Gary Marcus

That's exactly what they said about Go. Required intuition and long term planning. No brute force or simple algo could do it without understanding complex modeling of opponents' intentions.

Fair call, and therein lies the trap. Solving go was/is a massive achievement and demonstrates brute force success on that class of problems. It doesn't mean that all problems are amenable to that particular hammer.

On the contrary, I don't think DeepMind characterizes alpha zero as brute force. It it quickly develops an understanding of the problem at super human level without trying even a tiny fraction of the possibilities.

Albeit still using a ton of computation; depends on your definition. The point about hammers and nails holds, either way.

Well clearly the human brain uses an enormous amount of computation to develop an understanding of the game of go at professional levels. This is not a mark against intelligence or understanding.

I think when we thought of computation as a serial logical endeavor, those numbers seemed outlandish for AI. But in the Neural Network parallel training context, it means something more like 'more brain cells', not a deficiency. Brain seems to do petaflops of computation also.

Or you can think it of in terms of # games played or energy consumed...