Knowledge and Scepticism

BY ROBERT NOZICK
Robert Nozick (1938-2002) was Pellegrino University Professor at Harvard University. His early
book in political theory, Anarchy, State, and Utopia, was very influential, and he followed it with
Philosophical Explanations, The Examined Life, The Nature of Rationality, Socratic Puzzles,
and Invariances: The Structure of the Objective World.
_________________________________________________________________________
Reprinted from Robert Nozick. Philosophical Explanations (Cambridge, Mass.: Harvard
University Press, 1981), by permission of the publisher.
_________________________________________________________________________

You think you are seeing these words, but could you not be hallucinating or dreaming or having
your brain stimulated to give you the experience of seeing these marks on paper although no
such thing is before you? More extremely, could you not be floating in a tank while super-
psychologists stimulate your brain electrochemically to produce exactly the same experiences as
you are now having, or even to produce the whole sequence of experiences you have had in your
lifetime thus far? If one of these other things was happening, your experience would be exactly
the same as it now is. So how can you know none of them is happening? Yet if you do not know
these possibilities don't hold, how can you know you are reading this book now? If you do not
know you haven't always been floating in the tank at the mercy of the psychologists, how can
you know anything – what your name is, who your parents were, where you come from?

The sceptic argues that we do not know what we think we do. Even when he leaves us
unconverted, he leaves us confused. Granting that we do know, how can we? Given these other
possibilities he poses, how is knowledge possible? In answering this question. we do not seek to
convince the sceptic, but rather to formulate hypotheses about knowledge and our connection
to facts that show how knowledge can exist even given the sceptic's possibilities. These
hypotheses must reconcile our belief that we know things with our belief that the sceptical
possibilities are logical possibilities.

The sceptical possibilities, and the threats they pose to our knowledge, depend upon our
knowing things (if we do) mediately, through or by way of something else. Our thinking or
believing that some fact p holds is connected somehow to the fact that p, but is not itself identical
with that fact. Intermediate links establish the connection. This leaves room for the possibility
of these intermediate stages holding and producing our belief that p, without the fact that p being
at the other end. The intermediate stages arise in a completely different manner, one not
involving the fact that p although giving rise to the appearance that p holds true.

Are the sceptic's possibilities indeed logically possible? Imagine reading a science fiction story in
which someone is raised from birth floating in a tank with psychologists stimulating his brain.
The story could go on to tell of the person's reactions when he is brought out of the tank, of
how the psychologists convince him of what had been happening to him, or how they fail to do
so. This story is coherent, there is nothing self-contradictory or otherwise impossible about it.
Nor is there anything incoherent in imagining that you are now in this situation, at a time before
being taken out of the tank. To ease the transition out, to prepare the way, perhaps the
psychologists will give the person in the tank thoughts of whether floating in the tank is possible,
or the experience of reading a book that discusses this possibility, even one that discusses their
easing his transition. (Free will presents no insuperable problem for this possibility. Perhaps the
psychologists caused all your experiences of choice, including the feeling of freely choosing; or
perhaps you do freely choose to act while they, cutting the effector circuit, continue the scenario
from there.)

Some philosophers have attempted to demonstrate there is no such coherent possibility of this
sort.' However, for any reasoning that purports to show this sceptical possibility cannot occur,
we can imagine the psychologists of our science fiction story feeding it to their tank-subject,
along with the (inaccurate) feeling that the reasoning is cogent. So how much trust can be placed
in the apparent cogency of an argument to show the sceptical possibility isn't coherent? The
sceptic's possibility is a logically coherent one, in tension with the existence of (almost all)
knowledge; so we seek a hypothesis to explain how, even given the sceptic's possibilities,
knowledge is possible. We may worry that such explanatory hypotheses are ad hoc, but this
worry will lessen if they yield other facts as well, fit in with other things we believe, and so forth.
Indeed, the theory of knowledge that follows was not developed in order to explain how
knowledge is possible. Rather, the motivation was external to epistemology; only after the
account of knowledge was developed for another purpose did I notice its consequences for
scepticism, for understanding how knowledge is possible. So whatever other defects the
explanation might have, it can hardly be called ad hoc.

I. Knowledge
(Conditions for Knowledge)

Our task is to formulate further conditions to go alongside

(1) p is true
(2) S believes that p.

We would like each condition to be necessary for knowledge, so any case that fails to satisfy it
will not be an instance of knowledge. Furthermore, we would like the conditions to be jointly
sufficient for knowledge, so any case that satisfies all of them will be an instance of knowledge.
We first shall formulate conditions that seem to handle ordinary cases correctly, classifying as
knowledge cases which are knowledge, and as non-knowledge cases which are not; then we shall
check to see how these conditions handle some difficult cases discussed in the literature.

One plausible suggestion is causal, something like: the fact that p (partially) causes S to believe
that p, that is, (2) because (1). But this provides an inhospitable environment for mathematical
and ethical knowledge; also there are well-known difficulties in specifying the type of causal
connection. If someone floating in a tank oblivious to everything around him is given (by direct
electrical and chemical stimulation of the brain) the belief that he is floating in a tank with his
brain being stimulated, then even though that fact is part of the cause of his belief, still he does
not know that it is true. Let us consider a different third condition:

(3) If p were not true, S would not believe that p.

