Defending the Ignorance View of Sceptical Scenarios

Defending the Ignorance View ofSceptical Scenarios

Tim KraftUniversität Regensburg

[email protected]

forthcoming in International Journal for the Study of SkepticismFebruary 17, 2014

Abstract

What is the role of sceptical scenarios – dreams, evil demons, brains in a vat – in

sceptical arguments? According to the error view, sceptical scenarios illustrate the

possibility of massive falsity in one’s beliefs, whereas according to the ignorance

view, they illustrate the possibility of massive ignorance not necessarily due to falsity.

In this paper the ignorance view is defended by surveying the arguments in favour

of it and by replying to two pressing objections against it. According to the first

objection, the ignorance view illicitly introduces the KK-principle into sceptical

arguments. In reply I argue that KK is not less plausible than its main rival, the

closure principle. According to the second objection, relying on veridical ignorance-

possibilities contradicts the transparency of belief. In reply I introduce a version of

transparency that is consistent with the ignorance view.

Keywords

Cartesian scepticism, sceptical scenarios, brains in a vat, error-possibilities, ignorance-

possibilities, ignorance view, epistemic closure, KK-principle

1

1 Introduction

Cartesian sceptical arguments are sceptical arguments that rely essentially on sceptical

scenarios like the dreaming, the evil demon or the brain in a vat scenario. These are meant

to be possibilities one cannot, but need to rule out in order to know anything about the

external world (or the past or other minds). But the possibility of what are these scenarios

meant to illustrate? According to the error view, sceptical scenarios are error-possibilities,

that is, possibilities in which the target beliefs are false. According to the ignorance view,

sceptical scenarios are ignorance-possibilities, that is, possibilities in which the target

beliefs fall short of knowledge and do so not necessarily because they are false in the

scenario. Against a long-standing consensus that sceptical scenarios are error-possibilities

proponents of the ignorance view argue that they are ignorance-possibilities instead

(Winters 1981, Beebe 2010, 2011, Schaffer 2010, Murphy 2013 and me in Kraft 2012,

2013).1 In the first part of this paper, I survey the arguments for the ignorance view

that are brought forward at various places in the recent literature on sceptical arguments

(section 2). In the second part, I reply to two pressing objections against it (sections 3

and 4):

1. If sceptical scenarios are ignorance-possibilities, the inability to rule them out entails

scepticism only if the KK-principle is true. However, this principle is implausible

and by tying the sceptical argument to it the ignorance view turns an argument

that is intuitive to at least some extent into an argument that is not even prima

facie compelling (Brueckner 2011, Murphy 2013, Vahid 2013).

2. The ignorance view cannot explain how sceptical scenarios threaten their target

beliefs. For example, the possibility of being an envatted brain cannot threaten my

1 The ignorance view is at least foreshadowed by Stroud (1984: ch. 1). Hetherington (1996) and Reed

(2009) adopt it in the context of ‘Gettier-style’ sceptical arguments to which ignorance-possibilities are

crucial.

2

belief that I have a brain because the inability to rule out this possibility is not a

reason to revise my belief. Suggesting otherwise contradicts the transparency of

belief (Ballantyne and Evans 2013).

2 Defending the ignorance view

Are sceptical scenarios possibilities in which the target beliefs are false, as the error view

claims, or possibilities in which they fall short of knowledge, not necessarily because

they are false, as the ignorance view claims? The arguments put forward in favour of the

ignorance view fall into two groups. While the first group concludes that the ignorance

view provides a better understanding of how sceptical scenarios threaten empirical beliefs

(Winters 1981, Kraft 2012, 2013, Murphy 2013), the second group concludes that the

ignorance view is needed to explain how sceptical scenarios can threaten apriori or

necessary beliefs (Beebe 2010, 2011, Schaffer 2010). The first group is the topic of the

next two subsections and the other of the third subsection.

2.1 Problems of the error view

The argument in favour of the ignorance view begins with an observation due to Moore

and Wittgenstein. Both draw attention to the possibility of veridical dreams: It is possible

to dream that one is holding a speech while doing so (Moore 1959: 245) or to dream

that it is raining while it is raining with the sound of the rain even causing one’s

dream (Wittgenstein OC: § 676). This observation is not limited to dreams. The generic

descriptions of the three scenarios leave it entirely open how the world surrounding the

protagonist is like. Apparently, all empirical beliefs can be false or true in each of them.

This observation is commonplace in the literature on Cartesian scepticism (cf., for

example, Stroud 1984: 25–30, DeRose 1995: 32 fn. 33, Pryor 2000: 527, Vogel 2004: 435–

438). In itself it is insufficient to motivate the ignorance view. For even if dreaming

is compatible with truly believing that it is raining, it is also compatible with falsely

3

believing that it is raining. Since the three scenarios are generic possibilities, they can

be augmented in as many ways as needed. For example, if I happen to believe that it is

raining, the sceptical scenario can be augmented with the additional detail that the sun

is shining. If I happen to believe that the sun is shining, it can be augmented with the

additional detail that it is raining.

But the case for the ignorance view does not end here. There are three reasons why

such augmentation is bound to fail at some point. To begin with, the truth of some target

beliefs is entailed by the description of the sceptical scenario (Kraft 2013: 64, Murphy

2013: 276 fn. 9). No matter which scenario is chosen, some beliefs are not false in them

because their truth is essential to the scenario. In the case of the brain in a vat scenario

this includes at least the beliefs that I have a brain and that there are computers. In the

case of the evil demon scenario this includes the beliefs that other minds exist, intentional

deception occurs, and so on. To argue that these beliefs are exempt from sceptical doubt

is ad hoc. Why should my belief that there are airplanes, but not my belief that there are

computers be susceptible to sceptical doubt? If sceptical scenarios were error-possibilities,

scepticism would not target all empirical beliefs, but only the arbitrary subclass of those

empirical beliefs that are false in the chosen sceptical scenario.

Second, some target beliefs are necessarily true (if true) (Kraft 2013: 65, Murphy 2013:

276). If I am actually human, it is impossible that my belief that I am human is false.

Again, to argue that these beliefs are exempt from sceptical doubt is ad hoc: Why should

my belief that I am human, but not my belief that I have hands be exempt from sceptical

doubt? Again, if sceptical scenarios were error-possibilities, scepticism would not target

all empirical beliefs, but only contingent empirical beliefs.

Third, even if any target belief could be false, they cannot all be false at the same time.

Even if my beliefs are consistent, their negations are not. My beliefs that I have at least

one hand and that I have less than seven hands, my beliefs that there are animals and

that there are no unicorns, and so on are pairs of beliefs that cannot both be false. If

4

the scenario is changed so that I am a seven-handed creature, my belief that I have a

hand turns out to be true. If the scenario is changed to include unicorns jumping around

the vat, my belief that there are animals turns out to be true. Hence, no matter which

particular scenario is chosen, it will always turn out that some target beliefs are true.2

To sum up, no sceptical scenario is a possibility in which all of the target beliefs are false.

Some beliefs are true because their truth is entailed by the description of the scenario.

Some beliefs are true because they are necessarily true. Some beliefs are true because it is

impossible that all beliefs are false. Hence, if sceptical scenarios are meant to put pressure

on their target beliefs by illustrating the possibility of their falsity, they cannot succeed.3

2 For a discussion of this point as an argument against the error view cf. Müller 2003: 45–47 and Kraft

2013. It is also discussed by Gemes 2009, 2010 and Genova 2010, but not as an argument against the

error view. This argument also shows that not even most empirical beliefs of an envatted brain must be

false. Suppose the envatted brain is an astute logician and pairs every empirical belief with a carefully

chosen subcontrary belief; at least half of her empirical beliefs must be true. Therefore, it would not

save the error view to argue that most beliefs are false in a sceptical scenario. For this reply would

entail that, absurdly, the astute logician was safe from sceptical attacks.

