CHAPTER THREE

The Concept of Knowledge

Having examined Descartes’s epistemological view as a kind of prologue, we

will now turn to a more detailed consideration of a variety of more specific
epistemological issues, focusing mainly on those that naturally arise out of his
discussion. Our first specific concern will be to achieve a deeper understand-
ing of the concept of knowledge itself. What is it to know something? What,
that is, are we saying of a person when we ascribe knowledge to him or her?
A further set of questions, already briefly noticed in chapter 1, concerns the
significance of the concept of knowledge. Why does it involve the specific
conditions that it does? How do those conditions fit together or connect
with each other in an intelligible way? And, most fundamentally, why do or
should we care about knowledge? Why is having knowledge important and
valuable in the way that we normally take it to be (if indeed it really is)?
We have already encountered one specific account of the concept of
knowledge, the one that Descartes seems to have roughly in mind (though
without formulating it very explicitly) in the Meditations. According to that
account, for a person S to know some proposition1 P at some time t, the fol-
lowing three conditions must be satisfied (with the subscripts indicating that
these are the conditions of the Cartesian conception of knowledge):

1C. S must believe or accept P at t without any doubt.

2C. P must be true.
3C. S must have at t a reason or justification that guarantees that P is
true.2

23
24  Chapter Three

It is obvious on reflection that condition (3C) makes condition (2C) redun-

dant and so unnecessary: If S has a reason that guarantees P’s truth, then it
follows automatically that P is in fact true. But since there are other accounts
of knowledge that we will want to consider and compare with this one in
which the condition that is parallel to Descartes’s condition (3C) does not in
this way entail that the condition parallel (and in fact usually identical) to
(2C) is satisfied, it will be clearer to list condition (2C) separately in spite of
its redundancy in this case.
The Cartesian account of knowledge is in fact one specific version of a
more general account of knowledge that has come to be generally referred to
as “the traditional conception of knowledge.” According to this more gen-
eral account, knowledge requires the satisfaction of three conditions at least
roughly parallel to Descartes’s: (1) a belief or acceptance condition, (2) a
truth condition, and (3) a reason or justification condition (so that accounts
of this kind are often referred to as justified true belief accounts or definitions
of knowledge). Other specific versions of this general account almost always
share Descartes’s truth condition (2C), but differ somewhat in their specifica-
tion of the belief or acceptance condition parallel to (1C) and to a wider and
more serious extent in their specification of the reason or justification condi-
tion parallel to (3C). As we will eventually see, there are also many recent
accounts that add a further condition (4), while still retaining conditions
parallel to Descartes’s three.
We will begin by examining the three general kinds of conditions in-
cluded in the traditional account, considering the rationale for including
each of them as an essential part of the concept of knowledge, trying to
understand the general nature of each condition, and discussing, briefly for
condition (1) and more extensively for condition (3), some of the different
ways in which the general condition in question has been further specified
by different versions of the traditional account.

The Belief or Acceptance Condition

The basic rationale for this first general condition is quite straightforward:
Someone who is in serious doubt as to whether a particular proposition is true
or, perhaps even more obviously, who has never so much as considered or en-
tertained that proposition can surely not be correctly said to have knowledge
of it. If I am completely uncertain about whether it will rain tomorrow, then
I do not know that it will (or that it won’t). And if it has never so much as
occurred to me that my roof might be leaking, then again I plainly do not
know that it is—even if in fact it is leaking and even if I have what would
 The Concept of Knowledge  25

be good evidence for this being so if I were to recognize it as such (there are
damp spots on the rug and distinctive streaks on the walls).
The most obvious way to satisfy a condition of this general sort would be
for the person in question to be in the conscious state of explicitly consider-
ing and assenting to the proposition in question. This might involve, as the
formulation just given seems to suggest, a two-stage process: for example, my
wife suggests to me that perhaps the roof is leaking, and after considering the
evidence, I end up becoming convinced that this is indeed what is going on. It
is also possible, however, that the truth of the proposition strikes me as obvious
as soon as it enters my mind, without any preliminary stage of consideration.
But while this is one way in which a condition of the indicated sort might
be satisfied, it seems reasonably clear that it is not the only way, that people
can and do know many things at a particular time that they do not have
explicitly in mind at that time. I am about to give you an example of such a
piece of knowledge, something that it seems plain that you in fact know even
as you read these words though you do not at the moment have it explicitly
in mind. (It was, of course, precisely to produce this situation that I didn’t
initially specify the proposition in question.) Consider, then, the claim that
you are a human being, where what is intended is that each reader formulate
the appropriate version of this general sort of claim, the one that applies to
himself or herself. My suggestion is that the claim in question is something
that you knew to be true while reading the earlier part of the present para-
graph, even though you almost certainly did not have it explicitly in mind.
If this is right, then a correct formulation of the belief or acceptance condi-
tion for knowledge should not require explicit, conscious acceptance of the
relevant proposition at the time in question, even though this is clearly one
way in which such a condition might be satisfied.
Perhaps the most standard way of handling this point is to first formulate
the condition in question as the requirement that the person who has knowl-
edge believe the proposition in question and to then distinguish two different
kinds of belief (or, as it is often put, two “senses” of the term “belief”): oc-
current belief, which is what happens when the person has the proposition
explicitly in mind and accepts or assents to it; and dispositional belief, where
the person does not have the proposition explicitly in mind, but is disposed
to accept or assent to it, that is, would accept or assent to it if the issue were
raised. Thus in the case of the example just given, the suggestion would be
that each of the readers of this book had a dispositional belief (or believed
dispositionally) that he or she is a human being, though it is quite possible
that none of you had an occurrent belief to this effect at the time just before
the proposition was explicitly mentioned.
26  Chapter Three

