Canadian Journal of Philosophy

Defeasibility Theories of Knowledge

Author(s): Steven R. Levy
Source: Canadian Journal of Philosophy, Vol. 7, No. 1 (Mar., 1977), pp. 115-123
Published by: Canadian Journal of Philosophy
Stable URL: https://www.jstor.org/stable/40230676
Accessed: 28-01-2019 02:58 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Canadian Journal of Philosophy is collaborating with JSTOR to digitize, preserve and extend
access to Canadian Journal of Philosophy

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
CANADIAN JOURNAL OF PHILOSOPHY
Volume VII, Number 1, March 1977

Defeasibility
Theories of Knowledge

STEVEN R. LEVY, University of California, Riverside

There have been many attempts of late to formulate a satisfactory

theory of knowledge with which to replace the traditional justified
true belief analysis.1 Almost all agree that it must be the case that in
order for S to know that p; i.) p be true, and ii.) S believe that p.
Although many argue that there must be a condition stating that S has
adequate evidence for p, requirements other than i.) and ii.) are
controversial. The most popular approach taken, however, is the
addition of a requirement to the effect that for S to know that p there
must be no other evidence against p strong enough to undermine S's
belief that p, should this evidence come to S's attention. I shall call
such a requirement a "defeasibility condition" and any theory
containing such a requirement a defeasibility theory.2

* I am especially indebted to Professors John Perry, R. M. Yost, and Peter D. Klein

for helpful comments on earlier versions of this paper.

1 The analysis of knowledge as justified true belief was refuted in Edmund

Gettier's now famous article "Is Justified True Belief Knowlege?/' Analysis June,
1963.

2 Among those who propose defeasibility theories are Ernest Sosa ("The Analysis
of 'Knowledge that p'," Analysis, October, 1964), Keith Lehrer and Thomas
Paxson, Jr. ("Knowledge: Undefeated, Justified True Belief," The Journal of
Philosophy, April 24, 1969), Peter D. Klein ("A Proposed Definition of
Propositional Knowledge," The Journal of Philosophy, August 19, 1971), Risto
Hilpinen, ("Knowledge and Justification," Ajatus, 1971), and Marshall Swain
("Epistemic Defeasibility," American Philosophical Quarterly, January, 1974).

115

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
Steven R. Levy

Defeasibility theories are quite effective in handling a certain kind

of difficulty to which theories of knowledge are subject, namely
problems dealing with what Harman has termed the social aspect of
knowledge.3 However, I shall argue that in spite of this strong asset, it is
quite doubtful that any formulation of a defeasibility theory can give
us what we are looking for in an adequate analysis of knowledge. The
type of example that inspires defeasibility theories is exemplified by a
case devised by Lehrer and Paxson.4 In this case S sees a man remove a
book from the library by concealing it beneath his coat. S is sure the
man is Tom Grabit whom he has often seen before and thus reports
that he knows that Tom Grabit has removed the book. Now Mrs.
Grabit, Tom's mother, avers that on the day in question, Tom could not
have been in the library but that his identical twin brother, John
Grabit, was in the library. However, S does not know that Mrs. Grabit
has so averred. Had he known, he would no longer believe that Tom
removed the book. Now, as it turns out, Mrs. Grabit is a compulsive
and pathological liar and there is no John Grabit. Tom indeed
removed the book, just as S believed. Although Lehrer and Paxson
claim that in this particular case S does know that Tom Grabit removed
the book, it is suggested by Harman that should Mrs. Grabit's
testimony become widely known (although still unknown to S) and
the fact that she is a compulsive liar remain unknown, S would not
know that Tom Grabit removed the book.5 For if we were to grant S
knowledge in this case, we would find ourselves in the paradoxical
position of claiming that a person who does not have all of the
available evidence has knowledge while others who do have all of the
evidence do not have knowledge. S, then, would have knowledge in
virtue of his ignorance! How can such examples be ruled out as
instances of knowledge? S's belief is certainly justified and true. One
answer lies in the formation of a defeasibility condition.
Let us consider the following as representative of defeasibility
theories of knowledge:

S knows that p if and only if:

i.) p

3 Gilbert Harman, "Knowledge, Inference and Explanation/' America

Philosophical Quarterly, Vol. 3 (July, 1968). See VIM, The "Social Aspect"
Knowledge.

