knowing that one knows and the classical definition of ... that one knows and the classical...
TRANSCRIPT
Knowing That One Knows and the Classical Definition of KnowledgeAuthor(s): Risto HilpinenSource: Synthese, Vol. 21, No. 2 (Jun., 1970), pp. 109-132Published by: SpringerStable URL: http://www.jstor.org/stable/20114716 .
Accessed: 04/08/2013 17:04
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp
.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].
.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Synthese.
http://www.jstor.org
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
RISTO HILPINEN
KNOWING THAT ONE KNOWS AND THE CLASSICAL
DEFINITION OF KNOWLEDGE*
I. THE KK-THESIS
In his Essay in Modal Logic, G. H. von Wright suggested that the logic of knowledge
- or epistemic logic - is a branch of modal logic.1 E. J.
Lemmon has recommended a system of epistemic logic based on the
Feys-von Wright modal system M,2 and in Knowledge and Belief Jaakko
Hintikka proposed a stronger system which corresponds to Lewis's S4.3
The distinctive axiom of S4 is 'Np^NNp',* and the epistemic counter
part of this formula is the principle that knowing implies knowing that
one knows, that is,
(KK-thesis) Kap^KaKap,
where 'Ka9 stands for 6a knows that'. I shall call this principle 'the KK
thesis'. In recent literature on the theory of knowledge and epistemic
logic, this principle has been subjected to a great deal of discussion.5 In
this paper, I shall discuss the KK-thesis from the point of view of various
definitions of knowledge.
II. THE CLASSICAL DEFINITION OF KNOWLEDGE
In Perceiving: A Philosophical Study, Roderick M. Chisholm defined the
concept of knowledge in the following way:6
(DKi) 6a knows that /?' means:
(i) a accepts p,
(ii) a has adequate evidence for p, and
(iii) p is true.
This definition is a species of what I shall call the classical definition of
knowledge. According to the classical definition, knowledge is justified true belief, or true opinion combined with reason. A definition of this
Synthese 21 (1970) 109-132. All Rights Reserved Copyright ? 1970 by D. Reidel Publishing Company, Dordrecht-Holland
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
110 RISTO HILPINEN
type has been discussed by Plato in Theaetetus and in Meno,1 and it
underlies, e.g., Jaakko Hintikka's recent work on epistemic logic.8 Other
versions of the classical definition have been put forward by Roderick
M. Chisholm, David Rynin, and A. J. Ayer, among others.9 In all of
these definitions, the concept of knowledge is defined in terms of three
conditions, of which the first may be termed the condition of acceptance or belief the second, the condition of justification or evidence, and the
third, the condition of truth.
In Rynin's formulation of the classical definition, the first clause of
(DKi) is replaced by10
(DKi2) a believes that p.
(DKii) and (DKi2) can be regarded as equivalent. The second clause of
the classical definition, the requirement of justification, has been for
mulated in many different ways. Rynin has replaced (DK^i) by11
(DKii2) a can give adequate evidence that p9
and in Theory of Knowledge, Chisholm has used the condition12
(DKii3) p is evident for a.
Keith Lehrer has stated this condition in the form13
(DKii4) a is completely justified in believing that p.
There are differences of meaning between the conditions (DK^i^DKi^), but the common function of these conditions is to distinguish knowledge from true opinion.
A. J. Ayer's definition of knowledge is slightly different from the
definitions discussed above. According to Ayer,14
(DK2) a knows that p if and only if
(i) a is sure that p,
(ii) a has the right to be sure that p, and
(iii) p is true.
It seems natural to assume that a has the right to be sure that/7 if and only if a is completely justified in believing that p. Thus (DK2ii) can be taken
as equivalent to (DKii4), and if 'a is completely justified in believing that
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 111
/?' implies 'a has adequate (complete) evidence that /?' (or '/? is evident'),
(DK2ii) implies (DKxii) (or (DKii3)). (DK2i) implies ?DK^) (or (DKi2)): if a is sure that/?, a accepts p (or believes that/?), of course. The converse
implication does not seem to hold: (DKxi) does not assert anything about
the certainty of #'s beliefs. Many philosophers have criticized Ayer's condition of certainty, however, and argued that the requirement of
certainty included in (DK2ii) (or, e.g., (DKii4)) is sufficient.15 If this
view is accepted and Ayer's definition is modified accordingly, it comes
close to other formulations of the classical definition.
III. KK-THESIS AND THE CLASSICAL DEFINITION
If 6a believes that/?' (or 6a accepts/?') is expressed by 6Bap' and the justifi cation requirement of the classical definition by 6Eap\ (DKX) may be
expressed in short as
(DK3) Kap if and only if
(i) BaP, (ii) Eap,
and
(iii) p is true.
Here I shall not choose any specific interpretation for <,Eap> ; it can be
taken to represent any suitable formulation of the justification require
ment, e.g., one of the conditions (CK^iM^CKi^), or some other similar
condition. Various interpretations of 6Ea' will be discussed later on. In
the sequel, I shall discuss only the knowledge of one single person a; thus
the subscript 'a' may be omitted, and (DK3) can be expressed in the sim
plified form16
(DK4) Kp =
Bp8cEp&p.
According to (DK4),
(1) KKp = K(Bp &Ep &p).
In discussion of the logic of knowledge, it is normally assumed that the
operator K distributes over conjunctions, that is,17
(CK&) K{p&q) = Kp&Kq.
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
112 RISTO HILPINEN
(1) and (CK&) imply
(2) KKp =
KBp &KEp &Kp9
which is equivalent to
(3) KKp = BBp &EBp &Bp &BEp &EEp &Ep&p.
Thus the KK-thesis assumes (after obvious simplifications) the form
(4) Bp &Ep&p=> BBp &EBp &BEp &EEp.
It seems plausible to assume that (4) is not valid, unless the consequent is
implied by Bp&Ep; thus the validity of (4) depends upon the validity of
(5) Bp&Ep 3 BBp &EBp ScBEp &EEp.
The validity of (5) is certainly not obvious. Below, I shall consider it by
breaking (5) to simpler implications, of which the conjunction implies
(5). If these simpler implications are acceptable, (5) can also be accepted as valid, and counter-examples to these more elementary implications will be taken as evidence that (5) is not valid.
