Crumley II - Cap 2 - Anaysis of Knowledge

Review Copy
Chapter Two An Introduction to the

Analysis of Knowledge
Actually, the chapter title is a bit misleading. The focus of this

chapter is on the analysis of the concept of knowledge. Philosophers—
epistemologists, in particular—want to know the conditions that must
hold in order for a person to know. Very loosely, philosophers want to
know what distinctive properties a person must have or what distinctive
characteristics the world must have (or both) so that the person could be
said to know. But we will return to this in a moment.
Many contemporary epistemologists trace the general framework
for thinking about the concept of knowledge back some twenty-four
centuries.
Plato (c. 429–347 BCE ) developed an analysis of knowledge in the
larger context of his metaphysical and ethical concerns. Regrettably, we
must confine ourselves to a brief sketch of some of the more relevant epis-
temological themes. In the dialogue Meno, the title character asks Socrates
why knowledge should ever be preferred to true opinion or belief. Before
we let Socrates answer, it is worth trying to appreciate the force of this
question.
Suppose you are riding with a friend and are convinced that your friend
simply does not know how to get to your destination. Nonetheless, as she
changes lanes and turns down streets you would rather not travel, she
insists that she knows this is the way to go. And she smugly reminds you
of this when you arrive on time. If you find her reminder a bit frustrating,
perhaps it is because you are willing to concede that your friend had the
right opinion or a true belief, but that she really did not know that this
was indeed the way. Whether she had knowledge or simply true belief,
this much seems apparent: you reached your destination. In this context,
knowledge and true belief seem much alike in that they both get you to
53
Review Copy
54 an Introduction to Epistemology
the same place. If both knowledge and true opinion get you to the same
place, if they both get you to the right answer, why worry about having
knowledge? Why not simply settle for true belief? There doesn’t seem to be
any difference between them.
Socrates, however, is impressed by the fact that true opinions have a
disturbing tendency to “fly away”; under pressure we may give up true
opinions. Perhaps you have also had the experience of having a true belief,
but for various reasons, such as the influence of friends, you changed your
mind and subsequently had a false belief, or gave the wrong answer. Now,
Socrates thinks that our true beliefs are less likely to fly away if they are
appropriately tied down. Our ability to tie down a true belief is merely
our ability to give an account, or an explanation of why we think that our
belief is the right or correct belief. Knowledge is distinguished from true
belief by this ability to give an account or explanation. Thus, knowledge is
true belief plus the account, in Plato’s view.
Many twentieth-century epistemologists adapted Plato’s view and
explained knowledge as justified true belief. It is this analysis of the con-
cept of knowledge—the justified true belief analysis—that is called the tra-
ditional analysis of the concept of knowledge.1 Prior to 1963, hardly anyone
actually used the terminology of the traditional analysis; very few actually
explained knowledge as justified true belief. After 1963, the widespread,
but not unanimous, opinion has been that the traditional analysis is not
quite right. Epistemologists thus have a tradition with little history and not
much of a future! Nevertheless, it is plausible to think that the common
core of various theories of knowledge is justified true belief. Thus, in the
previous chapter, we noticed that one way of understanding the concept of
knowledge is as follows:
(1) P is true.
(2) The subject or agent believes that P .
(3) P is infallible.
Notice that conditions (1) and (2) require that knowledge be at least
true belief. But it is perhaps plausible to think of condition (3) as specifying
the requirement for the justification of a true belief. Descartes could thus
be read as claiming that for a true belief to count as knowledge, it must
not be possible for the belief to be mistaken. More succinctly, infallibility
is the type or kind of justification required for knowledge. So, we might
1 Like the Meno, the Theaetetus provides an apparent formulation of the justified true
belief analysis. At 201c and following, Socrates distinguishes knowledge from true
opinion, and he subsequently seems to identify knowledge with true opinion plus
speech or a rationale. Both are widely available, including in Plato 1961a and Plato
1961b. Of course, one cannot do better than reading Plato’s dialogues, but three
accessible surveys and interpretations are Taylor 1926, White 1976, and Gosling,
1973.
Review Copy
Chapter Two An Introduction to the Analysis of Knowledge 55
view Descartes’ understanding of the concept of knowledge as a variation

on a theme first articulated by Plato almost 2500 years ago. Perhaps there
is a bit of tradition in the traditional analysis after all.
Three matters concern us in this chapter. The first is to explain and
attempt to motivate the traditional or standard analysis. Despite the fact
that prior to recent decades few people actually defined knowledge as
justified true belief, it is worthwhile to see why the traditional analysis
provides a general framework for thinking about knowledge. The second
matter relates to more recent accounts of knowledge and concerns a type
of objection that led many to reject the idea that justified true belief is the
appropriate way to analyze the concept of knowledge. Third, we examine
three different theories of knowledge. Although our concern is the concept
of knowledge, these theories focus substantially on the notion of justifica-
tion and the sort of justification required for knowledge.
The Traditional Analysis
Truth Conditions
We are specifically interested in the analysis of the concept of propositional
knowledge. Propositional knowledge differs, for example, from procedural
knowledge, or knowing how. Knowing how seems to involve a particular
skill or ability, as in knowing how to ride a bike or how to find a bit of
information on the Internet. But the focus of the traditional analysis is
knowing that. For example, Sara knows that Gödel proved the incomplete-
ness theorems. The object of Sara’s knowledge is a certain proposition or
sentence. The content of her knowledge, what she knows, is given by the
proposition or sentence “Gödel proved the incompleteness theorems.”
More generally, and somewhat schematically, the traditional or stan-
dard analysis attempts to provide the truth conditions for the general sort
of statement we’ll call (K ): “S knows that P ,” where “S ” stands for a per-
son and “P ” stands for the object or content of the person’s knowledge.
Epistemologists are attempting to give a perfectly general analysis. They
want to say what is true of any person when that person has propositional
knowledge. Somewhat formally, the aim of an analysis of the concept of
propositional knowledge is to provide the necessary and sufficient condi-
tions for the truth of (K ). Less formally, the idea is to state under what con-
ditions an agent knows something. We want to know what it is about the
agent or the agent’s situation that makes it the case that the agent knows
and is not, say, just guessing.
Notice that when we ask this sort of question about the appropriate
analysis, we are asking a different sort of question than asking when an
agent is justified in claiming that he or she knows. For example, I might
be justified in claiming to know that the test will be given on Wednesday,
Review Copy
but for various reasons I cannot foresee, the test may be postponed until
Friday. Nobody knows P if P is false. Hence, although I was justified in
claiming to know, it turns out that I did not in fact know.
One way to express this distinction is to say that we are interested in
the truth conditions of (K ), but not necessarily the assertibility conditions
of (K ). The assertibility conditions of knowledge specify when someone is
justified in claiming to have knowledge, regardless of whether that person
actually knows. The truth conditions of (K ), the necessary and sufficient
conditions for knowledge, are the conditions under which someone does
in fact have knowledge. Analyses of the concept of knowledge are typically
concerned with the truth conditions of the concept, not the assertibility
conditions. Providing the truth conditions of (K ) is a way of explaining the
nature of knowledge, of explaining in what knowledge consists.
Motivating the Traditional Analysis

In 1956, A. J. Ayer (1910–1989) gave the following analysis:
(K A) S knows that P if and only if:
(K A-1) P is true.
(K A-2) S is sure that P.
(K A-3) S has the right to be sure that P .2
That condition (K A-1) is a necessary condition for knowledge has seemed

obvious to most, as it does to Ayer. This is not a recent development in
the understanding of the nature of knowledge. Recall that Plato thought
knowledge was at least true belief. But we might wonder whether we can
give any explanation for thinking that knowledge entails the truth of the
proposition.
One type of explanation might consist of an appeal to our normal prac-
tices, to our typical intuitions. Suppose Sam says to you, “I know that the
keys are on the desk.” Yet when you step over to the desk, there are no
keys to be found. You might protest that Sam doesn’t know, because there
are simply no keys there. You might insist that Sam thought he knew but
in fact he did not.
Perhaps one way to emphasize the oddness of this case is to think about
what we expect from someone who knows. Imagine, for a moment, that
your calculus teacher continually gives the wrong solutions to the assigned
problems. A natural reaction might be something along the lines of “He
doesn’t know what he is talking about!” We expect that someone who
knows can give us the right answer. That is, someone who knows has
locked on to a particular truth. Refusing to acknowledge that knowledge
implies truth is to collapse the distinction between knowledge and types
2 Ayer 1956, pp. 31–35.

