Documente Academic
Documente Profesional
Documente Cultură
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=black. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Blackwell Publishing is collaborating with JSTOR to digitize, preserve and extend access to Nos.
http://www.jstor.org
Philosophical Perspectives, 13, Epistemology, 1999
Stewart Cohen
Arizona State University
(I) Contextualism
Falliblism allows that we can know on the basis of non-entailing reasons. But
how good do the reasons have to be? Reflection on cases show that this can be a
difficult question to answer:
What should we say aboutthis case?3 Smith claims to know that the flight stops
in Chicago. Mary and John deny that Smith knows this. Mary and John seem to
be using a stricterstandardthan Smith for how good one's reasons have to be in
orderto know. Whose standardis correct?Let's consider several answers:
Is this a good answer? If we say that contraryto what both Mary and John pre-
suppose, the weaker standardis correct,then we would have to say that their use
of the word 'know' is incorrect. But then it is hard to see how Mary and John
should describe their situation. Certainly they are being prudentin refusing to
rely on the itinerary.They have a very importantmeeting in Chicago. Yet if Smith
knows on the basis of the itinerarythat the flight stops in Chicago, what should
Contextualism,
Skepticism,
andReasons / 59
they have said? "Okay,Smith knows thatthe flight stops in Chicago, but still, we
need to check further."To my ear,it is hardto make sense of thatclaim. Moreover
if what is printedin the itineraryis a good enough reason for Smith to know, then
it is a good enough reason for John and Mary to know. Thus John and Mary
should have said, "Okay,we know the plane stops in Chicago, but still, we need
to check further."Again it is hardto make sense of such a claim.
Perhapsthen the correct answer is:
2) John and Mary are right and so Smith's standardis too weak. (Smith can
not know, but John and Mary can know after checking furtherwith the
agent.)
I think this is a naturalresponse to this case as I have described it. But notice
that this contrasts with the standardswe typically use for knowledge ascrip-
tions. In everyday contexts, we readily ascribe knowledge to someone on the
basis of written information contained in things like flight itineraries. If we
deny that Smith knows, then we have to deny that we know in many of the
everyday cases in which we claim to know. We would have to say that a con-
siderable amount of the time in our everyday lives, we speak falsely when we
say we know things.
And it gets worse. We could describe a case where even Mary and John's
standarddoes not seem strict enough: If someone's life were at stake, we might
not even be willing to ascribe knowledge on the basis of the testimony of the
airline agent. We might insist on checking with the pilot. So it does not look
promising to say that Smith's standardis too weak.
We could, at this point, pursue a thirdoption, viz., all of these standardsare
too weak. This option leads, of course, to skepticism and presumably,this is a
result we want to avoid. (We will returnto this option in section III)
So far we have examined three different answers to the question of whose
standardis correct: (1) Smith's is correct and so John and Mary's standardis
too strong. (2) John and Mary's standardis correctand so Smith's standardis too
weak. (3) Neither Smith's nor John and Mary's standardis correct both are too
weak. None of these answers seems satisfactory.So let me say what I take to be
the best answer: Neither standardis simply correct or simply incorrect. Rather,
context determineswhich standardis correct.Since the standardsfor knowledge
ascriptions can vary across contexts, each claim, Smith's as well as Mary and
John's, can be correctin the context in which it was made. When Smith says, "I
know...",what he says is truegiven the weaker standardoperatingin thatcontext.
WhenMaryandJohnsay "Smithdoes not know...",what they say is truegiven the
stricterstandardoperatingin their context. Andthereis no contextindependent
correctstandard.
So I claim thatthis case, and others like it, strongly suggests that ascriptions
of knowledge arecontext-sensitive.The standardsthatdeterminehow good one's
reasonshave to be in orderto know are determinedby the context of ascription.
60 / Stewart Cohen
(III) Skepticism
might think that we have some reason to believe that we are not deceived in the
ways the skeptic suggests, but it is very hardto hold that we knowwe are not so
deceived.
Suppose, to use Dretske's example, that you are at the zoo looking at the
Zebraexhibit.13Considerthe possibility thatwhat you see is not a zebrabut rather
a cleverly-disguised mule. Though you may have some reason to deny you are
looking at a cleverly-disguised mule, it seems wrong to say you knowyou are not
looking at a cleverly-disguised mule. After all, that'sjust how it would look if it
were a cleverly-disguised mule.
The skeptic then appeals to a deductive closure principle for knowledge:
(3) IknowP.
is false.
Because I, and many others, share Fumerton's view, let's move on to consider
other responses to the skeptical argument.20
about epistemic priority and the structureof reasons. Note that on this view, it
looks as if my inductive evidence againstthe possibility of my seeing a cleverly-
disguised mule is irrelevant to my knowing I do not see a cleverly-disguised
mule. This is certainly a strangeresult.23We will take up this issue in more detail
later in the paper.At this point, I want to focus on a differentproblem for modus
ponensfallibilism. The view does not meet one of our criteriafor a satisfactory
resolution of the skeptical paradox,viz., explaining the appeal of skeptical argu-
ments. For if I can know on the basis of my evidence thatI see a Zebra,and know
on the basis of my seeing a zebra that I am not seeing a cleverly disguised mule,
then what explains the intuition that I fail to know I am not seeing a cleverly
disguised mule? Certainlynot that my reason is not strongenough. According to
this view, my reason for believing that I do not see a cleverly disguised mule is
that I see a zebra. And of course that is an entailing reason. So it remains a
mystery why the skeptical argumentshould have any cogency.24
dards for how strong those reasons have to be varies across contexts. By sup-
posing that knowledge ascriptions are context-sensitive in this way, we can do
justice both to our strong inclination to say we know and to the undeniable
appeal of skeptical arguments.
So on a contextualist approach,which proposition of the paradox gets de-
nied? As we have seen, some take the apparenttruthof both (2) and (3) to show
that (1) is false, thereby giving up the closure principle. But most find rejecting
the closure principleto be unacceptable.An advantageof contextualismis that it
can defend the closure principle while explaining why there is an appearanceof
closure failure.