There is, however, a problem lurking here which needs to be dealt with.
The following situation sometimes, perhaps even fairly frequently, occurs:
there is a proposition that a person has never consciously or explicitly con-
sidered, still less consciously assented to, but which is in some way obvious
enough that he or she would immediately accept or assent to it if it were
proposed. In such a case, the requirements of the so-called dispositional sense
of “belief,” as just given, seem to be satisfied, but it still seems plainly wrong
to say that the person believes the proposition in question—and even more
plainly wrong to say that he or she knows it, even if all of the other condi-
tions for knowledge should happen to be satisfied.3 Consider as an example
here the version of the leaking roof case discussed earlier in which I would
accept at once the proposition that my roof is leaking if proposed by my wife
or if it just happened to occur to me, but where I neither have it presently in
mind nor have accepted or assented to it earlier, and suppose also that Des-
cartes’s conditions are at least roughly on the right track. In such a situation,
even if the evidence I have (the wet spots on the floor and distinctive streaks
on the wall) would be enough to satisfy the correct version of the reason or
justification condition, and even if the claim in question is in fact true (and
even if whatever further conditions there might be in addition to or instead
of these two are also satisfied), it still seems plainly wrong to say that I know
that my roof is leaking—even though it does not seem wrong to say of the
readers of this book that they knew that they were human beings even at the
time prior to my explicitly suggesting this proposition to them.
For this reason, rather than defining dispositional belief in the way sug-
gested earlier, it should be specified instead as the dispositional state in
which (a) one has previously explicitly considered and consciously accepted
or assented to the proposition in question, and (b) as a direct result of this
prior acceptance or assent, would accept or assent to it again if the question
were explicitly raised. It is also perhaps a bit clearer not to use the term
“belief” for the first alternative of conscious or explicit acceptance or assent.
We can then say that condition (1) of the standard conception of knowledge
should be understood to require that the person in question either explicitly
and consciously accepts or else (dispositionally) believes the proposition in
question at the time in question. (Having clarified the point, I will some-
times for the sake of brevity follow fairly standard philosophical practice by
using the term “belief” to cover both dispositional belief under the corrected
specification and conscious, explicit acceptance or assent.)
One further issue, the main one that distinguishes different versions of
the belief or acceptance condition, is how strongly the person must accept
or believe the proposition in question, that is, how strongly they must
 The Concept of Knowledge  27

be convinced that it is true. The Cartesian view, as formulated earlier,

requires that the person have no doubt at all that the proposition is true,
a condition that is also sometimes formulated by saying that he must be
certain of it. This is a very strong version of the belief or acceptance require-
ment—one that many or probably most of the things that we seem ordinar-
ily to regard as instances of knowledge (see the list of examples in chapter
1) would not satisfy, though this fact is obscured somewhat by a tendency
to exaggerate when saying that claims are certain or indubitable. Thus I
might well say that I am certain that my dog is in the yard where I just left
him or certain that Obama was elected president in 2008 or certain that it
is very hot in the center of the sun; but if pressed, I would have to admit
that none of these claims is really beyond all possible doubt. (Try to think
in each of these cases and in various others of ways in they might be false
that you can see to be at least possible. Do so, if you can, without appealing
to anything as outlandish as the Cartesian evil genius.)
A significantly weaker version of the belief or acceptance condition
would say instead merely that the person must be fairly confident, reason-
ably sure in his or her belief or acceptance of the proposition in question.
This is a requirement that seems to agree much better with our common-
sense judgments about the extent of our knowledge (as reflected in the
list in chapter 1). Is this a good reason for thinking that it is this second,
weaker requirement that is in fact correct as one part of an account of the
concept of knowledge?

A Digression on Method
There is an important—and difficult—issue of philosophical method per-
taining to this last point, one that is indeed also relevant to the earlier
examples, and this is as good a place as any to discuss it. As we have already
seen, there are many claims of many different kinds, roughly indicated in the
earlier list, that from a “common-sense” or “intuitive” standpoint count as
cases of knowledge (with obviously some significant variation from person
to person). What this means is that most ordinary people and even most
philosophers, if asked to consider whether such an example is a case of
knowledge, would be inclined to say without much hesitation that it is. Thus
we have (a) a proposed requirement for knowledge, the Cartesian require-
ment that a proposition that is known must be believed without any doubt
or with certainty, together with (b) a large number of commonsensically or
intuitively accepted cases of knowledge that do not satisfy the requirement in
question—and so, if this requirement is correct, must not be genuine. Either
28  Chapter Three

the intuitive judgments about these particular cases or the proposed require-
ment must apparently be mistaken (assuming that the concept of knowledge
is unambiguous), but how are we to decide between these two alternatives?
There are at least two reasonably clear things to be said about this issue,
though they are not, alas, sufficient to resolve it. First, common-sense or
intuitive judgments about particular cases are a central and essential part
of our basis for understanding and delineating concepts like the concept of
knowledge. This is just to say that if all such judgments were dismissed as
undependable, we would have little handle left on such concepts. (Imagine
trying to figure out what knowledge is if you have no idea at all which par-
ticular examples in fact qualify as cases of knowledge. How would you begin?
What would you rely on?)
But, second, while common-sense or intuitive judgments of the sort in
question are in this way indispensable, there is no apparent reason to regard
them as somehow simply incapable of being mistaken. This would be so even
if it were not the case—as in fact it is—that the common-sense or intuitive
judgments of different people or of the same person at different times often
conflict with each other. And if it is possible for such judgments to some-
times be mistaken, then it is hard to rule out completely the possibility that
they might be largely or even entirely mistaken, so that some requirement
that most or all of the intuitive or commonsensical cases of alleged knowl-
edge fail to satisfy might still be correct.
The upshot of these considerations is thus rather inconclusive. It seems
right to say that the fact that the Cartesian version of the first condition
conflicts with our common-sense or intuitive judgments about cases of
knowledge counts against it and in favor of the weaker version mentioned
above. (Similarly, on an earlier issue, the fact that it seems intuitively wrong
to say that I know that my roof is leaking when the proposition in question
has never explicitly occurred to me counts in favor of a version of the first
condition that would not be satisfied in that case.) But this resolution of the
issue between the two versions of the first condition is not decisive, since
there is no guarantee that the relevant “intuitions” are correct. At least some
possibility remains that the Cartesian condition is correct after all—in which
case, hopefully, we might be able to find further reasons of some sort that
point in this direction.