Let us assume that a believes that /? and has complete (or adequate) evidence that /? (in other words, a is completely justified in believing that
/?).18 Is a also justified in claiming that he knows that/?, that is, is
(6) Bp&Ep =>EKp
valid? It is not implausible to answer in the affirmative, if a believes that/? and has complete evidence that/?, a knows that/?, if/? is in fact the case.
Thus a's evidence that p is ipso facto evidence that he knows that /?.
According to (DK4), (6) is equivalent to
(7) Bp&Ep => E{Bp &Ep &/?).
If it is assumed that the operator 'jE" is distributive with respect to con
junctions, that is,19
(CE&) E(p&q) = Ep &Eq9
(7) is equivalent to
(8) Bp&Ep id EBp&EEp.
(CE&) is a very plausible assumption, and is accepted by most students of
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 113
epistemic logic.20 The validity of (8) may be regarded as a consequence of
two principles,
(9) Bp z> EBp and
(10) EpzDEEp.
The first principle states that, whenever a believes that /?, he is also
completely justified in believing that he believes that /?. This seems fairly
plausible from the intuitive point of view; if a in fact believes that /?, he
need not collect any additional information in order to be in a position to state with full assurance that he believes that p.21 The second principle
expresses an important fact about the methodological role of 'adequate' or 'complete' evidence. Complete evidence must, indeed, be 'complete' in
the sense that it actually terminates the inquiry concerning /?. If a has
complete evidence to justify his belief that /?, no further inquiry (that is, no additional factual information) is needed to ensure that the evidence
is, indeed, complete. This is precisely what is expressed by (10).22 It also
seems plausible to strengthen (10) to an equivalence, i.e.
(11) Ep = EEp.
(11) is especially natural, if the adequacy or completeness of evidence is
thought of as being dependent upon criteria which are similar to logical or conventional criteria.23 If a has complete evidence that /?, then surely he possesses all the information he needs to determine whether he has
complete evidence or not.24 Furthermore, if a concludes that he has
adequate evidence, but this is not the case, a has made an error which is
(from the methodological viewpoint) similar to a logical error, and errors
of this type are presumably never justified. Thus (11) is valid.
On the basis of the argument given above, the validity of (8) seems very
plausible. If both (8) and
(12) Bp&Ep^BBp&BEp
are valid, (5) (and consequently the KK-thesis) is also valid. The simplest
way of studying (12) is to examine the implications
(13) Bp^BBp and
(14) Bp&Ep ^BEp.
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
114 RISTO HILPINEN
The conjunction of (13) and (14) implies (12). (13) is the doxastic counter
part of the KK-thesis, that is, the principle that believing implies be
lieving that one believes. This is not obvious, but not entirely implausible
either; perhaps (13) can be tentatively accepted.25 Thus the validity of (5)
depends upon (14). Under the most natural interpretation of 'Ep\ (14) seems definitely unacceptable. (14) states that ii a believes that/? and has
complete evidence that/?, a believes that he has complete evidence that/?.
Suppose that 'a has complete (or adequate) evidence that/?' means that a
possesses information which in fact is complete evidence that /?. Of
course, a may believe that /? and possess information which is adequate evidence that/?, without believing that he has adequate evidence; a may fail to recognize his evidence for what it is worth. In such a case the
antecedent of (14) is true, but the consequent false. This counter-example to (14) is also a counter-example to the present version of the KK-thesis.
This failure of the KK-thesis can also be explained as follows : Ac
cording to (DK4), the KK-thesis implies
(15) Kp^EKp&BKp,
that is,
(16) (Kp z> EKp)&(Kp ZD BKp).
In (16), the first conjunct is valid, but the second is not: a may know
something without believing that he knows, i.e., without being aware of
his knowledge, if he fails to recognize the adequacy of his evidence.26
IV. FIRST CRITICISM OF (DK^
The previous argument against 6Kp id KKp9 is based upon a case in which
a believes that/? and has complete evidence that/?, but does not recognize the adequacy of his evidence, and thus fails to believe that he knows that
/?. However, examples of this kind have also been taken as counter
examples to definition (DKJ. For instance, Keith Lehrer has expressed the justification requirement in the form (DKii4), and argued that
(DKxii) does not imply (DKii4): "If a person has evidence to completely
justify his belief he may still fail to be completely justified in believing what he does because his belief is not based on that evidence. What I
mean by saying that a person's belief is not based on certain evidence is
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 115
that he would not appeal to that evidence to justify his belief." Further
more, "even if a person has evidence adequate to completely justify his
belief and his belief is based on that evidence, he may still fail to be
completely justified in believing what he does. For he may be unable to
provide any plausible line of reasoning to show how one could reach the
conclusion he believes from the evidence that he has."27 According to
Lehrer, a knows that /? only if his belief that /? is actually based upon
adequate evidence in the sense that he would appeal to this evidence, if
he were demanded to justify his belief; a must recognize his evidence for
what it is worth. It is not obvious how this requirement should be for
malized in the notation employed here. However, it is natural to assume
that a would not appeal to his evidence, if he did not believe he has
evidence adequate to justify his belief that /?; thus it may be taken that
Lehrer's analysis of knowledge implies28
(17) Kp^BEp,
where 'is/?' is interpreted as (DKxii). An obvious way of fulfilling this
requirement is, of course, simply to add the consequent of (17) to the
conditions defining knowledge, and replace (DK4) by
(DK5) Kp =
Bp&Ep &BEp &p.
According to (DK5),
(18) KKp-KBp&KEp&KBEp&Kp,
which implies
(19) KKp = Kp &BBp &EBp &BEBp &EEp &BEEp &BBEp &EBEp&BEBEp.
Thus the KK-thesis assumes the form
(20) Bp &Ep&BEp &p z> BBp &EBp &BEBp &EEp &
BEEp & BBEp & EBEp & BEBEp.
By virtue of (9), (10) and (13), the antecedent of (20) implies
(21) BBp &EBp &EEp & BBEp & EBEp ;
thus (20) is valid, if its antecedent implies
(22) BEEp &BEBp &BEBEp
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
116 RISTO HILPINEN
as well. This can be proved, if we accept the following principle:
(CBimp) If /? 3 q is valid (logically true), Bp => Bq is valid.
By virtue of this principle, (10) implies
(23) BEp^BEEp;
and (9) implies
(24) BBp^BEBp,
and thus by (13),
(25) BpziBEBp.
By substituting 'jE/?' for '/?' in (25), we obtain
(26) BEp^BEBEp.
