williamson suggested earlier. And, if you take yourself not to know
because you take yourself to be in a Gettier case, we have already seen how
the evidential norm can explain the incoherence of that attitude. It seems that
we have our bases covered. If we were to assume that the solution of Moore’s
Paradox should be given in normative terms and assume that MA is true, while
the knowledge account might have the resources to explain the absurdity of the
thoughts we have thus considered it seems the knowledge account is unnecessary
for explaining why Moorean absurd thoughts strike us as contradictory. It seems
we can explain the same data using either the truth norm combined with the
evidential norm or the evidential norm taken on its own. So, while some might
think that MA is a dubious assumption and dispense with the very idea of
using Moore’s Paradox as a way of uncovering the norms of belief, MA properly
understood does not lead us to endorse the idea that the reason that the concept
of knowledge is significant is that it plays an essential role in the formulation of
the norms of belief.
6.4 Truth and Reasonable Belief

In dealing with lottery cases and cases of Moorean absurdity, the real work
is done by two assumptions. The first is that you cannot reasonably believe
outright what you know you are not in a position to know. The second is
that you should not assert what you cannot reasonably believe. Since you can
reasonably believe false propositions, say, if you happen to be the same on the
inside as someone who believes a true proposition on the basis of adequate
evidence, it is not at all clear what work the TNA can do in an account of
warranted assertion. Perhaps Williamson is right that TNA commits you to
ENA, but I can think of no reason why ENA would commit you to TNA.
Should we say that ENA is the norm that governs assertion and an assertion
is warranted iff the speaker is reasonable in believing that the proposition she
asserts is true?
The view that is emerging in the literature as the main rival to the knowl-
edge account of assertion takes seriously the idea that belief and assertion are
governed by common epistemic standards, notes that belief is governed by a
norm that says, in effect, that it is proper to believe p iff you justifiably believe
p, and then says something similar for assertion. The external components that
distinguish knowledge from justified belief are conditions that have nothing to
do with warranted assertion. The view denies TNW on the grounds that either
JA or RA is correct. Myself, I think this is a mistake.
How should we approach this issue? There is a debate between some writers
as to whether the point of assertion is to say something true or say something

you know. For reasons we have already discussed, I do not think that assertions
aim to express knowledge and do not think that there is any norm that tells
us that you should refrain from asserting what you do not know even in cases
where you have assumed responsibility for testifying to the truth of something
for the benefit of another. The debate between those who accept a truth norm
and those who deny that there is any such norm is not really about the point,
aim, purpose, or goal of assertion. Both parties agree that in telling someone
p, you tell them that p is true and so you fail to in your aim, in some sense, if
you speak falsely. One side thinks this is normatively significant and the other
does not. The side that denies TNW does not deny that the norms governing
assertion have something to do with truth. If I understand their view, they
think that we cannot fail to live up to our responsibilities or fail to meet our
duties simply because we speak falsely even if what we are trying to do is speak
the truth. The thought is that the right to do something cannot depend upon
something external like truth, not because there are overriding reasons to speak
falsely. There can be, but those are non-epistemic reasons and they are not
needed. You can have warrant to assert something that happens to be false
provided that you were reasonable in taking yourself to speak the truth.
The debate, then, between those who defend TNW and those who defend
RA as an alternative that allows for false, warranted assertions is similar to
debates about product liability. There are some who think that your obligation
is to exercise due care and that your obligation must end there. There are
some who think that a strict liability standard determines whether you have
lived up to your obligations. According to Thomas Nagel, strict liability is an
irrational moral doctrine.21 Perhaps he would not be terribly keen on it as an
epistemological doctrine, either. In its place, he and others might say we should
take reasonable care or caution but deny that the permissibility of anything can
turn on factors of which we have no knowledge if we are non-culpable in our
ignorance.
On what grounds would they say this? They seem to deny that a strict
liability standard governs assertion by offering a blanket denial that genuine
normative standards are ever strict. Those who defend RA sometimes say that
the what is impermissible always involves fault and what can reasonably taken
to be right cannot be wrong.22 Lackey writes:
[T]here is an intimate connection between our assessment of asserters
and our assessment of their assertions. In particular, asserters are
in violation of a norm of assertion and thereby subject to criticism
when their assertions are improper. An analogy with competitive
basketball may make this point clear: suppose a player steps over the
free throw line when making his foul shot. In such a case, there would
be an intimate connection between our assessment of the player and
our assessment of the free throw—we would, for instance, say that
21 See 1979, pp. 31
22 In addition to Kvanvig and Lackey, see Douven 2006, pp. 477 for an instance of this.
the player is subject to criticism for making an improper shot.23

She argues that TNW does not govern assertion precisely because it does not
follow from the mere fact that someone asserts something false that the speaker
is herself properly subject to criticism. While she is right that a speaker is not
properly criticized for asserting something false, per se, it seems her criticism of
TNW rests on the assumption that the reasonable is the mark of the permissible.
We have already seen that you cannot be fully excused for having φ’d unless it
was reasonable for you to think it is permissible to φ and the excusable is not the
mark of the permissible. There might be an argument that shows that assertion
is somehow special insofar as there is a tight connection between assessment of
the speaker and the speaker’s speech acts, but what we are looking for is an
argument.
Like Lackey, Kvanvig also thinks there is a close connection between the
assessment of the speaker and the assertion. He remarks:
This point should be self-evident . . . norms of assertion are norms
governing a certain type of human activity, and thus relate to the
speech act itself rather than the content of such an act. Notice that
when we look at the four conditions for knowledge above [i.e., truth,
belief, absence of defeaters, and justification], the only ones regard-
ing which apology or regret for the speech act itself is appropriate are
the belief and justification conditions. There is, therefore, a prima
facie case that knowledge is not the norm of assertion, but rather
justified belief is.24
The upshot is supposed to be that if some condition makes it wrongful to act or
assert, the very same condition makes it appropriate to regret having acted or
asserted and to apologize. From this it is supposed to follow that the internal
conditions necessary for knowledge determine whether an assertion is warranted.
The external conditions that distinguish reasonable belief from knowledge are
normatively insignificant because you need not apologize for having asserted
something false and need not regret having asserted something false.
I can see why someone would say that what you should apologize for is
what you can be faulted for and what you can be faulted for is what you should
apologize for. I agree with Kvanvig that you canot be faulted simply for asserting
something false. Having said that, I do not see how to convert his remarks into
an argument for the reasonable belief account of warranted assertion. You
should not engage in any wrongdoing, excusable or otherwise. If you do engage
in excusable wrongdoing, it is not entirely clear that an apology is in order.
An explanation might be in order and you might have an obligation to make
reparations, but it seems that you offer an apology in recognition of the fact that
someone else can properly take up reactive attitudes against you. If, however,
you ought to be excused for your action, it would be inappropriate for someone
to take up reactive attitudes against you. So, it seems that if there is a close
23 Lackey 2007, pp. 595.
24 Kvanvig 2009, pp. 147.
conceptual connection between what you should apologize for and what you can
be faulted for, there is not a close conceptual connection between the conditions
under which it is appropriate to apologize and the conditions under which you
act wrongfully. Perhaps whenever you ought to apologize, you committed some
wrong, but it is not at all obvious that whenever you commit some wrong, you
owe an apology. Having said that, if I am wrong on that point, it is likely
because I am wrong in thinking that you only ought to apologize when someone
can properly fault you for what you did. If we sever that connection and say that
you can owe an apology even for an excusable wrong, it seems that you should
sometimes apologize for having done something you could not have reasonably
expected at the time of action to be wrong. If this is so, there is no argument
for internalism here.
What about regret? Regret is complicated. We have already seen that an
agent can act with justification but regret having so acted when duty requires
her to act against some defeated reason. The agent might rightly regret having
φ’d even though she knows she must φ. The regret is not the recognition that
she ought to have done things differently. She regrets that she could not have
done things differently given what her obligations were and regrets that meeting
her obligations necessitated acting against the defeated reason. That you regret
having φ’d does not mean that you ought to have done something else. Perhaps
what Kvanvig is saying is this. Regret is not the mark of the impermissible;
rather, if something is not the sort of thing that you can properly regret, it is
not the sort of thing that has any bearing on whether to φ. To convert this point
into an argument for an internalist approach to warrant, we have to assume that
you cannot properly regret asserting something that you reasonably took to be
true. The problem with the argument seems to be this very assumption. I
think Oedipus can rationally regret that he married his mother. I think I would
have regretted telling Oedipus that Jocasta made quite the catch. Intuitively,
it seems that we can regret bringing about outcomes we did not know we were
bringing about. My own intuitions might be idiosyncratic, but if I am wrong on
this point, it is not because it is true in general that you should only regret that
you φ’d if you could have known at the time how to avoid φ-ing. In situations of
moral conflict, you know that you cannot avoid wrongdoing whatever you end
up doing. Still, you can regret acting against a reason however you decide to
act. if the unforseeable is not regrettable, it is not because you ought not regret
what you do not know how to avoid. The unavoidable can be regrettable.
One of the issues that separates those think you have sufficient warrant to
assert what you reasonably believe from those who do not seems to be an issue
about the relation between an assessment of the speaker and an assessment of
the speaker’s speech acts. DeRose and Weiner, for example, both deny that
you have sufficient warrant to assert p if ∼p, but agree that there is something
proper about asserting p even if ∼p. Weiner remarks:
Someone who reasonably believes that she is complying with a norm
is in some sense acting properly, even if she is in fact violating the
norm; and vice versa. If I have every reason to think that I know that
Alice is in her office, when in fact she has slipped out through the
window, you may not condemn me for asserting that Alice is in her
office, even though in so doing I violate the knowledge norm or the
truth norm (whichever applies). This is because my assertion was
secondarily proper even if it violated the primary norm of assertion.
We will deal largely with cases in which the distinction between
primary and secondary propriety does not arise because the speaker
knows exactly what her epistemic situation is.25
This talk of primary and secondary propriety is something Weiner picked up
from DeRose.26 The distinction they draw is an important one, but I wish they
would use different language for drawing it. So far as I can tell, this is just the
distinction between the permissible and the excusable. If something is merely
secondarily proper, it is wrong. It is not just pro tanto wrong, it is wrong all
things considered. The agent cannot be blamed for having committed a wrong,
so the agent ought to be excused or the agent cannot be held responsible. If,
however, the agent’s acts are primarily proper, they just are permissible. They
satisfy the demands that the norms governing the act place upon the agent.
Williamson offers us this defense of TNW:
Suppose that I rationally believe myself to know that there is snow
outside; in fact, there is no snow outside. On the BK and RBK
accounts [that say that you have warrant to assert what you believe
you know or you reasonably believe yourself to know], my assertion
‘There is snow outside’ satisfies the rule of assertion. Yet something
is wrong with my assertion; neither the BK nor RBK account implies
that it is. They can allow that something is wrong with my belief
that I know that there is snow outside, for it is false, but that is
another matter. The BK and RBK accounts lack the resources to
explain why we regard the false assertion itself, not just the asserter,
as faulty.27
While he is right that there is something defective with the assertion, those who
deny TNW might say that these defectiveness intuitions are weak evidence.
Someone who acts rightly might bring about regrettable side-effects and these
might be defects of a kind, but by hypothesis, the agent acted rightly. Not
only that, Williamson might have given his opponents just what they need, an
explanation as to why the assertion seems faulty. The false assertion would
dispose someone who trusted the speaker to believe something false and, as
Williamson notes, there is something wrong with the false belief. Since he
thinks such beliefs are nevertheless permissibly held, it is not at all clear why
the defect he has focused on makes the assertion wrongful.
I want to try to shore up the support for TNW by taking aim at this idea that
the test that determines whether someone has lived up to her responsibilities
25 2005, pp. 229.
26 See DeRose 2002.
is that the agent can be said to have been responsible in trying to meet them.
This seems to be the position of those who defend RA, and I think the view
does not get the cases right. Let’s start with an example:
Peacock just moved into the apartment next to Plum’s. To welcome
her to the building, Plum cooked her dinner. She did not realize
that the mushrooms she used in making her dinner were poisonous.
(So far as this is possible, imagine that she is not culpable or blame-
worthy for her ignorance. She used a field guide for distinguishing
safe from unsafe mushrooms, but it contained a few errors.) Plum
has on hand the stuff to give people who eat poisoned mushrooms,
but only enough for one person. It just so happens that her other
neighbor, Mustard, is suffering from food poisoning because he ate
a can of bad peaches. (So far as this is possible, imagine that he is
non-culpably ignorant). Plum’s stuff could help Mustard just as well
as it could help Peacock. It’s good stuff. Now, Mustard and Peacock
are equally sick and Plum can help only one. It seems intuitively
clear that Plum has a more stringent duty to assist Peacock than to
assist Mustard. She did poison Peacock, after all (Cook).
The example suggests that when it comes to an agent’s action, it is possible for
two agents to be internal duplicates up to the point of action but then differ
with respect to whether they acted permissibly. We could easily imagine a story
similar to Cook in which there were no poisonous mushrooms where it is clear
that some internal duplicate of Plum’s does nothing wrong in cooking a dish for
her new neighbor.
If Plum’s duty to Peacock was just some prima facie duty of beneficence, it
would be difficult to see how the duty to Peacock could be more stringent since
Mustard’s needs are just as great as hers. Thus, it’s tempting to think that
Plum’s duty is no mere duty of beneficence. My hypothesis is this. The reason
that Plum’s duty to Peacock is more stringent is that Plum is righting some
past wrong of hers by assisting Peacock. Whereas reasons having to do with
beneficence count in favor of helping both Mustard and Peacock, the reparative
duty gives a reason that breaks the tie. We cannot make sense of how there could
be this wrong on any internalist view for the simple reason that it seems there
is no ground for wrongdoing that is constituted by or strongly supervenes upon
the internal conditions that determine how things seem to Plum and Plum’s
counterpart. The reason she ought to assist Peacock first is that she poisoned
Peacock by serving her poisonous mushrooms, and this fact is something that
is not accessible to Plum.
Someone might say that while Plum has a more stringent duty to assist
Peacock, it doesn’t follow that this is a duty to address some prior wrong she’s
committed. Perhaps it is no mere duty of beneficence, but it is not a reparative
duty if such duties are understood as responses to past wrongs that the agent
has committed. To give this kind of duty a name, we can speak of reparative*
duties. A reparative* duty is similar to a reparative duty insofar as they are
duties one can be under only if the agent brought about some bad state of
affairs, but they are like the duty of beneficence insofar as they can arise without
any prior wrongdoing on the agent’s part. Why not say that the difference in
stringency is due to the fact that there is a prima facie duty to assist both
Peacock and Mustard, but a stronger duty to Peacock because there is the
additional reparative* duty that gives her a pro tanto reason to assist Peacock?
That way, we can accommodate intuition without giving up internalism about
the justification of action. The problem with this response is with this idea of
reparative* duties. If this is merely a reparative* duty, then we would have
to say that this is a case in which Plum did not act against any pro tanto
reason to refrain from giving Peacock the poisoned dish. (Otherwise, we would
have to say that this was a reparative duty.) But, then it seems quite odd to
think that Plum could have such a duty because it would have to combine two
features. First, it would have to give Plum a reason to act that a similarly
situated but causally idle agent would not have. (Otherwise, we would say that
the reparative* duty was really a mere duty of beneficence. It would be the very
duty that, say, Green would have if he had just the same amount of stuff to give
to someone who has been poisoned as Plum has.) Second, it would have to be a
reason for Plum to act over and above a reason associated with a mere duty of
beneficence to address some bad state of affairs when she could know full well
that she never had any reason not to bring that bad state of affairs about in
the first place.
On this account, there would be a resultant moral difference between Plum
and Green’s duties (i.e., both would have reasons of beneficence to assist either
subject but Plum would have the additional reason to discharge a reparative*
duty) that alters the range of permissible options available to them that arose in
virtue of a causal difference that was not coupled with any normative difference.
That sounds quite odd. Better, I think, to say that the reason that this causal
difference between Plum and Green makes a normative difference because it
was in virtue of a causal relation between Plum and the bad state of affairs that
she acted against a pro tanto reason unknowingly and now has the knowledge
necessary to see that her actions were wrongful and there is a wrong that needs
to be addressed. This is why Peacock has a stronger claim on Plum’s assistance
than Mustard does. But, this is why there is a reparative duty that Plum ought
to discharge, not a reparative* duty.
Let’s add a further detail to the story. Suppose Plum didn’t know what to
make Peacock to welcome her to the building. She asked White. White said
that she should use the mushrooms in the garden to make her dish and Plum
followed his advice. Should White have said this? Here’s a principle that seems
pretty plausible:
If an advisee oughtn’t φ and there is no reason to give insincere
advice, the advisor oughtn’t assert that the advisee ought to φ (Ad-
vice).
Why? If it is false, the reasons that speak against φ-ing do not constitute
reasons to refrain from encouraging someone to act against those reasons by
advising them to do so. That seems to go against everything we know about
giving sincere advice. Yes, sometimes we should give insincere advice but the
principle takes account of that. If the argument above is correct, a kind of
non-culpable ignorance works as an excusing condition. When such excusing
conditions obtain, the agent can only act rightly if there is some justifying
reason for giving the neighbor the poisoned dish. There is none in the story
I have just told. That is true of the action and the assertion that the action
should be performed.
Now we have our case against RA:
(1) Circumstances can arise in which a decisive case can be made
against φ-ing where the reasons not to φ are grounded in consid-
erations the agent is non-culpably ignorant of (e.g., Cook).
(2) In such cases, an advisor might also be reasonably ignorant of
the reasons that constitute a decisive case against φ-ing.
(3) In such cases, there is nevertheless a decisive case to be made
against the advisor’s asserting that the advisee ought to φ.
(C) Circumstances can arise in which a decisive case can be made
against the advisor’s assertion that the advisee ought to φ where
the considerations that constitute this case are considerations the
advisor is non-culpably ignorant of.
Since you cannot have warrant to assert that p is the case when there is a decisive
case to refrain from asserting that p is the case, Cook is a counterexample to RA.
Now, someone could resist this and say that the assertion was merely morally
defective, not epistemically defective. The effect of this would be to undercut
the motivation for RA. So far as I can tell, the motivation for RA is the general
thought that if an agent is fully responsible in how she conducted herself, she
could not have failed to live up to her responsibilities. As we have seen, it is
hard to square this view with the intuitive data. In the next chapter, we will see
further evidence that supports the view that a strict liability standard makes
good moral sense and epistemic sense, but insofar as the reason to reject TNW
seems to be the denial of this point, I deny that RA is well-motivated.
6.5 Justification and Warrant