We can illustratethis throughDretske'szebracase: My reasons for believing
I see a zebraconsist of the animal'slooking like zebras and being in pens marked
'Zebra'. My reason for believing I do not see a cleverly disguised mule consists
of the inductive evidence I have against the possibility of such a deception. It
looks as if I know I see a zebrabut I fail to know I do not see a cleverly-disguised
mule. We might be tempted to think my reasons for believing I see a zebra are
strongerthanmy reasons for thinkingI do not see a cleverly-disguised mule. But
surely if we accept the closure principle, we accept that where P entails Q, the
strengthof my reasons for believing P can be no greaterthan the strengthof my
reasons for believing Q. So my reasons for believing I see a zebracan be no
strongerthan my reasons for believing I do not see a cleverlydisguised-mule.
According to contextualism, however, the standardsfor how strong my reasons
have to be in orderfor me to know can vary across contexts. In contexts where we
consider whether I know I do not see a cleverly-disguised mule, the chance of
erroris salient, unlike in everyday contexts where we consider whetherI know I
see a zebra. And when the chance of erroris salient in a context, the standards
tend to rise. Thus we evaluate whether I know I do not see a cleverly-disguised
mule at a stricterstandardthan that at which we evaluate whetherI know I see a
zebra. This gives rise to the appearanceof closure failure. But if we hold the
context, and so the standards,fixed, we see that the closure principle is not
threatened.
So in everyday contexts the standardsare such that my reasons are good
enough for me to know I see a Zebra.And since my reasons for denying thatI see
a cleverly disguised mule can be no worse, my reasons are sufficient for me to
know that proposition as well, given the standardsof those contexts. Thus in
everyday contexts, I can know that I don't see a cleverly-disguised mule, on the
basis of the inductive evidence I have against such a scenario.
In skeptical contexts where the standardsare higher, I fail to know, on the
basis of the inductive evidence, that I do not see a cleverly-disguised mule. But
since my reasons my for believing that I see a Zebra can be no better, I fail to
know that proposition as well, given the standardsof that context. The appear-
ance of closure failure results from the shift in standardsthat occurs when we
move from consideringwhetherI know thatI see a Zebrato considering whether
I know that I do not see a cleverly-disguised mule.
Contextualism,
Skepticism,
andReasons / 67
evidence whatsoever against global alternatives, we can not hold that the evi-
dence we do have is good enough relative to everyday standards.So it looks as if
contextualismis of no use in respondingto global skepticism.32
We may have been too hasty, however, in deciding that I have no evidence
againstthe alternativethatI am not a brain-in-a-vat.Considerthe fact thatI have
no evidence in favor of this alternative.My experience has never been subjectto
radical unexplainedincongruities as might occur if the vat apparatussuffered a
power failure or in some way malfunctioned. The vatmeister has never "ap-
peared"to me, etc. Surely the fact that none of this has happenedcounts against
the alternativethatI am a brain-in-a-vat.So I do have some evidence againstthis
alternative,after all.33Now in skeptical contexts (such as the one we are in now
as we consider skeptical hypotheses), this evidence will not be sufficient for me
to know I am not a brain-in-a-vat.Nonetheless, the contextualistcan arguethatin
everyday contexts, this evidence is sufficient for me to know I am not a brain-
in-a-vat. In this way, the contextualistcan explain why, underskepticalpressure,
we are temptedto say we do not know we arebrains-in-a-vat,while allowing that
in everyday contexts we in fact know we are not.
But this contextualist strategy will go only so far. The absence of radical
incongruitiesin my experience, etc, counts as evidence againstmy being a brain-
in-a-vat only to the extent that the occurrenceof such things is probable,condi-
tional on my being a brain-in-a-vat.So consider the alternativethat I am a brain-
in-a-vat and l will never have evidence that l am. Call this the 'brain-in-a-vat8'
hypothesis. The fact that radical incongruities, etc., have not occurredis no evi-
dence whatsoever against the brain-in-a-vatehypothesis.
Can the contextualist treatmentbe extended to apply to the brain-in-a-vate
hypothesis? On the contextualist view we have been considering, context de-
termines how strong one's evidence must be, in order for one to know. But
suppose we think of context as determining,more generally,how rational one's
belief must be in orderfor one to know. While a belief can be rationalin virtue
of being supported by evidence, we need not hold that evidence is the only
source of rationality for a belief. Consider the belief that I am not a brain-in-
a-vat8. Although we may concede that we have no evidence in support of this
belief, it still seems intuitively compelling that the belief is rational at least to
some degree.3
Could this intuitionbe mistaken?It would be if we were compelled to accept
By my lights, (R) does not have the kind of axiomatic status that the closure
principle (C) has. The belief that I am not a brain-in-a-vatejust looks to be a
plausible counterexample.Moreover,many philosophershave held thatthereare
non-evidential(so called 'pragmatic')considerationslike simplicity and conser-
vatism relevant to rationalacceptance.35
Contextualism,Skepticism,and Reasons / 69
...one's belief as to whetherP is true match[es] the fact of the matteras to whetherP
is true, not only in the actual world, but also at the worlds sufficiently close to the
actual world...40
How far out from the actual world must one's belief in P matchthe fact as to
whetherP, in order for one to know P? On DeRose's view, this will depend on
context. According to what DeRose calls 'the rule of sensitivity', in contexts
where we are considering whether S knows P, the standardrequires that S not
believe P in the nearest not-P world(s). This leads to ostensible violations of
closure in just the way the tracking condition does. For example, I would not
believe I have a hand in the nearest world where I do not have a hand (a world
where I lose my hands in an accident). So I can know I have a hand. But I would
believe I am not a brain-in-a-vatin the nearestworld in which I am a brain-in-a-
vat. So I can not know I am not a brain-in-a-vat.