(22) can now be proved from (23), (25), (26) and the antecedent of (20) by
propositional logic. Thus (20) (i.e., the present form of the KK-thesis) is
valid.29
Certain qualifications are necessary here. (DK5) is probably not an
accurate formalization of Lehrer's notion of knowledge; thus the argu ment presented above is a persuasive rather than a strict proof of the
Lehrer-type variant of the KK-thesis. Nevertheless, it seems to me that
(DK5) catches certain important aspects of Lehrer's conditions on
knowledge. In fact, Lehrer has presented in another paper, 'Belief and
Knowledge',30 an argument which seems to support the KK-thesis. Some
philosophers have argued that knowledge that /? does not imply belief
that/?. Colin Radford has based this claim upon a case in which a person
gives correct answers to certain questions (and thus, according to
Radford, knows the correct answers), but does not believe that the
answers are correct. In such a case, Radford concludes, the person knows
something he does not believe.31 Lehrer's counter-argument is as fol
lows:32
(27a) If a does not believe that /?, a does not believe that he
knows that /?.
(27b) If a does not believe that he knows that /?, then, even
though a correctly says that/? and knows that he has said
that /?, a does not know that he correctly says that p.
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 117
(27c) If, even though a correctly says that/? and knows he has
said that/?, a does not know that he correctly says that/?, then a does not know that /?.
(27d) If a does not believe that /?, a does not know that /?.
This argument seems to have the form
(28a) ~Bpzi~BKp.
(28b) ~BKp=>s.
(28c) s=>~Kp.
(28d) ~Bpz>~Kp.
Conclusion (28d) follows from (28a)-(28c) by propositional logic, and is
equivalent to 6Kp^>Bp\ i.e. the conclusion Lehrer wants to prove. The
conjunction of (28b) and (28c) implies
(29) Kp^BKp.
This is precisely what is denied in the counter-example to the KK-thesis
presented in Section III. However, (29) follows from (DK5) (by (13) and
the doxastic counterpart of (CK&)). (DK5) is in this respect at least in
accord with Lehrer's intuitions.33
V. KK-THESIS AND THE LOGIC OF KNOWLEDGE
All definitions of knowledge discussed above imply
(CK) KpzDp.
In addition, we have assumed that the operator K distributes over con
junctions ((CK&)). The following principles seem plausible also :
(CKeq) If /? = q is valid (logically true), Kp = Kq is valid.
If/? and q are logically equivalent, they have the same truth-conditions
and the same (descriptive) content; thus knowing that /? is just knowing that q.
(CKlog) If/? is valid (logically true), Kp is valid.
A logical truth has no content; it asserts nothing. Knowing nothing but
logical truths represents a minimal amount of (factual) knowledge. Whatever you know, you know that /?, if/? is a truth of logic.34
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
118 RISTO HILPINEN
The principles (CK), (CK&), (CKeq) and (CKlog) define a system of
epistemic logic which corresponds to von Wright's modal system M. If
'K' is replaced by W, (CK), (CK&), (CKeq) and (CKlog) become
logical truths or valid rules of inference of M. If the KK-thesis is sub
joined to these axioms, we get the epistemic counterpart of S4. Among the consequences of (CK), (CK&), (CKeq) and (CKlog) are, e.g., the
principles
(CK 3) K(/? => q) => (Kp 3 KqY* and
(CKimp) If /? z> q is valid, K/? zd Kq is valid.
It has been claimed that epistemic logics of this type are 'unrealistic'. For
instance, (CKimp) states that a knows all the logical consequences of
whatever he knows. This is unrealistic, it has been argued; in the ordinary sense of 'know', people do not know everything implied by their know
ledge. Even (CK&) has been criticized on these grounds : "On reflection,
[(CK&)], though no doubt in general true, begins to look less like a
logical truth. For may not a particularly stupid person know that p&q and yet not realize that/? is a consequence of what he knows?"36 In the
same vein, it has been suggested that knowing that /? and knowing that q do not imply knowing that p and q, for a very stupid person may fail to
"put two and two together".
However, if we attempt to build our epistemic logic along the lines
suggested by the above criticism, the result is rather meager - our (propo
sitional) epistemic logic boils down to (CK); no other principles seem to
be immune to counter-examples from ordinary language.37 This ap
proach to the logic of knowledge does not appear very interesting. A more interesting approach is to take the principles of epistemic logic
as conditions of a certain type of 'rationality'. According to Hintikka's
interpretation of epistemic logic, statements which can be proved to be
false by means of epistemic principles, are not necessarily inconsistent in
the sense that their truth is a logical impossibility, but 'indefensible'.
Indefensible statements "depend for their truth on somebody's failure
(past, present, or future) to follow the implications of what he knows far
enough." 38
According to this interpretation, defensibility is "immunity to
certain kinds of criticism"39, viz. criticism of 'epistemic negligence'.40 Thus Hintikka restores the epistemic interpretation of modal logic by
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 119
reinterpreting the metalogical notions of consistency and inconsistency as defensibility and indefensibility. Statements whose negations are
indefensible are termed 'self-sustaining' ; thus the notion of self-suste nance corresponds to the customary metalogical concept of validity.
Alternatively, we may preserve the customary interpretation of our
metalogical notions and interpret 6K* in an unconventional way. Such
an interpretation is suggested by the observation that the alleged pe culiarities of the logical behavior of 'know' (e.g., the possible failure of
(CKimp)) seem to depend upon the condition of belief included in
definitions of knowledge. For instance, suppose that a knows that /?, and
/? implies q. According to (DKJ, a has complete evidence that /?, /? is
true, and consequently q is true as well. Furthermore, it is natural to
assume that a has complete evidence that q, if a has complete evidence
that/? and/? implies q; thus a has complete evidence that q. Nevertheless,
according to the criticisms mentioned above, a may fail to know that q. Given the definition (DKJ (or some variant thereof), this is, of course,
possible only if a does not believe that q, that is, if a's body of beliefs is
indefensible. And this appears perfectly possible; most philosophers
accept the possibility of people's holding wildly inconsistent, let alone
indefensible, bodies of belief.41 Perhaps the epistemic interpretation of
the system M (and possibly that of S4) can be 'saved', if we drop the
condition of belief from the definition of knowledge, and replace (DK4)
by
(DK6) Kp =
Ep&p.
In fact, it is easy to see that the concept of knowledge defined by (DK6) satisfies (CK), (CK&), (CKeq), and (CKlog), if the notion of complete evidence satisfies (CE&),
(CE) Ep => ~ E~p,
(CEeq) If /? = q is valid, Ep = Eq is valid,
and
(CElog) If/? is valid, Ep is valid.