Throughout this work, let us write the subjunctive "if-then" by an arrow, and the negation of a
sentence by prefacing "not-" to it. The above condition thus is rewritten as:
(3) not-p  not-(5 believes that p).

This subjunctive condition is not unrelated to the causal condition. Often when the fact that p
(partially) causes someone to believe that p, the fact also will be causally necessary for his having
the belief without the cause, the effect would not occur. In that case, the subjunctive condition
(3) also will be satisfied. Yet this condition is not equivalent to the causal condition. For the
causal condition will be satisfied in cases of causal overdetermination, where either two sufficient
causes of the effect actually operate, or a back-up cause (of the same effect) would operate if the
first one didn't; whereas the subjunctive condition need not hold for these cases. When the two
conditions do agree, causality indicates knowledge because it acts in a manner that makes the
subjunctive (3) true.

The subjunctive condition (3) serves to exclude cases of the sort first described by Edward
Gettier, such as the following. Two other people are in my office and I am justified on the basis
of much evidence in believing the first owns a Ford car; though he (now) does not, the second
person (a stranger to me) owns one. I believe truly and justifiably that someone (or other) in my
office owns a Ford car, but I do not know someone does. Concluded Gettier, knowledge is not
simply justified true belief.

The following subjunctive, which specifies condition (3) for this Gettier case, is not satisfied: if
no one in my office owned a Ford car, I wouldn't believe that someone did. The situation that
would obtain if no one in my office owned a Ford is one where the stranger does not (or where
he is not in the office); and in that situation I still would believe, as before, that someone in my
office does own a Ford, namely, the first person. So the subjunctive condition (3) excludes this
Gettier case as a case of knowledge.

The subjunctive condition is powerful and intuitive, not so easy to satisfy, yet not so powerful
as to rule out everything as an instance of knowledge. A subjunctive conditional "if p were true,
q would be true," p → q, does not say that p entails q or that it is logically impossible that p yet
not-q. It says that in the situation that would obtain if p were true, q also would be true. This
point is brought out especially clearly in recent "possible-worlds" accounts of subjunctives: the
subjunctive is true when (roughly) in all those worlds in which p holds true that are closest to
the actual world, q also is true. (Examine those worlds in which p holds true closest to the actual
world, and see if q holds true in all these.) Whether or not q is true in p worlds that are still farther
away from the actual world is irrelevant to the truth of the subjunctive. I do not mean to endorse
any particular possible-worlds account of subjunctives, nor am I committed to this type of
account. I sometimes shall use it, though, when it illustrates points in an especially clear way.

The subjunctive condition (3) also handles nicely cases that cause difficulties for the view that
you know that p when you can rule out the relevant alternatives to p in the context. For, as Gail
Stine writes, "what makes an alternative relevant in one context and not another? . . . if on the
basis of visual appearances obtained under optimum conditions while driving through the
countryside Henry identifies an object as a barn, normally we say that Henry knows that it is a
barn. Let us suppose, however, that unknown to Henry, the region is full of expertly made
papier-mache facsimiles of barns. In that case, we would not say that Henry knows that the
object is a barn, unless he has evidence against it being a papier-mache facsimile, which is now
a relevant alternative. So much is clear, but what if no such facsimiles exist in Henry's
surroundings, although they once did? Are either of these circumstances sufficient to make the
hypothesis (that it's a papier-mache object) relevant? Probably not, but the situation is not so
clear:” Let p be the statement that the object in the field is a (real) barn, and q the one that the
object in the field is a papier-mache barn. When papier-mache barns are scattered through the
area, if p were false, q would be true or might be. Since in this case (we are supposing) the person
still would believe p, the subjunctive

(3) not-p  not-(S believes that p)

is not satisfied, and so he doesn't know that p. However, when papier-mache barns are or were
scattered around another country, even if p were false q wouldn't be true, and so (for all we have
been told) the person may well know that p. A hypothesis q contrary to p clearly is relevant when
if p weren't true, q would be true; when not-p → q. It clearly is irrelevant when if p weren't true,
q also would not be true; when not-p → not-q. The remaining possibility is that neither of these
opposed subjunctives holds; q might (or might not) be true if p weren't true. In this case, q also
will be relevant, according to an account of knowledge incorporating condition (3) and treating
subjunctives along the lines sketched above. Thus, condition (3) handles cases that befuddle the
"relevant alternatives" account; though that account can adopt the above subjunctive criterion
for when an alternative is relevant, it then becomes merely an alternate and longer way of stating
condition (3).

Despite the power and intuitive force of the condition that if p weren't true the person would
not believe it, this condition does not (in conjunction with the first two conditions) rule out
every problem case. There remains, for example, the case of the person in the tank who is
brought to believe, by direct electrical and chemical stimulation of his brain, that he is in the tank
and is being brought to believe things in this way; he does not know this is true. However, the
subjunctive condition is satisfied: if he weren't floating in the tank, he wouldn't believe he was.