3 Vahid (2013: 247–249, cf. Vogel 2004: 437 f.) replies that sceptical scenarios are indirect error-possibilities,

that is, possibilities in which every target belief is either false (direct falsity) or easily might have been

false (indirect falsity). However, it is questionable whether every belief of, for example, an envatted

brain might easily have been false. Suppose an envatted brain in an otherwise empty universe believes

that all firefighters are courageous. This belief might not easily have been false in that possible world.

For this belief is false only if there is at least one timid firefighter, but in an almost empty universe a

firefighter might not easily have existed. Vahid also suggests that sceptical scenarios that are targeted

at necessarily true beliefs are also best understood as indirect error-possibillities. They would be

possibilities in which, say, the belief that modus ponens is deductively valid is “true only by luck”

and, if so, “could have been easily false” (2013: 249). This claim is ambiguous: On the one hand, such

beliefs cannot be true by luck, because as necessarily true beliefs they cannot be false. On the other

hand, they may be said to be true by luck only insofar as the believer easily might have believed a

different proposition. However, the latter kind of luck neither precludes knowledge nor can it be used

to construct error-possibilities for necessarily true beliefs. That someone might easily have believed

something else does not show that her actual belief easily could have been false.

5

2.2 Single scenario vs. multiple scenarios

Proponents of the error view may attempt a comeback: A common way of presenting

the Cartesian sceptical argument is the argument from ignorance (cf. DeRose 1995: 1,

SH stands for the hypothesis that the chosen sceptical scenario obtains, P for any target

proposition):

(1) I do not know that ¬SH.

(2) If I do not know that ¬SH, I do not know that P .

(3) ∴ I do not know that P .

There is an ambiguity hidden in this way of formulating the sceptical argument. SH

can be a placeholder for the particular hypothesis that a particular sceptical scenario

obtains or a variable ranging over a multitude of sceptical hypotheses. The distinction

between single and multiple scenario arguments makes these two readings explicit: Single

scenario arguments establish that all target beliefs fall short of knowledge by arguing

that there is one determinate scenario that I cannot but need to rule out. By contrast,

multiple scenario arguments do so by arguing that for each target belief there is a (possibly

different) sceptical scenario that I need to but cannot rule out.4 A multiple scenario

argument attacks, for example, the belief that I have two hands with the scenario of being

a handless brain in a vat, the belief that I have less than seven hands with the scenario

of being a seven-handed brain in a vat, the belief that intentional deception occurs with

4 As far as I know, only Cohen (1998: 155), Dodd (2012: 339), Pritchard (2012: 111) and Zalabardo

(2012: 11) explicitly allow the sceptical argument to rely on multiple scenarios. Dodd also suggests that

the sceptical hypothesis should be the disjunction that either sceptical scenario A obtains or sceptical

scenario B obtains and so on. But note that this proposal (when combined with the error view) leads

to multiple hypotheses that each contains multiple scenarios. Since the sceptical hypothesis must be

incompatible with the target belief and for a disjunction to be incompatible with the target belief each

disjunct must be incompatible with the target belief, there cannot be a single disjunctive sceptical

hypothesis for all target beliefs either.

6

being a brain in a vat in an otherwise empty universe, the belief that there are computers

with the evil demon scenario, and so on. The multiple scenario argument is congenial to

the error view: Altering the scenario relative to the belief under attack allows the error

view to bypass the problems from the last subsection.5 Thus, the debate between the

error and the ignorance view turns into a debate about whether the sceptical argument is

required to rely on a single, unchanging scenario. Since the argument for the ignorance

view depends on this requirement, my task is to defend it. I do so by arguing that giving

up the single scenario requirement leads into three serious difficulties.

2.2.1 The circumstantial evidence problem

The first problem is the circumstantial evidence problem (cf. Kraft 2013: 70). It is commonly

assumed that one cannot rule out a sceptical hypothesis by citing circumstantial evidence,

for example by arguing that no current computer is advanced enough to run the simulation

or that envatting a human brain incurs too many financial and legal risks. Citing this

as evidence is impermissible because these beliefs are themselves called into question by

the sceptical scenario. However, the error view lacks the resources to explain why these

background beliefs are disqualified as evidence. On the error view, all and only those

beliefs that are false in the scenario are under attack and may therefore not be cited as

evidence. As a consequence, the error view cannot block citing circumstantial evidence

whenever a sceptical scenario leaves too many beliefs true and thus unscathed.

As an illustration consider my belief that it is not the case that there are snakes in

my office. According to the error view, I do not know that this is so because I cannot

rule out being an envatted brain whose office contains at least one snake. In order to turn

this into a comprehensive error-possibility more details need to be added. For example, it

must be ensured that some room is the envatted brain’s office. That some room looks like5 Multiple scenario scepticism does not allow the error view to bypass the problem of necessarily true

beliefs. However, restricting the scope of sceptical arguments to contingent beliefs is a price defenders

of the error view might be willing to pay.

7

(what I take to be) my office is certainly not enough for a room to be the envatted brain’s

office. An error-possibility that would do the trick is the scenario of being a recently

envatted brain to whose office someone brought a snake. This would ensure both that

my office exists and that the target belief is false. However, then, I could rely on a lot

of circumstantial evidence to argue both in favour of my particular belief (snakes are

uncommon where I live, snakes do not climb easily to the third floor, nobody has a reason

to bring a snake to my office, the door of my office is closed when I am not in, and

so on) and against recent envattment (no computer is fast enough, no motivation for

envatting me, and so on). All of the cited evidence is true in the error-possibility under

consideration. Given the error view, there is no reason to disqualify these considerations

from being evidence. Of course, there are various good reasons for disqualifying these

propositions as evidence, but none of them fits well with the error view. For example, if a

proponent of the error view argues that she can easily present error-possibilities for the

alleged evidence for my belief that there is no snake in my office, she does not attack

my belief directly by offering an error-possibility, but only by offering a possibility in

which I do not know that this is so because my evidence for that belief is false; in other

words, by offering an ignorance-possibility for the target belief. To sum up, whenever a

sceptical scenario leaves too many beliefs true, the error view cannot prohibit relying on

them as circumstantial evidence. Therefore, the proponent of the error views owes us an

explanation of how to adjust the standard scenarios in a way that does not result in too

many true beliefs. This is a daunting task and it is difficult to see how it could be met.6

6 Briesen argues against citing circumstantial evidence by altering the brain in a vat scenario. In his

version the supercomputer not only feeds sense experiences, but also manipulates all memories (2011:

575 f.). Applying this idea to my example the new error-possibility is that I was recently envatted, but

spent my previous life in a millennium when supercomputers are cheap in a country where snakes are

common, and so on. The supercomputer manipulates my memory so that I seem to remember that, for

example, snakes are uncommon near my office although this is in fact false. In this error-possibility

both the target belief and all background beliefs are false and, therefore, the error view can explain

why I may not rely on the latter.

8

This sceptical scenario does not support the error view’s cause because it violates an important

requirement on sceptical scenarios. It is important to distinguish between experiential (or episodic)

and propositional (or semantic) memory (cf. Bernecker 2010: ch. 1 for a survey of these and other

distinctions). In a nutshell, experiential memory is accompanied by qualitative experiences whereas

propositional memory consists merely in the retrieval of stored propositions. To manipulate experiential

memory the computer could feed the brain experiences that deceive the brain into believing that (it

remembers that) P although it is false that (it remembers that) P . Of course, this possibility raises

questions of its own (sense experiences can be fed through the cranial nerves, but there are no nerves

for memorial experiences), but that they can be met should be granted for the sake of the argument.

However, the circumstantial evidence cited in the example does not involve experiential memory and,

therefore, the possibility of manipulating experiential memory is irrelevant to this kind of example.

Unlike manipulating experiential memory manipulating propositional memory should be disallowed.