These conditions seem to me entirely uncontroversial.42 (DK6) can be
said to define 'a is in a position to know that/?' or simply '/? is knowable to
a\ According to this interpretation, 'Kp' means that a knows that/?, if a
believes that p, recognizes the adequacy of his evidence, etc. This notion
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
120 RISTO HILPINEN
of knowledge is not without interest. In fact, so far as philosophers are
interested, not in the actual 'belief-performance' of persons, but in the
conditions under which people are justified in making knowledge-claims,
they are likely to be more interested in this notion of knowledge than in
the concept defined by (DK4) (or (DK5)).43 According to (DK6),
(30) KKp = EEp &Ep&p;
thus the KK-thesis is equivalent to
(31) p&EpzDEEp,
and follows from (10). The present form of the KK-thesis is obviously
valid, if 'complete evidence' is interpreted in the way described in Sec
tion III.44
In the arguments presented in Sections III and IV, the notions of
validity and consistency may be interpreted as self-sustenance and
defensibility. Thus, if the condition of belief is included in the definition
of knowledge and epistemic logic is interpreted in the way suggested by
Hintikka, the acceptability of the KK-thesis is very sensitive to shifts in
the formulation of the classical definition. However, there is another
sense of 'defensibility' in which the negation of the KK-thesis is clearly indefensible. I shall term this notion 'indefensibility in the generalized sense' or simply 'g-indefensibility'. According to Hintikka's doxastic
logic,
(32) If/? logically implies q, Bp&~ Bq is indefensible.
Suppose now that a believes that/?, and/? is complete evidence that q. p is
complete evidence that q, if q is a logical consequence ojf /? or p gives
'complete' (or adequate) nondemonstrative support to q. In the former
case Bp&~Bq is indefensible (in the customary sense), in the latter case
it is not. However, there is something 'wrong' with a person who refuses
to believe what has been proved 'beyond reasonable doubt'. I shall
express this by calling the belief sets of such persons 'g-indefensible'.
Indefensibility in the customary sense may be termed 'strict indefensi
bility' or 'logical indefensibility'. Thus,
(33) If/? is complete evidence that q,Bp&~ Bq is g-indefensi ble.45
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 121
g-defensibility implies strict defensibility, but the converse does not hold:
for instance, the belief sets of extreme sceptics may be strictly defensible, but g-indefensible.
If a has complete evidence that /?, he presumably knows (and thus
believes) something which gives complete evidence that/?. Thus, according to (33),
(34) Ep&~Bp is g-indefensible.
In other words, if a's beliefs are g-defensible,
(35) Ep => Bp
is true, and (DK4) is equivalent to (DK6). Again the KK-thesis follows
from (10). If knowledge is defined by (DK4), the negation of the KK
thesis is g-indefensible, but not strictly indefensible. In other words, the
KK-thesis is, if not 'self-sustaining', at least 'g-self-sustaining'. If the
completeness of evidence is thought of as being defined by inductive
rules of inference, the KK-thesis is true in a world populated by people who are not only 'deductively omniscient', but also 'inductively om
niscient'.46
VI. SECOND CRITICISM OF (DKj
Above, it was suggested that 'a has adequate evidence that/?' means that a possesses information which gives adequate or complete support to /?. In other words, if a has complete evidence that /?, there is an evidential
statement r such that r gives complete support to /?. Of course, a must
know that r, otherwise a could not be said to 'possess' the information.
Thus we have tacitly assumed the following definition of 'Ep' :
(DEX) Ep if and only if there is an evidential statement r such
that Kr and Crp,
where 'Crp' is short for V is complete evidence that/?'. This analysis of 'Ep' makes the definition (DK4) circular. (DEX) is also
circular: 'E' is defined in terms of 'K' and 'K' is defined in terms of 'E\
The classical definition of knowledge implies an infinite regress of
justification: knowledge is defined as true belief justified by knowledge, which again is supported by further knowledge, etc.47
The standard method of escaping this regress is to postulate the
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
122 RISTO HILPINEN
existence of 'basic knowledge'. Basic knowledge consists of beliefs which
are true and directly evident, and need not be supported by further
knowledge. The regress of justification stops at such directly evident or
'initially credible' beliefs.48
According to this classical view, all our knowledge is ultimately
justified by basic beliefs; thus it is assumed that the relation of complete
support is transitive :49
(CCtrans) Crs => (Csp zd Crp).
'Ep' can now be defined by
(DE2) Ep if and only if (i) Dp, or
(ii) there is a statement r such that
Dr & Crp,
where 'Dr' stands for V is directly evident (to a)'. If (DE2i) or (DE2ii) is
satisfied, a is completely justified in believing that /?. However, if 'Ep' is
defined by (DE2), definitions (DK4)-(DK6) are clearly inadequate
(though they might serve as definitions of basic knowledge, provided that
'E' is replaced by '/)'). It is normally assumed that Ep does not imply p ; one may be completely justified in believing a statement which is in fact
false. Let us assume that r is directly evident, but false,50 and /? is ade
quately supported by r. If/? is true and a believes that/?, then, according to (DK4) and (DE2), a knows that /?. Of course, in this case a does not
know that /?: 'knowledge' derived from false premisses is not really
knowledge. This difficulty does not arise, if 'Ep' is defined by the regress
entailing definition (DE^, as this definition implies that the evidential
statement r is true.
There are many different ways of escaping this difficulty. Above, it was
suggested that a knows that/? on the basis of r, if, apart form the obvious
requirements that /? be true and believed, /? is supported by r, and r is
basic knowledge (i.e., true, believed (by a), and directly evident (to a)). Thus we can define knowledge as a conditional concept, 'a knows that /? on the basis of r', briefly 'Krp', as follows:
(DK7) Krp = Br&Dr&r& Crp &Bp&p.
If V is substituted for '/?', the right-hand side of (DK7) follows from
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 123
Br&Dr&r; thus r itself is known on the basis of r, if it is basic knowledge.
(DK7) covers basic as well as non-basic knowledge.51 This definition can again be criticized on the same grounds as (DK4).
It may be argued that a knows that/? only if a's belief that/? is based upon
r, a can show that r supports /?, etc. This criticism will not be considered
here in any detail. It is not even relevant, if our main interest is in the
notion corresponding to (DK6), that is, 'a is in a position to know that /?