The person in the tank does not know he is there, because his belief is not sensitive to the truth.
Although it is caused by the fact that is its content, it is not sensitive to that fact. The operators
of the tank could have produced any belief, including the false belief that he wasn't in the tank;
if they had, he would have believed that. Perfect sensitivity would involve beliefs and facts
varying together. We already have one portion of that variation, subjunctively at least: if p were
false he wouldn't believe it. This sensitivity as specified by a subjunctive does not have the belief
vary with the truth or falsity of p in all possible situations, merely in the ones that would or might
obtain if p were false. The subjunctive condition

(3) not-p  not-(S believes that p)

tells us only half the story about how his belief is sensitive to the truth-value of p. It tells us how
his belief state is sensitive to p's falsity, but not how it is sensitive to p's truth; it tells us what his
belief state would be if p were false, but not what it would be if p were true. To be sure,
conditions (1) and (2) tell us that p is true and he does believe it, but it does not follow that his
believing p is sensitive to p's being true. This additional sensitivity is given to us by a further
subjunctive: if p were true, he would believe it.

(4) p  S believes that p.

Not only is p true and S believes it, but if it were true he would believe it. Compare: not only
was the photon emitted and did it go to the left, but (it was then true that): if it were emitted it
would go to the left. The truth of antecedent and consequent is not alone sufficient for the truth
of a subjunctive; (4) says more than (1) and (2). Thus, we presuppose some (or another) suitable
account of subjunctives. According to the suggestion tentatively made above, (4) holds true if
not only does he actually truly believe p, but in the 'close' worlds where p is true, he also believes
it. He believes that p for some distance out in the p . neighborhood of the actual world; similarly,
condition (3) speaks not of the whole not-p neighborhood of the actual world, but only of the
first portion of it. (1f, as is likely, these explanations do not help, please use your own intuitive
understanding of the subjunctives (3) and (4).)

The person in the tank does not satisfy the subjunctive condition (4). Imagine as actual a world
in which he is in the tank and is stimulated to believe he is, and consider what subjunctives are
true in that world. It is not true of him there that if he were in the tank he would believe it; for
in the close world (or situation) to his own where he is in the tank but they don't give him the
belief that he is (much less instill the belief that he isn't) he doesn't believe he is in the tank. Of
the person actually in the tank and believing it, it is not true to make the further statement that
if he were in the tank he would believe it--so he does not know he is in the tank.

The subjunctive condition (4) also handles a case presented by Gilbert Harman.8 The dictator
of a country is killed; in their first edition, newspapers print the story, but later all the country's
newspapers and other media deny the story, falsely. Everyone who encounters the denial believes
it (or does not know what to believe and so suspends judgement). Only one person in the country
fails to hear any denial and he continues to believe the truth. He satisfies conditions (1)-(3) (and
the causal condition about belief) yet we are reluctant to say he knows the truth. The reason is
that if he had heard the denials, he too would have believed them, just like everyone else. His
belief is not sensitively tuned to the truth, he doesn't satisfy the condition that if it were true he
would believe it. Condition (4) is not satisfied.

There is a pleasing symmetry about how this account of knowledge relates conditions (3) and
(4), and connects them to the first two conditions. The account has the following form.

(1)
(2)
(3) not-1  not-2
(4) 1  2

I am not inclined, however, to make too much of this symmetry, for I found also that with other
conditions experimented with as a possible fourth condition there was some way to construe the
resulting third and fourth conditions as symmetrical answers to some symmetrical looking
questions, so that they appeared to arise in parallel fashion from similar questions about the
components of true belief.

Symmetry, it seems, is a feature of a mode of presentation, not of the contents presented. A

uniform transformation of symmetrical statements can leave the results non-symmetrical. But if
symmetry attaches to mode of presentation, how can it possibly be a deep feature of, for
instance, laws of nature that they exhibit symmetry? (One of my favorite examples of symmetry
is due to Groucho Marx. On his radio programme he spoofed a commercial, and ended. "And
if you are not completely satisfied, return the unused portion of our product and we will return
the unused portion of your money.") Still, to present our subject symmetrically makes the
connection of knowledge to true belief especially perspicuous. It seems to me that a symmetrical
formulation is a sign of our understanding, rather than a mark of truth. If we cannot understand
an asymmetry as arising from an underlying symmetry through the operation of a particular
factor, we will not understand why that asymmetry exists in that direction. (But do we also need
to understand why the underlying asymmetrical factor holds instead of its opposite?)

A person knows that p when he not only does truly believe it, but also would truly believe it and
wouldn't falsely believe it. He not only actually has a true belief, he subjunctively has one. It is
true that p and he believes it; if it weren't true he wouldn't believe it, and if it were true he would
believe it. To know that p is to be someone who would believe it if it were true, and who wouldn't
believe it if it were false.

It will be useful to have a term for this situation when a person's belief is thus subjunctively
connected to the fact. Let us say of a person who believes that p, which is true, that when (3)
and (4) hold, his belief tracks the truth that p. To know is to have a belief that tracks the truth.
Knowledge is a particular way of being connected to the world, having a specific real factual
connection to the world: tracking it.