To manipulate propositional memory the supercomputer would need to manipulate the beliefs and

not the experiences of the envatted brain. In the original scenario the supercomputer feeds unreliable

sense experiences, but the belief itself is attributable to the envatted brain: The brain is not forced into

believing anything and, thus, it is the brain that does the believing. In Briesen’s scenario, however, the

supercomputer interferes with the envatted brain’s beliefs and those beliefs are no longer attributable to

the envatted brain (for a related notion of attributability cf. Reed 2007: 237–245). Cartesian sceptical

scenarios are to be distinguished from scenarios like being mad, being the victim of an evil demon who

directly imprints false beliefs or having a malfunctioning brain. In these scenarios the beliefs are not

attributable to the protagonist. But Cartesian scepticism does not argue that knowledge is impossible

because of the possibility of undetectable breakdowns of my cognitive capacities. Rather it claims that

even with idealised cognitive capacities (for example, no reasoning mistakes, comprehensive memory,

unlimited time) the knowledge in question is impossible. Unless one is willing to give up this distinctive

tenet of Cartesian scepticism, sceptical scenarios may not interfere with the protagonist’s cognitive

capacities in a way that undermines attributing the target beliefs to the protagonist of the scenario.

Note that I am not arguing that scenarios with internally impaired cognitive capacities are useless or

uninteresting; I am arguing that they lead to a different kind of scepticism that has nothing to do with

the (alleged) gap between mind and world but instead rests on the (alleged) impossibility to assure

oneself that one’s own mind is working properly ‘on the inside’.

9

2.2.2 The conjunction problem

A second problem is the conjunction problem (cf. Roush 2010, Atkins and Nance 2013).

The error view is committed to a bipartite conception of sceptical scenarios according

to which they consist of a generic part (e. g. being a brain in a vat) and an additional

stipulation (e. g. not having a hand): A sceptical scenario for the belief that P is being a

brain in a vat and ¬P .7 The sceptical scenarios cannot be known not to obtain due to

what is common to them, but they are error-possibilities not due to what is common to

them, but due to the specific additional stipulation. The bipartite conception of sceptical

scenarios is needed to support the two premises of the sceptical argument. The first

conjunct is meant to ensure that I cannot know whether the possibility obtains and the

second conjunct is meant to ensure that I need to know whether it obtains.

Splitting sceptical scenarios into two parts that play different roles is not as innocent as

it may seem.8 Incompatibility is preserved when a conjunct is added, but unknowability

7 The additional stipulation need not be ¬P , but may also be some Q that entails ¬P (this is an option

apparently overlooked by Atkins and Nance). For example, the bipartite sceptical scenario for the belief

that I have at least one hand may also be I am a brain in a vat all of whose hands are clean and not

clean, I am a brain in a vat in a world in which no limbs exist or I am a brain in a vat with less hands

than there are even prime numbers and so on. Since this option does not affect the need to understand

sceptical scenarios as being bipartite scenarios, I ignore it in the main text.

8 Roush (2010) and Atkins and Nance (2013) argue for the stronger claim that the bipartite conception

undermines the sceptical argument immediately. Roush does so by presenting a dilemma: Vis-à-vis

the possibility of being a handless brain in a vat there is no reason to give up the knowledge claim

that I have hands. That such specific possibilities do not obtain is “just not much to know” since

knowing this is neither difficult nor significant. And vis-à-vis the possibility of being any kind of

brain in a vat there is no reason to give up the knowledge claim either: Without incompatibility the

inability to rule out being a brain in vat poses no threat. Atkins and Nance (2013) argue that with

bipartite sceptical scenarios the sceptical argument is either question-begging, a transmission-failure or

structurally inefficient. The argument for the premise that I do not know that not (I am a brain in a

vat and ¬P ) must already presuppose its conclusion that I do not know that P . For, if I knew that P ,

I would, or easily could, know that not (I am a brain in a vat and ¬P ). Both versions of the objection

10

is not trivially preserved. Incompatibility is preserved because if the possibility that A is

incompatible with the target belief that P , the possibility that A∧B is incompatible with

the target belief as well. But unknowability need not be preserved when a conjunct is

added. Let us define that the hypothesis that A is unknowable iff it is impossible to rule out

that A, that is, iff one does not know that ¬A. Unknowability is not preserved whenever a

conjunct is added because even if A is unknowable, A∧B need not be unknowable as well.

To illustrate the problem consider an ordinary unknowable proposition: I cannot rule out

the possibility that the fly on my desk is moving whenever no one observes it. But this

does not entail that I cannot rule out the possibility that the fly is moving whenever no

one observes it and I have seven hands. Although this conjunctive possibility contains an

unknowable conjunct, it is quite easy to know that it does not obtain because I can rule

out its second conjunct (or, to be more precise, no reason is given for thinking otherwise).

It is tempting to reply that matters are different when dealing not with ordinary

unknowable propositions but with sceptical scenarios (cf. Avnur, Brueckner and Buford

2011: 445). One way of spelling out this reply is the repetition argument : Whatever

argument shows that I do not know that I am not a brain in a vat can be repeated to

show that I do not know that it is not the case that I am a brain in a vat and ¬P . For

example, suppose that the argument for the unknowability of being a brain in a vat is

based on the sameness of evidence lemma. The very same argument, or so it seems, can

be applied to the scenario of being a brain in a vat and ¬P . For example, if a real human

and a brain in a vat have the same evidence, then a real, handed human and a handless

brain in a vat have the same evidence, too.

are too quick. Roush does not explain why knowing that one is not a handless brain in a vat should

be easy. It is by no means obvious why knowing this should be less difficult than knowing that one is

not a brain in a vat (cf. Avnur, Brueckner and Buford 2011). Atkins and Nance do not explain why

arguing that I do not know that I am not a handless brain in a vat must rely on the premise that I do

not know that I have hands. Whether that is so depends on how the premise is actually defended, not

on its logical form.

11

There are three reasons for being sceptical about the repetition argument: To begin

with, adding a conjunct to the scenario often deprives the sceptical argument of its

intuitive power. Even if being a brain in a vat is a serious possibility, we need not admit

that being a seven-handed brain in a vat (error-possibility for my belief that I have less

than seven hands) or being a magically gifted brain in a vat (error-possibility for my belief

that I am not magically gifted) or being a brain in a vat that spent its teenage years on a

17th century pirate ship as a cook (error-possibility for my belief that I did not do that)

are serious possibilities. Certainly, if I am a brain in a vat, I am not such a highly specific

kind of brain in a vat! These possibilities are too outlandish and far-fetched to be worth

taking seriously.9

Second, as we have seen in the fly example, unknowability is not always preserved

when an ordinary proposition is added conjunctively. Consider some examples in which

sameness of evidence preservation fails: For example, I believe that my brain contains

more than 100 neurons, that I cannot observe distant stars without a telescope, that there

are computers that can compute ten digits of π per second, that not all virtual realities

are monochromatic. However, if I were a brain in a vat and one of those beliefs was false,

my evidence would be different from my actual evidence. The repetition argument has

obvious limits and, therefore, it must be explained which propositions are eligible as second

conjuncts. That task is not trivial. Depending on how one argues for the unknowability of

being in the generic sceptical scenario different propositions may be added to the generic

sceptical scenario.

Third, it is a common idea that ‘naked’ error-possibilities – “but your belief might be

false!” without an explanation of how or why the belief might be false – are ineffective

challenges: To amount to an effective challenge an error-possibility must also explain how or

9 This objection should especially worry those who think that the first premise of the argument from

ignorance is intuitively true. Even if it is intuitively true that I cannot know that I am not a brain in a

vat, there are target beliefs P for which it is not intuitively true that I cannot know the negation of

the sceptical hypothesis that is specific to the belief that P .

12

why the target belief is false. The repetition argument fails to show that bipartite sceptical

scenario meet this constraint. Even if the unknowability preservation problem could be

solved, the generic part of a bipartite sceptical scenario would still be explanatorily idle.