DeRose avoids this counterexampleto closure by holding that the standard
determinedby the rule of sensitivity applies to any knowledge ascriptionin the
context. Let us say that when one's belief regarding P matches the fact (at a
particularworld) as to whetherP, one's belief tracksthe truth(at that world). So
in a context where we are considering whetherS knows P, any belief, to count as
an instance of knowledge, must trackthe truthout to the nearestnot-P world. The
appearanceof closure failure results from the fact that in contexts where we are
consideringwhetherI know I am not a brain-in-a-vat,the standardrequirestruth-
tracking out to a more distant world than in contexts where we are considering
whetherI know I have a hand.Thus we end up evaluatingwhetherI know I am not
a brain-in-a-vat,at a higher standardthan that at which we evaluate whether I
Contextualism, andReasons / 71
Skepticism,
know I have a hand. But, as we will see, if we hold the context (and so the
standard)fixed, I know I have a hand only if I know I am not a brain-in-a-vat.
The skeptic, by raising the question of whether I know I am not a brain-in-
a-vat, creates a context where in order to know any proposition P, one's belief
regardingP must trackthe truthout to at least the nearestworld at which I am a
brain-in-a-vat.Since at thatworld I believe I am a brain-in-a-vat,I fail to know I
am not a brainin a vat, in thatcontext. But, in that same context, none (or at least
very few) of my empiricalbeliefs count as knowledge. For none of my empirical
beliefs will trackthe truthout to that distantworld. In the closest world(s) where
I am a brain-in-a-vat,all (or at least most of ) my actualempiricalbeliefs arefalse,
yet I still hold them.41
When I assert that I know I have a hand, the standardof this context will
require only that my belief that I have a hand track the truthout to the nearest
world at which I don't have a hand. Since at that world (where I lose my hands
in an accident) I do not believe I have a hand, I can know, in that context, that
I have a hand. But in that same everyday context, I know I am not a brain-in-
a-vat.42The standardfor that context requires that my belief that I am not a
brain-in-a-vattrack the truthout to the nearest world at which I do not have a
hand. And in such a world, I both am not a brain-in-a-vatand believe I am not
a brain-in-a-vat.
DeRose's contextualistresponse to the skepticalparadoxhas the same struc-
ture as my response. As we noted, I have an internalist construal of epistemic
position and DeRose has an externalistconstrual.Nonetheless, on both views, if
we hold the context, and so the standardsfixed, then the strengthof one's episte-
mic position will be sufficient for one to know one has a hand,just in case it is
sufficient for one to know one is not a brain-in-a-vat.In this way, both views yield
the same resolution of the skeptical paradox.In everyday contexts (3) is true and
(2) is false. And in skeptical contexts, (2) is true and (3) is false. But in every
context (1) remainstrue.As we have seen, this kind of treatmentcan explain the
appealof skeptical arguments,while preservingthe truthof our everyday knowl-
edge ascriptions.
Because it affirms (1) and so defends the closure principle, DeRose's con-
textualizedtrackingview faces the problemthatthe originaltrackingview avoids,
viz., accountingfor how we know we are not brains-in-a-vat.We saw that on my
internalist account, I know I am not a brain-in-a-vate independently of any
empiricalevidence that counts in it's favor. This yields the less than satisfactory
result that I have contingent a priori knowledge. Does DeRose's externalist ac-
count fare any better?
On DeRose's view, my belief thatI am not a brain-in-a-vatcounts as knowl-
edge in everyday contexts where I know I have a hand, because it meets the
standardfor truth-trackingthat governs those contexts. That is, my belief that I
am not a brain-in-a-vattracks the truthout to the nearest world(s) at which I do
not have a hand. This is so, simply because in every such world, I am not a
brain-in-a-vat,and I believe I am not a brain-in-a-vat.
72 / Stewart Cohen
But then DeRose's externalist account suffers from the same worry as my
internalistaccount. For on his account my knowing I am not a brain-in-a-vatis
not based on my having empiricalevidence againstmy being a brain-in-a-vat.My
knowing is secured by the mere fact that my believing I am not a brain-in-a-vat
persists out to the nearestworld at which I do not have a hand. So my knowing I
am not a brain-in-a-vatis a priorion DeRose's account in the same way it is a
* AA
prlorlon m1ne.
For DeRose's view, however, this points to a more general difficulty. For
example, on his view, in everyday contexts where I know I have a hand, I will
know any naturallaw that I merely believe very firmly that is believe in a way
that resists change across near worlds. For my belief in the law only has to track
the truthout to the relatively nearworld(s) at which I do not have a hand.And in
all such worlds, I will believe the law holds (ex hypothesi)and the law will hold
(since naturallaws only fail in distant worlds).
A more serious difficulty for DeRose's view is that it does not really avoid
the result that led us to reject the original trackingaccount. For the contextualist
mechanismDeRose invokes does not, in the end, preservethe closure principle.44
As we noted, Dretske and Nozick accept closure failure as part of their anti-
skeptical strategy.There are some cases, however, where a subjunctive condi-
tional accountleads to closure failures that are an embarrassmenteven for those
who acceptthatthe closure principleis not exceptionless. Considerthe following
case 45
I read the paperthe next morningand come to believe there was 6pm draw-
ing. On DeRose's view, can I know there was a 6pm drawing?This will depend
on whether my belief that there was a 6pm drawing tracks the truth out to the
nearestworld at which therewas no 6pm drawing.At thatworld, it is reportedthat
the drawingoccurs at 5pm, and so I would not believe there was a 6pm drawing.