II. Scepticism
The sceptic about knowledge argues that we know very little or nothing of what we think we
know, or at any rate that this position is no less reasonable than the belief in knowledge. The
history of philosophy exhibits a number of different attempts to refute the sceptic: to prove him
wrong or show that in arguing against knowledge he presupposes there is some and so refutes
himself. Others attempt to show that accepting scepticism is unreasonable, since it is more likely
that the sceptic's extreme conclusion is false than that all of his premises are true, or simply
because reasonableness of belief just means proceeding in an anti-sceptical way. Even when
these counterarguments satisfy their inventors, they fail to satisfy others, as is shown by the
persistent attempts against scepticism. The continuing felt need to refute scepticism, and the
difficulty in doing so, attests to the power of the sceptic's position, the depth of his worries.

An account of knowledge should illuminate sceptical arguments and show wherein lies their
force. If the account leads us to reject these arguments, this had better not happen too easily or
too glibly. To think the sceptic overlooks something obvious, to attribute to him a simple mistake
or confusion or fallacy, is to refuse to acknowledge the power of his position and the grip it can
have upon us. We thereby cheat ourselves of the opportunity to reap his insights and to gain
self-knowledge in understanding why his arguments lure us so. Moreover, in fact, we cannot lay
the spectre of scepticism to rest without first hearing what it shall unfold.

Our goal is not, however, to refute scepticism, to prove it is wrong or even to argue that it is
wrong. We have elsewhere distinguished between philosophy that attempts to prove, and
philosophy that attempts to explain how something is possible. Our task here is to explain how
knowledge is possible, given what the sceptic says that we do accept (for example, that it is
logically possible that we are dreaming or are floating in the tank). In doing this, we need not
convince the sceptic, and we may introduce explanatory hypotheses that he would reject. What
is important for our task of explanation and understanding is that we find those hypotheses
acceptable or plausible, and that they show us how the existence of knowledge fits together with
the logical possibilities the sceptic points to, so that these are reconciled within our own belief
system. These hypotheses are to explain to ourselves how knowledge is possible, not to prove
to someone else that knowledge is possible.

Sceptical Possibilities
The sceptic often refers to possibilities in which a person would believe something even though
it was false: really, the person is cleverly deceived by others, perhaps by an evil demon, or the
person is dreaming, or he is floating in a tank near Alpha Centauri with his brain being stimulated.
In each case, the p he believes is false, and he believes it even though it is false.

How do these possibilities adduced by the sceptic show that someone does not know that p?
Suppose that someone is you; how do these possibilities count against your knowing that p? One
way might be the following. (I shall consider other ways later.) If there is a possible situation
where p is false yet you believe that p, then in that situation you believe that p even though it is
false. So it appears you do not satisfy condition (3) for knowledge.

(3) If p were false, S wouldn’t believe that p.

For a situation has been described in which you do believe that p even though p is false. How
then can it also be true that if p were false, you wouldn't believe it? If the sceptic's possible
situation shows that (3) is false, and if (3) is a necessary condition for knowledge, then the
sceptic's possible situation shows that there isn't knowledge.

So construed, the sceptic's argument plays on condition (3); it aims to show that condition (3) is
not satisfied. The sceptic may seem to be putting forth

R: Even if p were false, S still would believe p.

This conditional, with the same antecedent as (3) and the contradictory consequent, is
incompatible with the truth of (3). If (3) is true, then R is not. However, R is stronger than the
sceptic needs in order to show (3) is false. For (3) is false when if p were false, S might believe
that p. This last conditional is weaker than R, and is merely (3)'s denial:

T: not-[not-p  not-(S believes that p)]

Whereas R does not simply deny (3), it asserts an opposing subjunctive of its own. Perhaps the
possibility the sceptic adduces is not enough to show that R is true, but it appears at least to
establish the weaker T; since this T denies (3), the sceptic's possibility appears to show that (3)
is false. However, the truth of (3) is not incompatible with the existence of a possible situation
where the person believes p though it is false. The subjunctive

(3) not-p  not-(S believes p)

does not talk of all possible situations in which p is false (in which not-p is true). It does not say
that in all possible situations where not-p holds, S doesn't believe p. To say there is no possible
situation in which not-p yet S believes p, would be to say that not-p entails not-(S believes p), or
logically implies it. But subjunctive conditionals differ from entailments; the subjunctive (3) is
not a statement of entailment. So the existence of a possible situation in which p is false yet S
believes p does not show that (3) is false; (3) can be true even though there is a possible situation
where not-p and S believes that p.

What the subjunctive (3) speaks of is the situation that would hold if p were false. Not every
possible situation in which p is false is the situation that would hold if p were false. To fall into
possible worlds talk, the subjunctive (3) speaks of the not-p world that is closest to the actual
world, or of those not-p worlds that are closest to the actual world. And it is of this or these not-
p worlds that it says (in them) S does not believe that p. What happens in yet other more distant
not-p worlds is no concern of the subjunctive (3). The sceptic's possibilities (let us refer to them
as SK), of the person's being deceived by a demon or dreaming or floating in a tank, count
against the subjunctive

(3) if p were false then S wouldn’t believe that p

only if (one of) these possibilities would or might obtain if p were false. Condition (3) says: if p
were false, S still would not believe p. And this can hold even though there is some situation SK
described by the sceptic in which p is false and S believes p. If p were false S still would not
believe p, even though there is a situation SK in which p is false and S does believe p, provided
that this situation SK wouldn't obtain if p were false. If the sceptic describes a situation SK
which would not hold even if p were false then this situation SK doesn't show that (3) is false
and so does not (in this way at least) undercut knowledge. Condition C acts to rule out sceptical
hypotheses.