In the previous section, I argued that there are cases in which someone reason-
ably believes p where they lack sufficient warrant to assert p. The argument
is supposed to show that there are cases in which the fact that an assertion is
false (or the facts by virtue of which it is false) constitute a decisive reason not
to assert that the proposition asserted is true. In this section, I shall argue that
if someone justifiably believes p, cannot lack sufficient warrant to assert p.
6.5.1 Justification Ascriptions

Here, I shall argue that if truth is necessary for warranted assertion, it is nec-
essary for justified belief. Of course, not everyone agrees with me that there
cannot be false, warranted assertions and that excuses are needed if someone
speaks falsely in a context where their overriding responsibility is to serve as
a testimonial source for another. There are a fair number of writers who do
agree with me.28 The aim of this section is to argue that they cannot have it
both ways. There cannot be false, justified beliefs because there cannot be false,
warranted assertions. The explanation for this is that you cannot lack warrant
for asserting what you have the right to believe. So, the right to believe requires
truth because it provides warrant and warrant requires truth.
Consider two further norms:
(TNJ) You should not believe p unless p is true.
(WNJ) You should not believe p if you lack sufficient warrant to
assert p.
In this section, I shall argue that anyone who violates TNJ violates WNJ (i.e.,
if you believe a false proposition, you cannot have sufficient warrant to assert
that what you believe is true). If you do not have warrant to assert that p is
true, you should not believe p. You might believe p even if you should not, but
not with justification. If among the norms of assertion is a norm that enjoins
us to refrain from asserting falsehoods, there is a norm that similarly enjoins
us to refrain from believing falsehoods. Since there are norms that enjoin us to
refrain from asserting falsehoods, FactivityJ is true. Key to the argument is the
assumption that TNA governs assertion and that you should not assert p unless
p is true.
An obvious argument for FactivityJ derives TNJ from the combined assump-
tions of TNA and WNJ. It then derives FactivityJ by means of the additional
assumptions that justification is a deontological notion and that there’s no norm
that takes precedence over TNJ. The argument depends on the assumption that
the conditions that determine whether our beliefs are justified can only ensure
that our beliefs are justified if they thereby give us the right to assert that our
beliefs are true. WNJ seems to me to be rather plausible. If you don’t have suf-
ficient warrant to assert p, you epistemically shouldn’t assert that p is the case.
If you shouldn’t assert that p is the case in this sense, there is an undefeated
reason for you to refrain from asserting that p is the case. If it is nevertheless
the case that your belief is justified, that is either because the reason to refrain
from asserting is not a reason to refrain from believing or because there is an
overriding reason to believe that does not provide a justification for asserting.
The suggestion that the (alleged) reason to refrain from asserting on epistemic
grounds would not constitute a reason to refrain from believing is obscure, as is
the suggestion that the (alleged) overriding reason to believe could not provide
an overriding reason to assert what there is (allegedly) a reason not to assert.
So, it seems plausible to maintain that belief and assertion are held to common
rather than divergent epistemic standards.
28 DeRose 2002, Weiner 2005, and Williamson 2000 all deny that there can be false, war-
ranted assertions but all believe there can be false, justified beliefs.
While this strikes me as a reasonable theoretical rationale for WNJ, there

is another route to FactivityJ to consider. If assertion is governed by TNA, it
seems the following principle is true:
If S were to assert that p is true and cause S’ an epistemic harm
by convincing S’ to believe p, S should not assert that p is true
(Non-Maleficence).
If someone’s assertion convinces you that p is true when in fact p is false, you
have suffered an epistemic harm. So, if Non-Maleficence is true, you should not
assert false propositions. Suppose I know are curious as to whether p is true
and I assert:
(1) There is sufficient justification for believing p.
As there cannot be sufficient justification for believing p unless it is permissible
for you to believe p, it follows from (1) that:
(2) It is permissible for you to believe p.

When p is false, it seems that my assertion that (1) is true or that (2) is true
could cause the very same epistemic harm as my assertion that p is true. Thus,
if you accept Non-Maleficence, it follows that you cannot have sufficient warrant
to assert (1) or (2) if p is false.
Suppose that this much is right. Let me add an additional assumption.
While it seems safe to say now that knowledge is not necessary for warranted
assertion, it is relatively uncontroversial that knowledge is sufficient for war-
ranted assertion. Remember that the notion of propriety that figures in an
account of warranted assertion is epistemic and it is hard to see what more than
knowledge that p is true could be needed for properly saying that p is true. If
knowledge, justified belief, or truth suffices for warranted assertion, knowledge
suffices for warranted assertion. If knowledge is not enough for warranted as-
sertion, situations should arise where we can properly say something like, ‘He
knew that p, but he was in no position to claim that p was true’. I doubt such
situations would arise with any frequency. Combine the Non-Maleficence with
the thesis that knowledge suffices for warranted assertion and you get the result
that you cannot know that (1) or (2) is true if p is not true. If p is not true,
Non-Maleficence says that you cannot have sufficient warrant to assert that (1)
or (2) is true. It follows from this and the thesis that knowledge suffices to
warrant assertion that you do not know that (1) or (2) is true.
Why can’t you know that (1) or (2) is true if p is false? Let’s run through
the potential explanations. The first is that your belief in (1) and (2) cannot
be justified unless p is true. This explanation assumes Factivity. We often
have sufficient warrant to ascribe justification as we do not infrequently ascribe
knowledge to one another. If the fact that p is false does not prevent someone
from justifiably believing p on its own, why would it prevent the speaker from
having sufficient justification to believe that there is sufficient justification for
the first-order belief that p is true? The second is that your belief in (1) and
(2) cannot be true unless p is true. This explanation also assumes FactivityJ .
The third potential explanation is that you cannot believe (1) and (2) unless p
is true. This does not assume FactivityJ , but it is crazy. The fourth potential
explanation is that if p is false, anyone who believes (1) or (2) fails to know (1)
or (2) for purely Gettierish reasons. This also does not assume FactivityJ , but
it has no plausibility. So, it seems that the only two plausible explanations as to
why it follows from the fact that p is false that someone cannot have sufficient
warrant to assert (1) or (2) assume that it cannot be that someone justifiably
believes p when ∼p.
Let’s take stock. If truth is required for warranted assertion and common
epistemic norms govern assertion and belief, truth is required for having the
right to believe. I would not say that this is because you have the right to
believe only what you have the right to assert. This seems to get the order of
explanation the wrong way around. The evidence suggests that assertion and
belief are governed by common standards, and so a sign that truth is required
for justified belief is that it is required for warranted assertion. At least, it
is in some cases. Since there cannot be more to having the right to assert
p than is involved in the right to believe p, both require truth. If epistemic
justification is a deontological notion in the seemingly trivial sense that you
should not believe without justification, it seems to follow that there can be no
false, justified beliefs. Even if we should not work from the assumption that
beliefs and assertions are governed by similar norms, it seems that our best
accounts of warranted assertion are committed to FactivityJ . If accepting our
assertions causes our audience to suffer a direct epistemic harm, we should not
have asserted what we did. When we assert that p is true when in fact p is
false, we can harm our audience in a way that is epistemically objectionable.
Intuitively, it seems that asserting that there is sufficient justification to believe
a proposition that turns out to be false causes the exact same harm. It follows
that if p is false we cannot have sufficient warrant to assert p and so cannot know
that there is sufficient justification to believe p when ∼p. This fact calls out for
explanation and the only remotely plausible explanations assume FactivityJ .
Chapter 7
Action
7.1 Introduction
Earlier in our discussion, I argued that we ought to think of justification in
externalist terms because we ought to think of reasons for action as constituted
by facts about the situation rather than facts about us. Belief is supposed to
provide us with reasons from which we can then reason. Beliefs that do not
fit the facts pass of spurious reasons as if they were genuine. In this section,
we look at another link between belief and action. Here, I shall argue that
internalism fails as an approach to justification because it cannot do justice to
our moral intuitions or the thought that there is an internal connection between
the normative standing of a belief and the actions that the belief rationalizes.
Rather than look for an argument for externalism in the external conditions
that bring reasonably held beliefs “closer” to knowledge, we should look inwards
towards the role that beliefs play in deliberation. We shall see that orthodox
internalist views fare poorly as do orthodox externalist views. For they too
cannot provide us with an adequate account of the relation between a belief’s
normative status and the normative status of the actions we perform in light of
what we believe.
7.2 Knowledge and Action