Rather I would believe there was a 5pm drawing. (We are assuming that not
holding a drawingat 6pm for any reason otherthancarryingout the plot would be
very unlikely.) So I can know there was a 6pm drawing.46
Now consider whether on DeRose's view, I know there was a drawing, at
some time or other?At this point, that seems like a strangequestion.After all, if
Contextualism,
Skepticism,
andReasons / 73
I know therewas a drawingat 6pm, how could I fail to know therewas a drawing
at some time or other?The strangenessof the questionreflects the intuitivenessof
the deductive closure principle.Does my belief that there was a drawingat some
time or other, track the truth out to the nearest world at which there was no
drawingat any time?At thatworld, I again would believe there was a drawingat
5pm, since thatis what the paperwould have reported.(Again, we are supposing
that not holding a drawingfor any reason other than carryingout the plot would
be very unlikely.) Thus I fail to know there was a drawingat some time or other.
So it looks as if I know therewas a 6pm drawing,but I do not know therewas
a drawing (at some time or other). So it looks as if we have a failure of closure.
This by itself is no problemfor DeRose quadefenderof closure.As we have seen,
he does not mean to deny the appearanceof closure failure; in fact, his view
predicts the appearanceof closure failure. Rather DeRose wants to follow the
contextualiststrategyfor explaining away the appearanceof closure failure. We
have seen how that strategy works in the case of I havea handand I am not a
brain-in-a-vat. Contextualshifts in the standardfor knowledge leads us to eval-
uate whether we know the latter proposition at a higher standardthan that at
which we evaluatewhetherwe know the former.But we know both propositions
at the lower standardand we know neitherproposition at the higher standard.
The problemfor DeRose is thatthis strategyfor saving closure can not han-
dle the presentcase. For the contextualist strategyto work, it must turnout that
we evaluate whether I know there was a drawing (at some time or other) at a
higher standardthan that at which we evaluate whetherI know there was a 6pm
drawing.On DeRose's account, this means that it must turn out that in contexts
where we consider whether I know that there was a drawing (at some time or
other) the standardfor knowledge requires truth tracking out to more distant
worlds thanthe standardof the context where we consider whetherI know there
was a 6pm drawing.According to DeRose's view, in any context where we con-
sider whetherI know P, the standardrequiresthat my belief in P trackthe truth
out to the nearestnot-P world. So for DeRose to apply the contextualiststrategy,
it must turn out that the nearest world at which there was no drawing is more
distantthan the nearestworld at which there was no 6pm drawing.The problem
is that the nearestworld at which there was no drawingjust is the nearest world
at which there was no 6pm drawing. The nearest world at which there is no
drawingis the world at which the conspiratorscarryout their plot and the paper
reportsthatthere was a Spm drawing.And thatis also the nearestworld at which
there is no 6pm drawing. So DeRose's version of the contextualist strategy can
not explain away the appearanceof closure failure. On his view, I can know there
was a drawingat 6pm relative to the very same standard(and so in the very same
context) at which I fail to know there was a drawing (at some time or other).
Presumably,not even the friendsof closure failurewould welcome this particular
instance.47
The principleadvantageof Dretske andNozick's originaltrackingaccountis
thatby denying closure, it escapes the problem of having to account for how we
74 / Stewart Cohen
know we are not brains-in-a-vat.We rejected that account on the grounds that
denying closure is too high a price to pay for this advantage.DeRose's contex-
tualist version of the trackingaccount ends up having to deny the closure prin-
ciple as well. But it does so withoutgaining the advantageof not having to explain
how we know we are not brains-in-a-vat.Because on DeRose's version of track-
ing, we do know, in everyday contexts, thatwe arenot brains-in-a-vat.And as we
have seen DeRose's view fares no betterthanmy internalistcontextualism(which
preserves closure) in explaining how we know this.
sentence thatsays Albany is the capital is not a misprint.So can I infer and come
to know Albany is the capital on the basis of the sentence saying so, and because
of the entailment, thereby be in a position to know that the sentence is not a
misprint?Surelymy reasonscannotbe structuredin this way. But then how do we
distinguish between this case and the reasons structureendorsed by moduspo-
nensfallibilism in the skeptical scenario cases.
In responseto this objection,Klein proposesa criterionfor distinguishingthe
atlas case from the disguised mule case and the brain-in-a-vatcase. According to
Klein, thoughsometimes an alternativeto P mustbe "eliminated"priorto coming
to know P, sometimes it can be eliminatedaftercoming to kno-wP, by appealing
to P itself. An alternativemust be eliminated prior to coming to know P just in
case "thereis some (even very minimal)evidence in its [thealternative's]favor".48
Let us say that when an alternativeH, to P is eliminated on the basis of P,
where the reasons for P are not reasons against H, that the reasons have an MPF
structure.In the atlas case, there is some evidence in favor of the alternativethat
the atlas contains a misprint the inductive evidence thatatlases sometimes con-
tain such mistakes. This explains accordingto Klein's criterion,why the reasons
can not have an MPF structure.According to Klein however, thereis no evidence
in favor of the alternativethat I see a cleverly disguised mule or the alternative
thatI am a brain-in-a-vat.So in these cases, the reasonscanhave an MPF structure.
I don't thinkthatKlein's criterioncan drawthe distinctionshe intends. Let's
grantthatwe have no evidence in favor of the brain-in-a-vathypothesis (although
one might say thatthis hypothesis is perfectly confirmedby ourexperience). Still
it's not clear that Klein's criterioncan explain why the reasons can have an MPF
structurein the zebra case. After all, people sometimes stage deceptions. And
why doesn't this count as "even very minimal"evidence for the disguised mule
deception?
Suppose we don't count the general fact thatpeople sometimes stage decep-
tions as inductive evidence in favor of the disguised mule hypothesis. Let's count
only the occurrenceof disguised mule deceptions. Still on Klein's view, when I
go to the zoo, my knowledge that I see zebras would appearto be very tenuous.
For if disguised mule deceptions have occurred,then by analogy with what Klein
says aboutthe atlas case, I can not rely on reasons with an MPF structureto know
that the disguised mule deception was not occurring.And since Klein says my
inductive evidence is not sufficient for me to know that such a deception is not
occurring,I could not know (in any context) thatthe deception was not occurring.