C: not-p -+ SK does not obtain.

Any sceptical situation SK which satisfies condition C is ruled out. For a sceptical situation SK
to show that we don't know that p, it must fail to satisfy C which excludes it; instead it must be
a situation that might obtain if p did not, and so satisfy C's denial

not-(not-p -+ SK doesn’t obtain).

Although the sceptic's imagined situations appear to show that (3) is false, they do not; they
satisfy condition C and so are excluded. The sceptic might go on to ask whether we know that
his imagined situations SK are excluded by condition C, whether we know that if p were false
SK would not obtain. However, typically he asks something stronger: do we know that his
imagined situation SK does not actually obtain? Do we know that we are not being deceived by
a demon, dreaming, or floating in a tank? And if we do not know this, how can we know that p?
Thus we are led to the second way his imagined situations might show that we do not know that
p.

Sceptical Results
According to our account of knowledge, S knows that the sceptic’s situation SK doesn’t hold if
and only if

(1) SK doesn’t hold

(2) S believes that SK doesn’t hold
(3) If SK were to hold, S would not believe that SK doesn’t hold
(4) If SK were not to hold, S would believe it does not.

Let us focus on the third of these conditions. The sceptic has carefully chosen his situation SK
so that if they held we (still) would believe they did not. We would believe we weren’t
dreaming, weren’t being deceived, and so on, even if we were. He has chosen situations SK
such that if SK were to hold, S would (still) believe that SK doesn’t hold – and this is
incompatible with the truth of (3).

Since condition (3) is necessary condition for knowledge, it follows that we do not know that
SK doesn’t hold. If it were true that an evil demon was deceiving us, if we were having a
particular dream, if we were floating in a tank with our brains stimulated in a specified way, we
would still believe we were not. So, we do not know we're not being deceived by an evil demon,
we do not know we're not in that tank, and we do not know we're not having that dream. So
says the sceptic, and so says our account. And also so we say--don't we? For how could we know
we are not being deceived that way, dreaming that dream? If those things were happening to us,
everything would seem the same to us. There is no way we can know it is not happening for
there is no way we could tell if it were happening; and if it were happening we would believe
exactly what we do now--in particular, we still would believe that it was not. For this reason, we
feel, and correctly, that we don't know--how could we?--that it is not happening to us. It is a
virtue of our account that it yields, and explains, this result.

The sceptic asserts we do not know his possibilities don't obtain, and he is right. Attempts to
avoid scepticism by claiming we do know these things are bound to fail. The sceptic's possibilities
make us uneasy because, as we deeply realize, we do not know they don't obtain; it is not
surprising that attempts to show we do know these things leave us suspicious, strike us even as
bad faith. Nor has the sceptic merely pointed out something obvious and trivial. It comes as a
surprise to realize that we do not know his possibilities don't obtain. It is startling, shocking. For
we would have thought, before the sceptic got us to focus on it, that we did know those things,
that we did know we were not being deceived by a demon, or dreaming that dream, or stimulated
that way in that tank. The sceptic has pointed out that we do not know things we would have
confidently said we knew. And if we don't know these things, what can we know? So much for
the supposed obviousness of what the sceptic tells us.

Let us say that a situation (or world) is doxically identical for S to the actual situation when if S
were in that situation, he would have exactly the beliefs (doxa) he actually does have. More
generally, two situations are doxically identical for S if and only if he would have exactly the same
beliefs in them. It might be merely a curiosity to be told there are non-actual situations doxical1y
identical to the actual one. The sceptic, however, describes worlds doxical1y identical to the
actual world in which almost everything believed is false.7

Such worlds are possible because we know mediately, not directly. This leaves room for a
divergence between our be1iefs and the truth. It is as though we possessed only two-dimensional
plane projections of three-dimensional objects. Different three-dimensional objects, oriented
appropriately, have the same two-dimensional plane projection. Similarly, different situations or
worlds will lead to our having the very same beliefs. What is surprising is how very different the
doxically identical world can be--different enough for almost everything believed in it to be false.
Whether or not the mere fact that knowledge is mediated always makes room for such a very
different doxically identical world, it does so in our case, as the sceptic’s possibilities show. To
be shown this is non-trivial, especially when we recall that we do not know the sceptic’s
possibility doesn’t obtain: we do not know that we are not living in a doxically identical world
wherein almost everything we believe is false.

What more could the sceptic ask for or hope to show? Even readers who sympathized with my
desire not to dismiss the sceptic too quickly may feel this has gone too far, that we have not
merely acknowledged the force of the sceptic’s position but have succumbed to it.