(on the basis of r)'. This concept may now be defined by
(DK8) Krp = Dr&r& Crp &p.
According to (DK8),
(36) KrKrp = Dr &r&CrKrp&Krp,
that is,
(37) KrKrp =
Dr&r&Cr(Dr&r& Crp &/?) & Crp &p.
Again it may be assumed that complete support is conjunctive, that is,
(CC&) Cr(/? &q) = Crp & Crq ;
thus (37) is equivalent to
(38) KrKrp =
Krp & Crr & CrDr & CrCrp.
According to (38), the KK-thesis is equivalent to
(39) Dr&r&Crp&rz>Crr& CrDr & CrCrp.
The first conjunct of the consequent of (39) may be regarded as analytic
(or valid); thus (39) is implied by the conjunction of
(40) Dr ZD CrDr
and
(41) Crp 3 CrCrp.
(40) and (41) can be defended by arguments similar to that presented above in support of (10) and (11). (40) states that, if r is directly evident, no further information (i.e., information in excess ofthat provided by r)
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
124 RISTO HILPINEN
is needed to justify that r is directly evident. This is plausible enough, and
in accord with the traditional view that directly evident statements are of
a special type, e.g., statements about the 'contents of experience' or
statements which contain only 'observational predicates'.52 Recognizing the evidential status of the facts expressed by such statements does not
presuppose any information about the world - except these facts them
selves. (41) can be defended on the basis of similar considerations. Thus
the present form of the KK-thesis may be accepted as valid.53
If knowledge is defined by (DK7), the KK-thesis can be refuted in the
same way as (5) was refuted in Section III. This counter-example consists
in pointing out that, even though p is adequately supported by r, a may fail to believe that /? is so supported, i.e., he may be unaware of Crp (or
Dr). However, if (DK7) is modified in accordance with Lehrer's criticism
of (DKO, i.e., by supplementing the definiens of (DK7) by 'BDr' and
'BCrp', the KK-thesis follows from (40), (41), (13), (CB&), (CBimp), and
(42) BpzDCrBp,
as the reader may verify. (42) is very plausible, if it is assumed that Bp is
directly evident (for a), and thus does not need any justification. It may thus be concluded that the modification of the classical defi
nition discussed above does not affect the acceptability of the KK-thesis.
VII. GETTIER'S COUNTER-EXAMPLE TO THE CLASSICAL
DEFINITION OF KNOWLEDGE
In a recent paper, Edmund Gettier presented an interesting counter
example to the classical definition of knowledge.54 Gettier's example is
relevant to all formulations of the classical definition discussed above; here I shall formalize it as a counter-example to (DK8).
Let us assume that
(43) Er&r&Crq,
and that
(44) /? is logically equivalent to q v s.
If the logic of knowledge is at least as strong as that of M, we have to
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 125
impose the following condition upon the notion of complete support:
(CCimp) If /? => q is logically true, Crp zd Crq.
This condition seems entirely uncontroversial. (43), (44) and (CCimp)
imply
(45) Crp.
Furthermore, if
(46) s is true and q is false,
/? is true. According to (43), (45), (46), and (DK8), a knows that/? on the
basis of r (or strictly speaking, a is in a position to know that /?). How
ever, in this case it is strongly counter-intuitive to say that a knows that/?, for a's claim to know that/? can be attacked (and presumably refuted) by
pointing out that the justification of/? involves the acceptance of a false
proposition, viz. q. As was mentioned above, inferential knowledge cannot be derived false premisses, and Gettier's example shows that the
intermediate steps through which p is derived from r must not include
false propositions either. In view of Gettier's example, Keith Lehrer and
Thomas Paxson have supplemented the classical definition by the re
quirement that the justification of /? by r be 'undefeated' (though the
expression 'undefeasible' may be more appropriate).55 In the case con
sidered above, a's claim to know that /? can be defeated by pointing out
that q is false. Thus the expression 'Crp' included in the definiens of
(DK8) must be reinterpreted as representing an 'undefeasible' justifi cation of/? by r. However, if Crp is undefeasible, there is no reason why
CrCrp should be defeasible, and thus (40) and (41) seem acceptable on
this interpretation, too. Hence Gettier's example does not seem very
interesting from the point of view of the KK-thesis.
VIII. CONCLUDING REMARKS
Finally, I wish to discuss certain criticisms of the KK-thesis which, so it
seems to me, are based upon misunderstanding the aims of epistemic
logic in general and the KK-thesis in particular. Two such criticisms are
presented in a recent article by David Rynin.56 First, Rynin argues as
follows: "If one assumes that if one knows that /? then one must know
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
126 RISTO HILPINEN
that one knows that /?, or be sure that /?, and if it be granted that one
could be mistaken unless one knows that one knows, it would seem to me
that there could be no knowledge, for in my view one could always be
mistaken, i.e., one could never be sure in the intended sense." According to the KK-thesis, the truth-conditions of 'Kp' and 'KKp' are identical.
Thus the KK-thesis implies that, whatever uncertainty there is included
in a's claim to know that p, the same uncertainty is involved in a's claim
that he knows that he knows that /?. If a could be mistaken in his justified claim that /? (in the sense intended by Rynin), though he knows that /?, he could also be mistaken in the claim that Kp, though he knows that he
knows that /?. Rynin's argument rests on a fallacy of equivocation.57
Secondly, Rynin argues that "if it is demonstrable that if one knows then
it follows that one knows that one knows it would seem to me very difficult if not impossible ever to know, for it seems clear to me that
this infinite process soon takes us beyond comprehension." Obviously
enough, no 'infinite processes' are involved here. If the truth-conditions
of 'Kp' and 'KKp' are identical, the latter statement yields no information
in excess of that provided by the former. According to the KK-thesis,
'Kp', and, say, 'KKKKp' are just two different ways of expressing the
same thing in writing (as far as the logic of knowledge is concerned),
though the former expression is usually preferable for obvious reasons.