The Truth Condition

The rationale for the truth condition is simply that one cannot know what is
not the case, something that almost no philosopher has seriously disputed. If
 The Concept of Knowledge  29

I know that my car is in the parking lot, then it must actually be there; if it is
not, then I did not in fact know that it was there, no matter how sure I may
have been and how strong my reasons or justification may have been.
One thing that sometimes makes people balk at accepting the truth con-
dition is that someone can, of course, think that he knows something when
in fact it is not true. Thus in the case just given, I may still think that I know
that my car is in the parking lot. Similarly, many people living prior to the
exploits of Columbus believed that the earth was flat and thought that this
was something that they knew. And many scientists and others living prior
the work of Einstein believed that Newtonian mechanics was an exactly
correct description of the behavior of material bodies and again thought that
this was a case of knowledge, indeed an exceptionally clear one. Moreover,
in describing cases of this kind, it is sometimes tempting, and perhaps even
useful in some ways, to temporarily take the point of view of the people in
question and thus describe the situation by saying that they knew the claim
in question—that is, that from their perspective it clearly seemed that they
knew. According to all versions of the traditional conception of knowledge,
however, such ascriptions of knowledge where the proposition in question
is false are always mistaken, however reasonable and obvious they may have
seemed to the people in question.
Here we have a somewhat more subtle example of the appeal to intuitive
or common-sense judgments. In general, it seems intuitively wrong to ascribe
knowledge where the claim in question is not in fact true. This is why a per-
son who claims to know something will normally withdraw that claim when
it is demonstrated in some way that the claim in question is mistaken and
will concede that he or she did not know after all. But there are also certain
cases, such as that of beliefs about the shape of the earth prior to Columbus,
where there seems to be something right about saying that the people in ques-
tion knew something that wasn’t so. This conflict is resolved by pointing out
that the ascription of knowledge in such cases in effect reflects the point of
view of the people in question, from which the proposition seemed true; thus
this ascription can still be said to be mistaken from a more objective stand-
point in which the falsity of the claim is acknowledged.
A related problem that you may perhaps have with the truth condition
arises from worrying about how you could ever tell that it is satisfied. As we
will see further below, a person does in a way have to determine that the
proposition is true, according to the traditional conception at least—some-
thing that is accomplished by appeal to the reasons or justification for it. But
it is tempting to make the mistake of thinking of the truth condition as one
whose satisfaction has to be somehow determined by the would-be knower
30  Chapter Three

independently of the satisfaction of the other two conditions, and the prob-
lem is then that there is no apparent way to do this. As the point is some-
times put, you cannot just “step outside” of your own subjective perspective
and observe independently that the claim that you believe and for which you
perhaps have good reasons or justification is also true—there is just no way
to occupy such a “God’s-eye” perspective. But a proponent of the traditional
conception will reply that what this shows is not that the truth condition is
mistaken, but rather that it is a mistake to think of it as a condition that a
person must determine independently to be satisfied in order to have knowl-
edge; instead, it is just a condition that must in fact be satisfied (something
that is in fact true of all of the conditions in question).
A useful way in which this point is sometimes put is to say that the
concept of knowledge is a “success” concept, that is, that it describes the
successful outcome of a certain kind of endeavor. The aim of the cognitive
enterprise is truth: we want our beliefs to correctly describe the world. And,
according to the traditional account of knowledge, we attempt to accomplish
this by seeking beliefs for which we have good reasons or strong justification.
When this endeavor is successful, that is, when the justified beliefs thus ar-
rived at are in fact also true, then we have knowledge; when it fails, when
the resulting strongly justified beliefs are not in fact true,4 we have only what
might be described as “attempted knowledge.” But the distinction between
genuine and merely attempted knowledge is not one that we have to, or
indeed in the short run could, independently draw. (A crude but still helpful
comparison: When shooting an arrow at a target, the aim is to hit the target,
and this is something that we attempt to achieve by aiming carefully. But
whether or not we succeed depends on whether the arrow does in fact hit
the target, and this may be so, in which case we have succeeded, even if we
have no independent way to establish that it is so—even if the target is only
briefly visible and cannot, for some reason, be examined later.)
A deeper and more difficult question, one that is rather more metaphysi-
cal than epistemological in character, concerns the nature of truth itself:
What does truth amount to? What does it mean to say that a particular
proposition is true? Here there is one answer that is both the most widely
accepted and also the one that is seemingly in accord with common sense.
But it is an answer that philosophers of very different persuasions have often
regarded as problematic or even as not fully intelligible. Thus we need to take
a brief look at this controversy, even though a full discussion of it is beyond
the scope of this book.
The widely accepted, commonsensical view is what has come to be
known as “the correspondence theory of truth.” It says that a proposition
 The Concept of Knowledge  31

is true if it corresponds to or agrees with the relevant aspect or part of reality.

Thus, for example, for the proposition that my car is in the parking lot to
be true, according to the correspondence theory, is for the content of this
proposition (that is, what I believe or accept when I believe or accept this
proposition) to agree with or match the appropriate aspect or chunk of in-
dependent reality—in this case, the physical configuration that involves a
certain complicated structure of metal, plastic, rubber, and so forth (my car)
being physically juxtaposed in the right way or not with a certain piece of
asphalt (the parking lot). The physical configuration that would make the
proposition true is something that one could point to or physically mark off
(with police tape or by building a box around it) quite independently of the
proposition or the various beliefs that involve it; the various ones that would
make it false are in general less localized, but could also be pointed at in a
way by indicating the two separate elements and their failure to realize the
indicated relation.
Many different sorts of problems and objections have been raised in rela-
tion to the correspondence theory, but probably the most widespread of these
involve doubts about how the relation of correspondence should itself be
explicated or clarified—or indeed whether it can be intelligibly explicated
at all. It has often been suggested that correspondence must be construed as
some sort of complicated structural isomorphism or relation of “picturing”
between (a) the components of the proposition, or perhaps of the linguistic
expression of the proposition, and (b) the relevant chunk or aspect of reality.
And intelligibly defining or specifying a relation of this sort has been argued
to be difficult or perhaps impossible. One reason offered for this claim is that
the relation would have be realized by propositions about very widely dif-
ferent sorts of subject matter, such as concrete physical situations (as in the
example just discussed), general physical laws, historical facts, facts about
mental states, abstract logical and mathematical facts, and perhaps norma-
tive or valuational facts. But how, it is asked, could there be one relation of
the sort in question that is realized in cases as different as these? How could
the very same relation that obtains between the proposition that my car is
in the parking lot and the physical configuration described earlier also ob-
tain between the proposition that 2 ⫹ 3 ⫽ 5 and the abstract mathematical
fact to which it would presumably have to correspond? A second, perhaps
even deeper reason often given for thinking that a specification of the cor-
respondence relation is impossible is that to formulate such a specification,
we would have to be able to talk about or indicate both sides of the relation:
both the conceptually formulated, linguistically expressible propositional
content and the mind-independent, nonconceptual aspect or chunk of
32  Chapter Three