Review Copy
of false belief. Yet we often desire to distinguish between someone who

has very good reasons for what he believes but believes something false,
and someone else who also has very good reasons but believes something
true.
Recall that part of the aim of epistemology is to identify the standards
by which we can make differential assessments of our beliefs. One natural
way to do so is to distinguish between true beliefs, which pick out genuine
features or characteristics of the world, and false beliefs. Moreover, to call
beliefs instances of knowledge suggests that we are epistemically better off
having such beliefs as opposed to false beliefs. These two thoughts together
seem to motivate the idea that knowledge requires the proposition at issue
to be true.
Still, we might be suspicious of this condition for at least two reasons.
First, we might think that knowledge merely represents doing the best the
agent can do and that in some of those cases the agent might have a false
belief. Notice, however, that we can distinguish two kinds of cases: (a)
one in which agents do the best they can but arrive at false beliefs, and (b)
one in which agents do the best they can and arrive at true beliefs. Now,
clearly, the latter case in some sense describes agents who are epistemically
better off.
A second reason relates to the claim that knowledge is in some sense rel-
ative to a perspective, either an individual’s or a community’s. Two things
can be noted in response. Even relative to a perspective, we can distinguish
true from false beliefs. More generally, the truth condition does not specify
what theory of truth is the correct theory. The condition is compatible with
any number of theories of truth, including what has been called a relativist
notion of truth. It is not unusual, however, to see many epistemological
discussions assume an objective notion of truth, and we will do so here.
This may not provide a compelling motivation for counting truth as a
necessary condition of knowledge. However, it at least makes the motiva-
tion for the truth condition somewhat more apparent.
More serious questions arise with (K A-2). Ayer suggests that the agent
must be sure that P . An implicit idea here is that the agent has some mental
state with a particular content, which can be characterized by the propo-
sition P . Our normal way of describing such matters is that the agent
believes that P . When Sara believes that the bank closed at five, she has a
mental state with a particular content. The mental state is about the bank.
Being sure is a stronger requirement—for example, not only do I believe
that the provost will retire, but I am sure, or perhaps convinced, that the
provost will retire. But we might wonder whether being sure that P is nec-
essary to know that P . Let’s see if we can sort this out.
There are two ways of understanding Ayer’s “being sure” requirement.
One is to interpret an agent’s “sureness” in a psychological sense. The other
is to construe “being sure” as indicating something about the agent’s view
Review Copy
of the epistemic status of the belief. Let us focus on the psychological

construal first.
Theories of Truth
Philosophers disagree about the nature of truth. One recent classification
distinguishes between nonepistemic and epistemic theories of truth. This
parallels, in some respects, the distinction between realist and nonrealist
views of the world. Nonepistemic theories of truth hold that the truth
of a proposition depends on the way the world is, independently of our
beliefs about the world. (Notice: Even your beliefs are part of the world.)
Epistemic theories hold that the truth of a proposition depends on our
beliefs about the world.
A more traditional classification distinguishes between correspondence,
coherence, and pragmatic views of truth. Roughly, correspondence theo-
ries hold that a proposition is true if and only if the proposition matches or
corresponds to the way the world is (or at least the part of the world rel-
evant to the proposition, sometimes called a “state of affairs”). Coherence
theories hold that a proposition is true if and only if it fits or coheres with
a system of propositions. Pragmatic theories claim that a proposition is
true if “it works,” that is, if acting or believing on the basis of that proposi-
tion produces the right sort of consequences.
For the purposes of this text, we assume a realist view and a correspon-
dence theory of truth. See “For Further Study.”
Notice that someone might “feel sure” of some belief, but for reasons
that have little or nothing to do with the truth of the belief. A person who
holds a belief based on mere superstition might nonetheless feel sure. Sam
might feel sure that his having a flat tire is due to his having walked under
a ladder. But such confidence is rarely a guide to the truth. Thus, one might
object to the psychological construal on the grounds that what we are
interested in is the epistemological issues concerning the concept of knowl-
edge, not the psychological issues. We need instead an explanation of why
a psychological state is an indication of the epistemic status of the belief.
Is the belief merely a good guess? Does it qualify as knowledge? Suppose
that Sam studied diligently for his upcoming logic test and answered all
the questions correctly. But suppose also that Sam has never been terribly
confident of his abilities in “abstract symbol” courses. Further suppose that
were you to show Sam incorrect uses of, say, De Morgan’s Rule, he would
not be inclined to change the answers he gave on the test. Indeed, Sam
checks his answers carefully and does not change any of them; nonetheless,
Sam is reluctant to say he did well.
Sam’s lack of confidence, in the preceding example, might indicate that
he does not think he has the best of reasons for giving the answers he has.
Review Copy
Thus, for this way of thinking about “being sure,” we might argue that a
person must believe that he has good reasons for holding the belief.
This line of thinking raises important issues, some of which we will
address shortly. But notice that this way of understanding “being sure” has,
in effect, introduced two new considerations for knowing: (a) the person
must have good reasons, and (b) the person must think that such reasons
are good reasons. Both of these must be considered in their turn. Note,
however, that the belief condition—that S believes that P —and consider-
ations (a) and (b) are logically independent of one another. Because they
are logically independent, it would perhaps be best to treat them separately.
We will, in fact, have to look at putative conditions for knowing like (a)
and (b), but for now, we rest content with the idea that (K A-2) should be
revised to read simply:
(K A 2*) S believes that P .
A few critics have objected to even this revised condition as a neces-

sary condition of knowledge.3 They argue that a person might be right
consistently about a certain matter, but only be willing to characterize the
claim as a guess, not a belief. Imagine Sara as a game show contestant who
gives the correct answer but is not completely confident. We might say she
knows the answer is P , but does not believe it.
But it might be better to say that Sara believes that P without being
confident that P is correct. In any event, this case might reasonably be con-
sidered a borderline case of knowledge. Sara may give the correct answer
to a question and not be inclined to give any other answer. Yet we might
not want to insist that Sara knows. Her lack of confidence may indicate
that she does not have very good reasons for her answer or that she is
simply guessing. Thus, we may plausibly claim that we have not yet been
given a clear-cut case of knowing but not believing.
It is worth mentioning one further reason for thinking that knowledge
does not imply belief: Suppose you ask a friend if she believes the presiden-
tial debate will be held at a local university. Her response is an emphatic “I
don’t just believe it; I know it!” This response is an example of a kind of
“conversational rule” that agents make the strongest epistemic claim they
are entitled to make. Thus, it might be claimed that if one knows that P ,
then one does not merely believe that P .4 But this sort of claim does not
show that knowledge does not imply belief. It merely shows that knowl-
edge is more than simple belief.
The final condition in Ayer’s analysis (K A-3) maintains that the agent
must have the right to be sure. An important feature must be noted here:
3 See, for example, Radford 1970, pp. 171–185.

4 See Hamlyn 1971, Chap. 4, and Harrison 1970. For a more detailed discussion, see
Shope 1983, pp. 180–187.
Review Copy
“Having the right to be sure” implies that whatever grounds the agent’s
confidence must be an adequate ground. That is, not just anything will
legitimate the agent’s confidence. This suggests that there is an important
normative component in knowledge, that an agent’s belief must in some
way satisfy a certain standard. How exactly this normative component is
to be explicated remains to be seen.
We noticed in considering condition (K A-2) that the idea of confidence
or sureness raised the issue of the reasons for the agent’s belief. So, one of
the things that we might view (K A-3) as suggesting is that in some as yet
unspecified sense, the agent must be justified in believing that P , where
“justification” means “having good reasons.”
A brief note on “having good reasons”: There are clearly inadequate
reasons for having a belief. Suppose Sam believes, as a result of consult-
ing a fortune-teller at a county fair, that he will someday be president.
Sam has a reason for his belief. But we might be reluctant to think that
this is a good reason, that the fortune-teller’s musings provide sufficient
evidence for Sam’s belief. More to the point, it is not a good reason if Sam
is interested in believing the truth and avoiding believing things that are
false. Compare this type of case with Sara’s belief that there will be a test
tomorrow. Imagine that she believes this because the test is scheduled on
the syllabus and the teacher reminded her of the test. We would be inclined
to think that Sara’s belief is based on good reasons.
At a minimum Ayer’s third condition suggests that the agent must have
good reasons for believing that P . That Sara studied diligently is a good
reason for her to believe that a particular answer is correct (assuming the
normal things we do about studying). However, condition (K A-3) leaves
open whether the agent must recognize that the reasons are good reasons
or whether it’s enough that the agent merely have good reasons, recognized
as such or not. Descartes offers an explicit version of the former sort of
view. His “indubitability” criterion indicates not only a reason for having
the belief but also the recognition by the agent of the indubitability as a
reason for having the belief. Indeed, the classical conception of knowledge
is one in which an agent not only has good reasons but recognizes that the
reasons are good.
On the other hand, to some people, this latter constraint, explicitly rec-
ognizing the reasons as good reasons, might seem too strong. We cannot
yet fully explain why some might want to opt for this weaker version of
the condition, but we should be aware of it in our formulation of the third
condition.
For the time being, we can avoid deciding between the stronger and the
weaker versions of condition (K A-3), and say merely that the agent must
be justified in believing that P . As we noticed early in the chapter, Socrates
distinguished mere true belief from knowledge. He did so by appealing to
the notion of a person having an account or explanation for the belief. This
account or explanation is roughly equivalent to our notion of justification.
Review Copy
The reasons for a person’s belief provide the justification for the belief.
Knowledge is more than a lucky guess; it is more than mere true belief.
We are unwilling to credit someone with knowledge unless the person
has acquired the belief by “one of the accredited routes to knowledge.”5
This provides us with a schema by which to approach the analysis of the
concept of knowledge:
(K ) S knows that P if and only if:
(1) P is true.
(2) S believes that P .
(3) S is justified in believing that P .
With the qualification that condition (3) was historically understood in