Thus by closure, I could not know (simply by looking) thatthe animalsI see at the
zoo are zebras.But even if some fraternityhas in the past staged a disguised mule
deception,surelywhen I go to the zoo, I can know I see a zebra(just by looking).49
MoreoverKlein's criteriondoes not explain why my reasons can not have an
MPF structurein the atlas case. Suppose there were no evidence in favor of the
sentence containinga misprint.50WouldI then be able to know the sentence does
not contain a misprinton thebasis of reasonswithan MPFstructure?5l That is,
would I then be able to infer that Albany is the capital from the fact that the
76 / Stewart Cohen
sentencein the atlas says so, andtherebybe in a position to know thatthe sentence
does not contain a misprint?The MPF structurestill seems objectionable. So
Klein's attemptto distinguishbetween applying the MPF reason structureto the
atlas case and applyingit to the brain-in-a-vatcase fails. Since the MPF structure
is objectionablein the atlas case it is objectionable in the brain-in-a-vatcase as
well.
So far, we have two views that provide a resolution of the skeptical paradox
that retainsthe closure principle the original internalistcontextualism and the
contextualizedversion of modusponensfallibilism. Any such view must provide
an accountof how we know we arenot brains-in-a-vat*.And this is a problemfor
both of these views. Perhapsthough, we can combine both views in a way that
avoids the problems of each.
Let us suppose that I can not have a priori knowledge of the contingent
propositionthat I am not a brain-in-a-vat*.Thus I can not know I am not a brain-
in-a-vat* on the basis of its non-evidentialrationality as proposedby the orig-
inal contextualistaccount.Still we can allow thatit is non-evidentiallyrationalto
deny thatI am a brain-in-a-vat* thoughthis by itself would not be sufficient for
me to know I am not a brain-in-a-vat*.
Let us also suppose that I can not come to know I have a hand simply on the
basis of my empiricalevidence (its seeming like I have a hand) and therebybe in
a position to know I am not a brain-in-a-vat.So I can not know I am not a brain-
in-a-vat on the basis of reasons with an MPF structure as proposed by modus
ponensfallibilism. Still we can say that my empirical evidence, in conjunction
with the non-evidentialrationalityof denying I am a brain-in-a-vat*is sufficient
for me to know I have a hand.
Then we can say that once I know I have a hand, I am in a position to know
that I am not a brain-in-a-vat,since my having a hand entails that I am not a
brain-in-a-vat.On this view, the rationalityof denying that I am a brain-in-a-vat
plays a role in my coming to know I have a hand. So the reasoningdoes not have
an MPF structure.Moreover, my knowledge that I am not a brain-in-a-vatis
based, in part,on my empiricalevidence (the evidence thatI have a hand), and so
is not a priori.
Though on this view my reasons do not have an MPF structure,one might
still object that the reasons have a circularstructure.For it looks as if I am nota
brain-in-a-vat is serving as (part of) my reason for I havea handand I havea
handis serving as my reason for I am not a brain-in-a-vat. But this is not quite
accurate.On this view, I amnota brain-in-a-vat has a degree of (non-evidential)
rationalitythatis independentof its being entailed by I havea hand.That degree
of rationalityis not sufficient, by itself, for me to know I am not a brain-in-a-vat.
But that degree of rationalityin conjunction with my empirical evidence is suf-
ficient for me to know I have a hand.And my knowing I have a handis sufficient
Contextualism,
Skepticism,
andReasons / 77
(1) Contextualism
as an errortheory
The fundamentalidea behind contextualism is that the truthconditions for
sentences of the form 'S knows P' include context-sensitive standards.According
to the contextualist treatmentof the skeptical paradox,competent speakers can
fail to be awareof these context-sensitive standards,at least explicitly, and so fail
to distinguish between the standardsthat apply in skeptical contexts, and the
standardsthat apply in everyday contexts. This misleads them into thinkingthat
certain knowledge ascriptions conflict, when in fact they are compatible. Con-
textualism thus combines a contextualist semantics for knowledge ascriptions
with a kind of errortheory a claim that competent speakers are systematically
misled by the contextualist semantics.
StephenSchiffer stronglydoubtsthe plausibilityof this story.Schiffer thinks
it is obvious that if ascriptions of knowledge are relative to context-sensitive
78 / Stewart Cohen
standards,then competent speakers will know this. And if they know this, then
they will not get confused about which standardsapply to their own ascriptions.
According to Schiffer
...speakerswould know what they were saying if knowledge sentences were indexical
in the way the Contextualistrequires.53
I think there is a very strong case for the claim that the truth-values of
knowledge ascriptionsare relative to context-sensitive standards.(It's not clear
whether Schiffer means to be denying this.54)First there are the cases like the
one I discuss in section (I) involving John and Mary at the LA airport.More-
over, as I argue in section (II), this kind of context-sensitivity seems to be
prevalent for naturallanguage predicates. Whenever a predicate can be satis-
fied to varying degrees, the degree to which the predicate must be satisfied in
orderto be satisfied simpliciterdepends on the standardsof the context. If this is
so, then ascriptionsof knowledge will be context-sensitive owing to the context-
sensitivity of justification.
So let us takefor grantedthe thesis thatascriptionsof knowledge arecontext-
sensitive in this way. Now I cite as a datum that many competent speakers of
English resist this thesis some fiercely.55Moreover, those who do accept the
thesis, generally do so only as a result of being convinced by philosophical re-
flection. It follows, contraryto what Schiffer asserts, that ascriptionsof knowl-
edge arerelative to context-sensitive standardsand yet many competentspeakers
do not know it at least not explicitly. And if there are these standards,but com-
petent speakersare unawareof them, then it is not surprisingthatspeakerscan be
misled by them.
But I need not suppose thatI am right aboutthe context-sensitivityof knowl-
edgeascriptionsto show that in general, there is nothing implausible aboutcom-
bining a contextualist semantics with an error theory. Consider ascriptions of
flatness. You can lead competent speakersto question their everyday ascriptions
of flatness by making salient "bumps"that ordinarilywe do not pay attentionto.
