I disagree. It's worth remembering that current trends in AI are taken from taking a model from the 1970's and adding huge amounts of data and compute power to make these models robust enough to classify high resolution photographs. It isn't the necessary outgrowth of the cognitive science(s) as is commonly touted.

It's a magic trick which preys on a person's lack of scientific understanding; neural networks seem to anticipate, "brains in vats" but we're still in the midst of learning what, "brains" "people" and, "minds" are and nothing in ML space seems to shed any meaningful light on these pressing questions. This might be partially because the questions of Mind(s) and Self(s) are open philosophical problems.

An easy refutation that, "ML can model meaning" from recent Chomsky talks: consider that a neural network can be trained on any statistically meaningful corpus of text; this includes Klingon and Elven; it also includes jibberish'. It also includes semi-random presses of a key by a person typing with their dick or the like; nothing meaningful is being captured. We have a machine that is capable of statistical analysis. This isn't a new invention.