"AI is hard to do because all the supposedly "easy" tasks proved to be the most computationally heavy for serial systems - and easy for a distributed, massively parallel system like a brain. The "higher" cognitive tasks proved to be the easiest of all because they're the most abstracted, and such symbol manipulation proved "easier" for giga-MIPS machines of the last generation of AI. Now we can do even better with peta-MIPS machines...

Minksy, Moravec and others pointed all this out years ago."

They did, didn't they. Yet they ALSO predicted AI within 20 years every decade since the 70s. Not a good track record. AI is almost a cult.

Polymath Roger Penrose's point is they'll ALWAYS be wrong as there are processes which are uncomputable and AI is just that. He contended decades ago in "The Emperor's New Mind" a book debunking the rampant AI claims of the 70s and 80s by the same Minsky et. al., that there is some quantum effect computing going on in the base of brain cells. Only in this decade have others begun to explore that idea.