The sceptic maintains that we know almost none of what we think we know. He has shown,
much to our initial surprise, that we do not know his (nontrivial) possibility SK doesn't obtain.
Thus, he has shown of one thing we thought we knew, that we didn't and don't. To the
conclusion that we know almost nothing, it appears but a short step. For if we do not know we
are not dreaming or being deceived by a demon or floating in a tank, then how can I know, for
example, that I am sitting before a page writing with a pen, and how can you know that you are
reading a page of a book?

However, although our account of knowledge agrees with the sceptic in saying that we do not
know that not-SK, it places no formidable barriers before my knowing that I am writing on a
page with a pen. It is true that I am, I believe I am, if! weren't I wouldn't believe I was, and if I
were, I would believe it. Also, it is true that you are reading a page (please, don't stop now!), you
believe you are, if you weren't reading a page you wouldn't be1ieve you were, and if you were
reading a page you would believe you were. So according to the account, I do know that I am
writing on a page with a pen, and you do know that you are reading a page. The account does
not lead to any general scepticism.

Yet we must grant that it appears that if the sceptic is right that we don't know we are not
dreaming or being deceived or floating in the tank, then it cannot be that I know I am writing
with a pen or that you know you are reading a page. So we must scrutinize with special care the
sceptic's "short step" to the conclusion that we don't know these things, for either this step
cannot be taken or our account of knowledge is incoherent.

Nonclosure
In taking the "short step," the sceptic assumes that if S knows that p and he knows that "p entails
q" then he also knows that q. In the terminology of the logicians, the sceptic assumes that
knowledge is closed under known logical implication; that the operation of moving from
something known to something else known to be entailed by it does not take us outside of the
(closed) area of knowledge. He intends, of course, to work things backwards, arguing that since
the person does not know that q, assuming (at least for the purposes of argument) that he does
know that p entails q. it follows that he does not know that p. For if he did know that p, he
would also know that q, which he doesn't.

The details of different sceptical arguments vary in their structure, but each one will assume
some variant of the principle that knowledge is closed under known logical implication. If we
abbreviate "knowledge that p" by "Kp" and abbreviate "entails" by the fishhook sign" -∃," we
can write this principle of closure as the subjunctive principle
P: K(p -∃ q) & Kp  Kq.

If a person were to know that p entails q and he were to know that p then he would know that
q. The statement that q follows by modus ponens from the other two stated as known in the
antecedent of the subjunctive principle P; this principle counts on the person to draw the
inference to q. You know that your being in a tank on Alpha Centauri entails your not being in
place X where you are. (I assume here a limited readership.) And you know also the
contrapositive, that your being at place X entails that you are not then in a tank on Alpha
Centauri. If you knew you were at X you would know you're not in a tank (of a specified sort)
at Alpha Centauri. But you do not know this last fact (the sceptic has argued and we have agreed)
and so (he argues) you don't know the first. Another intuitive way of putting the sceptic's
argument is as follows. If you know that two statements are incompatible and you know the first
is true then you know the denial of the second.

You know that your being at X and your being in a tank on Alpha Centauri are incompatible; so
if you knew you were at X you would know you were not in the (specified) tank on Alpha
Centauri. Since you do not know the second, you don't know the first.

No doubt, it is possible to argue over the details of principle P, to point out it is incorrect as it
stands. Perhaps, though Kp, the person does not know that he knows that p (that is, not-KKp)
and so does not draw the inference to q. Or perhaps he doesn't draw the inference because not-
KK(p -∃ q). Other similar principles face their own difficulties: for example, the principle that
K(p →q) → (Kp → Kq) fails if Kp stops p → q from being true, that is, if Kp →not-(p → q);
the principle that K(p -∃ q)-→ K(Kp → Kq) faces difficulties if Kp makes the person forget
that (p - ∃q) and so he fails to draw the inference to q. We seem forced to pile K upon K until
we reach something like KK(p -∃ q) & KKp → Kq; this involves strengthening considerably
the antecedent of P and so is not useful for the sceptic's argument that p is not known. (From a
principle altered thus it would follow at best that it is not known that p is known.)

We would be ill-advised, however, to quibble over the details of P. Although these details are
difficult to get straight, it will continue to appear that something like P is correct. If S knows that
"p entails q," and he knows that p and knows that "(p and p entails q) entails q" and he does
draw the inference to q from all this and believes q via the process of drawing this inference,
then will he not know that q? And what is wrong with simplifying this mass of detail by writing
merely principle P, provided we apply it only to cases where the mass of detail holds, as it surely
does in the sceptical cases under consideration? For example, I do realize that my being in the
Van Leer Foundation Building in Jerusalem entails that I am not in a tank on Alpha Centauri; I
am capable of drawing inferences now; I do believe I am not in a tank on Alpha Centauri (though
not solely via this inference, surely); and so forth. Won't this satisfy the correctly detailed
principle, and shouldn't it follow that I know I am not (in that tank) on Alpha Centauri? The
sceptic agrees it should follow; so he concludes from the fact that I don't know I am not floating
in the tank on Alpha Centauri that I don't know I am in Jerusalem. Uncovering difficulties in
the details of particular formulations of P will not weaken the principle's intuitive appeal; such
quibbling will seem at best like a wasp attacking a steamroller, at worst like an effort in bad faith
to avoid being pulled along by the sceptic's argument.
Principle P is wrong, however, and not merely in detail. Knowledge is not closed under known
logical implication. S knows that p when S has a true belief that p, and S wouldn't have a false
belief that p (condition (3)) and S would have a true belief that p (condition (4)). Neither of these
latter two conditions is closed under known logical implication. Let us begin with condition

(3) if p were false, S wouldn’t believe that p.