Sometimes the generic part helps to explain why the target belief is false. For example,

being a brain in a vat explains how I might have ended up being handless. But in many

cases being in the generic sceptical scenario does not contribute to an explanation of

how or why the target belief is false. Being a brain in a vat explains neither why it is

raining (although I believe that the sun is shining) nor why a snake ended up in my office

(although I believe that there is no snake in my office). Thus, if it is granted that ‘naked’

error-possibilities are ineffective, bipartite sceptical scenarios do not meet this constraint

because it is merely stipulated that the target belief is false. Of course, generic sceptical

scenarios explain something, for example they explain how it is possible that I believe

that it is raining independently of whether it is raining or not. But what is explained here

is not the potential falsity of my belief about the weather, but my potential ignorance

about the weather.

2.2.3 The collapse into fallibilism problem

A third problem is the collapse into fallibilism problem (cf. Kraft 2012, 2013: 69–71).

Both scepticism and fallibilism argue that any target belief can fall short of knowledge;

no target belief is guaranteed to be ‘good’. This thesis, however, exhibits a common scope

ambiguity: Is it possible that all beliefs are ‘bad’ at the same time? Or are just all beliefs

possibly ‘bad’? The scope ambiguity suggests a helpful way of distinguishing fallibilism

and scepticism: Scepticism asserts the first thesis (there is a possibility in which all target

beliefs fall short of knowledge), whereas fallibilism asserts only the second (for each target

belief there is a possibility in which it falls short of knowledge). What is distinctive of the

sceptical challenge is its unusually wide scope. It is threatening because, if the sceptical

scenario obtained, it would at once destroy a lot of knowledge. Fallibilism, however,

13

poses no such threat. It merely claims that any belief could fall short of knowledge,

but in no way suggests that a lot of beliefs could do so in one go. Therefore, fallibilism

allows a reassuring second-order thought: Since it is impossible that all of my beliefs are

‘bad’ at the same time, the worst that can happen in my epistemic affairs is that some

beliefs are ‘bad’. Of course, this thought does not entail that everything is fine from an

epistemological point of view and does not by itself amount to a positive, constructive

epistemological theory. Nevertheless, it is successful in shifting the burden of proof: If the

sceptical argument does not show more than the fallibilist thesis that some (but not all)

of my beliefs may turn out to be ‘bad’, scepticism is no more threatening than fallibilism.

The error view leads to multiple scenario scepticism and taken together they blur the

distinction between scepticism and fallibilism, whereas the ignorance view is consistent

with single scenario scepticism and can thus explain their distinctness: According to

fallibilism, for each target belief there is a (different) ignorance-possibility, but according

to scepticism, there is an ignorance-possibility for all target beliefs. Hence, the error view

establishes at most fallibilism. At least, it cannot rely on the scope distinction to explain

the difference between scepticism and fallibilism. In the absence of such an explanation it

is safe to object that it confuses scepticism with fallibilism.

2.2.4 Summary

To sum up, the error view is not rescued by giving up the requirement that Cartesian

sceptical arguments should be single scenario arguments. On the one hand, it is unlikely

that there is an appropriate sceptical scenario for each empirical belief (first and second

objection). On the other hand, even if there were, the error view needs an account of how

such scenarios help to establish scepticism instead of fallibilism (third objection).

14

2.3 The ignorance view and universal scepticism

The standard sceptical scenarios are used to attack (subclasses of) empirical knowledge,

but not conceptual knowledge, mathematical knowledge, self-knowledge, moral knowledge,

and so on. Beebe and Schaffer adopt the ignorance view in an attempt to revive scepticism

about apriori knowledge (Beebe 2010: 455–457, 2011) or even universal scepticism (Schaffer

2010). The driving thought, in outline, is that it is sheer prejudice to exclude those beliefs

from sceptical doubts. If scepticism is to be taken seriously, it is to be taken seriously in

all areas of knowledge. Hence, the error view is inadequate because it lets a lot of beliefs

off the hook without offering a convincing rationale for doing so.

The arguments for the ignorance view presented so far are independent of the issue

of whether there are convincing Cartesian arguments for apriori or universal scepticism.

However, the two lines of reasoning for the ignorance view complement each other: If the

ignorance view turns out to be the most plausible interpretation of the role of sceptical

scenarios in arguments for scepticism about empirical knowledge, it will be no surprise that

apriori or universal scepticism can and should rely on ignorance-possibilities, too. Thus,

the main issue concerning the viability of apriori and universal scepticism is not whether

scepticism may operate with ignorance-possibilities, but rather whether there actually

are convincing ignorance-possibilities for apriori or even all beliefs. Let us look briefly

at Beebe’s and Schaffer’s proposals for such scenarios: Both proposals have in common

that an evil demon feeds its victim non-sensory experiences. Beebe’s bumbling demon

(2013: 16 f., 2010: 456 f.) feeds rational intuitions. By mistake, it feeds true intuitions, for

example the intuition that modus ponens is deductively valid. Therefore, its victim does

not know that modus ponens is deductively valid although the belief is true. Schaffer’s

debasing demon first forces its victim into acquiring beliefs by guesswork (or in other

improper ways) and later feeds its victim the impression that the belief was properly

based. Again, its victim lacks knowledge no matter whether the belief is or even can be

false.

15

Both scenarios depend on controversial assumptions. Beebe’s scenario explicitly depends

on a rational intuitions account of apriority. The justificatory role of rational intuitions for

apriori beliefs is supposed to be similar to the role of sense experiences for empirical beliefs

and, consequently, in his scenario they are fed in ways similar to how sense experiences

are fed in the standard sceptical scenarios. Unsurprisingly, if rational intuitions play a

similar role to sense experiences, scepticism ensues. Schaffer’s scenario depends on an

impressions accounts of some beliefs, too. The debasing demon must make it seem to

its victim that a belief is properly based although it is not. Supposing that the demon

may not plant the second-order belief directly, this can only be achieved by feeding

impressions of properness. Unsurprisingly, if properness impressions play a similar role to

sense experiences, scepticism ensues.10

Thus, it seems that the scope of Cartesian scepticism depends on the applicability of

the experience/impression model to non-empirical beliefs. Fortunately, since the ignorance

view is not bound to universal scepticism, proponents of the ignorance view need not

decide whether this is a plausible model.

10 Furthermore, Schaffer entertains the possibility of having debased beliefs without recognising this only

with respect to beliefs one has already acquired. He does not argue that a belief that is produced

now on an improper basis can now seem to be properly based. This restriction is contained in one of

Schaffer’s three key assumptions:

“(3) It is always possible, when a belief is produced on an improper basis, for it to seem

later as if one had produced a belief properly based on the evidence.” (2010: 232, emphasis

added)

Many apriori beliefs, although originally acquired in the past, are epistemically overdetermined: I do

not only remember them, but can also reconvince myself of their truth. For example, even if I originally

came to believe the Pythagorean theorem by guessing, I can now base my belief properly by proving it.

Of course, there are also apriori beliefs I now only remember, but paradigmatic, basic apriori beliefs

are unlikely to be among them. This limits the scope of Schaffer’s scenario considerably. Apparently,

his scenario poses a threat only for merely remembered beliefs and, thus, only fuels scepticism about

memory.

16

2.4 The ignorance view and the structure of the sceptical argument

Adopting the ignorance view has important consequences for the reconstruction of the

Cartesian sceptical argument. It allows interpreting them as single scenario arguments.

The single sceptical scenario can be, for example, being a brain in a vat (since the

beginning of one’s existence). No further stipulations about the world besides the vat

and the supercomputer are needed. Such a brain in a vat does not have any empirical

knowledge, in particular an envatted brain knows neither that computers exist nor that it

has a brain although these beliefs are true (cf. Kraft 2012: 61 and 2013: 65 f. for reasons

why envatted brains do not know that they have a brain).

