Ages ago, I wanted to write a paper called ''Might' made right'. That's not going to happen, but I'm still working on epistemic possibility. Ordinarily, I think it would be pedantic to object to the following view in the ways I'm about to, but I have my reasons. First things first. The view:
(EPk) p is epistemically possible for S iff ~p isn’t obviously entailed by something S knows.
Think about cases of inductive knowledge. It seems odd to think that you only have knowledge of future events when it is not epistemically possible that these events do not occur. Myself, I don’t doubt that our beliefs about the future constitute knowledge. I doubt that it would be correct to say that it isn’t epistemically possible that these beliefs are mistaken.
Think about conversations where sceptical hypotheses are introduced. In such contexts, it seems proper to concede that we might be mistaken in just about any belief about the external world. Now, suppose that knowledge is necessary for warranted assertion and that concessions (e.g., ‘It might be that I’m a BIV’) are really assertions. It seems that given these assumptions and (EPk), the propriety of the concession would depend upon whether the speaker knew herself to be ignorant. But, it seems harder to know that you don’t know than it is to know that it’s proper to concede that you might be mistaken. Given (EPk), to assert knowingly that you might be mistaken, you either know that you don’t believe p, that your belief about p is mistaken, that the justification you have for your belief is insufficient, or that you are in some sort of Gettier case. I doubt that you know one of these to be true whenever you know that it’s proper to concede that you might be mistaken. Thus, you either should think that concessions aren’t really assertions, deny that knowledge is the norm of assertion, or say (as I do) that in conceding that you might be mistaken you might only be conceding that you are not completely certain.
If the example of inductive knowledge shows what I think it does, then we need to revise (EPk) as follows:
(EPx) p is epistemically possible for S iff ~p isn’t obviously entailed by something S knows w/X.
Whatever we put in for 'X', it just has to be something we don't always have when we know. We could put in 'out inference' for 'X', and we get that epistemic necessity is non-inferential knowledge. That gives us the induction case, but not the perception case. We could put in 'infallible grounds for believing' and that gives us the induction case and would show that CKAs can't possibly pose a threat to fallibilism. Given my views about perceptual justification, I don't think that gets the perception cases right. There are many things we know non-inferentially that I think we have infallible grounds for, but these are things we can properly concede we might be mistaken about when skeptical hypotheses are introduced. So, why not just say something like 'certainty' and be done with it. The context determines whether someone knows with certainty because the conversational context can determine whether certain possibilities are significant and we can say that something is certain for S when S's evidence rules out all the significant possibilities where S is mistaken. Assuming that 'knows' and 'certain' doesn't sway together, this view wouldn't motivate a contextualist account of 'knows'. 'Might' might have the power to derail conversations, but it doesn't threaten your knowledge or evidence.