When is it epistemically proper or permissible to treat something as a reason
for action? According to Hawthorne and Stanley, knowledge is the norm for
practical reasoning. They say that knowledge and action are related as follows:
Where your choice is a p-dependent choice, it is appropriate to treat
the proposition that p as a reason for action iff you know that p
(RKP).1
1 Hawthorne and Stanley 2008, pp. 578. A choice between options is p-dependent iff the
most preferable option conditional on p is not the most preferable option conditional on ∼p.
137
CHAPTER 7. ACTION 138
To some, the claim that nothing beyond knowledge of p’s truth could be nec-
essary for properly acting on p might seem perfectly harmless.2 The notion of
propriety we are concerned with is epistemic, not practical. What more could
we possibly need to properly treat p as a reason for action? Superknowledge?
Perhaps the main reason that RKP is controversial is that it asserts that noth-
ing short of knowledge of p’s truth could warrant acting on p. In this section, I
shall argue that you can have sufficient warrant to act on p even if you do not
know that p is true and address Hawthorne and Stanley’s arguments for RKP.
In support of RKP, Hawthorne and Stanley write:
Consider . . . how blame, judgments of negligence and so on interact
with knowledge. If a parent allows a child to play near a dog and
does not know whether the dog would bite the child, and if a doctor
uses a needle that he did not know to be safe, then they are prima
facie negligent.3
It is hard to know what to make of this passage because of the qualification
‘prima facie’. Let us ignore this qualification for now and consider the proposal
that:
In cases where you ought not φ unless p is true, you can be blamed
for φ-ing if you do not first know that p is true (Fault4 ).
Given the plausible assumption that it is not proper to treat p as a reason to φ
when you can be blamed for treating p as a reason to φ, it seems Fault1 does
lend support to RKP. Note that judgments of blame, negligence, and the like
also seem to interact with ascriptions of justification:
If you can be properly blamed for believing p, you are not justified
in believing p (Fault5 ).
To deny Fault5 , you would have to say that the facts in light of which someone
can be properly blamed for believing p do not threaten the justificatory status
of that belief. This would be an odd stance to take for someone who argues for
RKP by means of the assumption that if you can be blamed for treating p as a
reason for belief, it is not permissible to treat p as a reason for belief. Problems
arise for any view that incorporates both Fault4 and Fault5 . Combined, these
assumptions entail that if your belief that p is true is practically relevant, your
belief cannot be justified unless it constitutes knowledge.4 To use their example,
suppose that you should not use a needle unless it is clean. From Fault4 , it
follows that you can be blamed for using the needle if you use it but do not know
that it is clean. It seems, intuitively, that you can be properly blamed for using
the needle only if you are not blameless in the belief that it is clean. It follows
2 Brown 2008 criticizes the claim that knowledge that p is true is sufficient for properly
treating p as a reason for action. See Neta 2009 for a response to Brown’s criticism.
3 Hawthorne and Stanley 2008, pp. 572.
4 Let’s say that your belief that p is the case is practically relevant iff you faced with some
p-dependent choice.
from Fault5 , that you cannot be justified in believing that the needle is clean.
Two objections should suffice to show that we should not accept both Fault4
and Fault5 . Given that Fault5 is relatively uncontroversial it seems that the
objections below, if sound, give us good reason to deny Fault4 .
First, according to the JTB analysis of knowledge, if Audrey is justified in
believing p and her belief is true, she knows p. We all know that this analysis
will not do. Audrey and Cooper are on a cross-country trip and stopped this
afternoon to have lunch in the land of fake dollar bills. Neither knows that
they are in the land of fakes. That is why we can say that they are justified in
believing that they have cash and not counterfeit bills in their pockets. Audrey
recalls that she owes Coop ten dollars. She reaches into her pocket, pulls out
ten dollars, hands it to Coop, and says that they are now even. While her belief
that her debt is repaid is true and she is justified in that belief, she does not
know that her debt has been repaid. Or, so the story goes. If, however, she does
not know that her debt has been repaid, it follows from Fault1 that she can be
blamed for acting on her belief that p. In turn, she can be blamed for believing
p. In turn, it follows from Fault2 that her belief that p cannot be justified. In
turn, it follows that Gettier cases are not possible.
Here is a second objection. I do not think that epistemic justification super-
venes upon our non-factive mental states. Those who think that epistemic justi-
fication does supervene on our non-factive mental states will typically also assert
that the conditions that determine culpability and blameworthiness supervene
on these internal conditions. They will deny that it is possible for situations to
arise in which two subjects in precisely the same non-factive mental states φ and
only one of these subjects is properly blamed for φ-ing. Externalists about epis-
temic justification often accuse internalists of conflating this perfectly harmless
claim about blame and the perfectly false claim that justification supervenes on
the same internal conditions. As part of the error theory that purports to ex-
plain why the internalists are mistaken about epistemic justification, they will
say that the conditions that determine culpability and blameworthiness that
do supervene on the internal states are distinct from the conditions that de-
termine deontic status. The internalist’s mistake about justification is due to
their mistaken view that conditions you cannot be culpable for failing to take
account of cannot affect the justificatory status of your beliefs. Justification and
permissibility, they will say, can come apart from culpability.
Suppose, as seems plausible, that the conditions that determine blamewor-
thiness and culpability do supervene upon a subject’s non-factive mental states.
If you combine this supervenience thesis with Fault1 , you get the result that
you can only blamelessly believe p if every possible internal duplicate of you
knows p. This in turn commits you to an infallibilist conception of epistemic
justification according to which it is permissible to believe p only if the reasons
for which you believe entail p. Such a view about justification is quite clearly at
odds with ordinary intuition. Such a view, arguably, leads to skepticism given
the plausible additional assumption that next to nothing we believe about the
external world we believe on the basis of infallible grounds.
At this point, Hawthorne and Stanley might remind us of an important qual-

ification. They said that someone who acts on p without knowing p is, “prima
facie negligent”. If what they meant to say was that someone who violates RKP
appears negligent but need not in fact be negligent, I do not see how blame
judgments interact with knowledge. Their discussion of the interaction between
knowledge, blame, and negligence would have been a distraction. Perhaps what
they should say (and seem to say in some passages) is that anyone who violates
RKP fails to reason in the way that they ought to. However, they might add,
someone might not be blameworthy for having reasoned in a certain way if they
are non-culpably ignorant of the conditions in light of which they fail to reason
as they ought to. Let us assume that this is the picture that they are working
with.
If they reject Fault1 , they can avoid the two difficulties we have considered
thus far, but rejecting Fault4 will not save RKP. Consider this passage:
Consider also how knowledge interacts with conditional orders. Sup-
pose a prison guard is ordered to shoot a prisoner if and only if they
are trying to escape. If the guard knows someone is trying to escape
and yet does not shoot he will be held accountable. Suppose mean-
while he does not know that someone is trying to escape but shoots
them anyway, acting on a belief grounded in a baseless hunch that
they were trying to escape. Here again the person will be faulted,
even if the person is in fact trying to escape. Our common practice
is to require knowledge of the antecedent of a conditional order in
order to discharge it.5
If a guard shoots a prisoner on a baseless hunch they can be faulted for doing
this, but there is a world of difference between knowingly shooting a prisoner
trying to escape and doing so on a baseless hunch. You do not need RKP to
explain what is wrong with shooting an escaping prisoner on a baseless hunch.
If RKP is correct and the guard has good reason to believe mistakenly that a
prisoner is trying to escape, the guard ought not shoot the prisoner if the guard’s
belief is mistaken. The fact that the guard was reasonable in assuming that they
were doing what they ought is an excuse of the shooting, not a justification. This
seems right. Suppose that as the guard raises the rifle to take a shot at Tobias.
He looks just like a prisoner escaping. A second guard standing nearby the first
knows that Tobias is really an aspiring actor spending the weekend in the prison
preparing for his upcoming role in a film as frightened inmate number two. The
second guard might mace the first guard to stop him from shooting Tobias. The
reason it is not wrong for him to mace the guard knowing how painful it will
be for the guard to be sprayed with mace is that the first guard is about to
do something he ought not. He’s lost the right to non-interference as a result.
While RKP gets this sort of case right it seems it gets them right for the wrong
reasons.
To see this, forget about cases of reasonable but mistaken beliefs. Forget
about the cases of aspiring actors that look like prisoners trying to escape and
think about aspiring escapees who surround themselves with aspiring actors.
According to the order, George, who is a prisoner and not an actor, ought
to be shot if he tries to escape. According to the order, Tobias, who is an
actor but not a prisoner, ought not be shot. Assume George tries to escape.
According to the order, the guard ought to shoot him before he makes his escape.
Because unbeknownst to the guard there are aspiring actors like Tobias dressed
like prisoners, the guard does not know George ought to be shot. He merely
reasonably and correctly believes George ought to be shot. According to RKP,
it is wrong to act on the one premise that could justify shooting George (i.e.,
that he is a prisoner trying to escape). George ought not be shot. It seems to
follow that the guard ought to shoot George and ought not shoot George. That
seems like a contradiction.
You are going to run into trouble if you combine RKP with the view that it
is possible for there to be positive duties to φ if p is true if p is the sort of thing
someone can non-culpably fail to know to be true when it is. If it is possible
for circumstances to arise in which p is true where p cannot be known to be
true, it follows that you both ought to φ and ought not φ. That seems like a
contradiction. Similar difficulties arise if you combine RKP with the knowledge
account of assertion. According to the knowledge account, you ought not assert
p unless you know p. It follows that you have a conclusive reason to refrain from
asserting p if you do not know p. It follows from this and RKP that you ought
not assert p unless your belief that you know p constitutes knowledge. Since
not everything you know is something you are in a position to know that you
know, a problem arises, which is that knowledge of p’s truth is not invariably
going to ensure that you have sufficient warrant for asserting p. However, the
view that knowledge of p’s truth is sufficient for having epistemic warrant for
asserting p is surely more plausible than the view that knowledge of p’s truth
is necessary for having that warrant. So, it seems you ought not accept both
RKP and the knowledge account of assertion. It seems the easiest way to sort
out these messes is to deny RKP.
Earlier I suggested that if you combined RKP with Fault4 , you had to deny
that Gettier cases were possible. That seems pretty costly. You can avoid
paying that cost if you deny Fault4 , but by denying Fault4 you do not avoid
all the difficulties caused by Gettier cases. It seems that if RKP is true, there
is a prima facie reason for anyone who fails to know p to refrain from reasoning
from p. Now, if your belief about p is mistaken, I can see that the consequences
of acting on the mistaken belief might be terrible. Because of this, we might be
inclined to say that your having acted on p was wrongful. If your belief about p
is unreasonably held, I can see how your acting on p might manifest the kinds of
bad motives or intentions that show that you can be faulted for having acted on
p. It is not hard to see the normative significance of negligence or recklessness.
It is not hard to see why someone might think that RKP rightly says that you
ought not act from unreasonably held beliefs or mistaken beliefs.
What is hard to grasp is the idea that there are considerations beyond those
that have to do with the accuracy of your beliefs and the reasonableness of
holding those beliefs that have an additional kind of normative significance. If
Audrey hands Coop the ten dollars she owes him, her bills are genuine, and she
has no reason to think anything is amiss, precisely what is it that was wrong
with her acting from the belief that by handing that bill over she’d repay her
debt? I cannot fathom it. From Coop’s point of view, it is not as if he would
care whether she repaid the debt knowingly at home or unknowingly in the land
of fake bills. If I imagine myself as an outside observer who knows that Audrey
does not know she will repay the debt merely because she is trying to repay
that debt in the land of fake bills, I am not at all inclined to think that the
advisory judgment ‘You should not act from the assumption that you will repay
that debt’ is correct. It seems that the very same examples that show that we
cannot identify knowledge with true beliefs we are justified in holding show that
knowledge of p’s truth is not needed to properly rely on p in practical reasoning.
Surely we have all we need to rightly reason from p if our belief about p is true
and not unreasonably held.
We have seen reasons to think RKP must be wrong, so we have good reason
to be suspicious of arguments for RKP. The first argument we are offered draws
heavily on ordinary usage. Hawthorne and Stanley write:
Suppose . . . Hannah and Sarah are trying to find a restaurant, at
which they have time-limited reservations. Instead of asking some-
one for directions, Hannah goes on her hunch that the restaurant is
down a street on the left. After walking for some amount of time,
it becomes quite clear that they went down the wrong street. A
natural way for Sarah to point out that Hannah made the wrong
decision is to say, “You shouldn’t have gone down this street, since
you didn’t know that the restaurant was here”.6
It is natural enough for Sarah to say this and for us to construe this as criticism of
Hannah. The case provides little support for RKP however, because Hannah’s
belief fails to constitute knowledge for a variety of reasons (e.g., her belief is
really no better than a hunch and her hunch is mistaken). To test RKP properly,
it seems we should consider three variants on the example:
Hannah and Sarah are trying to find a restaurant, at which they have
time-limited reservations. Instead of asking someone for directions,
Hannah relies on her usually impeccable memory and decides to
go left. She has been eating at this restaurant regularly for years.
After walking for some amount of time, it becomes quite clear that
they went down the wrong street. Unbeknownst to Hannah, the
restaurant had caught fire three days ago and was working from an
alternative location two blocks away (Restaurant1 ).
Hannah and Sarah are trying to find a restaurant, at which they have
time-limited reservations. Instead of asking someone for directions,
Hannah goes on her hunch that the restaurant is down a street on the
left. They find the restaurant just in time when Hannah declares,
“That was lucky, I was just guessing that it would be this way.”
(Restaurant2 )
Hannah and Sarah are trying to find a restaurant, at which they
have time-limited reservations. Instead of asking someone for direc-
tions, Instead of asking someone for directions, Hannah relies on her
usually impeccable memory and decides to go left. She has been
eating at this restaurant regularly for years. After walking for some
amount of time, it becomes quite clear that they went down the
wrong street. Unbeknownst to Hannah, owners of a rival restaurant
managed to trick all the local papers and news outfits into running
a story according to which the restaurant burnt down and would be
serving at an alternative location (Restaurant3 ).
In Restaurant1-3 , Hannah does not know that the restaurant is to the left.
However, it is only in Restaurant1 and Restaurant2 that it seems natural for
Sarah to say, “You should not have gone down this street, since you did not know
that the restaurant was here”. The defender of RKP cannot say that the reason
it seems unnatural to say this in Restaurant3 is that Hannah is blameless in that
example, because she is blameless in Restaurant1 and yet Sarah’s remark seems
natural. The defender of RKP cannot say that the reason it seems unnatural
to say this in Restaurant3 is that there is no reason for Sarah to say this so
long as they arrived at the restaurant because it is natural for Sarah to say this
in Restaurant2 . I cannot see how either the original restaurant case or these
modified versions provide any more support to RKP than they to do the thesis
that knowledge is merely a matter of, say, true beliefs that are not baselessly
held.
There is a perfectly reasonable explanation for this pattern that does not
assume RKP. We often use “knows” loosely as if it served to pick out true beliefs
or not horribly unreasonable true beliefs.7 Unfaithful lovers will speak this way
when they think someone “knows” of their secret rendezvous. Thieves speak this
way of cops who “know” about the heist. In conversational contexts like this, the
propriety of using “knows” does not depend upon what is known. That Sarah’s
remarks only seem proper in Restaurant1 and Restaurant2 suggests that we are
dealing with conversational contexts like this. Observing how “knows” functions
in such contexts provides no real support for RKP. At least, no more support
than it does for the hopeless view that knowledge is merely a matter of firmly
held true belief.
7.3 Justification and Action