the stronger sense identified in the previous paragraphs, we have arrived at
the traditional or standard analysis, the justified-true-belief (JTB ) analysis,
of the concept of propositional knowledge. We will discover that criticisms
of this analysis divide epistemology in ways that might never have been
imagined.
Gettier and the Traditional Analysis
Although it is rare to find anything approaching consensus in philosophy,

there was considerable agreement that something like the traditional analy-
sis was along the right lines.6 This complacency came to a rather abrupt
halt in 1963, when Edmund Gettier published a paper entitled “Is Justified
True Belief Knowledge?”7 In this three-page article, Gettier provided two
examples suggesting that all was not quite right with the traditional, or
JTB , analysis. That is, the examples illustrated that a person might have
justified true belief but yet not have knowledge.
Gettier-type Counterexamples
Gettier’s own examples are straightforward enough, but we will consider
two other examples of the same type, the first offered by Keith Lehrer.8
Lehrer asks us to imagine a case in which an agent, the ubiquitous S ,
has a coworker, Ms. Nogot. Now, S has good reason to believe that Nogot
owns a Ford; imagine, for example, that S has seen Nogot driving a Ford,
S has been told by persons who have in the past been reliable that Nogot
5 Ayer 1956, p. 33.

6 This is not to suggest that there was universal agreement, of course.
7 The paper was originally published in Analysis 23 (1963): 121–123, but it is now
widely reprinted, for example, in Roth and Galis 1970.
8 See Lehrer 1970.
Review Copy
owns a Ford, and so on. Hence, S has justification for her belief that Nogot
owns a Ford, and from this she infers her belief that P —someone in the
office owns a Ford. But Ms. Nogot does not own a Ford. Nevertheless, S ’s
belief that someone in the office owns a Ford is true, because Ms. Havit
owns a Ford.
Notice that S has a justified true belief. The belief that someone in the
office owns a Ford is true. Moreover, S had good evidence for arriving at
this belief. That is, S is justified in believing that P . But it is not obvious
that we would want to say in this case that S has knowledge. We might
think that S has been a bit lucky. It just so happened that someone owned
a Ford, but not the person S thought owned the Ford. S has reasons, but
we might think that S doesn’t have the right sort of reasons.
Bertrand Russell and Gettier-type Counterexamples

Bertrand Russell, in 1912 and 1948, provided examples that look a great
deal like Gettier-type counterexamples. In The Problems of Philosophy,
Russell suggests the following example. If a man believes that the late
prime minister’s last name began with “B,” what he believes is true
because Mr. Bannerman was the late prime minister. But he will not
have knowledge because he believes that the late prime minister was
Mr. Balfour. In Human Knowledge, Russell uses the example of someone
looking at Big Ben and arriving at true belief about the time. However, the
clock has in fact stopped.
Russell uses these examples to show that knowledge cannot be identified
with true belief. Do you think these examples are Gettier examples? If not,
can you modify them so that they would serve as Gettier examples?
The second counterexample derives from Alvin Goldman.9 Imagine a

situation along the following lines. A man is taking his young son on a
tour of the country for the first time. The man points out various things to
his son and identifies them. The man points at random to one of several
things he takes to be barns, and says, “That’s a barn.” In fact, the man has
succeeded in pointing to a barn. Unbeknownst to him, however, a film
company has been in the area and has been constructing barn facsimiles.
Close by the actual barn the man pointed to is just such a barn facade.
Indeed, the man could not discriminate between the barn and the barn
facade. Again, the agent has a justified true belief: He sees the barn, can
recognize barns, and so on. But we might hesitate to claim that he has
knowledge. After all, the man just as easily could have pointed to a barn
facade and thus have had a justified, yet false belief.
9 Alvin Goldman 1988a, pp. 41–65.

Review Copy
Gettier-type counterexamples are abundant in the literature, but these

two will serve to illustrate the points we need to make for now.10 Two
quick words of caution: First, people engaging in this debate are not skep-
tics. These epistemologists think that we have knowledge; they disagree
rather about the appropriate analysis of knowledge. Second, we can alter
the counterexamples to make the justification as strong as we wish. These
examples are intended to show that a true belief might be based on ade-
quate—even excellent—evidence and yet fail to be a genuine instance of
knowledge.
In the Ford case, S clearly has reasons for his belief. Yet if we were to
explain why S ’s belief is true, we would refer not to Nogot, but rather to
Havit. S ’s justification does not direct us to what accounts for the truth of
his belief. In the barn case, the connection between the justification for the
belief and its truth is clearly more intimate. But we can still see that the
justification for the father’s belief does not suffice to explain why he just
happened to pick out a real barn rather than one of the facades. This bit of
good luck is in no way reflected in his justification. Sometimes, this feature
of Gettier-type counterexamples is expressed as a belief’s being accidentally
true with respect to the agent’s justification.
These remarks suggest a way of remedying the defect suggested by
Gettier cases. The remedy is to constrain or restrict the way in which the
justification condition is satisfied. One way to do this is to amend the
JTB analysis so that it includes a fourth condition that specifies additional
restrictions an agent’s justification must meet. That is, to have knowledge,
an agent must not only have a justified true belief but also be justified in a
particular way. Indeed, some epistemologists hold that the attempt to deal
with Gettier cases is merely an attempt to specify a fourth condition.
Alternatively, we might suggest that the way to remedy the apparent
defect of the JTB analysis is to change our understanding of what counts
as a “justified” belief. The thought here is that the kind of justification
necessary for knowledge might differ from our ordinary sense of what it
means for an agent to have reasons or evidence for a belief. Of course, a
response to Gettier cases might make use of both these strategies. That is, a
response might not only add some further fourth condition but also amend
our understanding of the justification condition. In what follows, we will
talk generically of amending the JTB analysis.
Before considering how theorists have attempted to amend the JTB
analysis, we need to consider a few dissenting words, for not everyone
agrees that Gettier has identified a problem with the traditional analysis.
10 Shope 1983 lists some ninety-eight such counterexamples that have appeared in the
literature.
Review Copy
A Defense of the Traditional Analysis

The main strategy of interest here is what we might call the entailment
strategy, which will be explained in a moment. Recall Descartes’ insistence
that the justification appropriate for knowledge requires us to show that
the belief at issue is true. In particular, Descartes insists that the justifica-
tory connections between one’s evidence and the belief in question are
always deductive connections. In this view, the justification condition and
the truth condition cannot be satisfied independently. If the justification
entails the belief, then it would be logically impossible for it to be true that
the belief is justified, but false that the belief is a true one.
Oliver Johnson defended the claim that, properly interpreted, sat-
isfaction of the justification condition entails satisfaction of the truth
condition.11 Johnson referred to this entailment relation as the logical inter-
pretation of the justification condition. According to Johnson, if we hold
the view that the justification condition does not entail the satisfaction
of the truth condition, what he calls the contingent interpretation, then
the standard analysis is rendered useless. We could not determine, for any
given belief, whether the belief qualifies as knowledge. This is due to the
fact that we could not tell whether the belief is in fact true; we would have
no decision procedure for determining the truth of the belief. Because the
truth condition is an uncontroversially necessary condition of knowledge,
Johnson held that our inability to demonstrate the truth of the belief ren-
ders the standard analysis useless.
On the other hand, the logical interpretation automatically guarantees
the satisfaction of the truth condition. The ability to justify one’s belief is
merely the ability to show that the belief is true. Hence, if we accept this
logical interpretation of the justification condition, Gettier-type counterex-
amples simply do not count against the standard analysis. After all, in the
logical interpretation, the justification condition is not satisfied in Gettier
cases.
The logical interpretation of the justification condition is a heroic
response to Gettier. Of particular interest is that in this interpretation
of the JTB view, most of the beliefs we would be inclined to count as
knowledge are simply not knowledge. Consider, for example, my belief
that Richard Nixon lost the 1960 presidential election. I have many beliefs
that support this belief. For example, I remember seeing interviews with
Nixon, reading articles about John Kennedy’s election in 1960. In the con-
tingent interpretation, and in our normal view, I know that Richard Nixon
lost the presidential election. Although I am clearly justified, none of my
supporting beliefs entail that my belief about the outcome of the election
is true. In the logical interpretation of the justification condition, I do not
know that Nixon lost the presidential election in 1960. It is not difficult
11 Johnson 1980.
Review Copy
to find innumerable other beliefs, which we typically count as knowledge,