reality. But, the argument goes, we have no way to get at the latter element
except via further conceptual, propositional descriptions, which thus, it is
claimed, merely presuppose the correspondence relation (assuming for the
sake of the argument that the correspondence theory is true) without really
helping to explain it. In effect, an attempted account of an instance of the
correspondence relation only exhibits a relation between two conceptual,
propositional descriptions and not one between such a description and a
hunk of independent, nonconceptual reality.
These reasons for doubting whether an intelligible specification of the
correspondence relation is possible raise difficult issues, and it would take
quite a bit of discussion, more than there is room for here, to get to the bot-
tom of them. Fortunately, however, there is a way of seeing that the problems
they raise, though perhaps important in other ways, need not be solved in
order to make sense of the correspondence theory of truth itself. The mistake
that is made by these reasons and the objection that they support is thinking
that the intelligibility of the correspondence theory requires a generally ap-
plicable specification of the relation of correspondence in the way that they
suppose, at least if such a specification is supposed to be more than utterly
straightforward and trivial. Any intelligible proposition, after all, says that
reality (in the broadest sense of the term) is a certain way or has certain fea-
tures that the content of the proposition specifies. And the best way to un-
derstand the correspondence theory, following Aristotle’s original statement
of it,5 is to construe it as saying no more than that such a proposition is true if
reality is whatever way or has whatever features the proposition describes it as
having. In some cases, the content of a supposed proposition may be less than
fully clear or intelligible, but that is a problem for that supposed proposition
and not for the correspondence theory. A way of putting this point is to say
that the only specification needed as to how reality would have to be to cor-
respond to a particular proposition and so of what correspondence for that
particular proposition involves is provided by the propositional content itself
and need not be independently specified by the correspondence theory—a
point that also allows for very different sort of propositions to describe and
so correspond to reality in their own distinctive ways. (There is no room
here for a consideration of the various other objections that have been raised
against the correspondence theory, though none of these seems to me in the
end to have any more force than the one just discussed.)
The belief that the correspondence theory is untenable has also led phi-
losophers to propose a variety of alternative theories or accounts of truth,
some of which have also been motivated by related doubts as to whether
truth would be knowable or accessible if understood in the way indicated by
 The Concept of Knowledge  33

the correspondence theory. These views cannot be discussed in any detail

here, but a brief enumeration of the most important alternatives and their
main problems may help to give you some idea of what they involve:
(1) The coherence theory of truth. According to this view, the truth of a
believed proposition simply consists in its fitting together coherently with
other propositions that are believed, where coherence involves both logi-
cal consistency and (usually) other relations of mutual support or explana-
tion. (It is important to understand that this is supposed to be what truth
ultimately amounts to, not merely—which would be substantially more
plausible—a test or criterion for determining what is true.6) Since this view
seems implicitly to deny the existence of any objects of knowledge beyond
beliefs and their propositional contents (for admitting such objects would
lead inevitably back to the correspondence theory), it seems to require an
idealist metaphysics in which only mental states (and the minds that have
them?7) genuinely exist.
In addition to the intuitive implausibility of this idealist view, there is
the further objection (among others) that it seems possible for there to be
many different and incompatible coherent systems of believed propositions,
all of the members of which would be true according to the coherence
theory—which appears (think about it) to be an absurd result. (This point
is sometimes made by suggesting that the propositions reflected in a well-
written novel might seemingly satisfy the requirement of coherence, so that
the beliefs of someone who accepted all of them would thereby be true ac-
cording to the coherence theory.)
(2) The pragmatic theory of truth. There are a number of different versions
of this view, but we will limit ourselves here to the simplest, advanced by
the American pragmatist William James,8 which holds that the truth is what
“works”: that is, that for a believed proposition to be true is for the holding
of that belief to lead in general to success in practice. Now there can be little
doubt that believing true propositions often leads to success in this way and
that believing false propositions often leads to failure: for example, if I have
a true rather than a false belief about the location of my car, then my efforts
to get to it and drive home are obviously much more likely to succeed.9 But
is such a belief true because it produces success, since producing success is
just what truth is (as the pragmatic theory claims)? Or isn’t it exactly the
other way around: doesn’t the belief lead to success because it is true (in the
correspondence sense)?
(3) The redundancy or “disappearance” theory of truth. Some recent phi-
losophers, seeking to avoid the problems that (as they see it) arise from the
correspondence theory and these other theories of truth, have suggested that
34  Chapter Three

there is really no need for any philosophical theory of the nature of truth.
They point out the necessary equivalence between assertions of the form
“P is true,” for some proposition P, and the simple assertion that P, for ex-
ample between the assertion that it is true that my car is in the parking lot and
the assertion simply that my car is in the parking lot: if one of these claims
is true, then the other must be true also, and vice versa (think about it).
But this equivalence means, they argue, that the assertion that a particular
proposition is true means or says no more than the simple assertion of that
proposition, in which case the former can always be replaced by the latter,
and any mention of truth thus disappears. Their conclusion is that talk of
truth is simply redundant—nothing more than a needlessly elaborate way of
asserting the propositions in question.
One problem with this view is that there are cases where the claim is made
that something is true but where the proposition in question is left unstated
(“what Tom said was true”), so that the proposed replacement doesn’t work.
A deeper problem is that the equivalence after all works both ways, and thus
could at least as reasonably be taken to show that all propositional assertions
are implicitly assertions that the proposition in question is true—which
would make an understanding of truth essential for even understanding the
idea of assertion or belief.
Having briefly canvassed these alternatives, I propose to follow common
sense and the main weight of philosophical opinion by assuming that it is
the correspondence theory that gives the correct account of truth, and un-
derstanding the second condition for knowledge accordingly.10

