Comment by cambalache

I agree with you. I was in the Michael Jordan camp, as in: Machine Learning is just statistical inference rebranded and with better marketing. There are lots of cool applications for sure but there is no "magic sauce" beyond your usual models obtained by error minimization. But lately, especially seeing the progress with the Attention/Transformer models I have been slowly letting myself convinced that maybe that is all there is behind human intelligence: Lots of priors encoded by evolution in the brain, and general quasi-brute-force mechanism for inference, deduction and imagination. Of course this is probably a naive view, but I think the main point will remain: Intelligence can be created or simulated by giving an agent the capability to extrapolate in a way that "makes sense" in the environment the agent will live.