that would not be knowledge if we accepted the logical interpretation.
But Johnson claimed that our ordinary conception of knowledge, which
presumably relies on the contingent interpretation, is an incoherent con-
ception of knowledge.
In the previous chapter, we saw some reason to reject such strong con-
straints on instances of knowledge. In such a view, it is difficult to see
how we would ever have beliefs about the world—for example, beliefs
about the kinds of objects and the properties they have—that would count
as instances of knowledge. Moreover, it is not entirely clear that having
knowledge requires that we be able to demonstrate the truth of the belief.
We might even suggest that this is an attempt to reintroduce the infallibil-
ity requirement. Pointing out that many of our beliefs would not count
as knowledge in Johnson’s view does not show that he is mistaken. But
it should at least make us curious as to whether there is some alternative
account of Gettier problems that might not be so restrictive.12
Some Strategies for Handling Gettier Problems
After the publication of Gettier’s paper, the task was to amend the JTB
analysis in such a way as to preclude Gettier cases. We are concerned here
with only a small and influential sample of the various specific proposals.
Our aim is to get a general sense of the motivation and structure of these
proposals, as well as with major difficulties each encountered. Here we
consider three views: causal theories, indefeasibility theories, and the no-
false-premise view. In the next chapter, we will consider in more detail a
view known as reliabilism.
A General Diagnosis
It will be helpful to begin by giving at least a general account of what goes
wrong in Gettier-type cases. As we already noted, Gettier cases show that
as it stands, the JTB analysis permits that what accounts for the justifica-
tion of the target belief (the belief that is a putative instance of knowledge)
is not what accounts for the truth of the target belief. An agent’s evidence
or reasons for a belief may not adequately reflect the circumstances that
underlie the truth of the belief. For example, S ’s reasons for believing that
someone owns a Ford have no obvious connection to the truth of that
belief. In a sense, S has evidence, but it is the wrong evidence for the belief
that she has.
12 Robert Shope, Shope 1983, suggests that there are Gettier-type examples to which
even Johnson’s logical interpretation falls prey. Such counterexamples do not depend
explicitly on the appeal to a false proposition.
Review Copy
But you might wonder why this is a bad thing, especially given that we
have apparently rejected the obvious solution: the entailment strategy. We
can perhaps see what is at issue here by considering what we expect from
our concept of knowledge. If the entailment strategy is too stringent a cri-
terion, what weaker demands can we make?
Gettier cases give us some hint of what we might expect from a satis-
factory analysis of the concept of knowledge. The moral of Gettier cases
might be this: Gettier-type cases suggest that a belief might be justified and
true, but had things been just a little different, we might have had a false
belief. In some sense, it was merely good fortune or a happy accident that
we arrived at a true belief.13 Thus, in the Lehrer case, for all the evidence
our omnipresent S had, it still might have been false that anyone, includ-
ing Havit, owned a Ford. Or, S might simply have chosen not to make the
existential generalization, in which case S would have had the false, yet
justified belief that Ms. Nogot owns a Ford.14 Similarly, in Goldman’s case,
it was just good fortune that the father pointed to an actual barn, rather
than to a mere barn facade. Consequently, a plausible approach for dealing
with the counterexamples is to find a way to amend the JTB analysis that
limits the possibility of our having a false belief. To put it slightly differ-
ently, we might look for an analysis that ties our justification more closely
to the truth of the belief, such that it is less of an accident that the belief
turns out true, given the evidence the agent has. With this in mind, let us
turn to the more specific proposals.
Causal Theories
Causal theories are a remarkably clever way of resolving the “accidentality
problem.” As we will see, they involve a subtly different understanding of
the justification condition. Roughly, causal theories of knowledge hold that
a belief is an instance of knowledge if the belief is true and is caused in
the right way. Of course, we need to know something more about what it
means to be caused in the right way. For the moment, we can motivate the
causal theory by explaining how we can move from our more traditional
understanding of the justification condition to the causal theorist’s under-
standing of the analysis of knowledge.15
13 For a recent example of a defense of the “anti-luck” view of knowledge, see Heller
1999, pp. 115–129.
14 Existential generalization is the move from a specific case (Nogot) to a more general,
indefinite “someone.” Some have thought that the way to preclude Gettier-type
examples is to restrict such logical moves. This seems a bit arbitrary, and there are
examples that do not clearly depend on such inferences.
15 It should be noted that Goldman does not consider his causal analysis to be an
interpretation of the justification condition. See Alvin Goldman 1970, pp. 67–87,
esp. p. 84. The aim in the following paragraphs is to motivate the causal analysis, to
see how one might move from the traditional analysis to something like the causal
analysis.
Review Copy
So far, we have been assuming that the justification condition requires

that agents have evidence for their beliefs. In other words, an agent has
other beliefs, which are the reasons for the target belief. As we noted pre-
viously, these are the reasons for which the agent holds the belief. It is not
too difficult to think of this evidence as the reason the agent has the belief
in the sense of being the cause of the agent’s having it. In a legitimate sense,
the evidence causes one to have the belief. The belief is an effect, or causal
result, of the reasons. Sara has the belief that she will be flying to London
tomorrow because she looks at her ticket and notices that the departure
date is the following day. It is by virtue of this evidence that she comes to
have the belief. Now comes the subtle shift. Instead of thinking of justifica-
tion in this more traditional “evidential” sense, we can perhaps understand
it in a more general “causal” sense. That is, a belief will be justified if the
reason the belief is held—what causes the belief to be held—is an appro-
priate reason.
You might be wondering how the shift from the evidential to the causal
sense of “reason” is supposed to help us with Gettier. Recall that the chal-
lenge presented by Gettier examples is to tie the justification and truth
conditions more closely. We can now marry these conditions if we say
that the reason the agent holds the belief is tied to the reason the belief is
true. That is, some of the conditions that account for the agent having the
belief must be the same conditions that account for the truth of the belief.
We want at least part of what causes the agent to have the belief also to
be what explains the truth of the belief. Is there a candidate for this dual
role? How about the “facts”? Intuitively, the facts—the way the world is,
or some features of the world—cause the agent to have the target belief,
and the facts explain why the belief is true.16
Consider a fairly mundane belief, that there is coffee in the cup. Why
do you have this belief? You look over there, where the cup is, and see the
coffee in the cup. You acquire the belief for the simple reason that you see
coffee in the cup. The fact that there is coffee in the cup (and your seeing
it) causes you to believe there is coffee in the cup. If this is what we mean
by being justified, then your belief clearly is justified.
On the other hand, what makes your belief true? Well, how about the
fact that there is coffee in the cup? Your belief is true only if there really
is coffee in the cup. So, the facts explain both the justification of the belief
and the truth of the belief. Moreover, this sort of analysis seems to fit with
our ordinary intuitions. How do you know there is coffee in the cup? You
look over there and you see it; your seeing it causes you to believe it. What
more could one want from an analysis? The justification and the truth
16 Some philosophers interested in the metaphysics of causation prefer to talk about

events causing other events. Here the use of “facts” is intended to be neutral between
different views of causation; “fact” points to features or characteristics of the
world.
Review Copy
condition can no longer be satisfied independently; the two conditions are