The Reason or Justification Condition

The easiest way to understand the need for this third general condition as a
part of the concept of knowledge is to consider briefly the suggestion that no
such condition (and no other condition beyond the two already discussed)
is necessary, that knowledge can be correctly understood merely as true be-
lief. In fact, there are a few kinds of situation where it does seem reasonably
natural to say that a person knows something, even though only these two
conditions are satisfied. Here the clearest examples are cases where (a) some
secret is being hidden from someone, and (b) in which the purpose for the
secrecy will be defeated if the person being kept in the dark comes to have
a confident true belief about the matter in question, whether or not he or
she has any reason or justification at all for this belief. Thus, for example, if
I am hiding from Susan (whether seriously or in a game) and she confidently
guesses my location and heads in that direction, I might say to myself or to
 The Concept of Knowledge  35

someone hiding with me “Susan knows where I am hiding”—even if it is just

a hunch for which she has no basis at all or even if it results somehow from
some sort of mistake or confusion on her part.
But apart from such relatively rare cases in which the truth of the belief
in question is in effect all that matters, it seems clear (a point that has been
recognized at least since Plato11) that a mere lucky guess or hunch does not
suffice for knowledge even though it undeniably may produce a true belief.
Really clear illustrations of this point are not easy to find because it is unusual
for a person to believe confidently that something is so even when he or she
lacks any real basis for the belief. But what is clear is that in a case where
the proposition believed happens to be true only by mere luck or accident, a
person does not come to know merely by somehow managing to have a suf-
ficiently confident belief. Thus, for example, if a person on a multiple-choice
type quiz show has no idea at all about the answer to a particular question
and simply hits the right answer by luck, it would be mistaken to ascribe
knowledge to them (prior to their being told that the answer was correct)
even if they did manage to believe confidently that the choice was correct.
(Here again we have an appeal to intuition.) Similarly, a rabid sports fan who
is utterly sure that his team will win a certain game even though there is no
real evidence or other basis for this claim did not know beforehand that his
team would win even if in fact it does. And even in the case discussed earlier,
it would be easy to challenge the claim that Susan really knows where I am.
The right account seems to be rather that in that specific case (and some
others) a true belief is just as good as knowledge and can therefore, in what
amounts to a kind of exaggeration, be described as such.
What more then is needed for knowledge than a true belief, perhaps a
very highly confident one? The answer offered by the traditional conception
of knowledge is that one further ingredient is needed: a sufficiently strong
reason or justification for thinking that the claim in question is true. Here the
last part of the specification is essential, for there are other sorts of reasons or
justification that I might have for holding a belief that would not be of the
right kind to yield knowledge. I might believe something out of loyalty to a
friend or out of commitment to a religious tradition (also a sort of loyalty)
or perhaps even just because it makes me happier to do so, but such beliefs
do not thereby constitute knowledge even if they should happen to be true.
What is needed for knowledge, according to the traditional conception, is
a reason or justification of a sort that is truth-conducive: one that increases
or enhances (to the appropriate degree—see below) the likelihood that the
belief is true. Such a reason or justification is standardly referred to as an
epistemic reason or as epistemic justification.
36  Chapter Three

The most familiar and obvious way to have an epistemic reason for some-
thing that I believe is to have evidence in favor of the truth of the proposi-
tion in question. In the clearest sort of case, evidence consists in further
information of some appropriate sort in light of which it becomes evident
that the proposition is true. Thus, for example, a police detective might have
evidence in the form of fingerprints, eyewitness testimony, surveillance pho-
tographs, and the like, pointing strongly to the conclusion that a particular
person is guilty of the crime he or she is investigating. A scientist might have
evidence in the form of instrumental readings and laboratory observations
in favor of the truth of a particular scientific theory. And a historian might
have evidence in the form of manuscripts and artifacts for the occurrence of
a particular historical event.
It is less clear whether the concept of evidence can be extended to en-
compass all cases in which someone has an epistemic reason or epistemic
justification. From an intuitive standpoint, it seems clear that my belief that
2 ⫹ 3 ⫽ 5 is epistemically justified, that I have a reason or basis of some sort
for thinking that it is true. (I am not merely guessing, nor am I accepting the
claim on the basis of authority; rather, I see or grasp directly why the claim is
true, indeed why it must be true.) But do I really have evidence that supports
the proposition in question? If so, what exactly is it? As we saw in the previ-
ous chapter, philosophers have spoken in cases of this kind of self-evidence,
where this seems to mean that the very content of the proposition in question
somehow provides or constitutes evidence for its own truth. We will investi-
gate this idea of self-evidence more fully later on,12 but it is clear at least that
self-evidence does not involve evidence in the most ordinary sense—that is,
it does not involve a separate body of information that supports the proposi-
tion in question, for otherwise it would not be self-evidence.
There are still other sorts of cases of apparent epistemic reasons or jus-
tification to which the concept of evidence does not comfortably apply.
What about cases of ordinary sensory perception, for example, my present
perception of a large green coniferous tree outside my window? Do I have
evidence for the existence and character of the tree, and if so what might it
be? Philosophers have sometimes spoken in such cases of “the evidence of
the senses,” but it is far from obvious how this idea should be understood
or, here again, that it involves a separate body of supporting information.
(Though it is worth noting that philosophers have also sometimes spoken of
“sensory information”—can you see anything in such a case that this might
refer to?) What about cases of memory? I believe and seem to know that I had
Grape-Nuts for breakfast this morning, but do I have evidence for this claim
when I simply remember it (as opposed to checking the traces left in the
 The Concept of Knowledge  37

