Suppose that after 99 draws, you've drawn only green M&M's from a bag you know contains only one remaining piece but before you can greedily gobble it down, you drop the bag and it falls into the busy street. Maybe you believe that the bag contained all green M&M's on the basis of your sampling. Do you have knowledge-level justification for believing that it will be green? Will modifying the numbers make a difference? Make it 999. Or, make it 9,999.
According to Fantl and McGrath:
(JKR) If p is knowledge-level justified for you, then p is epistemically eligible to be a reason you have for believing q, for any q.
Now, it seems that if you can justifiably believe that, say, all the M&M's are green on the basis of inductive grounds, the proposition that all the pieces in that bag were green could be a reason for you to believe that the remaining piece that was unseen wasn't blue.
Q: Is the proposition that all the pieces in that bag were green part of your evidence?
It seems to me intuitively obvious that the answer to this question is 'No'. So, where does that leave us? I think (JKR) looks pretty good, but I suppose you could try to save the idea that we can have knowledge-level justification for believing p on the basis of induction by distinguishing reasons for believing and evidence for belief. The degree to which something is justified will be a function of the evidence rather than the reasons. I don't like this maneuver.
Note that (JKR) uses the phrase '... is epistemically eligible to be a reason you have for believing q'. It might be that that which is a member of the set of things that are epistemically eligible to be a reason aren't reasons, and that might preserve the link between reasons and evidence. I don't think this will work, however, for internalists who I imagine will want there to be no epistemic gap between things that are reasons and thinks that are epistemically eligible to be a reason.
Any bright ideas?