indeed intimately connected.
Goldman’s solution to the Ford problem is succinctly stated in his “A
Causal Theory of Knowing.”17 There, Goldman claims that
S knows that p if and only if the fact p is causally connected in an
“appropriate” way with S’s believing p.
Clearly, we want to know the sense of “appropriate.” Goldman claims
that perception and memory are appropriate knowledge-producing causal
processes. But the Ford case is neither a case of mere perception nor a case
of memory. Thus, Goldman adds another appropriate process. He suggests
that (a) a causal chain must exist, leading from the fact to the belief, that is
correctly reconstructed by inferences, each of which is warranted, and (b)
the background beliefs that function in such inferences must be true.18 (He
also mentions that combinations of these three—perception, memory, and
inference—count as appropriate processes.)
In the Ford case, why does S fail to have knowledge? There is a causal
chain leading to the fact that someone owns a Ford. But this causal chain
involves Havit. S’s own inferences, however, reconstruct this causal chain
as involving Nogot. Thus, the inferences do not satisfactorily mirror the
causal chain: S does not know.
But this causal analysis does not adequately explain cases in which
there is more than one causal chain that might account for the truth of the
belief. Hence, the reconstructed inferences do not mirror the actual causal
chain.19
An example illustrates this problem. Sherlock Holmes’s faithful assis-
tant Watson might notice a gunshot wound in the chest of a victim and
pronounce him dead. Now, Watson surely knows the victim is dead. He
knows this even though his inferences do not reconstruct the actual causal
chain, say, a bite from an exotic and deadly spider released in the victim’s
bed. Watson is misled by the gunshot wound into thinking that this was
the cause of death. Watson is right that the gunshot wound is sufficient
to cause death; he is simply unaware of how many enemies the victim
had. But the gunshot is not the actual cause of the victim’s demise. Thus,
Watson’s inference that the victim is dead, an inference based on the rea-
sonable belief that a gunshot wound to the chest is sufficient to cause
death, does not correctly mirror the actual causal chain. This reconstruc-
tion of the actual chain will be left to Holmes.
More importantly, notice the beginning of a subtle shift in the notion of
“reason.” We ordinarily think of reasons as evidence one has. That is, we
often think of reasons as other beliefs the agent has, beliefs that the agent
could cite as required. But there is a shift in the causal theory toward think-
17 Alvin Goldman 1970, pp. 67–87.

18 Alvin Goldman 1970, pp. 72–75.
19 The technical name for this problem is “epistemic overdetermination.”
Review Copy
ing of reasons as causal—that is, what caused the agent to believe that p.
In a more traditional view of “reason,” we might grant that an appropriate
causal connection exists between a fact and a belief, but we might not be
so willing to grant that an agent has a reason unless she is aware of this
causal connection. We can understand why she believes what she does, but
we might still ask whether she is justified in believing as she does.
Causal analyses of knowledge are explicit about the move from eviden-
tial to causal reasons. Marshall Swain, for example, distinguishes between
“causal” and “evidential” reasons, the latter being what we would be more
ordinarily inclined to think of as a reason.20 Swain acknowledges that evi-
dential reasons are sometimes causal reasons. However, he insists that the
relevant reasons for justification are causal reasons. Consider again the
belief that there is coffee in the cup. Clearly, a causal connection exists
between agent and cup, namely, perception. In the causal view, the fact of
this connection is enough to justify the agent’s belief; there is a reason for
the agent’s belief. In the evidential sense of reason, however, for the agent
to be justified, the existence of this causal connection is not enough. The
agent must also at the very least believe that this connection exists. The
causal theory counts the notion of causal reason as the primary sense of
reason. We will be especially interested in this kind of move as we turn to
reliability theories in the following chapter.
Indefeasibility Theories
A widely favored response to Gettier-type problems is some type of inde-
feasibility theory. In this view, knowledge requires indefeasible justification.
Roughly, an indefeasible justification is one for which there is no under-
mining evidence.
The motivation for this view lies in holding that what goes wrong in
Gettier cases is that the agent’s justification is somehow defective. It is true
that the agent’s belief is justified, that the agent has some evidence for the
belief. But there is other evidence that the agent does not possess. There are
other true propositions that the agent does not believe or recognize but that
are nonetheless relevant to the truth of target belief. This other evidence is
such that if the agent possessed it, the agent’s current belief would no lon-
ger be justified; the agent’s justification would be undermined. Thus, there
is a sense in which the agent’s justification is incomplete. The agent has
some evidence but lacks other evidence that would render the target belief
insufficiently justified to count as knowledge. This undermining evidence is
called defeating evidence. A justification is defeasible if and only if there is
defeating evidence. Hence, the sort of justification required for knowledge
is not simply justified true belief but also indefeasible justification.
20 Swain 1981, p. 75.

Review Copy
We can begin to better understand this view by noticing how an inde-

feasibility theorist provides plausible diagnoses of the Ford and the barn
examples. In the Ford case, S believes that Nogot owns a Ford. On the
basis of this justified belief, S concludes that someone in the office owns a
Ford. We have not far to look to uncover a defeater: Ms. Nogot does not
own a Ford. If S believed this, then S would not be justified in believing
that someone owns a Ford (recall, after all, that S had no evidence regard-
ing Havit’s owning a Ford).
Note that the causal theory we examined earlier doesn’t help with the
barn example. Seeing the actual barn causes the father’s belief, and thus
he has a true belief, so why doesn’t he have knowledge? But indefeasibility
theory does appear to work better. In the barn case, we might identify the
defeater as the proposition that there are a number of fake barns in the
area that, from the highway, look exactly like real barns. Were the father
to believe the proposition that there are fake barns in the area, he would
no longer be justified in believing that he had succeeded in pointing to a
barn.
It is worth noting what it would take for the father to have knowl-
edge—that is, what might count as an indefeasible justification. His jus-
tification would need to be such that the information regarding the barn
facades no longer threatened his justification for stating, “That’s a barn.”
We might imagine, for example, that he pulls over and takes a closer look
from different perspectives to assure himself that he is indeed seeing a real
barn. Once he sees that the barn is in fact a barn, the information that
there are barn facades in the area is less troubling. Indeed, even if every
other barn-like object in the area is but a facade, the father seems indefea-
sibly justified in his belief about the real barn, for he now has reasons (spe-
cifically, his having seen the real barn from a variety of angles) that could
rule out the possibility that he is looking at a fake barn. Under these sorts
of conditions, we could say that the father has an indefeasible justification
for his belief, and hence has knowledge.
Indefeasibility strategies thus look promising. The difficulty arises,
however, in trying to formulate how the JTB analysis should be amended.
As one epistemologist has pointed out, the presence of the defeating evi-
dence—evidence that undermines one’s justification—is an entirely unex-
pected fact.21 Indeed, the existence of defeating evidence is not something
that the agent can simply introspect. As we will see, the unexpected nature
of the defeating evidence makes it difficult to say what should be required
for a justification to be indefeasible.
Perhaps we can approach the problem in this way: In the Ford case, S
does have an undefeated justified true belief that someone owns a Ford.
That is, S does not currently believe that Nogot does not own a Ford, so as
yet her justification remains intact. S ’s justification is undefeated, but she
21 Swain 1978, pp. 160–183.

Review Copy
does not have knowledge. Thus, we have to say that to have knowledge, S
must have a justification that is indefeasible—that there are no potential
defeaters lurking nearby. This means that S ’s actual or current justification
would have to be such that it screens out or blocks all potential defeaters.
Indefeasibility theorists do not require that an agent be able to show
that there are no defeaters. Nor do they require that the agent believe
there are no defeaters. Rather, for a justification to be indefeasible, these
theorists require merely that there in fact be no defeaters. Whether this fact
is reflected in the agent’s beliefs makes no difference to the defeasibility of
the agent’s justification. You might notice that this is a departure from a
more traditional conception of knowledge such as Descartes’. This sort of
departure proves to be somewhat controversial, and we will return to it.
But first, we need to consider how an indefeasibility condition might be
expressed.
A seemingly natural way to state such a condition is to claim that there
is no true proposition such that if it were added to S ’s justification, S would
no longer be justified. We should note just how strong a condition this is.
It claims that if there is any evidence that would, if believed by the agent,
undermine the justification, then the agent would no longer know. Thus,
according to this condition, the existence of even one defeater is sufficient
to preclude knowledge. Worse, it might seem as though even misleading
evidence could rule out knowledge.
Consider a slight modification of a case found in the literature. 22
Imagine that you attend a wedding of two close friends and that the cer-
emony is performed by a priest well known to you and the others at the
ceremony. The wedding is conducted and concluded without a hitch. On
the basis of this evidence, you come to believe that the couple is married.
Clearly, your belief is justified; you have every reason to think that you
know your friends are married. But let us suppose that unbeknownst to
you and the others present, including the priest, the bishop has gone crazy.
Among other things, the bishop falsely denounces the priest as a fraud.
Now, here is a true proposition: The bishop has denounced the priest as
a fraud. Were you aware of this proposition, you would not be justified
in believing your friends to be married, because fraudulent priests cannot
marry anyone, any more than fake barns can provide sanctuary for our
barnyard friends. Despite the apparent defeater, we are still tempted to say
that you know.
The crazy bishop case illustrates that a distinction must be drawn
between genuine and misleading defeaters. Misleading defeaters are defeat-
ers that can themselves be defeated; that is, still further evidence can be
obtained that would “restore” the original justification. What further evi-
dence defeats the defeater in the crazy bishop case? The bishop is crazy
seems a promising candidate.
22 Swain 1978, p. 165.