4 Lehrer and Paxson, op. cit., pp. 228-29.

5 Harman, loc. cit.

116

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
 Defeasibility Theories of Knowledge

ii.) S believes that p

iii.) S has adequate evidence for p (which can be characterized by

true statements)

iv.) There exists no other evidence (i.e. no other set of true

statements) such that if it became known to S, S would no
longer have adequate evidence for p.6

This analysis rules out the Tom Grabit case because although
conditions L), ii.), and iii.) are satisfied, condition iv.) is not. The true
statement that Mrs. Grabit testified as she did is one such that if it
became known to S, S would no longer have adequate evidence for p.
Thus the defeasibility requirement did the job that it was devised to
do. However I shall argue presently that this condition all too often
runs amuck.
Consider this reformulation of the Grabit case; S sees Tom Grabit
remove a copy of The Life and Times of Kant from the library and
comes to believe that he stole it. Under the circumstances this belief is
justified. Now suppose that the following propositions are all true:

1a Tom Grabit is an employee of the library.

1b Tom Grabit was hired to sweep out the stair-wells in the

library once a week.

2a Some employees of the library are authorized to remove

books from the library.

2b The only person who may remove a book from the library is
the head librarian.

3a Tom Grabit purchased a copy of The Life and Times of Kant

the day before S saw him remove one from the library.

3b Tom Grabit purchased a copy of The Life and Times of Kant

the day before S saw him remove one from the library and he
lost it on his way home.

6 This is, of course, representative in that it just gives the bare bones of
defeasibility theories. Examples of some early efforts can be found in Sosa, op.
cit, pp. 7-8, Lehrer and Paxson, op. cit., pp. 225-31, and Kline, op. cit, p. 475.

117

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
Steven R. Levy

Suppose further that S could reasonably have been expected to have

found out or otherwise have known these propositions but that he is
nevertheless completely justified in believing them to be false.7 Now
although 1a, 2a, and 3a are to varying degrees misleading, as can be
seen by comparing them to 1b, 2b, and 3b, respectively, all of which
are true by hypothesis, they can still be used to defeat S;s justification.
A skilful philosopher or a shady lawyer could come up with true
propositions such as 1a, 2a, and 3a for practically any knowledge claim,
yet to allow such "half-truths" to defeat knowledge claims (i.e., defeat
one's justification) leads directly to skepticism. The defeasibility
theories of Lehrer and Paxson, Sosa, and Klein are not equipped to
avoid this consequence and should be rejected for this reason alone.
The problem, however is not just that these theories do not prevent
"half-truths" or incomplete statements of relevant evidence from
defeating the justification for p, it is that formulating a condition
strong enough to rule out such propositions as defeating ones and yet
not demand the possession of total evidence (a demand which would
lead to skepticism) does not seem possible. For a theory of knowledge
to be a defeasibility theory and yet rule out these "half-truths" it must
allow the (a) set of propositions to act as defeating ones when there is
not a more complete set of true statements, such as the (b) set, which,
when taken with the (a) set do not undermine S's knowledge claim.
First of all I am not certain that any practical condition can be
formulated in which sets of statements such as the (b) set can be
described and hence picked out for use, although one very good
attempt will be discussed. Secondly, even if such a condition could be
manufactured it would not do the job for which it was intended,
namely allowing sets such as the (a) set to defeat a knowledge claim in
all and only those cases in which we want it to. Suppose, for the
moment, that we were able to construct a theory, call it DT, in which
the (a) set of propositions defeat S's justification when the (b) set does
not exist and does not defeat it when the (b) set does exist. Then, in the
last example S's justification for believing that Tom Grabit stole the
book is not defeated by the (a) statements. But now suppose that a
third set of true statements exists:

1c Tom Grabit was hired to sweep out the stairwells in the library
once a week and to act as an assistant to the head librarian.

7 I add this clause only to account for conditions added to defeasibility theories by
Sosa and Lehrer and Paxson.

118

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
 Defeasibility Theories of Knowledge

2c The only person who may remove a book from the library is
the head librarian, who may, and often does, authorize his
assistants to remove books.

3c Tom Grabit purchased a copy of The Life and Times of Kant

the day before S saw him remove one from the library, he lost
it on his way home, and then purchased another copy.

This new set of statements shows the (b) set to be the more misleading
of the (a) and (b) sets. In a case in which the (c) set of statements holds,
we should desire that the (a) set be allowed to defeat S's justification.
However our supposed theory, DT, makes no mention of this further
evidence. As long as the (b) set exists S's justification is not defeated.
Thus it cannot allow the (a) statements to defeat a knowledge claim in
all and only those cases in which we want it to.
That the existence of misleading evidence can play havoc with
defeasibility theories is a fact that has not gone unrecognized in the
literature. Indeed, a new crop of defeasibility theories is emerging to
combat this very problem. David Annis, for example, proposes a
solution, a summary of which follows:8

S knows that p if and only if

(i) p is true
(ii) S believes that p
(iii) There is a set of statements A that fully justifies S in believing that p and no
statement defeats this justification....