Taking this strategy to the extreme e.g., by calling attention to microscopic
irregularities one can lead competent speakers to worry whether anything is
really flat.56
Should we worrythatall along we have been speakingfalsely when we have
called things 'flat'? Surely not. Philosophical reflection will convince most that
ascriptionsof flatness arerelativeto context-sensitivestandards.57 Flatnesscomes
in degrees, and how flat a surface must be in order to count as flat simpliciter
depends on the context. Roads that count as flat in a conversation among Colo-
radans, do not generally count as flat in a conversations among Kansans. And
while one can truly ascribe flatness to a table in everyday conversations, one
might not be able to truly ascribe flatness to that same table when setting-up a
sensitive scientific experiment.If we implicitly raise the standardshigh enough
(by making salient microscopic bumps), then perhaps,relative to thatcontext, no
Contextualism,
Skepticism,
andReasons / 79
physical surface really is flat. But of course, thatdoes not impugnour ascriptions
of flatness in everyday contexts where the standardsare more lenient.
But then why can we get competent speakers to question their everyday
flatness ascriptionsby implicitly raising the standards?It must be that although
ascriptionsof flatness are context-sensitive, competent speakerscan fail to real-
ize this. And because they can fail to realize this, they can mistakenly think that
their reluctance to ascribe flatness, in a context where the standardsare at the
extreme, conflicts with their ascriptionsof flatness in everyday contexts.
So contraryto Schiffer's claim, competent speakerscan be unawareof, and
so misled by, the kind of context-sensitivity I have arguedis involved in ascrip-
tions of knowledge.
(2) Is contextualism
basedon a fallacy?
edge often differ (vis a vis the standards)from contexts where we investigate the
"extent"of *knowledge. In many epistemological contexts where we investigate
the nature and conditions of *knowledge, we are not concerned with skeptical
difficulties. Ratherwe are tryingto provide an analysis of the ordinaryconcept of
*knowledge. Here the standardsare the everyday ones and we rely on our every-
day intuitionsto guide us. We do consider what people would ordinarilysay, i.e.,
the ordinaryuse of sentences of the form 'S knows P', to be relevant to what we
are investigating.60If, as Sosa holds, we were concernedwith some non-ordinary
philosophical commodity, the intuitions of non-philosopherswould not be rele-
vant. So if Sosa were right, our practiceof soliciting the intuitions of undergrad-
uates, when we teach them epistemology, would not make sense. Instead we
should be introducing them to something new the philosophical notion of
*knowledge.
Now it is true that in contexts where we consider the extent of *knowledge,
we typicallyraiseskepticaldifficulties.In those contexts,the standardsfor *knowl-
edge ascriptionsare higher than the standardsfor *knowledge ascriptionsin ev-
eryday contexts. And as Sosa says, the contextualist'sclaim that sentences of the
form 'S knows P' are true in everyday contexts does not make contact with what
we say about *knowledge in those stricterphilosophical contexts. But this the
contextualistreadily concedes though crucially, the contextualistputs the em-
phasis in the other direction. For the point of contextualism is that what we say
about*knowledge in skepticalcontexts does not make contactwith our everyday
*knowledge ascriptions. Our inclination in skeptical contexts to deny that we
*know, does not conflict with our claims to *know in everyday contexts.
The way I see it, what is troublingand unacceptableabout skepticism is the
claim that all along in our everyday discourse, when we have been claiming to
*know, we have been speaking falsely. Contextualismattemptsto show how the
skeptical paradoxcan be resolved in a way that allows us to preservethe truthor
our everyday *knowledge ascriptions.This is not to say that a stricter *knowl-
edge of the sort thatcomes into play when we are thinkingabout skepticism is of
no interest. It's just that according to contextualism, we fail to have it. To this
extent, contextualism is a skeptical view. The point of contextualism is to give
skepticism its due, while blocking the troubling and unacceptableconsequence
that our everyday *knowledge ascriptionsare false.
According to contextualism, when I first think that I know, but then think about
skepticism and come to think that I don't know, I should look back on my previous
claim to knowledge and thinkthatit was correctas well. But I don't. I thinkthatI was
(or may have been) wrong.64
But the error theory component of contextualism predicts that one should
feel a conflict between one's skeptical inclinations and one's everyday knowl-
edge ascriptions. Feldman is assuming that competent speakers should not be
misled by shifts in context, that they should not mistakenlythink that their skep-
tical inclinations conflict with their everyday knowledge ascriptions.Again, the
point to be made in defense of contextualism is that some context-sensitivity is
such that competent speakersof the language are misled by it.
Again the analogy with flatness "skepticism"proves to be illuminating.As
we have noted, surely the controversy over whether anything is flat can be re-
solved by noting that ascriptionsof flatness are context-sensitive. We should not
say that we simply have cogent argumentsfor flatness skepticism and weighty
considerationsagainst it. But Unger's case for flatness skepticism is interesting
precisely because many who feel the pull of flatness skepticism look back on
their previous flatness ascriptionsand think they may have been wrong.
So the tendency of competent speakers to think that their skeptical inclina-
tions conflict with theireveryday knowledge ascriptionsdoes not count against a
contextualistinterpretationof those ascriptions.On the contrary,the errortheory
componentof contextualismpredictsthat competent speakerswill think there is
a conflict.
Assuming there are no serious problems with the contextualist model, is
there anythingreason to prefer it over Feldman's competing argumentsmodel?
As we have seen the contextualist model can explain the appeal of skeptical
arguments.On Feldman'scompeting argumentsmodel, what explains the appeal
of skepticism? It's just this: Despite the strong considerationsin favor of saying
we know, skeptical argumentsare quite cogent. Feldmanholds that "itis far from
obvious how to deal with skeptical arguments."