Review Copy
Apart from crazy bishops, misleading defeaters are illustrative of the

central intuition behind indefeasibility approaches and their most diffi-
cult obstacle. According to indefeasibility theorists, defeasible justifications
occur because an agent is less than ideally situated with respect to the
believed proposition.23 On the other hand, an agent’s justification is inde-
feasible only to the extent that the agent could acquire further evidence or
information without losing the original justification. An agent’s knowledge
ought to be extendible. According to indefeasibility theorists, an indefea-
sible justification cannot be undermined by the agent’s acquisition of new
evidence. There is something intuitively plausible about the extendibility of
knowledge. If one knows that some object is a barn or that some couple is
married, then one shouldn’t lose this knowledge merely because one learns
some further bit of information. The idea is that indefeasible justifications
are able to withstand the acquisition of new information.
Strong and Weak Conditions

In the literature, a particular condition is sometimes referred to as too
“strong” or too “weak.” A condition is too strong if it rules out cases
that we would be inclined to think of as knowledge. For example, we
suggested that Johnson’s logical interpretation of the justification condi-
tion is too strong because it rules out too many cases that we would be
inclined to consider knowledge. A condition is too weak if it counts cases
as knowledge that should not be so considered. Russell rules out identify-
ing knowledge with true belief because those conditions are too weak;
doing so would count the person who looked at the malfunctioning Big
Ben as having knowledge.
Perhaps now we can see more clearly the connection between misleading
defeaters and indefeasible justifications. Misleading defeaters are true prop-
ositions. Yet when they are properly understood, they should not undermine
an adequate justification. For example, the true proposition that the bishop
denounced the priest as a fraud is, in a sense, not the whole story; we must
also consider the bishop’s insanity. So, there is some piece of information
out there showing that the defeater is misleading, that it does not actually
undermine the agent’s knowledge. By itself, the misleading defeater appears
to show that the agent does not have knowledge. But once the whole story
is revealed, it becomes clear that the agent does know.
Yet it is not easy to see exactly how to articulate the intuition that
knowledge is extendible or that we can acquire new information without
losing our original justification. We need a condition that will identify the
defeasible justifications but will not be so strong as to rule out knowledge
in those cases in which the defeaters are misleading defeaters. We can rule
23 Swain 1978, p. 168.

Review Copy
out the misleading defeaters easily enough—for example, by requiring that

there be no true propositions that would undermine the agent’s justifica-
tion or that an agent’s knowledge be extendible—but this can come at a
very high price. Indeed, it has been pointed out that if we read such condi-
tions in a very strong sense, then it would seem to result in the claim that
we never have inductive or nondemonstrative knowledge.24 If any pocket
of evidence runs counter to our legitimate inductive inferences, then on
this strong reading of indefeasibility, beliefs based on such inferences could
never be knowledge.
The requirement that there are no true propositions that would under-
mine the agent’s justification is reminiscent of the skeptic’s apparent
requirement that to know, you must be able to rule out any alternative
explanations. But there is an important difference here. Indefeasibility
requires only that an agent’s justification not be susceptible to actually
true propositions that would provide counterevidence for the agent’s belief.
The skeptic requires that an agent be able to rule out even possibly true
competing explanations.
This difference should not mislead us. If we still require that our cur-
rent justification be such that it could survive our acquiring any true
proposition, then there may be very few cases in which we can be said to
have knowledge. And this seems to be due to the fact that much of our
knowledge is of the nondemonstrative or inductive sort. We come to know
things inductively even though our evidence does not entail the proposition
known, nor could it entail the proposition. This leaves open the possibility
that certain true propositions might count against our justification.
Suppose, however, we try to find an understanding of the indefeasibility
condition that does not seem to lead to this overly strong result. The idea
here is that even though learning of the misleading defeater would under-
mine our justification, we could come to learn something else that would
show us why the defeater is misleading. Although this may be a plausible
idea, it is difficult to accomplish.25
In learning some additional information that shows the defeater to be
misleading, we learn that our original justification was not defective, but
merely incomplete. Our original justification is thus saved. But if we are
said to have knowledge in such cases, this would seem to be due to our
now being justified in believing that our original justification was nonde-
fective. As Robert Meyers puts it, “What protects your knowledge is not
the original justification but the fact that you learned the counterevidence
along with the evidence that nullifies it.”26
24 See Meyers 1988, pp. 91ff.

25 See, for example, Shope 1983, Chaps. 2 and 3. Swain also addresses some of these
complexities in a number of places, including Swain 1978 and Swain 1981.
26 Meyers 1988, p. 96.
Review Copy
Let us see if we can understand what Meyers has in mind. Again, sup-
pose that you have just witnessed the wedding of your friends. You believe
that they are married, and your justification for this belief is your belief
that a priest has performed the ceremony. You would thus appear to have
a justified true belief. But now you learn that the bishop denounced the
priest as a fraud. This new bit of information undermines the justification
for your belief about the wedding. You can “regain” your knowledge by
learning that the bishop is no longer trustworthy about certain matters.
But now it seems as though your justification is different. Previously, your
justification consisted of your belief that a priest performed the ceremony.
Your “new” justification seems to consist of your belief that a priest per-
formed the ceremony and that the bishop cannot be trusted.
Why does Meyers believe that the change in your justification matters?
Perhaps for something like the following reasons. Initially your original
justification appears to be enough or sufficient to give you knowledge. But
attempts to rule out misleading defeaters seem to have the result that your
original justification is not sufficient. Thus, indefeasibility analyses, which
attempt to deal with the problem of misleading defeaters, fail to distin-
guish between justifications that are defective and justifications that are
merely incomplete. In the former case, the agent is undermined by genuine
defeaters; the agent does not have adequate evidence, as in the Ford case.
In the latter case, the agent’s justification is adequate, but it simply does
not include the misleading defeater and the relevant counterevidence.
One might wonder whether conditions of knowledge should require an
agent’s justification include evidence sufficient to rule out any misleading
defeater. To require this would seem to be too strong a constraint. But the
indefeasibility theorist might suggest the following. A justification need not
include evidence sufficient to rule out misleading defeaters; it is only neces-
sary that the agent could acquire that evidence without changing or giving
up the beliefs that comprised the original justification. For example, in
learning that the bishop is crazy, you do not need to change your original
justification comprising various beliefs, including that Pierre and the made-
moiselle were married by the priest. Indeed, the new evidence vindicates
your original justification. This strategy seems to have some promise, but
we leave the matter here.
We can briefly recount the main features of indefeasibility views. We
sometimes fail to have knowledge in cases in which we have a justified true
belief because our justification is not quite good enough. We might have
done the best we could, given the evidence we possessed. Unbeknownst to
us, however, additional evidence or information is relevant to the target
belief. Were we to come to know about that evidence or information, we
would recognize that our original justification was inadequate; we would
no longer be justified with respect to our original belief. Hence, we do not
have knowledge in such cases.
Review Copy
Although this approach may be intuitively plausible, it is difficult to find

a way to amend the JTB analysis that satisfactorily distinguishes between
genuine and misleading defeaters. On the one hand, the condition may be
too strong and rule out legitimate cases of non-demonstrative knowledge.
On the other hand, if the condition is weaker, it may fail to distinguish
between defective and merely incomplete justifications. However, if there
is a means of satisfactorily amending the condition, perhaps along the
lines of the “vindication” approach, then the indefeasibility approach to
the analysis of knowledge will have overcome one of its more significant
hurdles.27
No-False-Premise Views
Still another way to analyze Gettier cases is to note that the agent’s jus-
tification relies on a mistaken or false premise. In the Ford case, the false
premise is explicit: Nogot owns a Ford. In the barn case, the false premise
is more implicit but looks something like this: If it looks like a barn from
the highway, then it is a barn. (Another candidate is that there is nothing
unusual about the environment.) Thus, the fourth condition that might be
added to the JTB analysis is that the justification does not depend on any
false premises.28
As it stands, this proposed fourth condition is too strong. It may be that
an agent believes a false premise but that this premise does not, or need
not, play a central role in the agent’s justification, given what else the agent
believes. Consider a slight variation of the Ford case.
Suppose S believes that someone in the office owns a Ford because she
has seen the registration papers and seen Havit, who owns it, driving a
Ford. Suppose, however, S falsely believes that Havit bought the car at
East Ford. And suppose she infers that someone in the office owns a Ford
on the basis of her belief that Havit owns a Ford that was purchased from
East Ford.
We might nonetheless think that S knows that someone owns a Ford,
because where the car was purchased makes no difference to Havit own-
ing a Ford. The possibility of false but inessential premises requires some
explanation of what it means to say that a premise plays an essential
role in the justification. This further explanation centers on the idea that
the agent would not be justified in believing that P unless the agent also
believed some further proposition Q . The proposition Q is then said to be
essential to the agent’s justification.
27 Of course, internalists may worry about externalist features of the indefeasibility