The reasoning behind the first premise of the sceptical argument – I do not know

that I am not a brain in a vat – is not affected by the ignorance view. If the ignorance

view is adopted, the reasoning can be any of the arguments familiar from the literature

(indistinguishability, sameness of evidence, no sensitivity and so on). Yet even regarding

the first premise adopting the ignorance view is not irrelevant. It releases from the task

to explain of a multitude of slightly different scenarios why one cannot rule them out and

helps to explain why relying on circumstantial evidence is prohibited because all empirical

beliefs are called into question at once.

The reasoning behind the second premise – I do not know that P , if I do not know that

I am not a brain in a vat – is, however, considerably different if the ignorance view is

adopted. The error view usually defends it by appealing to Closure:11

Closure: If S knows that P and S knows that P entails Q, S knows that

Q.12

11 In this paper I ignore the debate about whether Closure or the underdetermination principle offers a

better support of the sceptical argument’s premises. Both arguments presuppose the error view and,

therefore, the differences are negligible for present purposes.

12 This formulation is only a preliminary one. It fails because of the problem of “not putting one and one

together”: Someone may fulfil the antecedent, yet not know that Q just because she does not believe

17

The connection between Closure and the error view can be made more explicit by restating

Closure as the principle of excluded error-possibilities: Knowledge requires knowing of

all known error-possibilities that they do not obtain. If sceptical scenarios are not error-

possibilities, a stronger principle is needed. An obvious candidate is the principle of

excluded ignorance-possibilities: Knowledge requires knowing of all known ignorance-

possibilities that they do not obtain.13 This principle is both close enough to Closure and

strong enough to be a promising candidate for the principle needed to get the sceptical

argument off the ground. Restating this principle yields:

KK*: If S knows that P and S knows that knowing P entails Q, S knows

that Q.

With the assumption that S knows that knowing P entails knowing P , this version of the

KK-principle entails the more common version of the KK-principle according to which

knowledge always iterates:

KK+: If S knows that P , S knows that she knows that P .

With the assumption that Closure is true, KK+ entails KK*. Since the two versions of

the KK-principle are closely related to each other, I use “KK” as a term that covers both

versions.

that Q. This problem can be circumvented by either weakening the consequent (‘position to know’) or

strengthening the antecedent (‘correctly deducing’). The formulation of KK* below faces the same

problem, but whatever amendment is chosen for Closure can and should be applied to KK* as well.

13 Of course, the fact that the second premise can be defended by relying on a version of KK does not

entail that it must be defended this way. For example, it might be argued that knowledge only requires

ruling out all relevant ignorance-possibilities, but that the brain in a vat scenario is among the relevant

ones. This strategy requires highlighting a distinctive feature of sceptical scenarios that accounts for

their relevance, for example their unusually wide scope (the strategy is mentioned by Pryor 2000:

526 f. and defended by me in Kraft 2012). The defence of the ignorance view offered in this paper is

compatible with this particular version of the ignorance view, but does not depend on it.

18

3 First objection: KK

The mere fact that the ignorance view depends on a version of the KK-principle is not

much news. But is this reliance illicit? Does it rob the sceptical problem of its seriousness?

Prima facie, introducing KK into the sceptical argument leads to an impasse: On the one

hand, introducing KK seems to be inevitable because of the arguments for the ignorance

view surveyed so far. On the other hand, relying on KK is a serious weakness of the

sceptical argument. Replying to the sceptical argument would become quite easy: Just

reject KK, a principle that is implausible anyways. This objection is raised – in different

versions and contexts – at least by Vogel (2004: 436), Luper (2007: 157 f.), Brueckner

(2011: 297), Murphy (2013: 277–279) and Vahid (2013: 244).

Let me review the dialectical situation: To begin with, the demise of the ignorance

view would not by itself reinstate the error view. At most the objection would show

that all sceptical arguments have their weak spot. The ignorance view might still offer a

better interpretation of sceptical scenarios than the error view. Since my claim is only

that the ignorance view offers a better interpretation of sceptical arguments and not that

scepticism is ultimately true, I do not need to argue that (a version of) KK is in fact true,

but merely that the resulting sceptical argument is worth taking seriously. Furthermore,

no sceptical argument relies on an epistemic principle that is unanimously accepted. Thus,

to defend the ignorance view I do not have to show that it rests on an epistemic principle

that is accepted by everyone. It is sufficient to show that (at least one version of) KK is

not inherently more problematic than Closure. Both Closure and KK are burdened with

their share of problems and counterexamples, but both are also intuitively compelling

and not easily refuted. It is a dangerous prejudice to think that relying on KK is illicit

whereas relying on Closure is not. It is a prejudice because the reasons against KK are far

from decisive and it is dangerous because it gets repeated so often that too many authors

do not bother stating their objections against KK anymore.

To sum up, I shall argue that there is no reason to allow the sceptical argument to rely

19

on Closure, but to disallow it to rely on (a version of) KK. I do so in two steps. In a first

step I argue that (at least one version of) KK is both as intuitive and explanatorily useful

as Closure. In a second step I argue that five common objections against KK are far from

decisive.

3.1 Arguments for KK

In this section I take a look at two prominent arguments for Closure and argue that there

are parallel arguments for KK.

The first argument is based on intuitions (cf. DeRose 1995: 28, Lewis 1996: 549 f.):

Intuitively, it is contradictory to suppose that someone knows something but does not

know what is obviously entailed. For example, it is contradictory to suppose that Anne

knows it is Max (and knows that if it is Max, it cannot be Moritz), but does not know it

is not Moritz. Intuitively, however, the same is true of ignorance-possibilities: For example,

it is as contradictory to suppose that Anne knows it is Max (and knows that she cannot

tell Max and Moritz apart when she does not have her glasses on), but does not know that

she has her glasses on. In fact, it is easy to give examples of ignorance-possibilities that are

more relevant than particular error-possibilities. For example, consider the error-possibility

that Max was abducted by aliens and replaced by an android who looks exactly like him.

It would be puzzling to claim that Anne must know of this remote error-possibility that it

does not obtain, but that it does not matter whether she knows of the ignorance-possibility

that it does not obtain. There is no reason to think that all error-possibilities no matter

how remote must be known not to obtain, whereas ignorance-possibilities no matter how

close need not just because they are not error-possibilities. If someone challenges Anne’s

belief by mentioning that she does not have her glasses on, she cannot reply that her

belief is about Max, not about her vision and that, therefore, the challenge is irrelevant to

her belief. KK offers a neat explanation of why this reply would be inadequate. Knowing

it is Max entails that she has her glasses on and, therefore, she cannot know it is Max

20

without knowing that she has her glasses on. By contrast, accepting Closure but rejecting

KK would predict that the remark that she does not have her classes on is not a challenge

at all.14

The second kind of argument focuses on the role of the two epistemic principles. Closure

connects knowledge with other pieces of knowledge, namely knowledge of a proposition

with knowledge of its (known) consequences. Analogously, KK connects knowledge of a

proposition with knowledge of one’s epistemic situation with respect to that proposition.

Since both principles connect knowledge with more knowledge, they are both a burden and

an opportunity. They are a burden because knowing one propositions requires knowing

other propositions. They are an opportunity because knowing one proposition gives rise

to knowledge of other propositions. Accordingly, both principles have two complementary

roles. Closure, on the one hand, underpins the possibility of inferential knowledge, that

is, of gaining new knowledge from what is already known. On the other hand, it raises

the bar for knowledge: If I do not know what is entailed by something I believe, I do not

know it and should give up the belief (I may, of course, assume, hypothesise or accept it

instead). Analogously, KK, on the one hand, underpins the possibility of second-order

knowledge, that is, of knowing what is known. On the other hand, it raises the bar for

knowledge: If I do not know that I know something, I do not know it and should give up

the belief (I may, of course, assume, hypothesise or accept it instead). Again this points to

a similarity between Closure and KK: Closure explains why one can gain new knowledge

from old knowledge or should give up a belief if one does not take oneself to know its

consequences. KK explains why one can gain knowledge of knowledge from knowledge

or should give up a belief if one does not take oneself to know it. Thereby, both connect

knowledge of some proposition with knowledge of other propositions. Proponents of the

KK objection owe an explanation of why they think relying on the one connection is14 That there are two different kinds of challenges or defeaters is defended, for example, by Pollock and

Cruz (1999: 196, rebutting/undercutting), Williams (2001: 149, non-epistemic/epistemic) and Casullo

(2003: 44–46, overriding/undermining).

