>>14413078ML has become unfortunately synonymous with large matrix multiplication and Markov chains, because for certain specific tasks they happen to be pretty good. they are basically impossible to generalize, though, because they are blackboxes - AI "research" got a little bit ahead of itself when it looked at the fact a trained set of matrix multiplication weights couldn't be reduced to a human-writeable algorithm and concluded that a) this was the key to intelligence, b) the singularity was fast approaching, and c) soon humanity would be able to construct beings with supernatural powers on silicon.
just a little bit ahead of itself.
in reality, we don't have that level of understanding of intelligence, Behaviorism was a failure, even animals have every indication of rich (if languageless) internal intellectual lives, and AI research has spent the better part of a decade going "i don't know how what i just built works so it must be God"