When a set of statements A fully justifies S in believing that p, this justification is

defeated if and only if for every fully justifying subset A', there is a statement q such
that,

(i) q is true
(ii) for some statement j, where j is a member of A' or j is p, and for the
corresponding subset Aj' that justifies S in believing j, the conjunction of the
members of Aj'and q does not justify S in believing that j -
(iii) q is not defective with respect to Aj1, j, and S.

A defeating statement q is defective with respect to A', p, and S if and only if there is a
true statement r which explains why q is misleading evidence with respect to A* and
p such that the conjunction A', q, and r justifies S in believing p but r alone does not
justify S in believing p.

8 David Annis, "Knowledge and Defeasibility/' Philosophical Studies, Vol. 24

(1973), pp. 199-203. (I have altered the notation slightly.)

119

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
Steven R. Levy

It can be seen that this analysis cannot work, when one applies the
above Tom Grabit example to it. Let p be the belief that Tom Grabit
stole the book. We can suppose that la defeats the justification S has
for p. Can we then say that S does not know that Tom Grabit stole the
book? We can if the members of the 'b-set' of statements are not true.
If the 'b' members are true then 1a is defective with respect to S's
justification, p and S. It is defective because there exists an r which
explains why 1a is misleading. Such an r might be:

r: 1a is misleading because it intimates that Tom's removing the

book from the library may be a legitimate function of his job,
but 1b suggests that this is not so.

At this point, we have an irreversible result that S does know that p

even if we now choose to allow the c statements to be true. But if we
allow the c statements to be true, 1b is seen to be the more misleading
and should not defeat the defeater. Thus, just as the hypothetical DT
could not specify when the b-set should be employed, Annis' theory
also cannot.
It may be claimed that if the c sentences are true then r is false
because it says that 1a is misleading when it, in fact, is not. This claim,
however, skirts the issue to which defeasibility theories are addressed.
Defeasibility theorists' aim is, and ought to be, to characterize
misleading evidence (whether it be justifying evidence or defeating
evidence) by providing criteria for its identification. An analysis of
knowledge which contains a non-explicit requirement concerning
misleading evidence puts us in a position only slightly better than that
of having no theory at all because the notion of what constitutes
misleading evidence is at least as fuzzy as that of what constitutes
knowledge.9 It seems clear that Annis' intent is not to become
involved in a circle by deciding what evidence is misleading by
determining whether or not we are willing to say that it can give us
knowledge, and then apply that criterion to deciding whether or not
we have knowledge.10 Thus, his characterization of a defective
defeating statement must be construed as an attempt to otherwise

9 Gilbert Harman ("Induction," in Induction, Acceptance and Rational Belief, ed.

by Marshall Swain, 1970, D. Reidel Publishing Company, Dordrecht, Holland,
pp. 83-99) constructs a theory of knowledge containing the principle that
"There must exist no special counter-evidence to ones conclusion," 'special'
evidence being that which undermines a claim to knowledge. In a footnote he
adds, "The word 'special' is and is meant to be a fudge."

10 Harman, it seems, does fall into this trap, ibid, and in Thought, 1973, Princeton
University Press, Princeton, New Jersey, U.S.A.

120

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
 Defeasibility Theories of Knowledge

pick out misleading evidence. And as I have argued, it does not do so

in a manner useful to constructing a theory of knowledge. The
statement that Grabit is an employee of the library is misleading in that
it doesn't tell the whole story. Whether it is misleading in a way that
undermines knowledge depends on whether the b statements tell the
whole story or the c statements do. Misleading evidence not captured
in Annis' theory can exist unless S has total evidence - an undesirable
consequence.
One other attempt at constructing a defeasibility theory of
knowledge must be mentioned. The most evolved theory to date
that of Marshall Swain. The intent of Swain's condition is that S's belief
that p is indefeasible if and only if his justification for p would hold
an enlarged "epistemic framework" in which he is justified in
believing all those true propositions that he is in fact justified i
believing to be false (Swain calls such a framework an evidence
restricted alternative [Fs*] to S's actual epistemic framework [Fs]. Th
ewcfence component of an epistemic framework is a subset of it whic
consists of all true expressions of the for "S is justified in believing

epistemic framework such that if that altern

obtain) then S would be more ideally situat
fact is."12 Swain's final formulation of his d

(ivg)

S's justification for p is indefeasible (that is, there is an evidence-restricted

alternative Fs* to S's epistemic framework Fs such that (i) "S is justified in believing
that p" is epistemically derivable from the other members of the evidence
component of Fs* and (ii) there is some subset of members of the evidence
component of Fs* such that (a) the members of this subset are also members of the
evidence component of Fs and (b) "S is justified in believing that p" is epistemically
derivable from the members of this subset).13

Condition (i) states that S must be justified in believing that p in the

specified enlarged epistemic framework. Condition (ii) requires that
the justification picked out by (i) be essentially the same as S's original
justification. This second condition is necessary to prevent the
granting of knowledge to S when, although his justification is

11 "Epistemic Defeasibility/' American Philosophical Quarterly, January, 1974.