21

admissible, but relying on the other is not. By default both should be on a par.

3.2 Objections against KK

A natural reply is that KK is rejected not because it is supposed to be completely

unfounded, but because of the severe problems that befall it but not Closure. Of course,

KK faces a lot of problems, even severe ones – but so does Closure. KK’s problems are

not more severe than Closure’s problems, or so I shall argue.

1. Objection: KK leads into an infinite regress: Knowledge would require knowing that

one knows that one knows that . . .

Reply: Consider the parallel problem for Closure. As a matter of fact I know that

any proposition P entails P ∧ P , P ∧ P ∧ P , P ∧ P ∧ P ∧ P , and so on. Thus,

knowledge that P would require knowing that P ∧ P , knowing that P ∧ P ∧ P , and

so on. Yet, this objection does not show that Closure leads into an infinite regress.

For knowledge is at most closed under known entailment and at some point the

propositions are so complex that I cannot even grasp, least of all know the relevant

entailments. The parallel objection to KK gets the parallel reply: The sceptical

argument only needs a version of KK according to which knowledge iterates only if

the relevant entailment is known, that is, KK* requires the n-th iteration of knowing

that one knows only if one knows that the n-th iteration entails the n-th iteration.

Since the proposition increases in complexity at each step, everyone will fail to know

the relevant entailment at some step. Even if one version of KK, KK+, should lead

into a problematic infinite regress, the version needed for the sceptical argument,

KK*, does not.

2. Objection: Knowing can be easy whereas knowing that one knows is an accomplish-

ment that requires cognitive powers not every knower has. Children, for example,

know a lot without knowing that they know.

Reply: Knowing that one knows need not be a bigger accomplishment than first-order

22

knowledge. In particular, if externalism is the correct view of first-order knowl-

edge, it is the correct view of higher order knowledge as well (cf. Wright 1991: 92,

Williamson 2005: 231, n. 14). Putting aside sceptical ignorance-possibilities for the

moment, ignorance-possibilities are usually easy to rule out. For example, it is easy

to know, even for children, that one has one’s glasses on (unblurred vision, feeling

of their weight, and so on). Moreover, since KK* requires only ruling out known

ignorance-possibilities and some children do not know the relevant entailments, they

are not counterexamples to KK* anyway.15

3. Objection: One cannot not know that one knows just by knowing something. For

example, just in virtue of knowing that the wall is red (and knowing that knowing

this entails that I am not colour-blind) I cannot know that I am not colour-blind.

Reply: Like Closure KK (in either version) is not a transmission principle. It does

not state that my epistemic support for believing P transmits to believing Q. In

the example I know that the wall is red just by looking at it, but I do not know

that I am not colour-blind just by looking at this wall. Compare this to a case

of Closure: I know that the animals are zebras just by looking at them, but I do

not know that they are not painted mules just by looking at them. In both cases

the additional knowledge – knowing that I am not colour-blind and knowing that

the animals are not painted mules – goes beyond mere perceptual discrimination.

Whatever explanation there is of how Closure is ensured in the zebra case can be

turned into an explanation of how KK is ensured.16

15 A consequence of this reply is that the sceptical argument applies only to those subjects who possess

the concept of knowledge since subjects who lack that concept cannot know the relevant entailments.

Again, this problem is not specific to the ignorance view. The sceptical argument relying on Closure

applies only to those who are sophisticated enough to know the relevant entailments, too. Hence, both

face the problem that they are committed to the view that those who know more entailments know

less.

16 For example, Pritchard (2012: pt. 2) argues that knowing that the animals are zebras requires perceptual

23

4. Objection: Since belief is a necessary condition for knowledge, it is conceptually

impossible to know that P without believing that P . However, it is conceptually

possible to know that P without believing that one knows that P . Therefore,

someone can know that P and know that knowing P entails Q, but fail to believe

that she knows that Q because she does not even believe that she knows that P .

This problem is different from the “not putting one and one together” problem: If

someone believes that knowing P entails Q, but does not believe that she knows

that P , there are not even two beliefs to be put together.

Reply: Although knowing that P does not entail believing that one knows that

P , knowing that P entails that one does not believe that one does not know that

P . A belief falls short of knowledge if the subject takes herself not to know it or

to be unable to rule out a known ignorance-possibility. As already argued, when

confronted with a defeater (like “but you don’t have your glasses on!”), it is irrational

to neither dispel the defeater nor revise one’s belief. To defend KK against the

objection it is useful to distinguish between two ways in which one may not believe

something. One may simply not have considered the issue or one may withhold belief

on the issue. The former consists in the absence of a doxastic attitude on whether P ,

whereas the latter is the doxastic attitude one entertains when after considering the

issue one nevertheless refuses to believe either P or ¬P (cf. Wedgwood 2002: 272).

Applying this distinction to the objection partly vindicates it and partly refutes it:

discrimination (discriminating epistemic support), whereas knowing that they are not painted mules

requires additional evidence (favouring support). However, “that there would be no point to such a

deception, that it would be costly and time-consuming without bringing any comparable benefit, that it

would be easily found out, and then the zoo-owner would be subject to penalties, and so on” (Pritchard

2012: 79) is sufficient favouring evidence for knowing that the animals are not painted mules. The same

kind of favouring epistemic support is available in the other example: Intersubjective agreement in

colour judgements and the high unlikeliness of spontaneous colour blindness favour believing that I am

not colour-blind.

24

If she lacks the belief that Q in the absence sense, there is no reason to suppose that

this must preclude her from knowing P . But, if she lacks the belief that Q in the

withholding sense, she has considered the issue whether Q and decided to believe

neither Q nor ¬Q. Since by assumption she knows that knowing P entails Q, she

takes herself to be unable to rule out an ignorance-possibility and, thus, knowing

that P is precluded. Thus, the version of KK the sceptical argument relies on must

be amended slightly. The antecedent must be strengthened to include in addition to

knowing P and knowing that knowing P entails Q that one has made up one’s mind

on whether Q. This amendment is harmless for my purposes because the additional

condition is clearly met with respect to sceptical scenarios. Moreover, the reply

is similar to replies from the debate about Closure. Proposals to strengthen the

antecedent in similar ways are common to protect Closure from counterexamples.

5. Objection: Williamson’s anti-luminosity argument is an objection to KK without

being an objection to Closure (cf. Williamson 2000: ch. 4).

Reply: Granting for the sake of the objection that Williamson’s argument is successful

it is not obvious that the sceptical argument falls with it. Even if Williamson’s

argument shows that for all (not-trivial) conditions there are circumstances under

which knowledge does not entail knowing that one knows, his argument does not

show that the particular instances of KK that the sceptical argument relies on are

problematic. At least, rejecting the second premise of the argument from ignorance

on the basis that knowing is not a luminous condition would be ill-motivated because

the anti-luminosity argument at most shows that some kind of ignorance-possibilities

need not be ruled out.

Of course, these cursory remarks are not enough to actually establish the truth of KK (cf.

Okasha 2013 and Greco 2013 for recent discussions pro KK). Yet, I have already done

more than the critics of the ignorance view who simply assume that KK is indefensible.