Any condition that distinguishes knowledge from justified belief is normatively
insignificant. If this is correct, justification is the norm of practical reason, not
7 See Goldman 2002, pp. 183.
knowledge:
Where your choice is a p-dependent choice, it is appropriate to treat
the proposition that p as a reason for action iff you justifiably believe
that p (RJP).
This view or views similar to it have been defended by a number of people as
an alternative to the knowledge account defended by Hawthorne and Stanley.
Those who have defended the view have worked with an orthodox account of
justification in the hopes of shedding some light on what it is to properly treat
something as a reason for action. We shall see below that RJP is true only if we
reject the orthodox view that insists that there can be false, justified beliefs. If
the arguments below are successful, the project of trying to use the concept of
justification to make sense of when it is proper to treat something as a reason
for action is a failure. We should work from the other direction. What it is for
a belief to be justified is for a belief to be fit for the purposes of deliberation.
If we can work out an account of when it is improper to treat something as a
reason for action, we can say what is required to believe with justification.
7.3.1 The Incoherence and the Subtraction Arguments

Suppose there is a difference between knowing p and justifiably believing p.
According to RKP, it is proper for you to treat p as a reason for action if you
know p and your choice is p-dependent, but improper to treat p as a reason if
you merely justifiably believe p even if you know your choice is p-dependent.
Suppose you know you ought to φ if p but ought to ψ rather than φ if ∼p. You
justifiably believe p, know that you ought to φ if p, but it seems RKP says that
you should not treat p as a reason for deciding to φ since you do not know p.
Suppose p is true. Suppose you reason from your belief in p, your belief that
you ought to φ if p, to the conclusion that you should φ. The conclusion is a
belief, not an action or intention. It is a belief with a practical subject matter,
but a belief nevertheless. Is it wrong to treat p as a reason to believe that? Of
course not, you might think, this is what J-Closure tells us. So, it seems that
taken in combination, RKP and J-Closure tell us that you should not treat p
as a reason for φ-ing even if you treat p as a reason for believing you should φ
and there is nothing at all wrong with doing that.
This seems incoherent. If you justifiably believe p, you can justifiably include
your belief about p in deliberation when you know p bears on the proper outcome
of that deliberation. According to RKP, if you can properly treat p as a reason
for action in the course of deliberating about whether to φ and either intending
to φ or φ-ing, you know p. According to J-Closure, you can properly treat p as
a reason for beliefs about whether to φ if you justifiably believe p and know that
you should φ if p. If justified belief sufficed for knowledge, there would be no
problem here, but justified belief is not knowledge. So, if you think that having
the right to treat p as a reason for judging that you should φ comes with the
right to treat p as a reason for intending to φ or for φ-ing, it seems you either
need to say that justified belief suffices for knowledge, deny J-Closure, or deny
RKP. Of the options, denying RKP seems best.
Three claims form an inconsistent triad. The first is that justified belief is
not sufficient for knowledge.8 The second is that justified belief is sufficient
for it to be epistemically proper to treat something as a reason for action.
The third is that you know p if it is epistemically proper for you to treat p
as a reason. It seems odd, to me, to acknowledge that justified belief both
falls short of knowledge and falls short of giving you the right to rely on p
in your deliberations, especially once we are clear that the kind of propriety,
right, permissibility, etc... at issue is epistemic. The incoherence argument
urges those who think justified belief falls short of knowledge to reject the idea
that knowledge rather than justified belief is the epistemic norm for practical
reasoning. It is strikingly similar to the incoherence argument from the previous
chapter that purported to show that it is a mistake to say that knowledge rather
than justified belief is the norm of assertion. If knowledge is justified belief, it
makes sense to say that knowledge determines a normative standard for belief,
assertion, action, etc..., but Gettier cases seem to show that the antecedent of
that conditional is false.
Hawthorne and Stanley float the suggestion that there is a principled link
between knowledge and reasons for belief that is akin to their principle linking
knowledge and reasons for action.9 This would be fine if they were willing to
go further and say that since you have to know p to properly treat p as a
reason for belief, you have to know p to justifiably believe p as Sutton does. I
see no indication that they are willing to say that knowledge is necessary for
justification and Gettier suggests that they would be wrong to do otherwise.
So, their view seems susceptible to the incoherence objection. Sutton avoids it,
but only by offering us an account of justification that is hard on intuition and
forces us to say awkward things about fake barn cases.
The incoherence argument is similar in some ways to an argument of Fantl
and McGrath’s. In support of something in the neighborhood of RJP, they run
the subtraction argument:
(1) If you know p, it is permissible for you to treat p as a reason for
action or for belief.10
(2) Holding fixed knowledge-level justification while subtracting from
knowledge any combination of truth, belief, and being unGet-
tiered makes no difference as to whether it is permissible to treat
p as a reason for action.
(C) If p is knowledge-level justified, it is permissible to treat p as a
reason for action or for belief.11
The key difference between their argument and mine is that they work with an
orthodox conception of justification on which you can have sufficient justification
8 Sutton 2005 and Unger 1975 deny this.
10 For a defense, see Fantl and McGrath 2009, pp. 72.
11 Fantl and McGrath 2009, pp. 99.
for believing p even if you do not happen to believe p, your belief is Gettiered,
or p is not true. In my argument, the notion of justification operative is a
purely deontic notion. A justified belief is a belief you can hold while fulfilling
your epistemic duties. In other words, the justified belief is the permissibly held
belief, whatever that happens to come to. I made no substantive assumptions
about what justification is, such assumptions are defended on the grounds that
we need them to understand how justified beliefs do what justified beliefs are
supposed to do. We have too tenuous a grip on the notion of justification to use
that concept to cash out the permissible use of a belief in deliberation and so
would be wary of using some independent notion of justification to argue that,
say, the truth of p does not matter when it comes to properly treating p as a
reason for action.
Below, I shall argue that the subtraction argument is unsound. While it
does not matter whether your belief is Gettiered when it comes to determining
whether you have sufficient warrant to treat what you believe as a reason for
action, the truth matters. Since the truth matters to determining whether it is
proper to treat p as a reason for action, beliefs must be true to be justifiably
held.
7.3.2 Segregationism
Fantl and McGrath defend RJP on the grounds of the subtraction argument as
well as an argument similar to my incoherence argument. As they see it, the
right to believe p comes with the right to treat p as a reason for action or belief.
They do not say, however, that p must itself be a reason if it is to be justifiably
treated as one. So, while we agree that RJP is true, we still find plenty to
disagree about. One of the primary points of disagreement between us has to
do with our understanding of the relation between the normative standing of
normative judgments and the normative standing of the intentions and actions
they rationalize. For reasons that will emerge, they defend Segregationism and
I defend Unificationism:
The demands of practical and theoretical reason can diverge in such
a way that it can be practically improper to treat p as a reason for
action even if it is epistemically proper to treat p as a reason for
action (Segregationism).
The demands of practical and theoretical reason cannot diverge and
so if it is epistemically proper to treat p as a reason for action, it is
practically proper to do so as well (Unificationism).
It might appear at first that they defend Unificationism because they say:
any proposition that is warranted enough to be a reason you have
for belief is also warranted enough to be a reason you have for action
or anything else. We can see the plausibility of the Unity thesis by
reflecting on our habits of deliberation. When trying to determine
what is true ... we draw conclusions from the reasons we have. The
same goes for trying to decide what to do ... We bring reasons into
our reasoning knowing that we might draw all sorts of conclusions
from them along the way, some practical and some theoretical.12
Appearances can be misleading. While they think you have sufficient epistemic
warrant to treat what you justifiably believe as a reason for action, belief, or
anything else, they think there are counterexamples to the stronger claim that
you also have sufficient practical warrant to treat what you justifiably believe
as a reason for action. Because of this, they cannot link the normative standing
of normative judgment to action in the way the unificationists do.
In the course of explaining why they think that you can justifiably believe
p even if p is not a genuine reason, they ask us to consider this example:
Coop tries to make two gin and tonics. He uses the last of the gin on
the first. He grabs a new bottle to make the second, but accidentally
mixes Audrey a Bernard (i.e., a petrol with tonic and fresh lime). Fill
in the details however you like so that Coop is perfectly reasonable in
thinking that he has just made two gin and tonics. He gives Audrey
her Bernard believing it to be a gin and tonic. She drinks and she
becomes violently ill. This date is not going well, Coop nearly killed
Audrey (Gin and Tonic).
For reasons already discussed, I would not say that Coop’s belief that the stuff
he gave to Audrey was a gin and tonic was justified and so I need not agree with
them in saying that it was proper for Coop to treat that this is a gin and tonic
as a reason to give Audrey the poisonous concoction. Myself, I think he should
not have given her the drink and so should not have thought he should, should
not have thought that the stuff was gin, etc... Predictably, they disagree. They
say that Coop’s action was perfectly justified:
Notice if we asked the unlucky fellow why he did such a thing, he
might reply with indignation: ’Well, it was the perfectly rational
thing to do; I had every reason to think the glass contained gin; why
in the world should I think that someone would be going around
putting petrol in the gin bottles!?’ Here the unlucky subject, in our
view, is not providing an excuse for his action ...; he is defending it
as the action that made the most sense for him to do and the propo-
sition that made most sense to treat as a reason. He is providing a
justification, not an excuse.13
I think this is all wrong, of course. The action could not be excused unless it
made sense for Coop to do what he did and he could have reasonably taken
himself to have acted rightly. This is something to return to later, but notice
that if Fantl and McGrath reject Unificationism it is not because of the way
that Unificationism treats cases of error or mistaken non-normative belief.
If they reject Unificationism and say that the epistemic warrant someone has
to treat p as a reason for action does not always come with a further practical
warrant to treat p as a reason for action, there must be cases where the epistemic
and practical warrant come apart. In conversation, they said that they did
not think cases of mistaken non-normative belief were the right sort of case to
provide a counterexample. They look to cases of mistaken normative belief to
motivate their segregationist view:
Coop has a prima facie duty to be in Austin and a prima facie duty
to be in Boston. He cannot be in both places. He knows of both
duties and their grounds. He thinks there is a weightier reason to be
in Austin and, let us assume, that this is something he is reasonable
to believe. There are, however, weightier reasons to be in Boston.
So, that is where he ought to be. (Austin and Boston).
Suppose Coop acts on his reasonable but mistaken belief about where he ought
to be. Fantl and McGrath are waiting for him in Boston and they accuse him of
wrongdoing. I can imagine Coop saying in response, “Well, it was the perfectly
rational thing to do; I had every reason to think I ought to be here in Boston
rather than Austin for as you both agree, this was the rational thing to believe.
Here I am, the unlucky subject, and I am not providing an excuse for an action.
I’m defending it as the thing it made most sense for me to do given not just what
I believe, but what I ought to believe.” Myself, I think that if Coop’s defense
works in Gin and Tonic, it works just as well in Austin and Boston. I think it
works as a way of defending himself from criticism (i.e., as an excuse), but think
it succeeds in neither case as a defense of what Coop did (i.e., a justification). I
might be wrong about this, but I doubt their defense succeeds in one case rather
than the other.
Perhaps they ought to be convinced by the defenses they offered on Coop’s
behalf initially in defending his behavior in Gin and Tonic. Perhaps they should
say that cases of mistaken belief are not counterexamples to the Unificationist
view. If they reject the Segregationist view, they can either opt for a view that
classifies Coop’s actions in both cases as justified or says that he didn’t act
with justification in either case. My preference is for the latter view, but such
a preference needs to be defended.
7.3.3 Unificationism
There has to be some connection between the normative status of the beliefs
that rationalize action and the actions and intentions such beliefs rationalize.
If you judge that you ought to φ and ψ instead knowing that you were ψ-ing
rather than φ-ing, it seems that you would be deeply irrational in acting in the
way that you have. It seems plausible that you should not be irrational in this
way. So, if there is some normative relation between beliefs that rationalize
actions and the actions rationalized, what is it? Could it be this?
If you believe you ought to φ, you ought to φ (NSO).
I think not. NSO allows for factual detachment. For counterexamples to NSO,
see any movie about Nazis.
If “ought” takes wide-scope, we can block factual detachment:
You ought to see to it that: if you believe you ought to φ, you φ
(WSO).
You cannot find counterexamples to WSO in movies about Nazis. If someone
believes he should φ but he should not believe this, WSO does not allow us
to detach the conclusion that he should φ. The trouble with WSO is not that
it allows factual detachment, it is that it is not at all clear that it represents
the normative relation between normative belief and the actions such beliefs
rationalize. Suppose that, in some sense, someone ought to believe she ought to
φ. If she ought to believe and does believe, can we detach the conclusion that
she ought to φ in accordance with her judgment? If she ought not φ and so
ought not φ in accordance with her judgment, can we say that she ought not
believe she ought to φ? It is unclear.
Those who like WSO might say that once you determine what you ought
to do and then you do it accordingly, this is not some fallacy of practical rea-
son. This is precisely how reasoning should go. As such, there has to be some
principle that allows for a kind of detachment. You cannot rightly detach the
conclusion that you ought to φ simply given that you believe you ought to φ, but
surely you can if you believe it and you ought to. The trouble here is making
sense of what happens when an agent’s reasoning can be represented as follows:
(1) She ought to believe she ought to φ and she does.
(2) She ought to see to it that: if she believes she ought to φ, she
φ’s. She does.
(3) She φ’s just as she ought to.
It looks like this kind of reasoning is good because it seems fine to say (6) follows
from (4) and (5):
(4) She ought to φ and she does.
(5) She ought to see to it that she ψ’s if she φ’s.
(6) She ought to ψ.
The trouble is that this seems different from:
(7) She ought to believe p and she does.
(8) She ought to see to it that: if she believes p, she φ’s.
(9) She ought to φ.
If (7) is true, it must be because there are certain kinds of epistemic reasons
by virtue of which the agent is obligated to believe something. If (7) and (8)
entail (9), how are we supposed to interpret (9)? She cannot have an epistemic
obligation to φ. There is no such thing as an epistemic obligation to act. If we
want to get a practical obligation out, we can try to put a practical obligation
in, but I have no idea what a practical obligation to believe would be. Whatever
they are supposed to be, they have nothing to do with the sorts of cases that
interest us. The move from a belief about what you ought to do to the action it
rationalizes is not some exotic thing that happens when, say, someone gives you
practical reason to induce a belief that creates a practical obligation to make
yourself into a believer.
The principles that capture the spirit of the unificationist view have to allow
for mixed deontic detachment. That is to say, they have to allow us to detach
a practical obligation from statements about epistemic obligation and some
linking premise that tells us how the epistemic and practical obligations mesh.
And, it has to allow us to detach an epistemic obligation from statements about
practical obligation and the same linking premise that tells us how the practical
and epistemic obligations mesh. Just so that we are clear, Unificationism should
accept the following conditionals that tell us when mixed deontic detachment is
allowed:
If you believe you oughtP to φ and oughtE to believe you oughtP to
φ, you oughtP to φ (MDD1 ).
If you oughtP not φ, you oughtE not believe you oughtP to φ (MDD2 ).14
The segregationist denies that these conditionals are true. Fantl and McGrath
denied that Gin and Tonic were counterexamples to MDD1 and MDD2 , but
they thought Austin and Boston constituted a counterexample to the claim. To
take one of these cases to be a counterexample requires cutting the link between
the reasonable and the permitted that suggested that the plea they offered on
Coop’s behalf would not be convincing in Gin and Tonic. So, one rationale for
MDD1 and MDD2 might be the thought that if the reasonable is the mark of
the permissible in either the practical or theoretical domain, it is the mark of
the permissible in both. Those who take it to be the mark of the permissible
in both might be attracted to these principles since it seems that denying them
requires allowing that it is possible that the thing you ought to do is to act
against your own impeccable normative judgment. Such a thing could never
be reasonable. If the reasonable is the mark of the permissible, you would be
permitted to act on your judgment. The putative counterexamples would be
defused.
Of course, some of us do not think that the reasonable is the mark of the
permissible. Those of us who deny that the reasonable is the mark of the
permissible have to defend MDD1 and MDD2 on different grounds. Some of
this will involve deflecting objections. One objection that I have heard that has
little force has it that MDD1 and MDD2 clash with the idea that the reasons that
bear on whether to believe or act either depend upon our perspective or have to
pass through some sort of “epistemic filter”. Defending MDD1 and MDD2 does
14 To say that you ought to φ is to say that in light of the practical reasons, you ought to
P
φ. To say that you oughtE to believe is to say that in light of the practical reasons, you ought
to believe.
not force you to deny that the reasons that bear on whether to act or believe
depend, in some sense, upon our perspective. It might seem that way because
MDD2 lets you say that you ought not believe something simply because there
is a practical obligation you are under not to act on that belief. Notice, however,
that all MDD2 says is that the reasons that oblige you not to act must have
passed through an epistemic filter if that is what reasons must do to oblige you.
To deny MDD2 it seems you would have to say that the reasons that determine
what to do do not thereby determine what to believe, and that would seem to
require reasons that determine your obligations that are not available to you.
So, if you think reasons depend upon perspectives, you should probably like
MDD1 and MDD2 .
Of course, some of us deny that reasons have to pass through an epistemic
filter to determine what you ought to do or believe. Some of us deny that the
reasons that apply to you depend upon your perspective. I think we have some
sense of how we can respond to putative counterexamples and know better than
to think that MDD1 and MDD2 force you to deny that reasons have to be avail-
able to you. I know of no other objections to MDD1 and MDD2 , but the reader
might harbor doubts. Let me offer two points in support of Unificationism.
First, think about the incoherence argument for RJP and against RKP.
Assuming that we do not need to know p to treat p as a reason for action,
I said that there was something strange to the view that says that you can
properly treat p as a reason in deliberation if that deliberation is concerned
with determining what to believe but not if it is concerned with determining
what to do. I think similar worries arise for Segregationism. With apologies to
Judith Thomson, please consider an example:
Plum: Mustard, I have a problem. We’re at war with a villainous
country called Bad, and my superiors have ordered me to drop some
bombs at Placetown in Bad. Now there is a munitions factory at
Placetown, but there is a children hospital there too. Some people
tell me that I should drop the bombs to help with the war effort
but some tell me that we should avoid killing innocents. I am so
confused, I just do not know what to think. Should I believe this is
a necessary evil or what?
Mustard: Look, Plum, given what you have said, it is clear that you
should appreciate that dropping the bombs is a necessary evil.
Weeks later:
White: Plum, you really should not have dropped those bombs. You
killed scores of children in that attack on the munitions factory.
Plum: Mustard, can I get a little help here?
Mustard: What’s the problem?
White: I told Plum that she should not have dropped the bombs.
Mustard: She’s right, you shouldn’t have.
Plum: But you told me that I should think of dropping the bombs
as a necessary evil.
Mustard: That’s right, and I stick by that. That is what you should
have believed. Isn’t that right, White?
White: Certainly, that’s just what you should have believed. But,
as I’m sure Mustard would agree, what you should have done is not
dropped the bombs. I’m really more concerned with action, Mustard
is more concerned with belief. We agreed to not disagree. I think
Mustard has the epistemology right and we both think I have the
ethics right.
Mustard: Precisely.15
I have little sympathy for this. I’m troubled by the thought that the epistemol-
ogist and the ethicist can agree not to disagree by denying that the oughts that
concern one have any bearing on the oughts that concern the others. If, like
me, you think that if someone ought to think of something as a necessary evil,
she should gnash her teeth and do what she thinks she must, you probably have
some sympathy for Unificationism.
Let me offer a second rationale for MDD1 and MDD2 . If either MDD1 or
MDD2 is false, it has to be possible for situations to arise where there is a decisive
case against acting that does not constitute a decisive case against believing you
should act in that way. In other words, there is sufficient epistemic reason to
believe you ought to φ but sufficient practical reason not to φ. Suppose, then, we
imagine two cases. In the first, the subject knows she ought to φ and knows that
the reasons by virtue of which she ought to φ are the reasons by virtue of which
she should believe she should φ. In the bad, the subject believes mistakenly that
she should φ. Really, she should not φ. Perhaps what the subject should believe
in the good case and bad is determined by the evidence, or her perspective, or
how things seem. What the subject should do, we might say, depends upon
the facts. Since the facts do not fit the beliefs in the bad case, we need a bad
case to cause trouble for MDD1 and MDD2 . At this point, however, it seems
the segregationists have to explain why one sort of reason depends upon things
available to us when the other does not. I take it that such differences cannot be
brute.16 The explanation, however, either has to appeal to something about the
reasonness of reasons, the epistemicness of epistemic reasons, or the practicality
of practical reasons. The explanation cannot be grounded in something having
to do with the concept of a reason, both epistemic and practical reasons are
reasons. We cannot say that the epistemicness of epistemic reasons or the
practicality of practical reasons will provide us with the explanation we seek.
The difference between the epistemic and the practical is that the former is
15 Inspired by example taken from Thomson 1991. She used her example to tell us something
about intention and permissibility, not something about the relation between the normative
standing of belief and action.
16 This expands on an argument from Gibbons Forthcoming. He thinks “Don’t be an idiot”
is a categorical imperative and that anyone who acts against their own justified judgment
about what to do is an idiot.
concerned with truth and knowledge whereas the latter is concerned with the
good. That difference does not help us see why one sort of reason applies to
us only if it is available to us when the other does not. So, since it cannot be
a brute fact that the reasons differ this way and it seems there is nothing that
could explain why the reasons bearing on belief and action would differ in this
way, perhaps the alleged fact about reasons is no fact at all. Either both sorts
of reason depend upon your perspective rather than the facts, depends upon
the facts rather than your perspective, or depends upon both.
7.4 From Unificationism to Externalism