The difficulty here is on the side of those philosophers rejecting the
KK-thesis, for they have the task of inventing for the latter expression a
meaning which differs from that of the former.58
Arthur C. Danto has presented, against the KK-thesis, an argument of
a different type.59 According to Danto, knowledge implies understanding: a knows that /? implies that a understands /?. It is possible that a knows
that a knows that /?, and consequently understands /?, without under
standing Kp. Thus a may know without knowing that he knows. By
'understanding /?' Danto seems to mean understanding the sentence '/?' ;
thus his argument presupposes that 'K' is treated as a predicate of sen
tences.60 Danto's argument implies that 'Kp' is a sentence in the syntax
language, and epistemic terms are prefixed, not to sentences, but to
names of sentences (i.e., '/?' here is the name of some sentence in a's
object language). However, if epistemic terms are understood in this way, all plausible epistemic interpretations of modal logic become incon
sistent.61 This difficulty does not arise, if 'K' is regarded as a sentential
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 127
operator. Thus there are good reasons for not interpreting 'K' in the way Danto does. If this interpretation is rejected, Danto's argument loses its
power.
Department of Philosophy,
University of Helsinki
REFERENCES
* This work has been supported by a grant form the Finnish Cultural Foundation
(Suomen Kulttuurirahasto). 1 An Essay in Modal Logic, North-Holland Publ. Co., Amsterdam, 1951, pp. 29-35.
Cf. also G. H. von Wright, 'Deontic Logic', Mind, N.S. 60 (1951) 1-15. 2 E. J. Lemmon, 'Is There Only One Correct System of Modal Logic?', Proceedings of the Aristotelian Society, Suppl. Vol. 23 (1959), pp. 39-40. 3
Knowledge and Belief : An Introduction to the Logic of the Two Notions, Cornell Uni v.
Press, Ithaca, N.Y., 1962. Cf. also Jaakko Hintikka, 'The Modes of Modality', in
Proceedings of a Colloquium on Modal and Many-Valued Logics (Acta Philosophica Fennica 16), Helsinki 1963, pp. 65-81 ; especially p. 78. 4 This axiom is 'distinctive' in the sense that we obtain S4, if it is added to the axioms
of M. 5
See, e.g., Roderick M. Chisholm, 'The Logic of Knowing', Journal of Philosophy 60
(1963) 773-95; Arthur Danto, 'On Knowing That We Know', in Epistemology: New
Essays in the Theory of Knowledge (ed. by A. Stroll) (Sources in Contemporary
Philosophy; F. A. Tillman, ed.), Harper & Row, New York, 1967, pp. 32-53, and
Analytical Philosophy of Knowledge, Cambridge Univ. Press, Cambridge, 1968, es
pecially pp. 147-58; Jaakko Hintikka, 'Epistemic Logic and the Methods of Philo
sophical Analysis', Australasian Journal of Philosophy 46 (1968) 37-51 ; E. J. Lemmon, 'If I Know, Do I Know That I Know?', in Epistemology (ed. by A. Stroll), pp. 54-82;
Charles Pailthorp, 'Hintikka and Knowing that One Knows', Journal of Philosophy 64
(1967) 487-500; David Rynin, 'Knowledge, Sensation, and Certainty', in Epistemology
(ed. by A. Stroll), pp. 10-31, especially pp. 29-30. 6 Cornell Univ. Press, Ithaca, N.Y., 1957, p. 16. Later Chisholm has rejected this
definition; cf. his Theory of Knowledge, Prentice-Hall, Englewood Cliffs, N.J., 1966,
pp. 6-7, and the discussion below. 7
Theaetetus, 201 c-d; Meno 98. In this classical approach to epistemology, the funda
mental question is : What must be added to true opinion to yield knowlegde? 8 See Knowledge and Belief, especially pp. 20-1, 50. 9 See Roderick M. Chisholm, Theory of Knowledge, p. 23 ; David Rynin, 'Knowledge,
Sensation, and Certainty', p. 10; A. J. Ayer, The Problem of Knowledge, Penguin
Books, Edinburgh 1956, p. 35, and 'Knowledge, Belief, and Evidence', in Danish
Yearbook of Philosophy, Vol. I, Munksgaard, Copenhagen, 1964, especially pp. 18-9; Richard B. Braithwaite, 'The Nature of Believing', Proceedings of the Aristotelian
Society 33 (1932-33) 129-46, especially pp. 129-30. 10 David Rynin, op. cit., p. 10. 11
Rynin, loc. cit. 12 See p. 23. 13
'Knowledge, Truth, and Evidence', Analysis 25 (1965) 168-75, see p. 168.
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
128 RISTO HILPINEN
14 The Problem of Knowledge, p. 35. 15
See, e.g., E. J. Lemmon, 'If I Know, Do I Know That I Know?', pp. 70-1 ; cf. also
A. D. Woozley, 'Knowing and Not Knowing', Proceedings of the Aristotelian Society 53
(1952-3) 151-72. 16 In what follows, I shall adopt the customary conventions regarding the combining force of various logical connectives, that is,
~ is the strongest connective, & is stronger than V, V than ==>, and => is stronger than =. 17 If the logic of knowledge is assumed to be similar to the system M (or S4), (CK&) is logically true. 18
Here, the expressions '? has complete (or adequate) evidence that /?' and 'a is com
pletely justified in believing that/?' are used interchangeably. It may be argued, however, that they are not synonymous; cf. Section IV. 19 For an intuitive justification of this principle, see, e.g., Carl G. Hempel, 'Deductive
Nomological versus Statistical Explanation', in Scientific Explanantion, Space and
Time (Minnesota Studies in the Philosophy of Science, Vol. 3) (ed. by H. Feigl and G.
Maxwell), Univ. of Minnesota Press, Minneapolis, 1962, especially pp. 150-1 ((CE&) is implied by Hempel's condition (CR1)), and Risto Hilpinen, Rules of Acceptance and
Inductive Logic (Acta Philosophica Fennica 22), North-Holland Publ. Co., Amsterdam,
1968, pp. 37-9, 44-6. 20
Henry E. Kyburg, Jr., is an exception. See, e.g., his 'Probability, Rationality, and a
Rule of Detachment', m Proceedings of the 1964 Congress for Logic, Methodology, and
Philosophy of Science (ed. by Y. Bar-Hillel), North-Holland Publ. Co., Amsterdam,
1965, pp. 301-10. Kyburg's rejection of (CE&) is motivated by the fact that no purely
probabilistic definition of 'complete evidence' satisfies (CE&). For a discussion of this
issue, see Jaakko Hintikka and Risto Hilpinen, 'Knowledge, Acceptance and Inductive
Logic', in Aspects of Inductive Logic (ed. by. J. Hintikka and P. Suppes), North
Holland Publ. Co., Amsterdam, 1966, pp. 1-20, and Risto Hilpinen, Rules of Ac
ceptance and Inductive Logic. 21 This view is accepted by Chisholm. According to Chisholm, believing is a 'self
presenting' state: "If I do believe that Socrates is mortal, then, ipso facto, it is evident
to me that I believe that Socrates is mortal" (Theory of Knowledge, p. 28). This is
precisely what (9) expresses. In this context, the expression 'evident' is, of course, more
natural than 'has adequate evidence'. For other proponents of this view, see Chisholm,
op. cit., p. 28n. 22 This view of the methodological role of complete (or conclusive) evidence is defended
and discussed in detail by Douglas Arner in 'On Knowing', Philosophical Review 68
(1959) 84-92. Cf. also Jaakko Hintikka, 'Epistemic Logic and the Methods of Philo
sophical Analysis', pp. 48-9. 23
According to Arner, "what counts as conclusive evidence is a matter of tacit,
continuing agreement among the users of language", (op. cit., p. 87). Cf. also S?ren
Halld?n, 'A Pragmatic Approach to Model Theory' in Proceedings of a Colloquium on
Modal and Many-Valued Logics (Acta Philosophica Fennica 16), Helsinki 1963, pp. 56-8.