Recall that the aim is not to argue that KK is true, but to show that (at least one version

25

of) KK is not inherently more problematic than Closure. That it makes the sceptical

argument dependent on (a version of) KK is an objection against the ignorance view only

if KK is obviously untenable in any version and there is an interpretation of sceptical

argument that avoids relying on KK. I have argued that neither is the case.

4 Second objection: Transparency

Ballantyne and Evans’ target is Schaffer’s debasing demon, but their objection – the

transparency objection – applies to the ignorance view in general (Ballantyne and Evans

2013). Let us call an ignorance-possibility in which the target belief is true a veridical

ignorance-possibility. Ballantyne and Evans argue that veridical ignorance-possibilities

do not threaten their target beliefs because challenging a belief with such an ignorance-

possibility contradicts the transparency of belief. The version of the transparency thesis

they have in mind is the following:

“The question whether one is justified to believe p is transparent to the question

whether p, in the sense that one’s answer to the latter always determines one’s

answer to the former.” (2013: 555)

According to the transparency objection, veridical ignorance-possibilities cannot be

relevant to the question whether believing that P is justified. That is so because veridical

ignorance-possibilities are not relevant to the question whether P . Since P is true no

matter whether the ignorance-possibility obtains, one need not revise one’s answer to the

question whether P in the light of the ignorance-possibility:

“Insofar as your question is just whether p, it does not matter which possibility

actually holds. Either way, p is true.” (2013: 555)

But if veridical ignorance-possibilities are not relevant to the question whether P and

belief is transparent, they are not relevant to the question whether believing P is justified

either.

26

Before replying to the transparency objection I want to offer two observations on it.

First, the transparency objection does not only show that veridical ignorance-possibilities

are no threat. It also shows that the actual obtaining of a veridical ignorance-possibility

(and knowing that it obtains) is irrelevant. Even if I discovered that a veridical ignorance-

possibility obtained, I would not need to revise my beliefs.

Second, Ballantyne and Evans assume that in a sceptical ignorance-possibility the target

beliefs are true. But if the description of the scenario does not specify which target beliefs

are true, the objection cannot be raised. For example, Schaffer’s debasing demon ensures

that all of my beliefs were acquired by guesswork or are otherwise improperly based. Since

guessing can lead to both true and false beliefs, the description of the scenario does not

entail anything about the truth-value of (most of) the victim’s beliefs. The only specific

beliefs that must be true in a sceptical ignorance-possibility are necessarily true beliefs

and those beliefs whose truth is entailed by the description of the scenario. Yet, although

this observation limits the scope of the transparency objection, there are still enough

beliefs to which the objection applies.

Turning from observations to criticism, my reply to the transparency objection consists

of two steps. In a first step, I complement Ballantyne and Evans’ version of transparency

(whether-transparency) with another version of transparency (why-transparency). Veridical

ignorance-possibilities do not contradict transparency because, although they are irrelevant

to the whether-questions, they are relevant to the why-questions. In a second step, I

explain how this distinction applies to sceptical scenarios.

The first step of my reply is best explained by giving an example: Suppose Ben believes

that all novels written by Shakespeare are love stories. He came to believe this when he

read a quotation he took to be from Shakespeare’s only novel. Later he finds out that

Shakespeare did not write any novels. Learning this does not lead Ben to revise his belief

since his answer to the question whether all of Shakespeare’s novels are love stories does

not change. After all, his belief that Shakespeare’s novels are love stories turns out to

27

be (vacuously) true. Ballantyne and Evans are committed to the claim that the new

information would be irrelevant to Ben’s belief.

This is an unconvincing claim: Although Ben’s reply to the question whether Shake-

speare’s novels are love stories does not change, his reply to the question why to give this

answer changes. In other words, learning that Shakespeare did not write any novels does

not force Ben to revise his belief that Shakespeare’s novels are love stories, but forces

him to rebase his belief, that is, believe it for different reasons. The relevance is best

explained by distinguishing whether-relevance from why-relevance. The new evidence is

not relevant to whether believing that P is justified, but it is relevant to why believing

that P is justified. At the same time, it is not relevant to which answer to give to whether

P , but relevant to why to give it. This distinction allows to rescue both transparency

and the relevance of veridical ignorance-possibilities. Veridical ignorance-possibilities

take away the epistemic support one thought one had and thereby force to rebase one’s

belief, that is, to seek different epistemic support for one’s belief. To sum up, veridical

ignorance-possibilities are why-relevant (not whether-relevant) and require rebasing (not

revising) the target beliefs.

The first step of my reply is not sufficient for explaining how sceptical scenarios threaten

their target beliefs: Like Ben’s belief my belief that I have a brain could survive the

sceptical attack. After becoming aware of the brain in a vat scenario and my (apparent)

inability to rule it out, all I needed to do would be to believe that I have a brain for

different reasons. Instead of believing this because of my original empirical evidence I

could argue by disjunction elimination: Either I am a brain in a vat or I am not. If I am

a brain in a vat, I have a brain. If I am not a brain in a vat, my original evidence stands.

Hence, I have a brain. Thus, considering the brain in a vat scenario would not destroy my

knowledge that I have a brain, but lead me to an impeccable justification for this belief

thereby refuting scepticism. It gets even worse for scepticism: If this new justification of

believing that I have a brain is available, it allows refuting scepticism on an even larger

28

scale. Each empirical belief that P could be replaced by the belief that if I am not brain

in a vat, P . The conditional beliefs are empirical beliefs, but would be immune to the

sceptical attack.

The second step of my reply complements the first step by taking away this alleged new

justification. The strategy depends on a common constraint on sceptical scenarios, the

possibility constraint : Both the scenario and its normal counterpart must be metaphysically

possible (cf. Kung 2011: 388–391 for references and a defence). If being an envatted brain

and being a normal brain are exhaustive metaphysical possibilities, the strategy of basing

my belief that I have a brain on the disjunctive justification is feasible. But the possibility

constraint is overstated: It is sufficient that the sceptical scenario and its counterpart

are believed to be metaphysically possible. The sceptical argument is not based on the

assumption that the sceptical scenario and its counterpart are metaphysically possible.

It merely chooses a scenario that I must believe to be possible thereby blocking the

objection that it is merely a fantasy. The revised constraint suffices for an explanation

of why scepticism relies on the brain in a vat scenario and not, say, on the magic teapot

scenario (cf. Kung 2011: 398): Suppose I am a teapot in a wizard’s study whose sense

experiences are magically copied from the wizard’s mind with random changes caused by

the liquid I am filled with. Given all my beliefs about myself and my epistemic situation

in the world, the brain in a vat scenario is possible, but the teapot scenario is not. But if

I believed in magic, the teapot scenario might well be a better choice than the brain in a

vat scenario.17

17 A different argument against the possibility constraint is offered by Beebe 2010: 457–461. The revised

constraint is also corroborated by Descartes’ argument in the first meditation. He introduces the last

step of doubt by connecting it to what he believes thereby making the scenario seem possible: “And yet

firmly rooted in my mind is the long-standing opinion that there is an omnipotent God” (1641: AT VII

21). He does not in fact argue for the possibility of the scenario, but is content with noting that he and

his contemporaries already implicitly believe its possibility. (In case someone happens to have different

beliefs in theological matters he adds a couple of sentences for those, too, cf. 1641: AT VII 21 f.)

29

The revised possibility constraint has two virtues. It saves the sceptical argument from

self-contradiction (to which it allegedly succumbs, cf. Kung 2011). Since I can know

that being an envatted brain is metaphysically possible only by relying on empirical

knowledge and since the latter knowledge is called into question by the sceptical argument,

the sceptical argument should better not depend on the assumption that being an

envatted brain is metaphysically possible. The self-contradiction is avoided because the

sceptical argument only exploits what I already believe and does not itself depend on any

assumptions about what is possible.