In this section, I shall argue that Unificationism requires a kind of externalism.
Unificationists should deny that there can be false, justified beliefs. The cases of
mistaken belief and ignorance we shall consider below are better conceptualized
as cases of excusable wrongdoing rather than regrettable right action. Given
MDD1 and MDD2 , we shall see that if an agent who reasonably judges she should
φ nevertheless fails to meet her practical obligations, she has failed to meet her
epistemic obligations as well. For among those obligations is the obligation not
to include beliefs in deliberation that pass off non-reasons as if they are reasons.
Mistaken normative judgments ought to be excluded from deliberation and facts
about what you should do depend (in part) upon contingent facts external to
you, not just facts about you and your own mental states.
Let’s begin with a story:
Mustard is behind in his payments to Green, the loan shark. Green
gave Mustard a severe beating last week and a warning that if he
missed another payment, Green would kill him. The payment is
due today and Mustard does not have the money. He borrows a
revolver and hangs out at Peacock’s restaurant hoping that Green
will leave him alone in public. He is shocked when he sees a man
with a menacing look he takes to be Green come in and walk straight
towards him. Mustard says ‘I won’t let you get me Green!’ and he
pulls out his revolver and takes aim (Loan Shark).17
Suppose the story continues as follows:
Peacock has a pipe. She knows that the man Mustard takes to be
Green is really Green’s twin brother. She knows that while he might
look dangerous, he is a threat to no one. She knows that to stop
Mustard from firing at Green’s twin, she will need to club him with
her pipe. She does so, intervening on behalf of Green’s brother (Loan
Shark A).
Concerning Loan Shark A, it seems natural to say this:
17 The example is taken from Robinson 1996. The argument is similar to his in some respects,
but we shall see that it needs supplementation and modification.

(1) Peacock would act rightly if she intervened on behalf of the man
Mustard intends to shoot (i.e., on behalf of Green’s twin).
The story could have unfolded differently. We could have changed our cast of
characters a bit:
Peacock knows that the man Mustard takes to be Green is really
Green, not Green’s twin. As Green approaches, Mustard produces
the revolver, pulls the trigger, but nothing happens. He is out of
bullets. Peacock intervenes and knocks Green unconscious with her
pipe (Loan Shark B).
Concerning Loan Shark B, it seems intuitive to say:
(2) Peacock would only act rightly if she decided to intervene on
behalf of Mustard.
What explains these intuitions? The reason that the range of permissible options
open to Peacock differs in these stories is that Mustard loses the right to non-
interference in Loan Shark A but not in Loan Shark B. I think that this tells
us something about the deontic status of the acts Mustard intends to perform
in Loan Shark A and B. This can be contested, of course, and we shall look at
the way that it has been contested momentarily.
Suppose for now that I am right. Suppose that Mustard is permitted to use
force in one case but not the other. What follows? Consider:
(3) It is consistent with views that deny FactivityJ that Mustard’s
beliefs are justified in both cases.
(4) So, views that deny FactivityJ should allow that if we took the
trouble to fill out the details of the case, we could fill them out
in such a way that Mustard justifiably judges that he should use
force to defend himself from the man approaching in both Loan
Shark A and B.
(5) Given Unificationism, however, that would mean that Mustard
is permitted to use force in Loan Shark A and B.
(6) But, this simply is not so. In one case, he is permitted to use
force and in the other he is not.
(7) Thus, given Unificationism, views that deny FactivityJ have to
say that reasonably held mistaken beliefs do not merely excuse,
they obviate the need to justify acting against undefeated rea-
sons.
(8) Thus, given Unificationism, views that deny FactivityJ cannot
do justice to our moral intuitions.
We might agree that there is certainly something bad about shooting someone
who just happens to look just like a loan shark, but how do we decide whether the
shooting would have been bad and impermissible or merely regrettable? From
my own experience, the intuitions that favor internalist views are strongest when
we ask what someone should do or believe and do not let the story continue from
there.
Externalist intuitions start to get their grip on us when the let the story
continue and ask what should be done in the wake of some untoward chain of
events. Think about the injuries that Green suffers in Loan Shark B. He gets
a nasty bump on the head and a massive headache waiting for him if he ever
wakes up. Suppose we add in some detail to Loan Shark A. In Loan Shark A,
White sees Peacock swinging her pipe at Mustard. White believes that Green’s
twin is Green and believes that Peacock must be helping the loan shark rub
Mustard out. White grabs a pipe and swings it at Peacock on the reasonable
but mistaken belief that she is helping to defend Mustard from a loan shark and
an accomplice. White connects and knocks Peacock unconscious, but only after
Peacock connects and knocks Mustard unconscious. The story is complicated,
but at the end of the story we have two unconscious subjects. Mustard was
knocked unconscious by Peacock because she knew that Mustard was going
to shoot an innocent person if she did not intervene. White knocked Peacock
unconscious just as she was striking Mustard. After the police come, White sees
Green and Peacock begin to stir. She has enough pain-reliever to help one of
these subjects but not enough to help both. It seems intuitive to say:
(9) Given just enough pain-reliever to help one, White ought to assist
Peacock rather than Mustard.
If this is right, why is her duty to Peacock stronger than her duty to Mustard?
If her duties to Peacock and Mustard were both just duties of beneficence, it
seems that given that I’ve stipulated that she can help both equally and both are
equally badly off, we would be at a loss to explain why (9) is true. If, however,
her duty to Peacock is a duty of reparation, it is easy to see why (9) is true.
However, duties of reparation are duties to respond to previous wrongs. I don’t
see that a view that denies that the justification of action depends upon certain
facts external to the agent can easily explain how it could be that White’s
actions were wrongful. White, we might assume, was perfectly reasonable in
her beliefs and had her beliefs been correct we might all agree that she did just
the thing that she should have done. Unfortunately for Peacock, however, her
beliefs were not correct and if we say that she acted impermissibly we can say
that the intuitions that support (1), (2), and (9) will cause trouble for views
that deny that the truth of a belief matters to whether it can properly figure in
deliberation.
In conversation, some have suggested that the intuition that White owes
something to Peacock rather than Mustard has everything to do with the fact
that there is a causal chain that connects White to Peacock.18 Perhaps the
thought is that our intuition is sensitive to this fact and this fact about the source
of our intuition helps explain it away and undercut the support it is supposed to
provide for (9). While someone could say this, I think it does little to blunt the
18 Both Robert Howell and Sarah Wright suggested a response along these lines.
force of the intuition. We might easily imagine that White is ambidextrous and
wields two clubs. Striking both someone engaged in wrongdoing and someone
trying to help stop wrongdoing, we can imagine there are causal chains leading
from one agent to two bumps. We can ask again which of these bumps White
should do something about first, which bump she has stronger reason to respond
to, etc... If your intuitions are anything like mine, they suggest that White has
a stronger duty to the party trying to help rather than the party up to no good.
Now it seems we have two sorts of intuition putting pressure on the unifi-
cationist to accept FactivityJ . There are intuitions about the justified use of
force in intervention and intuitions about reparative duties owed in the wake
of an action. With the argument now before us, we can consider a number of
ways in which someone might try to resist the conclusion. Some might contest
the intuitions, some might contest their significance, and I suspect many will
contest both. Below, I do my best to dispel these various doubts.
7.4.1 Conflicts of Justification