So is skepticismtrue?HereFeldmanjoins me in assumingthatskepticismcan
not be true,thoughI of course hold this only for everydaycontexts. So the advan-
tage of contextualismover Feldman'scompeting argumentsmodel is that,ceterus
paribus,thecompetingargumentsmodelis simplya statementof the skepticalprob-
lem, whereasthe contextualistmodel, if correct,is a solution to the problem.65
Contextualism,
Skepticism,
andReasons / 83
(4) Cancontextualism
be usedto supportskepticism?
The skeptic holds that all of our everyday knowledge ascriptionsare false.
We reject skepticism because we find it intuitively compelling that we know
things. But just as we use contextualismto explain away the appeal of skeptical
arguments,the skeptic could use contextualismto explain away the appealof our
everydayknowledge ascriptions.Why not say thatwhat is governedby context is
not when one can truly say that someone knows but ratherwhen one can appro-
priatelysay that one knows? On this pragmaticview, though all knowledge as-
criptionsare false, it can serve a useful function, in everyday contexts, to assert
thatwe know certainthings. This explains the appealof our everydayknowledge
ascriptions.Such a view could usurpthe entirecontextualistmachineryinterpret-
ing context as governing merely the appropriateness-conditionsfor knowledge
ascriptionsratherthan their truth-conditions.66
Is thereany argumentthatfavors the semanticversion of contextualismover
the skepticalpragmaticview? I thinkthe strongestargumentagainstthe skeptical
pragmaticview is precisely that it is a skeptical view. The advantageof the se-
mantic version of contextualism is that it enables us to avoid skepticism (about
everydaycontexts). And I take this to be a desirable result.
Does this argumentbeg the question against skepticism?Certainlyit does-
but again, no more than the skeptical pragmaticinterpretationbegs the question
againstcommon sense. I do not think either side of this dispute can demonstrate
the correctnessof its view to the other side. But if we are antecedentlyconvinced
of the falsity of skepticism, the semantic version of contextualism allows us to
explain away our own skeptical inclinations inclinations that give rise to the
skepticalparadox.And that is enough to recommendit.67
Notes
warranttransmission according to which one has the very same warrantfor not-H
as one has for P.
23. This seems to be a problemfor the so-calledprimafaciereasons view of justification
advocatedby John Pollock (1974)&(1986) On such a view, e.g., I can know the table
is red on the basis of it's looking red, even if I have no evidence against e.g., the table
being white with red lights shining on it. But then by closure if I know the table is red,
I can know the table is not white with red lights shining on it. But then any inductive
evidence I have againstthe possibility of such a deception turnsout to be irrelevantto
my knowing the deception doesn't obtain.
24. Klein (forthcoming)arguesthatthe "plausibility"and seemingly "compelling"nature
of the skeptical argumentstems from the various ways in which the closure principle
can be satisfied. (Simplifying slightly), he notes thatit can be satisfied by not-H being
evidentiallypriorto P, or by P being evidentially priorto not-H.Whereasthe skeptical
argumentassumes the formeris the case, in fact the latteris the case.
But this provides no explanationfor the appealof the skeptical argument.If P is
evidentially prior to not-H and we have sufficient evidence to know P, why do we
have the intuition that we don't know not-H? What tempts us to think that not-H is
evidentially prior to P?
In section (X) I take up this issue concerning evidential priority.
25. See Cohen (1988). I originally developed contextualism as a kind of relevant alter-
natives theory.David Lewis (1996) often speaks of contextualismin this way as well.
But I now think contextualismcan be presentedmore perspicuouslyif detachedfrom
the relevant alternativesmachineryand that is the approachI follow here.
26. One might think that contextualism gains some of its plausibility from the failure to
realize that it is, in part, a metalinguistic thesis. I respond to this concern in section
(XII)(2) of this paper.
27. But this is not to say that when the chance of erroris salient we inevitably shift to a
skeptical context. As I mentioned in note 12, though when skeptical alternativesare
salient, there is a strong upward pressure on the standards,this can sometimes be
resisted. One device for doing this is adoptinga certain tone of voice. So in response
to the skeptic, one might say, "C'mon, you've got to be kidding I know I am not a
brain-in-a-vat!".If this is the dominantresponse among the conversationalpartici-
pants, then everyday standardsmay remain in effect.
In many cases, we vacillate between skeptical and non-skepticalcontexts.
28. As Stine (1976) notes, to deny closure on this basis would be to commit a fallacy akin
to equivocation.
29. Sometimes the context-sensitivity for these kinds of predicates will be obvious to
competentspeakersand sometimes it will not. In the lattercases, such speakerscan be
misled. In section (XII)(1&3), I discuss objections to contextualism that says it's
implausible to suppose that competent speakerscould be misled in this way.
30. G.C. Stine (1976) proposes a similar way of respondingto the closure argument.In
Cohen (1988), I criticize Stine's particularway of saving closure relative to a context
and propose the view presentedhere.
31. My being a brain-in-a-vatis certainly compatible with the truthof many empirical
propositions.Strictly speaking, to make it an alternativeto these propositionsit must
include that the propositionis false. So where P is such a prop. the alt is not-P and I
am a brainin a vat being stimulated...asif P.
32. I first raised this problem in Cohen (1988). It has also been raised in conversationby
JonathanVogel, and by Hilary Kornblith,Michael Williams, and Peter Klein in dis-
86 / Stewart Cohen
cussion of a paperI gave at the 1998 SOFIAconference in Oviedo, Spain. Klein also
raises it in Klein (forthcoming).
33. This was suggested to me by David Lewis in conversation.
34. To say "itis rationalto some degree"means thatit is to some degree morerationalthan
not. Of course this rationalityis primafacie. If the vatmeisterwere to appearto me, it
would no longer be rationalto believe I am not a brain-in-a-vat*.
35. See, for example, Quine (1960),Harman(1986), and JonathanVogel (1990).
In Cohen (1988) I arguethat denying I am a brain-in-a-vat*is intrinsicallyrational.I
now characterizeit as non-evidentially rational partly in response to criticisms by
JonathanVogel (1993).
