say of any subject, S, who believes that P and who has conclusive reasons for believing that P, that, given these reasons, he could not be wrong about P or, given these reasons, it is false that he might be mistaken about P.

Suppose, then, that

(1) S knows that P and he knows this on the basis (simply) of R entails

(2) R would not be the case unless P were the case.

The latter formula expresses a connection between R and P which is strong enough, I submit, to permit us to say that if (2) is true, then R is a conclusive reason for P. For if (2) is true, we are entitled, not only to deny that, given R, not-P is the case, but also that, given R, not-P might be the case. That is to say, (2) eliminates R and not-P as a possible (joint) state of affairs and, when we are given R, it eliminates not-P as a possible state of affairs. This is so because (2) entails the falsity of,

(3) Although R is the case P might not be the case.

How can Dretske's claim that the evidence, grounds, or reasons must be "conclusive" strengthen the case for knowledge? He admits there are many examples of mistaken knowledge claims, one where the thermometer used is known to stick at readings above 98.6 degrees and another example of mistaken testimony.

And he cites Gilbert Harman's example of the lottery player with a friend that will provide the $100,00 prize in the event the player gets a losing ticket. This contrived case is similar to the highly unlikely but possible Edmund Gettier's 1963 and Keith Lehrer's sophistical examples.

Dretske approved Alvin Goldman's development of a causal account of knowledge. (This idea that causes matter goes back to Frank Ramsey.)

Dretske's work was just a few years after Willard van Orman Quine had proposed the "naturalization" of epistemology, by which he meant seeing epistemology of a part of empirical science. This was the end of searching for a priori justifications of true belief

In 1981 Dretske called for epistemology to be put on an information-science basis.

At the time when cognitive science was proposing mind models based on input and output from digital computers, Dretske likened the problem of knowledge to an analysis of the information communicated by statements or propositions (which he called "digital") and by pictures or images (which he called "analog").

In his book Knowledge and the Flow of Information, Dretske reviewed Claude Shannon's mathematical treatment of the amount of information that can be communicated over a channel between a source s and a receiver r.

He showed that alternative possibilities must exist for messages, otherwise no information is transmitted. (In a deterministic world, the total information is conserved over time.)

Information, Dretske claimed, can causally sustain belief, although he asked himself "How can an abstract commodity like information be causally efficacious".

For his answer, Dretske fell back to the standard epistemological arguments. S knows F because he has received a signal s over an information channel that F is the case. But an information-theoretic argument as to how S learns something does not advance the quality or verifiability of the presumed knowledge.

As Charles Sanders Peirce and Frank Ramsey knew, that requires a connection between the belief or knowledge and the efficacious behavior of the agent.