One way to resist the argument is to push back against the idea that intuitions
about the permissible use of force in intervention are a way of determining
something about the justificatory status of an agent’s actions. It might seem
that the argument rested on the following assumption about the right to non-
interference:
It is permissible for an agent to use force to prevent another agent
from acting only if that agent’s intended course of action was imper-
missible (Non-Interference).19
Unfortunately, it seems that Non-Interference is false. Even if the only way for
two agents to resolve some conflict is by the use of force, it is hardly obvious
that at least one of these agents had intended to something impermissible even
if we include using force to resolve the conflict. If this is so, it is hard to see
how someone could argue from the observation that an agent lost the right to
non-interference by intending to φ to the conclusion that it would have been
wrong for her to φ.
Various examples make it difficult to defend the view that justifications
cannot conflict. Suppose Plum’s niece has been kidnapped and hypnotized. She
has been sent to steal the medication from Mustard. Without this medication,
Mustard will surely die. Mustard is quite frail and he can only defend himself
from Plum’s niece by shooting her. Plum’s niece, we might imagine, would not
survive the shooting and Plum feels obliged to protect her niece. Must we really
say that morality obliges either Plum to do nothing and so do nothing to protect
her innocent niece if her intervention means Mustard’s death? Must we really
say that morality obliges Mustard to do nothing to protect himself from this
19 I believe Robinson 1996 appeals to a principle in the neighbrhood of this one to argue that
cases of mistaken belief are cases of excusable wrongdoing. He defends the deeds theory of
justification on which the justification of the action depends upon its objective characteristics
and not (just) the reasons the agent took to count in favor of its performance.
little innocent aggressor? Or, less realistically, suppose there is only one life
preserver left and both Green and White need it to survive.20 Does morality
really require them to find a coin to flip and condemn them if they wrestle for
it? Morality might condemn one of them for kicking, biting, or gouging during
the match and would surely condemn Green if he threw sand in White’s eyes,
but I do not think it would condemn White for being quicker and more agile
if she used agility and speed to get the last remaining life preserver in a fair
match.
What to do? While I agree that Non-Interference is questionable, cases that
call it into question might give us a clue as to how to present the argument for
FactivityJ . Consider:
Cooper intends to bomb a munitions factory. Destroying the fac-
tory is an important step towards winning a just war. If the bombs
hit the factory, the explosion will destroy the apartments that abut
the factory. Audrey is among the non-combatants who live in those
apartments whose deaths count as acceptable losses. Audrey re-
alizes that she can defend herself and her family by manning an
anti-aircraft gun and firing at Cooper’s plane. Someone could say
that Audrey’s decision to try to stop Coop from carrying out his
mission was justified and that Coop’s decision to bomb the factory
was justified (Anti-Aircraft).
According to Non-Interference, Audrey’s intervention would be justified only
if Coop’s intended course of action was not justified. But that seems like a
mistake. It is not wrong for those non-combatants whose deaths are permitted
by a just war theory to take arms against just actors who would cause their
deaths unintentionally if they weren’t stopped by aggressive means. Surely
if Audrey decides to allow herself to be killed by Coop’s actions because she
hoped that his side would win, she has gone beyond the call of duty. Surely
circumstances can arise in which Coop could justifiably act in a way that would
predictably cause Audrey’s death.
Notice that in this example, Audrey has a range of permissible options avail-
able to her. She can intervene on behalf of Coop and die for the cause or she
can intervene to protect herself and her loved ones from Coop. It would be
wrong for her to intervene on behalf of, say, the forces of the despotic tyrant
that is sending planes to intercept Coop. In Loan Shark A and B, there is a
range of permissible options available as well. The agent can rightly do nothing
or rightly intervene on behalf of one party, but they do not have free choice as
to which party to assist. So, perhaps the way to state the challenge is this. The
best explanation as to why agents are permitted to assist different parties in
Loan Shark A and B has to do with the fact that it is wrong to assist someone
engaged in wrongdoing and the parties engaged in wrongdoing differ in these
cases. We can free ourselves from relying on the mistaken thought that you can
20 For a discussion of these kinds of examples (and whether they actually involve conflicting
justifications), see Husak 1999.

never justifiably use force to interfere with someone who is also acting (or trying
to act) with justification. So, we can remain agnostic as to whether there can
be conflicts of justification.
7.4.2 Obligations and Outcomes

In trying to bolster the intuition that the conditions that determine whether it
is permissible to act include facts having to do with the circumstances in which
the agent acts and not (just) facts about the agent’s take on things, I appealed
to intuitions about reparative duties. It is possible, I claimed, for two agents to
be perfectly alike in all respects that matter to the justification of her beliefs
apart from the truth of her belief, for these agents to judge that they ought
to φ, for them to φ accordingly, and for only one of them to owe reparations
for having φ’d . Thus, whether someone acted permissibly or not can depend
upon the facts that determine whether the agent’s reasonably held attitudes
are correct and not just whether they are indeed reasonably held. This kind of
argument can be resisted in two ways. First, it might be possible to account for
the intuition that these reparative duties are owed in an internalist framework.
Second, it is possible to argue that we have no such reparative duties in these
cases.
Herman has described a way that someone could reconcile the Kantian view
that all moral evaluation is concerned with the agent’s will rather than the
states of affairs the agent brings about with the further thought that an agent’s
obligations can depend, in part, upon the results of her actions. Her view is not
that the character of someone’s will is determined by the states of affairs she
brings about. Whatever her general views about moral luck might be, she is
not advocating the view that simply by virtue of having caused some bad state
of affairs, your will must be defective in some way. The Kantian thinks that the
deontic status of an action depends upon the maxims on which an agent acts
rather than the outcomes of having acted on that maxim. In cases where good
intentions go awry, we should focus both on both the maxim on which the agent
acted as well as supplementary maxims of response. We can think of maxims of
response as the measures that the agent will take if things do not go as planned.
So, let us consider her example:
Suppose someone fully intends to return a borrowed clock and has
a maxim of so acting that is adequate to her intentions. On the
occassion of executing the return, however, she trips and the clock
breaks. If the moral assessment of actions is based on the assessment
of the agent’s maxim, the maxim we have to work with in this case is
the maxim of good intentions. And if, as in this case, the maxim of
good intentions is itself without fault, there appears to be no way for
the theory even to register (no less assess) the failure of execution:
a failure to bring about what was intended or willed, a failure to
return what was owed.21
21 Herman 1993, pp. 97.
Her suggestion is that we can capture the thought that the agent’s duties do not
end with the breaking of the clock by saying that the agent who is required to
adopt the end is required to take the means adequate to bring that end about,
and that will require adopting further strategies when the initial attempt is not
met with success. Built into the obligation to, say, return the clock or keep a
promise is the obligation to pursue adequate means, and in this way the Kantian
can explain how the moral story continues.
This all seems perfectly sensible, but I wonder if her treatment of this case
really covers the cases we need it to. In her example, the agent had an obligation
to φ, could anticipate that her attempts to φ could fail, knows that she must
pursue some means by which she could φ and so is under a persistent obligation
to φ if her initial attempts failed. What happens, however, if the agent misses
her opportunity to φ? Can she then wash her hands of the situation? Or, what
happens if the agent was not obligated to φ in the first place?
In Herman’s example, you are obligated to return a clock. The necessity of
the end makes it necessary to pursue means effective to that end, and so when
you know that the duty persists and know that your initial attempt fails, you
know that you must either adopt a new end or adopt new means. However,
since you know the end is non-optional, you know that you must adopt new
means. This, I take it, is how the Kantian would look at it. Suppose, however,
that the clock in question cannot be replaced and so the end intended is one
you know now cannot be achieved. Now there is no resultant obligation to
pursue sufficient means to the end because you know there are no such means.
Intuitively, it seems there is still a resultant obligation to make amends the best
you can. But, it seems that this second best thing you must do (whatever that
is) is not a way of discharging your initial obligation. So, how can the Kantian
explain why there is this residual duty to do the next best thing when the initial
end is impossible and so no longer necessary?
This worry suggests all might not be well with the Kantian approach to
reparative duty, but focusing on this sort of case distracts us from the fact that
there is an important difference between my cases (i.e., Loan Shark and Cook)
and Herman’s case. In my cases, my agents were never under any obligation to
perform the acts they performed. There was no obligation to make dinner for
the new neighbor, that was just a thoughtful gesture. There was no obligation
to grab a pipe and protect an innocent person from an armed loan shark, that is
clearly going above and beyond the call of duty. In my cases, the agent adopts
an end she is free not to have adopted in the first place, pursues that end by
means reasonably taken to be sufficient, but then fails. My agents are free to
walk away. If the poisoned dish fell to the floor before it was handed over the
neighbor, there was no necessary end adopted in the first place that requires the
agent to make another. If our heroic individual grabs a pipe, tries to help, but
slips on a wet floor and fails to render assistance, she can just stay down. With
no end these agents must pursue, there is nothing internal to the Kantian story
that could explain the intuition that they must do something about the messes
that they made. The Kantian story is partial at best, and while it is a good
story for the cases it covers, it does nothing to undermine the argument offered.
In Loan Shark and in Cook, an agent comes to have reparative duties that are
duties to right some past wrong and the wrong is not the failure to fulfill some
prior obligation. So the cases do pose a serious threat to the view that moral
evaluation is limited to an evaluation of the quality of the agent’s will. That is
surely part of it, but outcomes also matter.
Or, perhaps they do not. Perhaps there is no way to account for the thought
that an agent who pursues an optional end and takes due care can come to have
reparative duties simply by virtue of harming another. According to Zimmer-
man, none of us have the right not to be harmed by others and we have no right
to be compensated for having been harmed. At most, he says, we have the right
not to be put at risk of harm. So, he rejects the first thesis but accepts the
second:
We have moral rights against others that they not cause us harm
(Harm Thesis).
We have moral rights against others that they not impose risks of
harms on us (Risk Thesis).22
Whether someone is put at risk of harm is determined not by the epistemic
position of the victim, but of the agent who I would allege owes compensation.
If this is right, the problem with my argument was not that I drew the wrong
moral from a perfectly sound intuition. The problem with my argument is
simply that there are not reparative duties owed to a victim in cases like Cook.
Against the claim that someone is owed compensation by those who harm
them, Zimmerman says three things. First, that this leaves some needy parties
(e.g., White) “out in the cold” even if this party is just as deserving of compen-
sation. So, for example, in Cook, we had two parties who were equally deserving
of assistance and it seems strange to him to suggest that one of these parties has
a stronger claim on receiving that assistance. Zimmerman thinks that rights are
correlative with duties:
One party has a moral right against another agent that this agent
φ iff this agent has an obligation to this party to φ (Correlativity
Thesis).23
If there is no duty to the party harmed, they had no right not to be harmed.
If they had no right not to be harmed, there might be duties in the wake of an
action, but not duties to the party that I have claimed in my examples. Since
reparative duties are duties that relate agents to particular parties and so differ
from mere duties of beneficence, it looks like an attack on the Harm Thesis puts
me in an awkward spot. Second, he says that the party that harmed may have
been just as innocent as the party harmed. The significance of this, I take it, is
that it makes no sense to hold one party accountable for making reparations to
another if both are equally innocent. Third, he says that there might be some
further party who is just as much at fault as the party that causes the injury
that is just as deserving to be made to make amends who we know should not
be made to do so. Why not haul them into the picture and make them pay
some reparations?
I do not find these responses altogether convincing. We cannot determine
what an agent’s obligations are by determining whether we think there is in-
dependent reason to think that they deserve to be under these obligations or
made to live up to those obligations. Against the second and third point, no
one deserves to be under a duty of beneficence. Remember, if you have a duty
of beneficence, this can arise without there being any prior relation between you
and the benefactor and so the duty does not require that there is any relation
between you and anyone else by virtue of which you deserve to be on the hook
for their welfare. We are for that often duty bound to assist others at an ex-
pense to ourselves and when we are perfectly innocent in terms of what brought
it about that they need our assistance. Against the first, I think we cannot
rest too much weight on this point. Suppose Mustard had tried to poison Plum
and succeeded in so doing. If White and Plum are equally faultless in finding
themselves poisoned, surely they are equally deserving of assistance, but nobody
would say that Mustard’s obligations to Plum are for that reason not stronger
than the duties he has to those he has not tried to kill. If the first point were
applied consistently, I think it would essentially prevent us from saying that
victims are owed compensation by those who put them at risk of harm for no
good reason just as surely as it would prevent us from saying that victims are
owed compensation for being harmed with no overriding reason to have done
that. The Risk Thesis would be at the same risk as the Harm Thesis. Since it
is uncontroversial that one of these theses is true, this objection cannot succeed
in establishing that it is the Risk Thesis rather than the Harm Thesis that is
true.
7.4.3 Against Internalist Unificationism

