AlphaGo wins with Lee Sedol

Google Deep Mind‘s AlphaGo won two games against the world go champion, Lee Sedol. This is a ginormous triumph of statistical methods in general and machine learning in particular over “symbolic AI.”

I remember writing an essay for a class in philosophy some years ago about the progress of AI game engines and the somewhat unimpressive achievements of Deep Blue. It was of course exciting to see a computer beat a reigning chess world champion, but underneath all the heuristics IBM implemented for chess, it was all “brute force.” Chess has a game tree complexity of 10123, which is huge, but still “traversable” by modern computers using good heuristic functions. Go, on the other hand, was deemed unsolvable by any “brute force” methods, because its game tree complexity is 10360—far too big. I don’t think anyone in 2006 expected that within ten years a computer program will beat the best Go player (I know I didn’t), yet it just happened. Continue reading “AlphaGo wins with Lee Sedol”