Can a subject’s beliefs, desires, fears, or goals causally influence the way things seem to her? Suppose that the answer is yes. That is, suppose that a subject’s mental states can penetrate the way things seem to her. What are the epistemic implications of this cognitive penetration? This question is becoming a ‘hot’ topic in epistemology these days, though it has been around in various forms for decades at least. My tentative plan is to write two blog posts on this topic. Both are responses to Jack Lyons’ very helpful paper, but I hope the wider significance of the posts will be apparent. In this post, I will argue, contra Lyons, that we have no reason to think reliability explains which instances of cognitive penetration are benign and which are malignant. In the next blog post, I’ll argue that reliabilism has cognitive penetration problems of its own.
It’s clear that some instances of cognitive penetration are perfectly benign and, sometimes, even epistemically beneficial. It seems to me that I ate cereal for breakfast now because this morning I (rationally) formed the belief that I ate cereal for breakfast. The previous belief makes it seem to me that I ate cereal for breakfast now. If this kind of penetration isn’t benign (or beneficial), then one wonders whether any of our memorial beliefs are justified. On the other hand, some cases of cognitive penetration apparently pose a problem for anyone who, like me, wants to hold that all (perceptual) seemings prima facie justify. If my lust for gold causes it to seem that the yellow nugget is gold, then most of us feel some hesitancy in allowing that such a seeming provides even prima facie justification that the nugget is gold.
So there are good cases of cognitive penetration and there are (apparently) bad cases. What explains this difference? Jack holds that it is explained by reliability (see section 4). In the “good cases”, such as ordinary memorial belief formation, the reliability of the relevant process is high. In the bad cases, the reliability of the relevant process is low. If Jack could make good on these claims, it would be a real win for reliabilism; however, Jack is writing checks that reliabilism can’t cash (yet). The problem is that whether reliability can explain the difference between the good and bad cases depends almost entirely on how one individuates processes, and it is far from clear that there is a sensible way of individuating processes that allows reliability to track the goodness/badness of cognitive penetration.
To see the problem, consider this question: are cognitively penetrated processes different processes than the ones that aren’t cognitively penetrated? Since I don’t think an “it depends” answer will help the reliabilist much, I’m going to ignore it. Suppose that the reliabilist says “no”. This means that cognitively penetrated processes are identical to the ones that aren’t penetrated. Hence, the reliability of those processes also will be identical to the reliability of the processes that aren’t penetrated. Presumably, those processes will be reliable. But then the bad cases of penetration would yield justified beliefs because they are produced by reliable processes. So if the reliabilist is trying to explain the difference between the good and bad penetration, he shouldn’t say “no” to this question.
Suppose the reliabilist says, “yes, cognitively penetrated processes are different processes than non-penetrated processes.” Now we are faced with a different question: Do the good cases of cognitive penetration involve different processes than the bad cases? Again, I ignore the “it depends” answer. If the reliabilist says “no,” she is stuck with a similar problem to what she faced by saying “no” to the previous question. If the good and bad cases involve the same process, then the good and bad cases will be equally reliable and either the good cases will involve unreliable processes or the bad cases will involve reliable processes. Either way, reliability can’t explain the difference between the good and bad cases of cognitive penetration. So if the reliabilist is trying to explain this difference, she shouldn’t say “no” to this question either.
If the reliabilist is trying to explain the relevant difference, she needs to hold that, “yes, the good cases of cognitive penetration involve different processes than the bad cases of cognitive penetration.” For the sake of keeping this blog post less long, I’ll just declare three things I think are true: 1) The required account of process individuation would likely need to individuate processes fairly narrowly. 2) No one, Jack included, has identified a way of individuating processes (i) that meets the above requirements for explaining the relevant difference; (ii) that is not ad hoc; and (iii) that is empirically adequate. 3) We won’t be in position to affirm that reliability (of processes) explains the difference between good and bad penetration until we have discovered a way of individuating process that at least plausibly satisfies (i)-(iii). I conclude that we have no reason to think that reliability really does explain the difference between the good and bad cases of cognitive penetration.