12 Ibid., pp. 21-22.

13 Swain, op. cit., p. 25. 1 have altered Swain's notation slightly. Where I have 'p' he
has 'h\

121

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
Steven R. Levey

defeated, there exists some discoverable justification which is perhaps

totally unrelated to the original and the premises of which S is justified
in believing to be false. This formulation fails, however, because it rests
on the assumption that believing a true statement always puts one in a
better epistemic position than believing a comparable false one.
However, since true statements can be misleading there is no reason
to suppose that in some situations certain false statements cannot be
more beneficial epistemically. As an example, consider the above
Grabit case with a slight variation to make the point more graphically. S
sees Tom remove the book. Suppose that S is justified in believing it
false that Tom is an employee of the library (S inspected the University
records the day before and found that Tom was employed by the
maintenance division and not the library). Suppose, also, that S is
justified in believing it false that some employees of the library may
remove books (such was printed in the latest bulletin). Given these
beliefs, the inference to p, that Tom's removal of the book was
unauthorized, is certainly a warranted one. When we add that Tom did
steal the book there should be little hesitation in saying that S knows
that he did. Now let us suppose that on the morning in question, for
reasons only understood by administrators, the names of some
maintenance employees were transferred to the library's budget,
although there was no change in duties. Tom, of course, was one of
those transferred and hence is technically an employee of the library.
Also, for purely bureaucratic reasons, and unknown to all but a few,
the rule that no employees may remove books was rescinded. S's
beliefs in these two areas, then, are false. But it seems that they are false
on a technicality. S still knows that Tom was unauthorized in removing
the book. In terms of Swain's theory, the following picture emerges.
The evidence component of Fs, S's actual epistemic framework,
contains the following statements:

S is justified in believing:

1) Tom removed the book.

2) It is not the case that Tom is an employee of the library.

3) It is not the case that some employees of the library may

remove books.

The evidence component of Fs*, the "more ideally situated"

epistemic framework contains:

S is justified in believing that:

122

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms
 Defeasibility Theories of Knowledge

1) Tom removed the book.

2r) Tom is an employee of the library.

3') Some employees of the library may remove books.

Since p is not derivable in Fs*, Swain would be forced to say, counter

to intuition, that S does not know that p. It may be argued that Fs also
contains a statement to the effect that:

4) This is not an instance in which Tom is authorized to remove a

book.

and that this is also contained in Fs* - hence p is derivable in Fs *. But

if 4) is in Fs , it is derived from 2) and 3). and since 2) and 3) are not in Fs *,
there is no reason to include 4). So once again, because S is not privy to
all of the evidence relevant to p, a defeasibility theory is forced to say
that S does not know that p.
I have argued that defeasibility theories of knowledge must avoid
cases in which a statement is defeating only because it is misleading. I
have argued further that in contending with this difficulty, the
defeasibility theorist must come up with a characterization of
misleading statements which is independent of intuitions concerning
when a person has knowledge (at least insofar as he desires to
incorporate such a characterization into a theory of knowledge). I do
not claim to have shown conclusively that such a task cannot be
completed successfully. However, it does seem that there are
enormous obstacles standing in the way. One obstacle is that true
statements can be misleading to varying degrees with respect to each
other and that only some should be allowed to defeat a knowledge
claim and those only sometimes. The other major obstacle is that
sometimes a false statement may be less misleading than a true one.
Because of this, attempts such as Swain's to rule out misleading
evidence by an appeal to truth cannot work. If my suspicion is true that
these obstacles cannot be overcome, then for a defeasibility theory to
work it must demand that total evidence be possessed by S in order for
him to have knowledge. But for empirical propositions total evidence
is never possessed, and the true defeasibility theorist is led to
skepticism. If this outcome is to be avoided, then it seems that another
approach must be taken in constructing an analysis of knowledge

January 1976

123

This content downloaded from 164.73.224.2 on Mon, 28 Jan 2019 02:58:08 UTC
All use subject to https://about.jstor.org/terms