Margolis and Laurence, "Learning Matters" (2011), pp. 528–529

Perhaps something can be said about some of the characteristics of typical instances of learning. This isn't to give a definition of learning, just to note a few of the features that implicitly guide the recognition of certain clear-cut cases. We'd suggest three.

The first and most basic is that learning generally involves a cognitive change as a response to causal interactions with the environment. Of course, not all changes that trace back to an organism's environment will count as learning — that is the point of Fodor's examples where concepts are acquired through futuristic neurosurgery or through a miraculously fortuitous hit on the head. And not all cases of learning will involve environmental sensitivity, as some learning may be wholly a priori. Nevertheless, one important feature of learning is that it often, perhaps typically, involves a sensitivity to the environment.

The other two features highlight some aspects of the causal interaction that occur in paradigmatic cases of learning. One is that learning often implicates a cognitive system that isn't just altered by the environment but, in some sense, has the function to respond as it does. For example, learning facts about the locations of various objects when entering a room isn't just a matter of having your mind altered upon perceiving the situation. The changes presumably are of the sort that our perceptual systems and related belief-fixation mechanisms are designed to subserve. In contrast, when you get hit in the head, as in Fodor's example, your mind might be miraculously altered in a useful way, but the intervening mechanisms don't have the function of subserving these sorts of changes; it's just a matter of blind luck.

The other suggestion is that learning processes are ones that connect the content of an experience with the content of what is learned. The two aren't merely causally related. They are semantically related. Hypothesis testing exemplifies one type of semantic relatedness, but it hardly exhausts the possibilities. For example, the rote learning of a list of numbers may involve reciting the numbers several times out loud, chunking the numbers in thought, and associating the numbers with other memorable items. By any reasonable standard, the processes integral to these activities are ones in which the outcome is semantically related to the preceding experience — the cognitive processes that bring about the change in thought are ones that turn on the contents of the mental states involved in the transition.