Learning in the Natural World

A couple of thoughts concerning a recent release on learning, published on the Science Daily site. One is that this is somewhat revealing of how some folks still think of learning potential and semantic intelligence as though these were, in some sense, independent of the environments in which they are embedded. We've known for quite some time that animals, humans being no exception, are predisposed to abstract certain generalizations from certain evidence. Such predispositions are the only way it can be possible for humans to learn, e.g., subclasses of grammar classes which are provably unlearnable in the limit on the paucity of input data which humans evidently rely on. It also seems indisputable that these predispositions are rooted in the evolutionary lineage of the species; our epistemology is built on a foundation of syntheses that represent, as it were, an archaeology of what our ancestors needed to induce given the requirements of the times.

It's also important not to discount the role of the secondary effects of modeling other minds. Humans don't just induce rules in the sense of adapting their behavior so as to conform with identifiable rule - sets: they are able to model this behavior in themselves and others in a way that results in further adaptation of behavior, and to appeal to models as justification for behavior. This opens the door to things like culture and narrative playing a role in the learning process.

These considerations add up to the result that the process whereby humans arrive in development at semantic awareness sufficient for allowing things like society and acculturation to play a role in learning can't happen any old way. And the way that it does happen is likely to be a reflection of the development of the species as a whole. For this reason, it behooves us to be very careful of assuming that features that we readily perceive as being causally or transparently connected appear so or could appear so to brains in earlier stages of development.

Here's a practical upshot for one of the things that was looked at by Kuhl. It's a pretty safe assumption that television and computer monitors weren't on the scene fifty-and-some-odd thousand years ago when the bootstrap-your-way-into-semantic awareness procedure was in the process of working itself out on the African veldt. And when one considers the mess of cultural and artistic conventions which support interpreting a bunch of phosphorescent pixels on a 2-D surface as a 3-D representation of a narrative involving a person, there is therefore no reason to think that applying this set of conventions would be a native human ability, or anything like a native human ability. In short, we really shouldn't be surprised that human infants can't inherently learn things from people in television images: expecting otherwise is, frankly, as crazy as expecting that a flat red octagon with the letters S-T-O-P printed on it in white would 'automatically' qualify as a stop sign for a Chaldean shepherd in the 6th century B.C..