"I’m encouraged by this. I think it’s really cool that
It’s possible to learn an evaluation function directly from raw data, with no preprocessing
A fairly slow evaluation function (several orders of magnitude slower) can still play well if it’s more accurate
I’m pretty curious to see if this could fare well for Go or other games where AI’s still don’t play well"
I think AlphaGo has kinda demonstrated that the answer is... yes?