bowl)? And what about my apparent knowledge of my own states of mind, of

my “immediate experience”? I believe and seem to know that I am currently
thinking about the concept of knowledge, that there is a large patch of dark
green in my visual field, that I have an itch in my left elbow, and that I am
determined to finish this chapter today, but do I have evidence for any of
these claims? In this last sort of case, philosophers have also sometimes ap-
pealed to the idea of self-evidence. This too will be considered later, but we
can see immediately that this is again a rather strained use of the ordinary
notion of evidence.
All of these matters will require further discussion later on in this book.
For the moment, we can say that the concepts of an epistemic reason or of
epistemic justification as they figure in the traditional concept of knowl-
edge are, if not simply identical to the concept of evidence, at least fairly
straightforward generalizations of that concept. First, they involve a basis of
some sort for thinking that the proposition in question is true or likely to be
true, even if not necessarily the sort of separate body of information that the
idea of evidence most naturally suggests. Second, on the most standard and
obvious interpretation, these concepts also seem to involve the idea that this
truth-conducive basis is something that is within the cognitive possession of
the person whose belief thereby comes to be justified, that is, that it is some-
thing that he or she is aware of in some way that would allow it to be cited
as a reason or as giving justification for the belief in question.13
Yet a further issue pertaining to the reason or justification condition for
knowledge is how strong the reason or justification must be, that is, how
likely it must make it that the proposition in question is true, for knowledge
to result. We have already seen Descartes’s apparent view on this point: the
reason must be conclusive, must guarantee the truth of the proposition in the
sense that it is impossible for the proposition to be false, given that reason.
Accounts of knowledge that, like the Cartesian account, involve this strong
version of the reason or justification condition are sometimes referred to as
versions of the “strong conception of knowledge” (or the strong sense of the
term “knowledge”).14
It is fairly easy to see the appeal of the strong version of the reason or
justification condition and the strong conception of knowledge that results.
If, as suggested earlier, the aim of our cognitive endeavors is truth and our
reasons or justification are our means for achieving this goal, then only a
reason or justification that satisfies the strong version of the condition allows
us to be sure that the goal has in fact been achieved; with anything less than
this, success would be to some extent uncertain. Moreover, this interpreta-
tion of the third condition for knowledge seems to agree with at least some
38  Chapter Three

of the ways in which we ordinarily use the term “know”: given inconclusive
evidence, it is natural for a person to say, at least if pushed, that he or she
doesn’t really know that the claim in question is true despite having a fairly
good reason for believing it.
But the main problem with the strong conception of knowledge is that
there seem to be many, many cases that we commonsensically or intuitively
regard as cases of knowledge where the strong version of the reason or justi-
fication condition is clearly not satisfied. As we have learned from Descartes
(even though he himself seems sometimes to lose sight of this lesson in the
later stages of the Meditations), it is very hard to find beliefs for which there
is not some possible way in which the proposition in question could be false
in spite of the reasons or justification for thinking that it is true. Given pos-
sibilities like the evil genius, it is doubtful whether any beliefs about the
material world outside of our minds or about the past will count as knowl-
edge, according to the strong conception. Indeed, contrary to Descartes, it
can even be questioned whether beliefs about our own states of minds will
constitute knowledge according to this strong standard: is it really impossible
(given my evidence or basis, whatever exactly it is) that I could be mistaken
about whether I am experiencing a specific shade of color or about how se-
vere a sensation of pain is? Thus if the strong conception is the right account
of knowledge, it may well follow that we have virtually no knowledge at all,
perhaps nothing beyond the minimal knowledge for each of us of his or her
own existence. And this result seems to conflict both with common-sense in-
tuition and with our ordinary usage of the terms “know” and “knowledge.”
It is this sort of objection that has led most recent philosophers to adopt
versions of what is sometimes referred to as the “weak conception of knowl-
edge” (or the weak sense of “knowledge”).15 According to these views, the
correct version of the reason or justification condition does not require con-
clusive reasons or justification for there to be knowledge. What is required is
instead only reasonably strong reasons or justification, strong enough to make
it quite likely that the proposition in question is true, but not necessarily
strong enough to guarantee its truth. It is at least fairly plausible to suppose
that most or all of the beliefs that we intuitively regard as cases of knowledge
do in fact satisfy this less demanding condition.16
In fact, there is a connection here between the first and the third condi-
tions of any particular version of the traditional conception of knowledge.
It seems to be plainly irrational for a person to believe something more
strongly than the strength of their reason or justification would warrant
(and perhaps also, though less obviously so, to believe it less strongly).
Thus if we assume, reasonably enough it would seem, that knowledge in-
 The Concept of Knowledge  39

volves beliefs that are rationally held, then accepting the weak version of
the reason or justification condition is a good reason for also accepting the
weaker version of the belief or acceptance condition that was mentioned
at the end of the discussion of that condition; whereas one who accepts the
strong version of the reason or justification condition has no reason not to
accept (and perhaps good reason in favor of accepting) a comparably strong
version of the belief or acceptance condition.
One very obvious question to ask about the weak conception is how likely
the truth of the proposition must be to satisfy this weaker version of the
reason or justification condition. If, as seems at least initially reasonable, the
level of likelihood can correctly be thought of as something like a level of
probability, then just how probable must it be in light of the reasons or justi-
fication available that the proposition is true in order for it to be adequately
justified to count as knowledge? Presumably more than mere 51 percent
probability is required, since it seems intuitively wrong to say that a person
knows something that is only barely more likely to be true than false—and,
of course, obviously wrong to say that something that is less likely to be true
than false is known. But how much more is required? Is 80 percent probabil-
ity adequate or is that still too low? Should it be 90 percent, or 95 percent,
or 99 percent, or 99.9 percent? There is no very obvious way of answering
this question, and the even more striking fact is that almost none of the ad-
vocates of the weak conception of knowledge have ever seriously tried to do
so.17 Even more important, it is simply unclear what sort of basis or rationale
there might be for fixing this level of justification in a nonarbitrary way.
However problematic the strong conception of justification may be in other
ways, its intuitive significance and importance is clear. But nothing like this
seems to be true for the weak conception.
This last problem calls into serious question whether any clearly moti-
vated version of the supposed weak conception of knowledge even exists as
an alternative to the strong conception. But it will be convenient to defer
further discussion of this issue until we have considered a quite different and
somewhat surprising problem that has been recently (by philosophical stan-
dards) raised in relation to the traditional conception of knowledge.