The revised possibility constraint also explains what is wrong with the alleged new

disjunctive justification for believing that I have a brain. Once I concede that I need to but

cannot know that I am not an envatted brain, I do not even know that it is metaphysically

possible that I have a brain. The sceptical argument does not mention brains because it

relies on the assumption that I must have either an envatted or a normal brain. Instead,

the argument, if sound, reveals that even this assumption is unfounded. Therefore, the

assumption the disjunctive justification rests on is illicit once the sceptical conclusion

is conceded. Thus, there is a crucial difference between Ben’s belief that Shakespeare’s

novels are love stories and my belief that I have a brain. Unlike Ben who can continue

with his belief if only he bases it on different reasons, I cannot rebase my belief in the

light of the sceptical challenge.

5 Conclusion

To conclude, the error view does not offer a convincing interpretation of the Cartesian

sceptical argument. In its single scenario version its conclusion falls short of the desired

general sceptical conclusion. This argument at most shows that we lack knowledge of

some empirical matters and its conclusion is compatible with our having a lot of empirical

knowledge. In its multiple scenario version its premises are difficult to defend because there

still is not an appropriate sceptical scenario for each empirical belief and its conclusion is

30

not sufficiently different from fallibilism. The ignorance view offers a better interpretation

of Cartesian sceptical arguments than the error view. It allows to interpret the Cartesian

sceptical argument as an argument that relies on a single scenario that we intuitively

need to but cannot rule out. But the ignorance view also has several consequences for

our understanding of Cartesian sceptical arguments. One notable consequence is that

the sceptical argument relies on (a version of) the KK-principle. I have argued that

this consequence is not devastating since the version of KK needed is not less plausible

than closure. A second consequence is that an explanation is needed of how veridical

ignorance-possibilities can be used to challenge a belief. I have suggested an interpretation

of the transparency of belief and of the possibility requirement on sceptical scenarios

that offers such an explanation. Whether the resulting sceptical argument is ultimately

compelling is, of course, open to debate. But recall that my aim is not to defend scepticism,

but to offer the best interpretation of it. If the sceptical argument ultimately fails and

the ignorance view helps to reveal its weak spots, I will be as thrilled as everyone else.18

References

Atkins, Philip and Ian Nance (2013): “A Problem for the Closure Argument”, in: Inter-national Journal for the Study of Skepticism forthcoming: 1–13.

Avnur, Yuval, Anthony Brueckner and Christopher Buford (2011): “No Closure onSkepticism”, in: Pacific Philosophical Quarterly 92: 439–447.

Ballantyne, Nathan and Ian Evans (2013): “Schaffer’s Demon”, in: Pacific PhilosophicalQuarterly 94: 552–559.

Beebe, James (2010): “Constraints on Sceptical Hypotheses”, in: Philosophical Quarterly60: 449–470.

Beebe, James (2011): “A Priori Skepticism”, in: Philosophy and Phenomenological Re-search 83: 583–602.

Bernecker, Sven (2010): Memory. A Philosophical Study. Oxford: OUP.Briesen, Jochen (2011): “Antiskeptische Trittbrettfahrer des semantischen Externalismus”,

18 I am grateful to an audience at Regensburg University and two anonymous reviewers for very helpfulcomments on earlier versions of this paper.

31

in: Zeitschrift für philosophische Forschung 65: 563–585.Brueckner, Anthony (2011): “Debasing Scepticism”, in: Analysis 71: 295–297.Casullo, Albert (2003): A Priori Justification. Oxford: OUP.Cohen, Stewart (1998): “Two Kinds of Skeptical Argument”, in: Philosophy and Phe-

nomenological Research 58: 143–159.DeRose, Keith (1995): “Solving the Skeptical Problem”, in: Philosophical Review 104:

1–52.Descartes, René (1641): Meditations on First Philosophy, in: The Philosophical Writings

of Descartes. Vol. 2. Translated by John Cottingham, Robert Stoothoff and DugaldMurdoch. Cambridge: CUP, 1984: 3–62.

Dodd, Dylan (2012): “Evidentialism and Skeptical Arguments”, in: Synthese 189: 337–352.Gemes, Ken (2009): “A Refutation of Global Scepticism”, in: Analysis 69: 218–219.Gemes, Ken (2010): “A Vindication of a Refutation of Global Scepticism, a Refutation

of Global Perceptual Scepticism and a Refutation of Global Existential Scepticism”,in: Analysis 70: 63–70.

Genova, A.C. (2010): “Has Gemes Refuted Global Scepticism?”, in: Analysis 70: 59–63.Greco, Daniel (2013): “Could KK be OK?”, in: Journal of Philosophy forthcoming: 1–37.Hetherington, Stephen (1996): “Gettieristic Scepticism”, in: Australasian Journal of

Philosophy 74: 83–97.Kraft, Tim (2012): “Scepticism, Infallibilism, Fallibilism”, in: Discipline Filosofiche 22:

49–70.Kraft, Tim (2013): “Sceptical Scenarios are not Error-Possibilities”, in: Erkenntnis 78:

59–72.Kung, Peter (2011): “On the Possibility of Skeptical Scenarios”, in: European Journal of

Philosophy 19: 387–407.Lewis, David (1996): “Elusive Knowledge”, in: Australasian Journal of Philosophy 74:

549–567.Luper, Steven (2007): “Moore’s Missing Principle”, in: Philosophical Papers 36: 151–161.Moore, George Edward (1959): “Certainty”, in: Philosophical Papers. London: Allen &

Unwin, 1959: 226–251.Müller, Olaf (2003): Wirklichkeit ohne Illusionen. Vol. 1, Paderborn: Mentis.Murphy, Peter (2013): “The Defect in Effective Skeptical Scenarios”, in: International

Journal for the Study of Skepticism 3: 271–281.Okasha, Samir (2013): “On a Flawed Argument Against the KK Principle”, in: Analysis

73: 80–86.Pollock, John and Joseph Cruz (1999): Contemporary theories of knowledge. 2nd ed.,

32

Lanham: Rowman & Littlefield.Pritchard, Duncan (2012): Epistemological Disjunctivism. Oxford: OUP, 2012.Pryor, James (2000): “The Skeptic and the Dogmatist”, in: Noûs 34: 517–549.Reed, Baron (2007): “The Long Road to Skepticism”, in: Journal of Philosophy 104:

236–262.Reed, Baron (2009): “A New Argument for Skepticism”, in: Philosophical Studies 142:

91–104.Roush, Sherrilyn (2010): “Closure on Skepticism”, in: Journal of Philosophy 107: 243–256.Schaffer, Jonathan (2010): “The Debasing Demon”, in: Analysis 70: 228–237.Stroud, Barry (1984): The Significance of Philosophical Scepticism. Oxford: Clarendon.Vahid, Hamid (2013): “Skepticism, A Priori Skepticism, and the Possibility of Error”, in:

International Journal for the Study of Skepticism 3: 235–252.Vogel, Jonathan (2004): “Skeptical Arguments”, in: Philosophical Issues 14: 426–455.Wedgwood, Ralph (2002): “The Aim of Belief”, in: Philosophical Perspectives 16: 267–297.Williams, Michael (2001): Problems of Knowledge. Oxford: OUP.Williamson, Timothy (2000): Knowledge and its Limits. Oxford: OUP.Williamson, Timothy (2005): “Contextualism, Subject-Sensitive Invariantism and Knowl-

edge of Knowledge”, in: Philosophical Quarterly 55: 213–35.Winters, Barbara (1981): “Sceptical Counterpossibilities”, in: Pacific Philosophical Quar-

terly 62: 30–38.Wittgenstein, Ludwig (OC): On Certainty. Edited by G.E.M. Anscombe and G.H. von

Wright, translated by G.E.M. Anscombe and Denis Paul. Oxford: Blackwell, 1969.Wright, Crispin (1991): “Scepticism and Dreaming: Imploding the Demon”, in: Mind

100: 87–116.Zalabardo, José (2012): Scepticism and Reliable Belief. Oxford: OUP.

33