view, but we cannot pursue these matters here.
28 Meyers is explicit about holding such a view, at least for cases of inferential
knowledge. Others have held a similar view—see, for example, Harman 1973—but
this seems not to be the strategy of choice.
Review Copy
So, the fourth condition in the no-false-premise view is that there is no

false proposition essential to the agent’s justification. Or, as Meyers states
the condition:
S is justified in believing p on the basis of q only if: every prop-

osition essential to S being justified in believing p on the basis
of q is true.29
There is a standard objection to this view. The idea behind this objec-
tion is that if an agent reasons cleverly enough, then the agent will not
be committed to believing any false premises. But such objections seem
to trade on the idea that if no false premise is explicitly articulated in the
agent’s reasoning, then the agent’s justification does not depend on a false
premise. Richard Feldman suggests an example that appears to undermine
the no-false-premise view.30 In the Ford case, Feldman argues that if S were
to use existential generalization prior to her inferring that Nogot owned
a Ford, then S ’s justification would not depend on the false proposition
that Nogot owns a Ford. That is, if on the basis of her evidence that she
had seen Nogot driving a Ford and she had been told that Nogot owned
a Ford, she had immediately inferred that someone in the office owned a
Ford, then S ’s justification would not depend on a false premise.
In response, we might reply that whether S articulates or reasons explic-
itly from a false proposition, S clearly seems to believe something false. In
particular, S clearly seems to believe that Nogot owns a Ford. Indeed, were
it not for S having this belief, it would seem strange to us that she would
infer that someone owns a Ford.
The point here is that the explicit premises in one’s justification may
well depend on further background beliefs. Feldman’s counterexample
seems to ignore these background beliefs. Doing so permits him to claim
that we need only consider the explicit premises of an agent’s justification.
But once we recognize that some background beliefs are essential to an
agent’s explicit reasoning, we can see that in a legitimate sense what goes
wrong in Gettier-type cases is that some false belief underlies the agent’s
justification.
This line of thought may be problematic. We do not want to understand
this condition so strongly that every background belief held by the agent
is an essential proposition for the agent’s justification. But the notion of
an essential proposition seems to trade on our intuition that some beliefs
matter more than others for particular cases of reasoning and that some
beliefs matter not at all. If this intuition proves to be well grounded, then
29 Meyers 1988, p. 100.

30 Feldman 1996.
Review Copy
there is much to recommend the no-false-premise view.31 For the moment,

we will leave matters here.
The Significance of Gettier
The Gettier problem and subsequent work on it have not been univer-
sally prized. As counterexample spawned counterexample, as rebuttal met
with rebuttal, and as yet a further subclause was added to the analysis of
knowledge, some suspected that the forest had been lost, not for the trees,
but for the pine needles. However, important issues have arisen as a result
of this process. We will restrict our discussion to two main issues: (a) the
nature of knowledge and (b) the nature of justification.
First, you may have already asked yourself what the proper response
to Gettier should be. Each of the three strategies presented previously was
motivated by intuitively plausible considerations, yet each faced certain
difficulties. So, can we expect a return to the widespread agreement in
evidence prior to Gettier?
Notice that the indefeasibility and no-false-premise views agree that
justified true belief is necessary, if not sufficient, for knowledge. Depending
on how we interpret the causal theory, it, too, might be viewed as holding
that justified true belief is necessary for knowledge. We should also notice
that our justified true beliefs can be “gettierized” in many ways. These
observations suggest two things. First, there may be no single account that
deals with each and every counterexample. In turn, we are led to a second,
perhaps more important consideration.
As characterized here, the Gettier problem results from a potential gap
between the satisfaction of the justification condition and the satisfaction
of the truth condition. The challenge is to find a way to close this gap.
It may be, however, that this gap can arise in different ways. Indeed, we
might view our three accounts as suggesting different ways in which a gap
might arise between the justification and the truth conditions. Thus, dif-
ferent ways of closing the gap may be necessary, just as is suggested by the
three views in the previous section. That is, different types of justification
may be appropriate under different circumstances.
We can elaborate the idea that different responses to Gettier cases might
address different ways in which the traditional analysis is inadequate.
Ernest Sosa has suggested that knowledge is apt or appropriate (true)
belief.32 We can view this suggestion as a schema whereby the challenge
31 It is perhaps worth noting that more recently Feldman suggests a version of the
no-false-premise view, which takes account of essential premises. See Feldman 2003,
pp. 36–37. Feldman does not explicitly mention “premises,” but condition (iv) of his
“modest proposal” reads “S ’s justification for p does not depend essentially on any
falsehood.”
32 Sosa 1991b, pp. 245–246.
Review Copy
is to identify the circumstances under which a belief is apt. We may want

to distinguish, for example, between simple perceptual beliefs and beliefs
arrived at as the result of a complicated chain of inductive reasoning. Thus,
there may be no clear choice among the analyses presented. Still, we should
keep in mind the extent to which a particular view can be extended to
cover a broad range of types of counterexamples.
A second issue about knowledge arises. Do Gettier cases show, as
Mark Kaplan’s article title suggests, that “It’s Not What You Know That
Counts”?33 Kaplan argues that finding the appropriate response to Gettier
would make no difference in how we conducted our inquiries. Once an
agent has made the best possible use of his or her evidence, Kaplan argues
that whether the agent knows makes no difference. That further condition
seemingly required by Gettier cases, which would make a justified true
belief an instance of knowledge, is not something the agent can check to
see if it is satisfied. What goes on in Gettier cases is that the environment
has been “tricked up” in some way that normally exceeds an agent’s dis-
criminative capacities. Because such cases lie outside the agent’s ability to
recognize them, it is simply not clear what further requirements should be
made of the agent. But then, whether the agent has knowledge seems to
leave the fact that the agent has a justified true belief unaffected.
One might concede some of Kaplan’s suggestion but still resist the con-
clusion that a solution to Gettier is unimportant.34 In response to Kaplan’s
claim, it is argued that the goal of inquiry is not simply to attain a justified
belief, but rather to attain knowledge. Moreover, although Gettier prob-
lems may indeed often be undetectable by the agent, in some cases further
investigation may lead to genuine knowledge.
Another line of response to Kaplan is the following: At the outset of this
chapter, we distinguished between the assertibility and truth conditions
of knowledge. Here, we might add the notion of verification conditions.
Kaplan seems to be suggesting, among other things, that an analysis of
knowledge is useful only if we can verify that we have knowledge, only if
we can somehow check to see that we have knowledge. The verification
conditions of knowledge are a separate issue from the truth conditions of
knowledge. There clearly is an independent interest in whether an agent
knows, even if the agent cannot verify that knowledge.
We might want to understand the concept of knowledge despite the fact
that we might not always be able to recognize every occasion that a person
has knowledge. Moreover, there is reason to think that post-Gettier analy-
ses of knowledge can make a difference to our future inquiries. Suppose,
for example, that we understand the concept of knowledge along the lines
of indefeasibility. Imagine that Sara knows, even though she cannot verify
that her justification is indefeasible. Later, however, she may learn addi-
33 Kaplan 1985.
34 See Conee 1996.
Review Copy
tional facts that explain why her justification is indefeasible, and she might
be able to use these new facts to guide her future inquiries.
Two final points can be made about Gettier cases. First, although there
may be an air of unreality about some of them, cases such as Goldman’s
barn example surely seem to involve genuine instances of justified belief.
Natural Gettier cases may be rare, but there are such cases. A recent issue
of Science News reports that researchers may have had a justified true
belief that “yo-yo dieting” can cause cancer but that recent studies sug-
gest they believed it for the wrong reasons.35 Such a case might count as
a “natural” Gettier case. However, although such cases might not be the
norm, they indicate that we are sometimes epistemically better off when
our justification reflects the facts, that is, when our justification does in fact
explain why the belief is true.
Second, Gettier cases can be used to draw our attention to certain aspects
of knowledge. At the very least, we might think that some Gettier cases sug-
gest that we can have knowledge only in a generally cooperative environ-
ment. If we rely on our generally “accredited routes to knowledge,” we count
on the world being more or less normal. We can fail to have knowledge due
to the global complications imagined by Descartes, but knowledge may also
elude us when the world is less dramatically abnormal. This “failure of nor-
mality” can go undetected, and thus preclude our having knowledge, even
though we have done all that we might reasonably be expected to do.
A distinction, which will be important to us in subsequent chapters, is
that between externalist and internalist accounts of knowledge and jus-
tification. Externalism is roughly the idea that an agent’s justification or
knowledge depends on some condition that need not be reflected in an
agent’s beliefs.36 Internalism, on the other hand, roughly holds that a belief
is justified only if the relevant justifying conditions are in some respect
reflected in the agent’s beliefs or cognitive perspective.
Some of the responses to Gettier may signal a move toward external-
ism.37 In discussing the causal theory, we noticed the shift from evidential
reasons to causal reasons. What matters is the causal mechanism that pro-
duces the agent’s belief. The causal theory does not appear to insist that
knowledge depends on the reasons for which an agent holds the belief.
This has led some to think that knowledge is externalist in nature, while
justification is internalist.38
35 Science News 1997.

