Ruse of the century.

It would make sense that AI could one day reason in ways not too dissimilar from human cognition (just far faster). In the way a person learning a new language has to consciously translate every word. They will search their brain for what the translation of "bathroom" is, until one day there is no noticeable translation, the word just comes. It seems reasonable that training will increasingly compress things that used to require long chains and compresses it into pseudo-intuition with common answers coming almost automatically and novel problems requiring deliberate thought. We'd see changes from "what did this AI think?" to "what internal state produced this decision?" It'd be like a Chess champion telling you "that move loses" and you're like "why?" and they're just speaking from years of intuition and pattern recognition, and actually have to reconstruct the line afterwards.