The Gettier Problem

It is reasonable to say the some version or other of the traditional conception
of knowledge was taken for granted, often without very much in the way of
detailed specification, by virtually all philosophers seriously concerned with
knowledge in the period from the time of Descartes until the middle of the
40  Chapter Three

twentieth century. In 1963, however, Edmund Gettier published a remark-

ably short (three-page) paper that seemed to many to show clearly that the
traditional conception was at the very least seriously incomplete and quite
possibly even more badly mistaken.18
Gettier’s argument relies on examples (so-called “Gettier cases”) in which
the conditions required by the traditional conception of knowledge are sup-
posedly satisfied, but which are nonetheless intuitively not cases of knowl-
edge. Here are two such examples:19

Case 1:
Eleanor works in an office in which one of the other workers, Tom,
drives a Mercedes, talks about how much fun it is to own a Mercedes, wears
Mercedes T-shirts, receives mail from the Mercedes owners club, and so
forth. She infers and comes to strongly believe on this basis the proposition
that one of her co-workers owns a Mercedes. In fact, however, Tom does
not own a Mercedes: the car he has been seen driving is rented and all of
the other evidence is part of an elaborate hoax aimed at convincing people
that he owns a Mercedes. In fact, however, one of Eleanor’s other co-work-
ers, Samantha, does own a Mercedes, which she keeps garaged, hardly ever
drives, and does not mention to anyone, though Eleanor has no evidence of
this at all. (Note carefully: the belief at issue is the general belief that one or
another of Eleanor’s co-workers owns a Mercedes, not the specific belief that
co-worker Tom does, though Eleanor of course has the latter belief as well.)
Case 2:
Driving in the country, Alvin sees what looks like several sheep stand-
ing behind a fence beside the road and hence believes strongly that there
are sheep in that field. There are indeed sheep in the field in question, but
they are out of sight behind a grove of trees, and the animals that Alvin sees
are in fact large dogs bred and groomed so as to resemble sheep very closely.
(Note carefully: the belief at issue is the general belief that there are sheep in
the field in question, not the belief, which Alvin also has, that the particular
animals he sees are sheep and are in the field.)

Gettier’s first claim is that in cases of this sort (which are surprisingly easy
to construct), the three conditions of the traditional conception of knowl-
edge are satisfied. Clearly this is so for the truth condition, but it is plausibly
so for the other conditions only if it is the weaker versions of those condi-
tions and thus the weak conception of knowledge as discussed above that
is in question—which is clearly what Gettier has in mind.20 But, he claims
further (think very carefully about this point), neither Eleanor nor Alvin has
 The Concept of Knowledge  41

knowledge of the specific claim in question when this issue is judged from a
common-sense or intuitive standpoint. Intuitively, though their beliefs are
both justified (in the weak sense) and true, they are not true in the way that
their reasons or justification suggest, but rather as a matter of something
like a lucky accident. It is merely a lucky accident (without which her belief
would have been justified but false) that one of Eleanor’s other co-workers
happens to own a Mercedes, even though the specific one to whom her
evidence pertains does not. And the same sort of point is true in a different
way of Alvin.
Think again of the archery analogy mentioned earlier. The analogy to a
Gettier case would be one in which someone aims well but, perhaps because
of the difficult conditions, would still have missed the target, and then hits
it by accident, due, for example, to a random gust of wind at the last instant;
such a person has indeed hit the target, but not as a result of his or her
skill—the endeavor to hit the target by using the person’s skill has in fact
not succeeded. And analogously, in a Gettier case, the person in question has
indeed achieved true belief, but not in the right way for knowledge: not as a
result of his or her reasons or justification. (Here is a good place to stop and
think: Do you see the problem with the traditional conception of knowledge
clearly? If so, can you see any way around it? Does it show that the concep-
tion in question is mistaken, and if so, in what way?)
The conclusion reached by most of the philosophers who have discussed
the Gettier problem is that the traditional conception of knowledge is in-
complete, that a fourth condition has to be added to the standard three in or-
der to rule out such cases as cases of knowledge. Many such conditions have
been proposed, but we may focus here on one that has the virtue of being
closely related to the intuitive account just given of what goes wrong in such
cases. The proposed condition is that for a person to have knowledge, given
the satisfaction of the other three conditions of the traditional conception
in its weak version, it must also not be an accident, in relation to the person’s
justification, that their belief is true.21
Thus we would have the following modified version of the weak concep-
tion of knowledge. For person S to know proposition P at time t:

1W. S must confidently believe or accept P at t.

2W. P must be true.
3W. S must have at t a reason or justification that makes it highly likely
that P is true.
4W. It must not be an accident, in relation to S’s reason or justification,
that P is true.
42  Chapter Three

The Modified Weak Conception versus the Strong Conception

How are we to decide between the weak conception of knowledge (as modi-
fied to handle the Gettier problem) and the strong conception of knowledge,
that is, the Cartesian conception formulated earlier? The main argument for
the modified weak conception is that it seems to accord pretty well with our
common-sense or intuitive judgments about whether or not we have knowl-
edge in various particular cases, whereas the strong conception seems to lead
to the skeptical conclusion that we have almost no knowledge, perhaps even
that each person can only know of his or her own existence. Given that
these intuitive judgments represent at the very least a large part of our basis
for delineating the concept of knowledge, this is a strong objection to the
Cartesian conception and so a strong argument in favor of the weak concep-
tion—one that may indeed seem at first to be totally decisive. But things are
not quite this simple, as I will now attempt to show. For one thing, as we
will see toward the end of our discussion, there is a way to mitigate at least
somewhat the apparently decisive objection to the Cartesian conception
just mentioned. And, moreover, it turns out that there are also the following
pretty serious objections to the modified weak conception.
First. Though condition (4W) was added to solve the Gettier problem, it is
not clear that it entirely works. The problem is that in relation to the weak
version (3W) of the reason or justification condition, it could be argued that
the truth of the belief is always to some extent an accident. There is always
some chance that a belief that is only weakly justified will turn out to be false
(since weak justification does not guarantee truth), and it thus seems to be
always to some extent a matter of luck or accident whether this chance of
falsity is realized in any specific case, as it will in fact inevitably be in some.
And if this is so, then no case in which merely the weak justification condi-
tion (3W) is satisfied will be able to fully satisfy condition (4W) and qualify
as knowledge.
Second. As we noticed earlier, there is a problem about the precise degree
of justification that the weak conception requires, that is, about how likely
the truth of the proposition must be in relation to the reason or justifica-
tion that the person has. This is a very serious and quite possibly unsolvable
problem. One thing that it calls into question is whether the concept of
knowledge has the importance that is often attributed to it: how important
could it be (and why) that the strength of one’s reason or justification for a
claim is above rather than below a line that cannot be clearly and nonarbi-
trarily defined?
 The Concept of Knowledge  43