When S knows that p, his belief that p is contingent on the truth of p. contingent in the way the
subjunctive condition (3) describes. Now it might be that p entails q (and S knows this), that S's
belief that p is subjunctively contingent on the truth of p, that S believes q, yet his belief that q
is not subjunctively dependent on the truth of q. in that it (or he) does not satisfy:

(3’) if q were false, S wouldn’t believe that q.

For (3') talks of what S would believe if q were false, and this may be a very different situation
from the one that would hold if p were false, even though p entails q. That you were born in a
certain city entails that you were born on earth.8 Yet contemplating what (actually) would be the
situation if you were not born in that city is very different from contemplating what situation
would hold if you weren't born on earth. Just as those possibilities are very different, so what is
believed in them may be very different. When p entails q (and not the other way around) p will
be a stronger statement than q, and so not-q (which is the antecedent of (3')) will be a stronger
statement than not-p (which is the antecedent of (3)). There is no reason to assume you will have
the same beliefs in these two cases, under these suppositions of differing strengths.

There is no reason to assume the (closest) not-p world and the (closest) not-q world are doxically
identical for you, and no reason to assume, even though p entails q, that your beliefs in one of
these worlds would be a (proper) subset of your beliefs in the other.

Consider now the two statements:

p = I am awake and sitting on a chair in Jerusalem;

q = I am not floating in a tank on Alpha Centauri being stimulated by electrochemical means to
believe that p.

The first one entails the second: p entails q. Also, I know that p entails q; and I know that p. If
p were false, I would be standing or lying down in the same city, or perhaps sleeping there, or
perhaps in a neighboring city or town. If q were false, I would be floating in a tank on Alpha
Centauri. Clearly these are very different situations, leading to great differences in what I then
would believe. If p were false, if 1 weren't awake and sitting on a chair in Jerusalem, 1 would not
believe that p. Yet if q were false, if I was floating in a tank on Alpha Centauri, 1 would believe
that q, that I was not in the tank, and indeed, in that case, I would still believe that p. According
to our account of knowledge, I know that p yet I do not know that q, even though (1 know) p
entails q.

This failure of knowledge to be closed under known logical implication stems from the fact that
condition (3) is not closed under known logical implication; condition (3) can hold of one
statement believed while not of another known to be entailed by the first. It is clear that any
account that includes as a necessary condition for knowledge the subjunctive condition (3), not-
p → not-(S believes that p), will have the consequence that knowledge is not closed under known
logical implication. When p entails q and you believe each of them, if you do not have a false
belief that p (since p is true) then you do not have a false belief that q. However, if you are to
know something not only don't you have a false belief about it, but also you wouldn't have a
false belief about it. Yet, we have seen how it may be that p entails q and you believe each and
you wouldn't have a false belief that p yet you might have a false belief that q (that is, it is not
the case that you wouldn't have one). Knowledge is not closed under the known logical
implication because 'wouldn't have a false belief that' is not closed under known logical
implication.

If knowledge were the same as (simply) true belief then it would be closed under known logical
implication (provided the implied statements were believed). Knowledge is not simply true
belief, however; additional conditions are needed. These further conditions will make knowledge
open under known logical implication, even when the entailed statement is believed, when at
least one of the further conditions itself is open. Knowledge stays closed (only) if all of the
additional conditions are closed. I lack a general non-trivial characterization of those conditions
that are closed under known logical implication: possessing such an illuminating characterization,
one might attempt to prove that no additional conditions of that sort could provide an adequate
analysis of knowledge.

Still, we can say the following. A belief that p is knowledge that p only if it somehow varies with
the truth of p. The causal condition for knowledge specified that the belief was "produced by"
the fact, but that condition did not provide the right sort of varying with the fact. The subjunctive
conditions (3) and (4) are our attempt to specify that varying. But however an account spells this
out, it will hold that whether a belief that p is knowledge partly depends on what goes on with
the belief in some situations when p is false. An account that says nothing about what is believed
in any situation when p is false cannot give us any mode of varying with the fact.

Because what is preserved under logical implication is truth, any condition that is preserved
under known logical implication is most likely to speak only of what happens when p, and q, are
true, without speaking at all of what happens when either one is false. Such a condition is
incapable of providing "varies with"; so adding only such conditions to true belief cannot yield
an adequate account of knowledge.