Halld?n's arguments concern the notion of logical truth, but they can also be applied to the notion of complete evidence. 24 'Information' here refers to factual information, i.e., information about the world. 25 If (13) is accepted, can it be strengthened to an equivalence, i.e., is
(i) BBp =5 Bp
also acceptable? According to Nicholas Rescher, (i) is ,"from a strictly intuitive point of
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 129
view, a feasible candidate for acceptance by a logical theory of belief" ('The Logic of
Belief Statements', in Topics in Philosophical Logic (by N. Rescher), D. Reidel Publ.
Co., Dordrecht, 1968, p. 48). Nevertheless, Rescher is not ready to accept (i), for it
implies, if subjoined to certain other seemingly acceptable doxastic principles,
(ii) Bp^p,
which is clearly unacceptable. However, it is possible to construct a plausible logical
theory of belief in which (i), but not (ii), is valid. In fact, such a theory is already available: If the semantical theory of deontic modalities presented by Jaakko Hintikka
in 'Deontic Logic and Its Philosophical Morals' (in Jaakko Hintikka, Models for
Modalities, Selected Essays, D. Reidel Publishing Co., Dordrecht, 1969) is reinter
preted as logic of belief, (i) is valid, but no untoward consequences (like (ii)) are forth
coming. The crucial semantical conditions are:
(iii) If BpepeQ, and if v e Q is a doxastic alternative to //, then Bp e v.
(iv) If Bp eveQ, and if v is a doxastic alternative to at least one fieQ,
thenpev.
Condition (iii) yields (13), and (iv) yields (i). For the terminology employed in (iii) and (iv), see Hintikka, Knowledge and Belief, Ch. 3. It may be pointed out, however, that in
Knowledge and Belief, Jaakko Hintikka rejects (i) (p. 123). These observations are also relevant to (11). If (iii) and (iv) are modified in such a
way that they concern adequate evidence (or justification of belief), they yield (11). In
fact, on this interpretation (iii) and (iv) seem more plausible than in the case of the
notion of belief. 26 Cf. Chisholm, Theory of Knowledge, p. 6. This failure of the KK-thesis is described
by Hintikka as a case in which "somebody is in a position to know something without
being aware of it" (Knowledge and Belief, p. 118n). 27
'Knowledge, Truth, and Evidence', p. 169. 28 This formalization is especially natural, if belief that/? is interpreted as a disposition to act as if p were true. For surely justifying one's belief by appealing to certain
evidence is to act as if the evidence were adequate to justify the belief. 29 (9), (13), and (25) imply
(i) Bp 3 EBp &BBp ScBEBp ;
thus, according to (DK5), the principles accepted here imply
(ii) Bp => KBp. In Knowledge and Belief, Jaakko Hintikka rejects (ii), although he accepts (13) and the
KK-thesis (pp. 51-3). This view can be reconciled with the present analysis of 'know'
for instance by rejecting (9), and accepting instead only the weaker principle
(iii) Bp&Ep=>EBp. A detailed discussion of the implications of (9) and (ii) cannot be attempted here.
30 Philosophical Review 76 (1968) 491-9.
31 'Knowledge
- By Examples', Analysis 27 (1966) 1-11.
32 Lehrer, op. cit., p. 498.
33 The sentence s is somewhat puzzling. It cannot be analyzed by propositional logic in such a way that Lehrer's argument would be acceptable (i.e., that all premisses
would be seemingly true and imply the conclusion), s seems to be a subjunctive con
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
130 RISTO HILPINEN
ditional 'If a should correctly say that/? and know that the has said that/?, he would not
know that he correctly says that /?'. Lehrer's explanation of the meaning of 'a knows that he correctly says that/?' (p. 498n).
cannot be correct, as it makes s inconsistent. If *Ap9 stands for 'a says that /?', then 'a
correctly says that/?', 'a knows that he says that/?' and 'a knows that he correctly says
that/?' are expressed by 'Ap &/?', 6KAp\ and 6K(Ap &/?)', respectively. 34 These arguments are plausible in so far as we are speaking of factual knowledge,
i.e., knowledge that something is the case, and they are especially natural, if the ex
pression '/?' occurring in the scope of a modal operator is taken to represent, not the
statement/?, but the descriptive content of /?, i.e., in Wittgenstein's terminology, the
sentence radical Cf. Erik Stenius, Wittgenstein s 'Tractatus', Basil Blackwell, Oxford,
1960, Ch. IX; 'The Principles of a Logic of Normative Systems', in Proceedings of a
Colloquium on Modal and Many- Valued Logics (Acta Philosophica Fennica 16), Hel
sinki 1963, pp. 247-60; 'Mood and Language-Game', Synthese 17 (1967) 304-33;
reprinted in Philosophical Logic (ed. by J. W. Davis, D. J. Hockney, and W. K. Wilson), D. Reidel Publ. Co., Dordrecht, 1969, pp. 251-71. 35 An axiom system consisting of (CK), (CK^) and (CKlog) is also sufficient for the
epistemic counterpart of M. Cf. Robert Feys, Modal Logics (ed. by J. Dopp), Gauthier
Villars, Paris, 1965, p. 124. 36 E. J. Lemmon, 'If I Know, Do I Know That I Know?', p. 77. 37 Cf. E. J. Lemmon, 'Is There Only One Correct System of Modal Logic', p. 38. 38 Jaakko Hintikka, Knowledge and Belief, p. 32. 39
Hintikka, op. cit., p. 31. 40 This expression has been used by R. M. Chisholm in 'The Logic of Knowing' ; see
p. 784. 41 In 'The Logic of Belief Statements', Nicholas Rescher has emphasized the
" 'irra
tional' or 'illogical' nature of belief" (p. 42 in Topics in Philosophical Logic). It may be
argued, however, that philosophical logicians should not be too much concerned
about this 'irrationality'; cf. Alan Ross Anderson, 'Comments on von Wright's
"Logic and Ontology of Norms" ', in Philosophical Logic (ed. by J. W. Davis et al),
p. 110. 42 Cf. Risto Hilpinen, Rules of Acceptance and Inductive Logic, pp. 38-9. 43 Cf. Hilpinen, op. cit., p. 28. 44 In 'If I Know, Do I Know That I Know?', E. J. Lemmon has presented an example
which he takes to be a counter-example to the present form of the KK-thesis (pp. 69-70).