Third. Another problem for the weak conception grows out of an el-
ementary fact of probability theory, on the assumption again that levels of
justification can be regarded as probabilities (or at least as behaving like
probabilities). According to the weak conception, a person achieves knowl-
edge (assuming that the other conditions are satisfied) when the level of
their justification reaches a certain specific (though not yet clearly specified)
level—that is, we are assuming, when the believed proposition is probable to
that degree or greater in relation to their reason or justification. Suppose now
that a person has knowledge, according to this account, of two propositions,
P and Q. One of the strongest intuitions about knowledge is that he or she
should then be able to infer the conjunctive proposition P-and-Q, together
with any further consequences that follow from P-and-Q, and thereby have
knowledge of these further results. What, after all, is knowledge good for
except to draw further conclusions that will usually involve also appealing
to other known premises? But it is a fact of probability theory that the prob-
ability of a conjunction is equal to the product of the probabilities of the
conjuncts, which means that if the probabilities of P and Q separately just
barely meet the required level of probability, whatever it is, the probability
of the conjunction P-and-Q is guaranteed not to meet it. (For example, if
the required level is 0.9, then the probability of the conjunction will be only
0.81, and similarly for any level of probability short of certainty.) Thus if
the weak conception were the correct one, one would not in general have
knowledge of the consequences of one’s knowledge, making it again unclear
whether and why the concept of knowledge has any real importance.
Fourth. A final problem for the weak conception grows out of what has
become known as “the lottery paradox.” Suppose that the weak concep-
tion is correct and, just to make the presentation of the argument simpler,
that the “magic” level of probability required for adequate justification is
0.99. Suppose further now that a lottery is going to be held in which a prize
(perhaps a turkey) will be awarded to the holder of the winning ticket (one
ticket that is drawn out of the 100 tickets sold). It follows (assuming that
the drawing is fair) that the probability that any particular ticket will win is
0.01 and the probability that it will lose is 0.99. Suppose then that I believe
strongly of each ticket that it will lose (so that I thereby have 100 separate
beliefs). Out of these, 99 are true, and for each of these true beliefs I have the
“magic” level of justification. It is perhaps less clear, for the reason discussed
in connection with the first of this series of objections, that condition (4W)
is satisfied; but if that condition is ever satisfied in a case of less than con-
clusive justification, it seems reasonable to suppose that it is satisfied here.
44  Chapter Three

Thus a proponent of the modified weak conception must apparently agree

that I know that each of the losing tickets will lose—though I obviously do
not know this about the winning ticket, since neither condition (2W) nor
condition (4W) is satisfied there. But from an intuitive standpoint this result
seems plainly mistaken: in the case as described, I have no knowledge at all
concerning which specific ticket will win (though I do know that each of
them is quite unlikely to win).
These problems, of which the second is in my judgment the most impor-
tant, seem to suggest strongly that all is not well with the modified weak
conception of knowledge, thus raising the further question of whether any
way can be found to make the strong, Cartesian conception more palatable
from an intuitive standpoint.22 The best suggestion that I know of in this
connection is the following.23 Think of the concept of knowledge as charac-
terizing an ideal cognitive state: I am completely certain that a proposition
is true, and I have reasons or justification adequate to guarantee that I am
correct. Like many ideals, this ideal state is rarely achieved in practice; but,
also as with many ideals, other states of the same general kind can be use-
fully viewed and assessed as approximating, in varying degrees, the conditions
realized in the ideal state. And further, again in a way that seems to be true
of other sorts of ideal states, common sense is characteristically inclined to
underestimate the various reasons that make it difficult to achieve the ideal
state and so to judge that it has been achieved in cases that actually fall short,
perhaps even very significantly short.
Such a view of the concept of knowledge is supported to some extent at
least by the fact that initially confident common-sense ascriptions of knowl-
edge, to others or to oneself, often tend to be withdrawn in the face of serious
challenge or especially when the issue at stake turns out to be very important.
Thus, for example, I might be willing to say that I know that the liquid in a
certain container is water or that the pills in a certain bottle are ibuprofen,
but if challenged (“are you really sure?”) or if the issue is whether the liquid
can safely be poured on a fire or the pills safely consumed to relieve a head-
ache by someone who is allergic to aspirin, I may well be willing to admit
that I don’t really know.
Nonetheless, in spite of this point, it still remains highly doubtful that
common-sense assessments of knowledge, even in these relatively serious
cases, ever come very close to employing the extremely high standard laid
down by the Cartesian conception. It thus remains the case that the Carte-
sian conception is radically incompatible with our common-sense intuitions
about cases of knowledge—which are, to repeat, our main and indispensable
basis for deciding what the concept of knowledge really amounts to.
 The Concept of Knowledge  45

The apparent upshot of this discussion is that the traditional conception

of knowledge is seriously problematic with regard to the strength of the rea-
son or justification that should be required for knowledge (and, correlatively,
with regard to the proper strength of the belief or acceptance condition). We
seem forced to choose between (a) a view of knowledge that is so demanding
that few if any of our ordinary beliefs even come close to satisfying it and
(b) a view that leaves the required level of justification unspecified and prob-
ably unspecifiable, and that has other serious problems as well. In this way,
the concept of knowledge turns out to be something of a mess.
This result might seem to seriously threaten the whole enterprise of epis-
temology, leaving it without any clearly defined subject matter. I believe,
however, that the correct conclusion is substantially less dire. What reflec-
tion on this problem seems to me to suggest is instead that the concept of
knowledge, though it provides a necessary starting point for epistemological
reflection, is much less ultimately important in relation to the main episte-
mological issues than it has usually been thought to be. For whichever of the
two main candidates for an account of the concept of knowledge should turn
out to be correct, the main issues will be whether and how we have reasons or
justification for our beliefs of various kinds and how just strong such reasons
or justification in fact turn out to be. This will be so whether we think of our
cognitive goal as approximating as closely as possible to the Olympian ideal
of the strong, Cartesian conception or as seeking to achieve the ill-defined
level prescribed by the modified weak conception.24 And if we are unable
to decide firmly between those two conceptions (or even come to suspect
that there is no clearly correct choice to be made), the main questions just
mentioned—(a) whether we have reasons or justification in light of which
our various beliefs are likely to be true and (b) how strong or compelling such
justification is—will be no less urgent or important.
For this reason, most of our concern in the succeeding chapters will be with
issues pertaining to reasons and justification, with the concept of knowledge
falling very much into the background. We begin in the next chapter by con-
sidering an issue that Descartes does not raise and that, surprisingly enough,
neither he nor the philosophers that came before and even immediately after
him seem to have even noticed clearly, even though it is quite important for
other issues that they do discuss: the problem of induction.