One of the lessons of Plantinga’s argument against evolutionary naturalism is that the mere fact that a claim is improbable on a certain piece of information doesn’t imply that the latter information is a defeater of any evidence in favor of the claim in question. In the context of his argument, this point plays out as follows. Even if the claim that our faculties are reliable is improbable given only the assumption of evolutionary naturalism doesn’t imply that these assumptions defeat any and every defense of the reliability of our cognitive faculties.
The debate about Plantinga’s argument thus turned to the interesting question of how to determine when such a conditional improbability would count as a defeater and when it wouldn’t. There’s some interesting literature on that question, but I’m more interested in the analogy on the other side: if conditional improbability doesn’t signal defeat, then conditional probability may not signal support, either.
Here’s what I mean. Consider the following detachment rule:
DR: if Pr(p) exceeds some threshold X (less than 1), then one is epistemically justified in believing p.
The idea behind DR is as follows. You learn some information, and by some complicated rule, (perhaps Bayesian conditionalization, perhaps something else), that information teaches you that the probability of p is high (meets the threshold in question but is less than 1). What should you believe? Well, for one thing, it’s surely OK to believe that p is likely to be true. But what of p itself? If DR is correct, then in such a case one can detach the probability operator and believe p as well.
I expect this proposal to fail, for much the same reasons that negative probabilistic relevance fails to imply that something is a defeater. That is, I expect positive probabilistic relevance sometimes tracks evidential support and sometimes not. I’m working out details on specific cases, which I’ll post a bit later, but wondered if others have been thinking about this much.