Generalization and other innateness

I don’t particularly look up to any AI boffin, but so far I found three folks’ opinions carrying weight, when talking about how to achieve True AI and they’re all Jewish: Marvin Minsky, Gary Marcus and Jehuda Pearl. Even in the set of bad AI-related ideas, the best ones of those often also originate in Jewish brains. I conclude it may be that the cleverest people are disproportionately Jewish.

In spite of decades of being urged not to engage in it, generalization is perhaps the most useful things brains and all neural networks do. Bayesian learning. It’s innate even in the brain of the Leopard slug, which can learn to avoid a location with tasty food, when it has associated it with electric shocks. One could argue that it keeps in the back of its mind: “Thrice a shock, likely always a shock at those beets – not worth the crawl.”

Gary Marcus said that innateness is a concept to consider, and that jives with my own ideas. Apart from the ability to generalize, I’ve started to make a list of features, relevant to AI, that human brains come “standard” with:

– Understanding concretized higher-level sensory input such as colors, shapes, movement, sounds, smells

– Understanding concepts of personal possession, threats to personal safety, emotional states, genetic in-group preference, the fact that tools can be made/used to achieve goals, the need to breathe, eat, drink, sleep, have access to shelter and procreate

– The ability to learn language

– At least a basic understanding of Algebra. At the very least, a grasp of “none, one, a few, many” and simple set theory as in “x belongs to group y and group y has property z”. Some people have an incredible talent for math.

Disqus Comments