* The author is indebted to Gary Malinas and D. S. Mannison for comments on an earlier draft of this paper, and to the University of Queensland and the Academy of Finland for support of research.

1 In Problems in the Theory of Knowledge,ed. by Wright, G. H. von Nijhoff, Martinus The Hague 1972, pp. 12–13.Google Scholar

2 Williams, op. cit., p. 10.

3 ‘Comments on Professor Williams’ “Knowledge and Reasons“', in Problems in the Theory of Knowledge, pp. 12–13.

4 Williams, op. cit., pp. 10–11.

5 For a discussion of the concept of collective action, see section VI ('Collective Agency’) of my paper ‘On the Semantics of Personal Directives', in Semantics and Communication,ed. by Heidrich, C. North-Holland, Amsterdam 1974, pp. 162–179.Google Scholar

6 For the concept of primitive action, see Davidson, Donald ‘Agency', in Agent,Action, and Reason,ed. by Binkley, R. Bronaugh, R. and Marras, A. Basil Blackwell, Oxford 1971, pp. 3–25.Google Scholar

7 The qualification ‘by deductive inference’ is important here. The joint knowledge of a group of persons can justify inductively propositions which are not supported by the evidence available to various members of the group separately, and which are not known in the sense of (5). If the members of the group pool their evidence, they can come to know, on the basis of inductive inference, that such propositions are true. However, according to (5), these propositions are not known before someone actually draws the relevant inductive conclusions.

8 Cf. Hintikka, Jaakko Knowledge and Belief, Cornell University Press, Ithaca 1962, pp. 31–39.Google Scholar

9 In his paper ‘Knowledge, Belief and Logical Consequence’ (Ajatus 32 (1970). pp. 32–47) Jaakko Hintikka has argued that (7) holds for the concept of personal knowledge only if ‘p ﬤ q’ is what Hintikka calls a “surface tautology”. The restriction of (7) to surface tautologies can be defended as follows: It seems plausible to assume that ‘a knows that p’ implies ‘a understands p'.lf we could identify understanding p with knowing all logical consequences of p, this principle would justify (7): a person could know that p only if he knew all its logical consequences, since he could not understand p without knowing what its logical consequences are. However, this proposal involves an unrealistic and impossible standard of understanding. It is more plausible to assume that understanding p only involves knowledge of certain simple or ‘trivial' consequences of p. Hintikka's concept of surface tautology is intended to capture such trivial consequences of a proposition. Thus, by virtue of the principle that knowledge implies understanding, (7) can be regarded as valid (in the case of personal knowledge) only if ‘p ﬤ q’ is a surface tautology. The principles (5) and (7) concerning the concept of impersonal knowledge can perhaps be weakened along similar lines.

10 Cf. Barbara Hall-Partee, ‘On the Requirement that Transformations Preserve Meaning', in Studies in Linguistic Semantics,ed., by Fillmore, J. and Langendoen, D. T. Holt, Rinehard and Winston, New York 1971, pp. 1–21Google Scholar (d. pp. 14– 15).

11 Impersonal knowledge can also be interpreted in a more restricted sense which does not satisfy (4). For instance, in the paper referred to above A. J. Ayer says that “a person's reports of his present thoughts, feelings, or sensations” are not represented in the corpus of impersonal (or public) knowledge ('Comments on Professor Williams’ “Knowledge and Reasons” ‘, pp. 14–15). However, the 'private’ knowledge of various persons cannot be incompatible with public knowledge, and it should be possible, in principle, to extend the latter in such a way that it includes everyone's ‘private’ knowledge. Thus we can assume that (4) holds for impersonal knowledge in the present extended sense.

12 Cf. Williams, op. cit., p. 10, and Ayer, op. cit., p. 13.

13 Here I shall by-pass many traditional problems concerning the justificationist analysis of knowledge, for instance, the tacit circularity involved in the usual formulations of the justification condition (cf. my paper ‘Knowing That One Knows and the Classical Definition of Knowledge', Synthese 21 (1970), pp. 109– 132; especially pp. 121–123). I assume only that a person can in some cases come to know the truth of a proposition on the basis of nondemonstrative inference.

14 Inductive arguments are defeasible, and can be undermined by new evidence; cf. Hilpinen, Risto ‘Knowledge and Justification', Ajatus 33 (1971), pp. 7–39.Google Scholar

15 ‘Knowledge and Justification', pp. 25–32.

16 Knowledge and Belief, pp. 20–21.

17 American Philosophical Quarterly 5 (1968), pp. 164–173.

16 ‘Knowledge, Inference, and Explanation', p. 172. Harman has discussed similar examples in ‘Induction', in Induction, Acceptance, and Rational Belief,ed. by Swain, Marshall Reidel, D. Dordrecht 1970, pp. 63–99CrossRefGoogle Scholar, and in his recent book Thought, Princeton University Press, Princeton 1973, pp. 142–154.

19 ‘Knowledge, Inference, and Explanation', p. 172.

20 Ibid., p. 173, note 10.

21 I do not assume that this is the only way in which a prima facie justified true belief can fail to be knowledge. For instance, if an inference involves false intermediate steps or mistaken assumptions about the relevance of one's evidence, it cannot yield knowledge. Many Gettier-type counter-examples to the classical analysis of knowledge are of this kind and do not depend on the social aspect of knowledge. However, in Harman's example the inference by which Tom comes to believe that the civil rights leader has been assassinated does not seem to involve any false assumptions or presuppositions.

22 ‘Knowledge, Inference, and Explanation', p. 173.