36 This is only a rough account; the next chapter will address the needed refinements.
Also, in a sense, even the traditional analysis is externalist, because whether the belief
is in fact true is not necessarily cognitively accessible to the agent. But classification
of a view as externalist (or internalist) does not refer to this feature.
37 In Kitcher 1992 Philip Kitcher suggests that one of the sources of naturalized
epistemology, an externalist view, is the reaction to Gettier.
38 Representative examples of this view may be found in Audi 1988 and Sosa 1991a.
See also Chapter Six.
Review Copy
As a way of illustrating this view, we might briefly consider a view of

knowledge recently outlined by Richard Foley.39 Foley suggests that an
agent knows when the agent has a true belief and the agent is not missing
significant or relevant information about the target belief and the related
matters. For example, in the familiar barn case, the father is missing sig-
nificant and relevant information about the vicinity. And the significance
or relevance of some information depends on context.
In some narrowly defined contexts, simply having a true belief is suf-
ficient for knowledge. Foley mentions the narrow context in which college
students are asked quite specific questions about dates in history, e.g., the
year World War II began. Foley suggests that possession of that informa-
tion is enough to be said to know in what year World War II began. Larger
or broader contexts of course will require possession of a larger number
of true beliefs. It is also difficult precisely to identify the relevant context
or the “neighborhood” of a target belief, but this, Foley suggests, is true
of any account of knowledge. We saw something similar in our consider-
ation of the indefeasibility analysis. There the difficulty lay in specifying
precisely how much of the story, relevant to a particular belief, an agent
must have.
The significant feature of Foley’s account for our present purposes is
that justification is severed from the analysis of knowledge. Foley suggests
that consideration of Gettier cases draws attention to the presumed con-
nection between knowledge and justified belief. It is a dubious connection,
in Foley’s view. In asking whether a person knows, we are asking whether,
relative to the “neighborhood” of the target belief, the person has sufficient
true beliefs. If we think that a person does not know, then, according to
Foley, we should look for those true beliefs that the person lacks, those
which explain why the person lacks knowledge. Again, for example, in
the crazy bishop case, what true beliefs does the person lack in the rel-
evant context? That the bishop is crazy, of course, and cannot be trusted.
In determining, however, whether a person has sufficient true beliefs, we
are not addressing the issue of whether the person is justified. Assessing
the justificatory status of a belief, argues Foley, is a task separate from
assessing whether a belief is an instance of knowledge. And Foley further
suggests that externalists are principally interested in knowledge, while
internalists are principally concerned about the nature of justification.
The important point is that we notice at least the beginning of the divi-
sion between internalist and externalist conceptions of knowledge and
justification. In Chapter One, the traditional conception of knowledge
and justification seemed largely internalist. But in the next chapter, we
shall see a type of causal theory that explicitly repudiates that traditional,
internalist conception of knowledge and justification. Indeed, the contrast
39 Foley forthcoming. Used with permission.

Review Copy
between internalism and externalism will concern us in several subsequent

chapters.
Key Concepts
Traditional analysis of
No-false-premise theories
knowledge
Gettier-type counterexamples Internalism
Causal theories of knowledge Externalism
Indefeasibility theories of
knowledge
Review Questions
1 Do you think that either the Ford case or the barn case is a case of
knowledge? Explain. If not, explain why each is a case of justified
true belief, but not a case of knowledge. Can you construct your own
Gettier case?
2 How do causal theories attempt to avoid the problem of the indepen-

dent satisfaction of the justification and the truth conditions?
3 What is the difference between genuine and misleading defeaters? Do

you think indefeasibility theories can resolve the problem presented by
misleading defeaters? Explain.
4 Explain Feldman’s objection to the no-false-premise view. Do you

think this objection shows that this view is an inadequate response to
Gettier? Why or why not?
For Further Study
The debate surrounding Gettier examples mainly took place in journal

articles. The most comprehensive treatment of the various theories and
variations on theories is Shope 1983. Early responses to Gettier, including
Gettier’s original paper “Is Justified True Belief Knowledge?” are reprinted
in Roth and Galis 1970. Three other worthwhile collections are Pappas
and Swain 1978, Crumley 1999, and Midwest Studies in Philosophy 5
Review Copy
1980. John Pollock also develops a version of indefeasibility theory in

Pollock 1986, pp. 180–193.
In Williamson 2000, Timothy Williamson argues that the concept of
knowledge is unanalysable; that is, knowledge is a kind of primitive or
basic notion, which cannot be explained further. Frank Jackson in Jackson
2002 responds to Williamson’s argument.
A view not discussed in the text is Robert Nozick’s “tracking view.”
Critical appraisals of this view can be found in Luper-Foy 1987.
A useful introduction to theories of truth is Horwich 1992. An acces-
sible, yet comprehensive examination is Schmitt 1995b; Chapter Seven
examines truth in connection with the analysis of knowledge and
skepticism.

Crumley II - Cap 2 - Anaysis of Knowledge

Uploaded by

Copyright:

Available Formats

Crumley II - Cap 2 - Anaysis of Knowledge

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Crumley II - Cap 2 - Anaysis of Knowledge

Uploaded by

Copyright:

Available Formats

Review Copy

Chapter Two An Introduction to the

Actually, the chapter title is a bit misleading. The focus of this

view Descartes’ understanding of the concept of knowledge as a variation

The Traditional Analysis

Motivating the Traditional Analysis

That condition (K A-1) is a necessary condition for knowledge has seemed

2 Ayer 1956, pp. 31–35.

of false belief. Yet we often desire to distinguish between someone who

of the epistemic status of the belief. Let us focus on the psychological

(K A 2*) S believes that P .

A few critics have objected to even this revised condition as a neces-

3 See, for example, Radford 1970, pp. 171–185.

With the qualification that condition (3) was historically understood in

Gettier and the Traditional Analysis

Although it is rare to find anything approaching consensus in philosophy,

5 Ayer 1956, p. 33.

Bertrand Russell and Gettier-type Counterexamples

The second counterexample derives from Alvin Goldman.9 Imagine a

9 Alvin Goldman 1988a, pp. 41–65.

Gettier-type counterexamples are abundant in the literature, but these

A Defense of the Traditional Analysis

to find innumerable other beliefs, which we typically count as knowledge,

Some Strategies for Handling Gettier Problems

So far, we have been assuming that the justification condition requires

16 Some philosophers interested in the metaphysics of causation prefer to talk about

condition can no longer be satisfied independently; the two conditions are

17 Alvin Goldman 1970, pp. 67–87.

20 Swain 1981, p. 75.

We can begin to better understand this view by noticing how an inde-

21 Swain 1978, pp. 160–183.

22 Swain 1978, p. 165.

Apart from crazy bishops, misleading defeaters are illustrative of the

Strong and Weak Conditions

23 Swain 1978, p. 168.

out the misleading defeaters easily enough—for example, by requiring that

24 See Meyers 1988, pp. 91ff.

Although this approach may be intuitively plausible, it is difficult to find

27 Of course, internalists may worry about externalist features of the indefeasibility

So, the fourth condition in the no-false-premise view is that there is no

S is justified in believing p on the basis of q only if: every prop-

29 Meyers 1988, p. 100.

there is much to recommend the no-false-premise view.31 For the moment,

The Significance of Gettier

is to identify the circumstances under which a belief is apt. We may want

35 Science News 1997.

As a way of illustrating this view, we might briefly consider a view of

39 Foley forthcoming. Used with permission.

between internalism and externalism will concern us in several subsequent

Gettier-type counterexamples Internalism

Causal theories of knowledge Externalism

2 How do causal theories attempt to avoid the problem of the indepen-

3 What is the difference between genuine and misleading defeaters? Do

4 Explain Feldman’s objection to the no-false-premise view. Do you

For Further Study

The debate surrounding Gettier examples mainly took place in journal

1980. John Pollock also develops a version of indefeasibility theory in

You might also like