There are those who harbor internalist sympathies who like Unificationism and
think that we ought to argue from a more internalist conception of justified
belief than I would defend to a more internalist account of justified action than
I would defend.24 They either do not share the intuitions we discussed above
or they overriding reason to think their view is correct in spite of the intuitions
that support my view. In this section, I want to argue that it is a mistake
to reject FactivityJ if you accept Unificationism. One reason I worry about
such a combination of views is that it would force us to sanction wrongdoing.
At least, it seems to. If the view tries to accomodate the intuition that a
subject is justified in her beliefs when it is reasonable for her to hold them,
24 Gibbons is the chief advocate of such a view. He rejects Supervenience Internalism, but
he also rejects FactivityJ and thinks that you cannot be obligated to do something if you
could not reasonably work out that it is your obligation. He does, however, endorse mixed
deontic detachment. Zimmerman 1996 notes that some subjective views of “ought” have to
deny that “ought” implies “can”, and we shall see that this sort of problem arises for Gibbons’
version of Unificationism.
insofar as subjects with defective moral views can be reasonable in holding their
views because they reason to them carefully and they are based on firmly held
intuitions, the unificationist view would force us to sign off on their actions if we
sign off on the attitudes that rationalize them. This was a worry we discussed
earlier in discussing phenomenal conservatism. That view, we saw, condones all
manner of morally abhorrent behavior. If that was right and that was a worry,
I do not see how that worry does not arise again here.
There is a further objection to unificationist views that deny FactivityJ . To
see what the problem is, let us sketch an account of “ought to believe”. Perhaps
we can say that:
You ought to believe p if you have sufficient evidence, are concerned
to settle the question whether p, and you have given the matter
sufficient reflection (SE1 ).
Let us say that:
You have sufficient evidence if you have precisely the same evidence
as someone who knows p (SE2 ).
There are probably ways of refining these claims, but for our purposes, they
should do.
Now, we need an example to cause trouble for the internalist. Coop gets
in line to buy a snack from the vending machine. He sees that there is an
infant, a puppy, and a kitten trapped inside. He knows the machine is in good
working order because he has seen people using it all morning and knows that
the machine was serviced yesterday. He knows he has just enough change to
save either the infant, the kitten, or the puppy but not enough to save two.
He thinks that puppies are worth more than kittens but thinks that infants are
worth more than either puppies or kittens. So, let us say:
(1) In w1 , Coop believes correctly and on exceptionally good evi-
dence that it is better to save the infant.
dence that he can save the infant.
dence that he can save the infant.
This much, I can stipulate. It is tempting to say that if he reasons from his belief
that it is best to save the infant and his belief that he can save the infant that
he ought to save the infant. Indeed, it is tempting to say that he knows that he
ought to save the infant. But, imagine that in some possible world where Coop
is in the same mental states as he is in w1 that the following is true:
(4) In w2 , Coop has just the same evidence for his beliefs as he does
in w1 .
It seems that it follows that:
(5) If Coop knows in w1 that it is best to save the infant, that he can
do that, and that is what he ought to do, Coop ought to believe
these things in w2 ((4), SE1 , and SE2 ).
(6) If Coop knows in w1 that it is best to save the infant, that he
can do that, and that is what he ought to do, Coop ought to save
the infant in w2 ((5), MMD1 ).
It is consistent with everything that has been said that:
(7) In w2 , Coop cannot save the infant because the vending machine
is broken.
It follows that if “ought” implies “can”:
(8) It is false that Coop ought to save the infant in w2 ((7)).
(9) It is false that Coop ought to believe he ough to save the infant
in w2 ((8), MMD1 ).
(10) It is false that Coop could know both that he ought to save the
infant and that the best thing to do is save the infant ((9), SE1 ,
and SE2 ).
So, the unificationists who accept SE1 and SE2 end up saying that those of us
who do know that infants matter more than kittens or puppies cannot have
sufficient evidence to believe we can get infants out of vending machines. Now,
someone did say that we can never know what we will get out of a vending
machine once we put our money in. This is a very high standard for knowledge,
but really, it does not matter. Or, someone could say that we never have
obligations to bring about states of affairs in which, say, babies are saved from
vending machines. We could change (7) to deal with such worries. Coop cannot
save the infant because there is a transparent piece of glass that covers the
coin slot so he cannot get his coins into the machine. Or, Coop cannot save
the infant because he cannot move his arm because the sight of the infant set
of a strange chain of events in his nervous system that left him temporarily
paralyzed without his feeling anything strange at all. Or, Coop cannot save the
infant because he cannot so much as try to do so owing to some even stranger
events taking place in his nervous system.
The cost of combining the unificationist view with the view that denies
FactivityJ is a surprising form of skepticism. Why is it that Coop cannot know
that he can free the infant? Because an epistemic counterpart of his could
have just his evidence for believing this proposition and be mistaken where this
mistaken belief would rationalize an action that the agent cannot perform. So,
you know p only if you have no epistemic counterparts in any possible world who
falsely believe p and whose belief that p would rationalize forming the intention
to perform an action the agent cannot perform under those circumstances. If
that’s right, is there anything we can know about the external world? Little.
All it took to show that Coop did not have knowledge was to find some possible
world where he had the same evidence as he did actually but had a mistaken
belief. Now it is starting to look like this view is committed to the view that
we have sufficient evidence to believe p only when our evidence for believing
p entails p. Seems like a strange view to adopt if you are opposed to the less
demanding view that says that you ought never believe a false proposition.
Unlike this view, my view does not demand that you have entailing evidence for
your beliefs. If the choice is between FactivityJ and the view that is committed
to a kind of infallibilism that seems to make it impossible to have justified beliefs
on the grounds that such beliefs could have been mistaken and been based on
the same evidence, I think my view is an attractive one to choose.
Chapter 8
Justification
This chapter will build on “Reasons and Belief’s Justification”. In this chapter, I
discuss the significance of the results from the previous chapters for the orthodox
internalist and externalist views and develop a positive account of justification.
On the account I defend, to justifiably believe p is to believe p on the basis
of something that shows that p is true. The view is similar to the knowledge
account of justified belief insofar as both views deny that there can be false,
justified beliefs. It differs, however, insofar as there are Gettier cases in which
it seems intuitive to say that a belief that fails to constitute knowledge is not
normatively defective in any way.
165
Bibliography
[1]
1. Adler, Jonathan 2002. Belief ’s Own Ethics. Cambridge, MA: MIT Uni-
versity Press.
2. Almeida, Claudio de 2001. What Moore’s Paradox is About. Philosophy
and Phenomenological Research 62: 33-58.
3. Audi, Robert 2001. An Internalist Theory of Normative Grounds. Philo-
sophical Topics 29: 31–45.
4. Bach, Kent. 1985. A Rationale for Reliabilism. The Monist 68: 246-63.
5. Baron, Marcia 1995. Kantian Ethics Almost Without Apology. Ithaca,
NY: Cornell University Press.
6. Bird, Alexander 2007. Justified Judging. Philosophy and Phenomenolog-
ical Research 74: 81-110.
7. BonJour, Laurence 1985. The Structure of Empirical Knowledge. Cam-
bridge, MA: Harvard University Press.
8. Brewer, Bill 1999. Perception and Reason. New York: Oxford University
Press.
9. Brown, Jessica 2008. Subject-Sensitive Invariantism and the Knowledge
Norm for Practical Reasoning Nous, 42: 167-89.
10. Brueckner, Anthony 1992. What an Anti-Individualist Knows A Priori.
Analysis 52 :111-18.
11. Burge, Tyler 1979. Individualism and the Mental. Midwest Studies in
Philosophy 4: 73-121.
12. Byrne, Alex and Heather Logue 2008 “Either/Or,” in A. Haddock and
F. Macpherson, Disjunctivism: Perception, Action, and Knowledge. New
York: Oxford University Press.
13. Carlson, Erik 1995. Consequentialism Reconsidered. Boston, MA: Kluwer.
166
BIBLIOGRAPHY 167
14. Cohen, Stewart. 1984. Justification and Truth. Philosophical Studies 46:
279-95.
15. Cohen, Stewart 1988. How to Be a Fallibilist. Philosophical Perspectives
2: 91-123.
16. Collins, Anthony 1997. The Psychological Reality of Reasons. Ratio 10:
108-23.
17. Conee, Earl. 2007. Disjunctivism and Anti-Skepticism. Philosophical
Issues 17: 16-36.
18. Conee, Earl and Richard Feldman 2004. Evidentialism. New York: Oxford
University Press.
19. Conee, Earl and Richard Feldman 2008. “Evidence,” In Q. Smith, Episte-
mology: New Essays. (New York: Oxford University Press): 83-104.
20. Comesaña, Juan. 2005. Justified versus Warranted Perceptual Belief: A
Case Against Disjunctivism. Philosophy and Phenomenological Research:
367-83.
21. Comesaña, Juan and Holly Kantin. 2010. Is Evidence Knowledge? Phi-
losophy and Phenomenological Research 80: 447-54.
22. Dancy, Jonathan 1995. Why There Is Really No Such Thing as the Theory
of Motivation. Proceedings of the Aristotelian Society, New Series 95: pp.
1-18.
23. Dancy, Jonathan 2000. Practical Reality. New York: Oxford University
Press.
24. Dancy, Jonathan 2004. Ethics Without Principles. New York: Oxford
University Press.
25. D’Agostino, Fred 1988. Incommensurability and Commensuration: The
Common Denominator Ashgate Publishing.
26. DePaul, Michael 2004. “Truth Consequentialism, Withholding, and Pro-
portioning Belief to the Evidence,” Philosophical Issues 14: 91-112.
27. DeRose, Keith 2002. Assertion, Knowledge, and Context. Philosophical
Review 111:167-203.
28. Dodd, Dylan 2007. Why Williamson Should Be a Sceptic. The Philosoph-
ical Quarterly 57: 635-49.
29. Douven, Igor 2006. Assertion, Knowledge, and Rational Credibility. Philo-
sophical Review 115: 449-85.
30. Dreher, Rod. 1974. Evidence and Justified Belief. Philosophical Studies
25: 435-39.
BIBLIOGRAPHY 168
31. Engel, Mylan. 1992. Personal and Doxastic Justification. Philosophical

Studies 67: 133-51.
32. Fantl, Jeremy and Matt McGrath 2009. Knowledge in an Uncertain
World. New York: Oxford University Press.
33. Feldman, Richard. 1974. An Alleged Defect in Gettier Counter-Examples.
Australasian Journal of Philosophy 52: 68-92.
34. Feldman, Richard 1988. Subjective and Objective Justification in Ethics
and Epistemology. Monist 71: 405-19.
35. Feldman, Richard 2000. The Ethics of Belief. Philosophy and Phenomeno-
logical Research 60: 667-95.
36. Feldman, Richard 2004. Foundational Beliefs and Empirical Possibilities.
Philosophical Issues 14: 132-48.
37. Foley, Richard 2001. “The Foundational Role of Epistemology in a Gen-
eral Theory of Rationality,” in A. Fairweather and L.Zagzebski, Virtue
Epistemology: Essays on Epistemic Virtue and Responsibility (New York:
Oxford University Press): 214-31.
38. Foot, Philippa 1985. Utilitarianism and the Virtues. Mind 94: : 196-209.
39. Gardner, John. 2004. The Wrongdoing That Gets Results. Philosophical
Perspectives 18: 53-88.
40. Gardner, John 2007. Offenses and Defenses. New York: Oxford Univer-
sity Press.
41. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23:
121-3.
42. Gibbons, John 2009. You Gotta Do What You Gotta Do. Nous 43: 157-
77.
43. Gibbons, John. Forthcoming. Things That Make Things Reasonable.
Philosophy and Phenomenological Research.
44. Ginet, Carl. 1975. Knowledge, Perception, and Memory. Dordrecht:
Reidel.
45. Goldman, A. 1979. “What is Justified Belief?” in G. Pappas, Justification
and Knowledge. Dordrecht Reidel.
46. Goldman, Alvin 1986. Epistemology and Cognition. Cambridge, MA:
Harvard University Press.
47. Goldman, Alvin 1999. Knowledge in a Social World. New York: Oxford
University Press.
BIBLIOGRAPHY 169
48. Goldman, A. 2009a. “Williamson on Knowledge and Evidence,” in P. Gree-