36. Have I now switched from talking abouta single scale of justification to two different
kinds evidentialandnon-evidential?Insofaras thepriorrationalityof denying skep-
tical alternativesis a necessary condition for knowing any empiricalproposition,the
non-evidentialrationalityis a componentof the overall rationalityorjustification for
any empirical proposition.
37. These cases do not fit the structureof the reference-fixingcases called to our attention
by Kripke.
As nearas I can tell, David Lewis (1996) is committedto this kind of contingent
a priori knowledge when he allows that you can know you're not a brain-in-a-vatby
properlyignoring it.
38. DeRose (1995).
39. I don't pretendthatthese notions of internalismand externalismarewell-defined. For
the best discussion I know of this distinction, see Fumerton(1996).
40. DeRose (1995), (p.3).
41. We are assumingthatthe brain-in-a-vathypothesis specifies thatmy actualbeliefs are
false, as it must to be an alternativeto those beliefs. See note 31.
42. Of course, I can not assertthatI know I am not a brain-in-a-vatwithouttherebyraising
the standards.This is also a consequence of my view in Cohen (1988). Lewis stresses
this same consequence of his view in Lewis (1996).
43. Presumably,my belief that I am not a brain-in-a-vatwill be causally based on my
experience.But given thatmy experiencedoes not count as evidence againstmy being
a brain-in-a-vat,my belief is not evidentially based on my experience. That is the
relevant sense in which the knowledge is a priori.
44. DeRose himself characterizesthe denial of the principle as "abominable"and "intu-
itively bizarre"andtakesgreatpainsto distinguishhis view fromNozick's andDretske's
on this point. (1995) p.38-39.
45. Both Kripke, in unpublishedlectures, and Fumerton(1987), present cases with this
kind of structureagainst Nozick's (1981) tracking view. These cases show that the
trackingview leads to instances of closure failure that presumablyare unacceptable
even for friends of closure failure like Dretske and Nozick. My aim here is to show
that althoughDeRose's contextualizedversion of Nozick's view preserves closure in
the cases involving skeptical alternatives, it still leads to these other instances of
closure failure. JonathanVogel presents a similar kind of case against DeRose in
"Subjunctivitis"(forthcoming).
46. DeRose, at one point, says "truth-tracking" is one component of epistemic strength.
But he doesn't mention anything else and he takes truth-trackingthat meets the
standardsof the context to be sufficient for my knowing, in that context, that I am
not a brain-in-a-vat.This is crucial for his response to skepticism. So if DeRose
says truth-trackingthat meets the standardsof the context, is not sufficient for me
andReasons / 87
Skepticism,
Contextualism,
to know, in that context, that there was a 6pm drawing, he owes us an account of
the difference.
47. The issues here are somewhatmore complicatedthan I have made them out to be. For
one thing, DeRose (1998) following Nozick notes thatwhen we apply the subjunctive
test, we must hold the method of belief forming fixed. Now it is somewhat obscure
what the method is in this case. But it seems reasonable to say that the method by
which I form my belief in this case, is the method of inferringfrom (the newspaper's)
testimony.But if thatis the method, then we do hold the method fixed when we apply
the subjunctivetest. If my belief were false, I would still have formed it by this same
method.It'sjust thatthe newspaperactuallysays therewas a 6pm drawingandcounter
factually says there was a 5pm drawing.
48. He gave this response in a comment on my paper "Contextualismand Skepticism"
(forthcoming) at the 1998 SOFIA conference in Oviedo, Spain.
49. On my view, so long as the occurrenceof such deceptions is infrequent,when I am at
the zoo, I have sufficient (in everyday contexts) inductive evidence to know that such
a deception is not occurring.
50. Of course it is hardto know what counts as (even very minimal) evidence in favor of
a hypothesis. Would the occurrence of misprints in books or other written material
count as evidence in favor of the atlas containing a misprint?How about the very
generalfact thatthere are sometimes mistakes in the attemptsto transmitinformation
(throughwhatevermedium). Of course insofar as we count these sorts of generalfacts
as evidence we createproblemsfor applying the criterionto the Zebracase. For more
criticisms of Klein's view see Brueckner(forthcoming).
51. We must take care to distinguish this case from a case where I know directly that the
sentence does not contain a misprint on the basis of inductive evidence against the
possibility of a misprint. For the point of the MPF structure,is that in some cases
where H in alternativeto P, I can infer P on the basis of evidence that does not count
against H, and therebycome to know P. Then on the basis of knowing P, I can come
to know not-H.
52. This kind of bootstrappingreasons structurewould be permissible only in the special
case where our reasons are partly non-evidential.
53. Schiffer (1996) p.328. The characterizationof it as an errortheory is Schiffer's.
54. He does concede there is "contextualvariability"but only in the "dilationand con-
striction"of the penumbraof vague terms. See note 7 of this paper.
55. I have personally witnessed this resistance on many occasions.
56. Unger (1975).
57. Lewis (1979), Dretske (1981), Unger (1984).
58. Sosa (forthcoming)I cannot,in the space I have here, do justice to the many important
and subtle points Sosa raises. I think that what I say responds to the spirit of his
remarkseven if I neglect some of the details.
59. Sosa (forthcoming).
60. As Sosa notes, some ordinaryfolk will say things like "Themedievals knew the Earth
was flat." He cites this as evidence that our everyday assent to sentences of the form
'S knows P' is not relevant to our philosophical concerns. But even if, as Sosa notes,
the OED allows this as a correctuse of 'know', most of those same ordinaryfolk will
concede that this is loose talk, i.e., not what we ordinarilymean by 'know'.
61. Feldman (forthcoming). Feldman'spaper is very comprehensive and contains many
subtle points to which I can not do justice in the space I have here. I hope that my
remarksrespondto the main points of his paper.
88 / Stewart Cohen
References