Cohen’s contextualism starts from a the holism of reasons. (So note that Cohen is working within epistemic notions. He is not, like Goldman and Nozick, trying to define knowledge in more basic reductionist terms.)
Edward Craig (next session) gives this example of the way reasons interact:
When we hear that Dancing Bear has won his last five races against top-class opposition, we regard it as very probable that he will win again this afternoon. If we add the evidence that a stranger was seen tipping white powder into his drinking-water, we cease to regard a win as likely. If we hear that it was only glucose we change back. Told that on his way out the stranger passed a bundle of notes to the jockey we lower the probability again, and so on. [Craig 1987: 226]
A reason for a belief can be undermined by a defeating reason (a ‘defeater’) which can itself be undermined by another reason, thus re-instating the first belief. In the face of this, what are the conditions for knowledge?
Cohen uncontentiously argues that there is more to having good enough reasons for knowledge than thinking one does, oneself. If S believes q on the basis r whilst holding d which obviously (to us) undermines r as a reason for q (and assuming no other relevant reason in play) then we will not say S knows that q. S ought to have spotted that d undermines q. If as a matter of lucky fact, d is itself but unobviously itself undermined, still S did not know that q.
What is less clear is the argument that S does not need to have epistemically ideal reasons in order to have knowledge.
One might have thought that the first argument (that there is more to having good enough reasons for knowledge than thinking one does) should lead to the conclusion that possessing any ‘undefeated defeaters’ undermines knowledge. But Cohen argues that it is correct to ascribe knowledge if someone holds a defeater (providing their belief is true, of course, so a misleading defeater) but one whose relevance it would take a genius, ie more than the usual social standard, to understand.
I think that there may be two main reasons for this.
1: ideal reasons would place the bar too high and knowledge would be impossible.
2: as a matter of fact, we do ascribe knowledge without access to super geniuses to vouch for us.
Against 2, we might argue that there is a difference between a reasonable use of the phrase ‘S knows’ and a correct use. Perhaps it is more correct to say that we thought that S knew (and S thought that she knew) when in fact she didn’t, because she held a defeating reason. Cohen’s response is to press contextualism and say that ascriptions can be correct in context in the way that ascriptions of flatness are contextually sensitive. But this is to stress the conclusion not an argument for contextualism.
Barn Façade County may help motivate this appeal. Does its mere possibility undermine normal barn recognitional knowledge? Context may play two roles. First, where is S actually situated? Second, what is the nature of our discussion of S’s knowledge? Since Barn Façade County does not exist, S is never there. And if we are not discussing it, then normally we can ascribe knowledge of barns to S. In court, that is the correct thing to say. But change the context to a philosophy discussion – eg merely by introducing Barn Façade County – and now it is not correct to say that S knows.