nough and D. Pritchard, Williamson on Knowledge (New York: Oxford
University Press): 73-92.
49. Goldman, Alvin and Olsson, Erik 2009b. “Reliabilism and the Value of
Knowledge,” in A. Haddock, A. Millar, and D. Pritchard, Epistemic Value
(New York: Oxford University Press): 19-42.
50. Graham, Peter 2010. Theorizing Justification.
51. Greco, John 2000. Putting Skeptics in their Place. New York: Cambridge
University Press.
52. Hawthorne, John 2004. Knowledge and Lotteries. New York: Oxford
University Press.
53. Hawthorne, John and Stanley, Jason 2008. Knowledge and Action. Jour-
nal of Philosophy 105: 571-90.
54. Herman, Barbara 1993. The Practice of Moral Judgment. Cambridge,
MA: Harvard University Press.
55. Hieronymi, Pamela 2005. The Wrong Kind of Reasons. Journal of Phi-
losophy 102: 37-57.
56. Hooker, Brad 2000. Ideal Code, Real World. New York: Oxford University
Press.
57. Horder, J. 2004. Excusing Crime. New York: Oxford University Press.
58. Hornsby, Jennifer 2007. Knowledge in Action. In A. Leist (ed.), Action
in Context. (Berlin: De Gruyter), pp. 285-302.
59. Huemer, Michael. 2006. Phenomenal Conservatism and the Internalist
Intuition. American Philosophical Quarterly 43: 147-58.
60. Huemer, Michael 2007. “Moore’s Paradox and the Norm of Belief,” in S.
Nuccetelli and G. Seay, Themes from G.E. Moore (New York: Oxford
61. Husak, Doug 1999. Conflicts of Justification. Law and Philosophy 18:
41–68.
62. Hyman, John 1999. How Knowledge Works. Philosophical Quarterly 49:
433-51.
63. Hyman, John 2006. Knowledge and Evidence. Mind 115: 891-916.
64. Kelly, Thomas 2008. Evidence: Fundamental Concepts and the Phenom-
enal Conception. Philosophy Compass 3: 933-55.
BIBLIOGRAPHY 170
65. Kvanvig, Jonathan 2003. The Value of Knowledge and the Pursuit of
Understanding. New York: Cambridge University Press.
66. Kvanvig, Jonathan 2009. “Assertion, Knowledge, and Lotteries,” in P.
Greenough and D. Pritchard, Williamson on Knowledge (New York: Ox-
ford University Press): 140-61.
67. Lackey, Jennifer 2007. Norms of Assertion. Nous 41: 594-626.
68. Langsam, Harold. 2008. Rationality, Justification, and the Internal-
ism/Externalism Debate. Erkenntnis 68: 79-101.
69. Lord, Errol 2008. Dancy on Acting for the Right Reason. Journal of
Ethics and Social Philosophy.
70. Lowy, Catherine. 1978. Gettier’s Notion of Justification. Mind 87: 105-8.
71. Lyons, David 1965. Forms and Limits of Consequentialism. New York:
Oxford University Press.
72. Maher, Patrick 1996. Subjective and Objective Confirmation. Philosophy
of Science 63: 149-74.
73. Maitzen, Stephen 1995. Our Errant Epistemic Aim. Philosophy and Phe-
nomenological Research 55: 869-76.
74. McDowell, John 1998. “Criteria, Defeasibility, and Knowledge,” in his
Meaning, Knowledge, and Reality. Cambridge, MA: Harvard University
Press.
75. McDowell, John 2008. “The Disjunctive Conception of Experience as Ma-
terial for a Transcendental Argument,” in A. Haddock and F. Macpherson,
Disjunctivism: Perception, Action, and Knowledge. New York: Oxford
University Press.
76. McKinsey, Michael 1991. Anti-Individualism and Privileged Access. Anal-
ysis 51: 9-16.
77. McNaughton, David 1996. An Unconnected Heap of Duties? The Philo-
sophical Quarterly 46: 433-47.
78. Moore, Michael. 1993. Placing Blame. New York: Oxford University
Press.
79. Miller, Christian 2008. Motivation in Agents. Nous 42: 222-66.
80. Nagel, Thomas 1979. Mortal Questions. New York: Cambridge University
Press.
81. Nelson, Mark 2002. What Justification Could Not Be. International Jour-
nal of Philosophical Studies 10: 265-81.
BIBLIOGRAPHY 171
82. Nelson, Mark 2010. We Have No Positive Epistemic Duties. Mind 119:
83-102.
83. Neta, Ram and Pritchard, Duncan 2007. McDowell and the New Evil
Genius. Philosophy and Phenomenological Research 74: 381 – 396.
84. Neta, Ram 2008. What Evidence Do You Have? The British Journal for
the Philosophy of Science 59: 89-119.
85. Neta, Ram 2009. Treating Something as a Reason for Action. Nous 43:
684-99.
86. Norcross, Alistair 2008. Two Dogmas of Deontology: Aggregation, Rights,
and the Separateness of Persons. Social Philosophy and Policy 26:76-95.
87. Parfit, Derek 1997. Reasons and Motivation. Proceedings of the Aris-
totelian Society, Supp. Volume 71: 99-130.
88. Pollock, John 1975. Knowledge and Justification. Princeton, NJ: Prince-
ton University Press.
89. Pollock, John and Joseph Cruz 1999. Contemporary Theories of Knowl-
edge. Totowa, NJ: Rowman and Littlefield.
90. Pritchard, Duncan 2007. Recent Work on Epistemic Value. American
Philosophical Quarterly 44: 85-110.
91. Pryor, James. 2000. The Skeptic and the Dogmatist. Nous 34: 517-49.
92. Rawls, John 1955. Two Concepts of Rules. The Philosophical Review 64:
3-32.
93. Raz, Joseph 1990. Practical Reason and Norms. New York: Oxford Uni-
versity Press.
94. Reynolds, Stephen 2002. Testimony, Knowledge, and Epistemic Goals.
Philosophical Studies 110: 139-61.
95. Riggs, Wayne 2008. “The Value Turn in Epistemology,” in V. Hendricks
and D. Pritchard, New Waves in Epistemology (New York: Palgrave
MacMillan): 300-24.
96. Rizzieri, Aaron. Forthcoming. Evidence Does not Equal Knowledge.
Philosophical Studies.
97. Robinson, Paul 1996. “Competing Theories of Justification: Deeds v.
Reasons,” in A. Simester and A. Smith, Harm and Culpability (New York:
Oxford University Press): 45-70.
98. Ross, W.D. 1930. The Right and the Good. New York: Oxford University
Press.
BIBLIOGRAPHY 172
99. Sadock, Jerrold. 1978. On Testing Conversational Implicature. Syntax

and Semantics 9: Pragmatics: 281-97.
100. Scanlon, Timothy 1999. What We Owe to Each Other. Cambridge, MA:
Harvard University Press.
101. Schroeder, Mark 2008. Having Reasons. Philosophical Studies 139: 57-71.
102. Shope, Robert. 1983. The Analysis of Knowing. Princeton, NJ: Princeton
University Press.
103. Silins, Nico. 2005. Deception and Evidence. Philosophical Perspectives
19: 375-404.
104. Smith, Michael 1987. The Humean Theory of Motivation. Mind 96, pp.
36-61.
105. Sorensen, Roy 1988. Blindspots. New York: Oxford University Press.
106. Stanley, Jason 2005. Knowledge and Practical Interests. New York: Ox-
ford University Press.
107. Stanley, Jason 2008. Knowledge and Certainty. Philosophical Issues 18:
33-55.
108. Stratton-Lake, Philip 2000. Kant, Duty, and Moral Worth. New York:
Routledge.
109. Stocker, Michael 1990. Plural and Conflicting Values. New York: Oxford
University Press.
110. Sutton, Jonathan 2005. Stick To What You Know. Nous 39: 359-96.
111. Sutton, Jonathan 2007. Without Justification. Cambridge, MA: MIT
University Press.
112. Sverdlik, Steven 2001. Kant, Nonaccidentalness and the Availability of
Moral Worth. The Journal of Ethics 5: 293-313.
113. Swanton, Christine 2003. Virtue Ethics: A Pluralistic View. New York:
114. Swinburne, Richard 1999. Providence and the Problem of Evil. New York:
115. Thomson, Judith 1991. Self-Defense. Philosophy and Public Affairs 20:
283-310.
116. Thomson, Judith 2008. Normativity. Chicago, IL: Open Court.
117. Turri, John 2009. The Ontology of Epistemic Reasons. Nous 43: 490-512.
BIBLIOGRAPHY 173
118. Unger, Peter 1975. Ignorance. New York: Oxford University Press.
119. Van Cleve, James 2004. “Externalism and Disjunctivism,” in R. Schantz,
The Externalist Challenge (de Gruyter): 481-95.
120. Velleman, David 2000. The Possibility of Practical Reason. New York:
121. Wedgwood, Ralph 2002a. Internalism Explained. Philosophy and Phe-
nomenological Research 65: 349-69.
122. Wedgwood, Ralph 2002b. The Aim of Belief. Philosophical Perspectives
16: 267-97.
123. Weiner, Matthew 2005. Must We Know What We Say? The Philosophical
Review 114: 227-51.
124. Wiland, Eric 2002. Theories of Practical Reason. Metaphilosophy 33, pp.
450-67.
125. Williams, Bernard 1965. Ethical Consistency. Proceedings of the Aris-
totelian Society, supplementary volume 39: 103-38.
126. Williams, Bernard. 1973. Problems of the Self. New York: Cambridge
University Press.
127. Williams, Bernard 1981. Moral Luck. New York: Cambridge University
Press.
128. Williamson, Timothy 1990. Identity and Discrimination. Malden, MA:
Blackwell.
129. Williamson, Timothy 2000. Knowledge and its Limits. New York: Oxford
University Press.
130. Williamson, Timothy 2007. “On Being Justified in One’s Head,” in J.
Greco, A. Mele, and M. Timmons, Rationality and the Good: Critical
Essays on the Ethics and Epistemology of Robert Audi (New York: Oxford
131. Williamson, Timothy 2009. “Replies to Critics,” in P. Greenough and
D. Pritchard, Williamson on Knowledge (New York: Oxford University
Press): 279-385.
132. Zagzebski, Linda 2003. “The Search for the Source of Epistemic Good,”
Metaphilosophy 34: 12-28.
133. Zimmerman, Michael 1988. An Essay on Moral Responsibility. Totowa,
NJ: Rowman and Littlefield.
134. Zimmerman, Michael 1996. The Concept of Moral Obligation. New York:
Cambridge University Press.
BIBLIOGRAPHY 174
135. Zimmerman, Michael 2008. Living with Uncertainty. New York: Cam-
bridge University Press.

Littlejohn Justification Manuscript

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Littlejohn Justification Manuscript

Uploaded by

Copyright:

Available Formats

Chapter 1

The Orthodox Views

3.2 Epistemic Consequentialism

consequentialist approach to right action.

Goldman defends something similar to ERC in his work.5 To my knowledge, no

3.2.1 Epistemic Rule-Consequentialism

On his view, a belief is justified if it is formed in accordance with a right system

3.2.2 Can We Unify the J-Rules?

non-consequentialism (PENC), a belief is jusitifed if it is formed by someone

reliably determine the locations of various individuals using clairvoyant powers

3.2.3 The Case Against Epistemic Consequentialism

no false, justified beliefs in later chapters.

If the nominal aim is the reason for having, or pursuing, justifica-

3.3 Evidentialism and Epistemic Value

argument persuasive. Suppose S reasons competently from p and if p then q to q. If knowledge

As we saw earlier in our discussion of phenomenal conservatism, there seem to

A plan (decision, action, strategy, belief, etc. . . ) is rational in sense

a different response, such as honoring, commending, respecting, and forgiving.

3.4 The Right and the Good

have is instrumental value, when something of intrinsic value is realized, the

We can say the following about the case:

4.2 Getting a Fix on the Notion

4.3 The Refutation of Mentalism

reason, a reason to believe rather than something as opposed to (just) a reason

4.3.1.1 The Linguistic Case for FactivityE

4.3.1.2 The Objection that Proves the View

the neighborhood. Therefore, again, if E=K1 is true then closure

But, these entail:

4.3.2 Having Evidence

constitute evidence in the strict sense. If you justifiably believe p on inductive

p is part of your ultimate evidence only if p is the content of some

Mentalists are free to deny Broad Experientiality, but doing so comes at a

4.4 Evidence and Knowledge

If you know p, p is part of your evidence (ESK).

the basis of non-entailing evidence, is p an extra bit of evidence you acquire by

Although the principle is controversial, it is unlikely that the argument fails

of the problem of reconciling content externalism with privileged access.

not justifiably believe p without independent evidence, your belief in p would

Your evidence includes p iff you know p non-inferentially (E=IK).

for (e) is true in virtue of the number of coins in Smith’s pocket,

4.5 Evidence and Epistemic Rationality

4.6 The Refutation of Supervenience Internalism

5.2 Reasons for Belief

things appear to be that way. The mountain appears to be that

5.3 Externalism and Epistemological Disjunctivism

He says that this Cartesian view faces a decisive objection:

The evidence veridical experience provides is better than the evi-

depends upon whether you are hallucinating or your experience is

phenomenal conservatism. Externalist views such as Goldman’s 1986 reliabilism, Bergmann’s

5.3.1 Preliminary Objections

5.3.2 The Epistemological Argument for Disjunctivism

against acting can contribute to normative standing of an action even if the

rightly held belief and entailing evidence.

5.4 Reasons for Action

(5) A belief cannot be justified if it cannot justifiably be included in

5.5 Reasons and Motivation

that cases of error cause trouble for Dancy’s view.

5.5.1 Motivational and Normative Psychologism

5.5.2 The Argument from Error

This is why he denies that it is a consequence of (7). It had better not be a

In the debate between Dancy and the defenders of PsychologismP, both

5.5.3 Acting for a Reason as an Achievement

6.2 Warrant and Normativity

6.3 The Knowledge Account