However, Lemmon's example is concerned with much too weak senses of 'a knows'
and 'a has adequate evidence' to count as a genuine counter-example to the present
form of the KK-thesis. 45 Of course, this condition cannot be formulated in terms of knowledge, as complete evidence does not imply truth. 46 The expression 'logically omniscient' has been used by Chisholm; see 'The Logic of
Knowing', p. 781. 47 For a detailed discussion of this regress and various reactions to it, see S. G. O'Hair,
Foundations in Epistemology, Cambridge Univ. Press (forthcoming). 48 An up-to-date version of this Cartesian view has been presented by Roderick
Chisholm in Theory of Knowledge; see especially pp. 24-9. 49 The requirement of transitivity is not obvious in the case of inductive support. For
instance, Isaac Levi has proposed an inductive acceptance rule which does not satisfy this condition (See Gambling with Truth, Alfred A. Knopf, New York, 1967, pp. 151-2).
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
KNOWING THAT ONE KNOWS 131
The transitivity requirement has been defended by Risto Hilpinen in Rules of Acceptance and Inductive Logic, pp. 102-4. 50 This is not possible, if it is assumed that directly evident propositions are 'absolutely certain' or 'incorrigible'. This view of the basis of knowledge encounters many difficul
ties, however, and is not accepted here. Cf. O'Hair, op. cit., especially Ch. VII. 51 Here we assume that Crr is valid. 52
See, e.g., C. I. Lewis, An Analysis of Knowledge and Valuation, Open Court Publ.
Co., La Salle, 111., 1946, Ch. VII; Carl G. Hempel, 'Fundamentals of Concept Formation in Empirical Science', in International Encyclopedia of Unified Science, Vol. II: 7, University of Chicago Press, Chicago, 1952, pp. 20-3, and Alfred J. Ayer, The Foundations of Empirical Knowledge, Macmillan, London, 1940.
J. L. Austin has critized the view that there are basic statements in the present sense
(Sense and Sensibilia, Oxford University Press, Oxford, 1962, pp. 114-5). Austin
speaks of 'incorrigibility', but a similar criticism may be applied to the wider notion of
'directly evident'. According to Austin, no type of sentence is incorrigible as such, some
statements are incorrigible in fact under certain circumstances. Statements which are
incorrigible only in this sense cannot serve as the 'basis' of knowledge, for, in order to
be in a position to assert that, say, r is incorrigible (in Austin's sense), we have to know
that the circumstances are appropriate, and this presupposes factual information not
conveyed by r. (Cf. O'Hair, op cit. Ch. V.) But this is just another way of saying that (40) is not satisfied.
If the absolute notion of 'directly evident' is rejected, we may still give an interesting
interpretation to (D8). 'Dr' can be taken to mean that r is perfectly free from doubt, or
that no justification for r is demanded. The resulting sense of 'knowledge' is genuinely conditional or 'local' : In so far as r is taken for granted, a knows that /?. For this local
sense of knowledge and of justification, see C. S. Peirce, 'The Fixation of Belief, in
Collected Papers of Charles Peirce, Vol. 5 (ed. by C. Hartshorne and P. Weiss), Har
vard Univ. Press, Cambridge, Mass., 1932, p. 233, and Isaac Levi, Gambling with Truth, Alfred A. Knopf, New York, 1967, pp. 3-6. 53 In view of (DK8), we could now define '?/?' by
Ep=Dr&r&Crp.
Given (i), (DK?) and (DKs) are equivalent (apart from the fact that Ep and Kp must
now be taken as conditional on r). 54 'Is Justified True Belief Knowledge?', Analysis 23 (1963) 121-3, Gettier presents two examples; the formalization below corresponds to the second example. The first
example involves a definite description, and is more problematic than the
second. 55 Keith Lehrer and Thomas Paxson, Jr., 'Knowledge: Undefeated Justified True
Belief, Journal of Philosophy 66 (1969) 225-7; see especially pp. 225-7. 56 David Rynin, 'Knowledge, Sensation, and Certainty', p. 29. 57 This fallacy is natural here, for '? is certain that he knows' is one of the 'residual
meanings' of 'a knows that he knows'; cf. Hintikka, Knowledge and Belief, p. 116. 58 Cf. also Jaakko Hintikka, 'Epistemic Logic and the Methods of Philosophical
Analysis', p. 49. This difficulty does not arise in the case of the counter-example to the
KK-thesis discussed in Section II; according to (DK4) and the principles of doxastic
logic accepted here,
(i) KKp&~KKKp
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions
132 RISTO HILPINEN
is indefensible, as the reader may easily verify. Thus all instances of the schema
(ii) KKp=K...Kp
are valid (or self-sustaining). 59 'On Knowing That We Know', pp. 49-50, and Analytical Philosophy of Knowledge, pp. 147-52. 60 See especially Analytical Philosophy of Knowledge, p. 148. 61 Cf. Richard Montague, 'Syntactical Treatments of Modality, with Corollaries on
Reflexion Principles and Finite Axiomatizability', in Proceedings of a Colloquium on
Modal and Many-Valued Logics (Acta Philosophica Fennica 16), Helsinki 1963, pp.
153-67; see especially pp. 158-61. This inconsistency result follows, if the language under consideration includes arithmetic.
This content downloaded from 130.15.241.167 on Sun, 4 Aug 2013 17:04:44 PMAll use subject to JSTOR Terms and Conditions