A belief's somehow varying with the truth of what is believed is not closed under known logical
implication. Since knowledge that p involves such variation, knowledge also is not closed under
known logical implication. The sceptic cannot easily deny that knowledge involves such
variation, for his argument that we don't know that we're not floating in that tank, for example,
uses the fact that knowledge does involve variation. ("If you were floating in the tank you would
still think you weren't, so you don't know that you're not.") Yet, though one part of his argument
uses that fact that knowledge involves such variation, another part of his argument presupposes
that knowledge does not involve any such variation. This latter is the part that depends upon
knowledge being closed under known logical implication. as when the sceptic argues that since
you don't know that not-SK, you don't know you are not floating in the tank, then you also don't
know, for example, that you are now reading a book. That closure can hold only if the variation
does not. The sceptic cannot be right both times. According to our view he is right when he
holds that knowledge involves such variation and so concludes that we don't know, for example,
that we are not floating in that tank: but he is wrong when he assumes knowledge is closed under
known logical implication and concludes that we know hardly anything.9

Knowledge is a real factual relation, subjunctively specifiable, whose structure admits our
standing in this relation, tracking, to p without standing in it to some q which we know p to
entail. Any relation embodying some variation of belief with this fact. with the truth (value), will
exhibit this structural feature. The sceptic is right that we don't track Some particular truths--the
ones stating that his sceptical possibilities SK don't hold--but wrong that we don't stand in the
real knowledge-relation of tracking to many other truths, including ones that entail these first
mentioned truths we believe but don't know.

The literature on scepticism contains writers who endorse these sceptical arguments (or similar
narrower ones), but confess their inability to maintain their sceptical beliefs at times when they
are not focusing explicitly on the reasoning that led them to sceptical conclusions. The most
notable example of this is Hume:

I am ready to reject all belief and reasoning, and can look upon no opinion even as
more probable or likely than another. . . Most fortunately it happens that since reason
is incapable of dispelling these clouds, nature herself suffices to that purpose, and cures
me of this philosophical melancholy and delirium, either by relaxing this bent of mind,
or by some avocation, and lively impression of my senses, which obliterate all these
chimeras. I dine, I play a game of backgammon, I converse, and am merry with my
friends; and when after three or four hours' amusement, I would return to these
speculations, they appear so cold, and strained, and ridiculous, that I cannot find in my
heart to enter into them any farther. (A Treatise of Human Nature, Book I, Part IV,
section VII.)

The great subverter of Pyrrhonism or the excessive principles of skepticism is action,

and employment, and the occupations of common life. These principles may flourish
and triumph in the schools; where it is, indeed, difficult, if not impossible, to refute
them. But as soon as they leave the shade, and by the presence of the real objects, which
actuate our passions and sentiments, are put in opposition to the more powerful
principles of our nature, they vanish like smoke, and leave the most determined skeptic
in the same condition as other mortals. . . And though a Pyrrhonian may throw himself
or others into a momentary amazement and confusion by his profound reasonings; the
first and most trivial event in life will put to flight all his doubts and scruples, and leave
him the same, in every point of action and speculation, with the philosophers of every
other sect, or with those who never concerned themselves in any philosophical
researches. When he awakes from his dream, he will be the first to join in the laugh
against himself, and to confess that all his objections are mere amusement. (An Enquiry
Concerning Human Understanding, Section XII, Part II.)

The theory of knowledge we have presented explains why sceptics of various sorts have had
such difficulties in sticking to their far-reaching sceptical conclusions "outside the study," or
even inside it when they are not thinking specifically about sceptical arguments and possibilities
SK.
The sceptic's arguments do show (but show only) that we don't know the sceptic's possibilities
SK do not hold; and he is right that we don't track the fact that SK does not hold. (If it were to
hold, we would still think it didn't.) However, the sceptic's arguments don't show we do not
know other facts (including facts that entail not-SK) for we do track these other facts (and
knowledge is not closed under known logical entailment). Since we do track these other facts--
you, for example, the fact that you are reading a book; I, the fact that I am writing on a page--
and the sceptic tracks such facts too, it is not surprising that when he focuses on them, on his
relationship to such facts, the sceptic finds it hard to remember or maintain his view that he does
not know those facts. Only by shifting his attention back to his relationship to the (different)
fact that not-SK, which relationship is not tracking, can he revive his sceptical belief and make
it salient. However, this sceptical triumph is evanescent, it vanishes when his attention turns to
other facts. Only by fixating on the sceptical possibilities SK can he maintain his sceptical virtue;
otherwise, unsurprisingly, he is forced to confess to sins of credulity.

NOTES

1. See Hilary Putnam, Reason, Truth and History (Cambridge, 1981), ch. I.
2. I should note here that I assume bivalence throughout this chapter, and consider only
statements that are true if and only if their negations are false.
3. See Robert, Stalnaker, “A Theory of Conditionals”, in N. Rescher, ed., Studies in Logical
Theory (Oxford 1968); David Lewis, Counteifactuals (Cambridge 1973); and Jonathan
Bennett’s critical review of Lewis, “Counterfactuals and Possible Worlds”, Canadian
Journal of Philosophy, 4/2 (Dec. 1974), 381-402. Our purposes require, for the most
part, no more than an intuitive understanding of subjunctives.
4. G. C. Stine, “Skepticism, Relevant Alternatives and Deductive Closure”, Philosophical
Studies, 29 (1976), 252, who attributes the example to Carl Ginet.