Comments on: AlphaGo https://www.isegoria.net/2016/01/alphago/ From the ancient Greek for equality in freedom of speech; an eclectic mix of thoughts, large and small Sat, 18 Apr 2026 00:55:29 +0000 hourly 1 http://wordpress.org/?v=3.6.1 By: Candide III https://www.isegoria.net/2016/01/alphago/comment-page-1/#comment-2455521 Candide III Sat, 30 Jan 2016 15:59:11 +0000 http://www.isegoria.net/?p=39733#comment-2455521 There is both Monte Carlo tree-search using deep learning results and extensive use of fixed programmed features and patterns. The games look sane, not like CrazyStone, and beating European amateur champion is quite an impressive feat, though I’d like to go over the games with a stronger player. Still, from how the games feel, I wonder how will AlphaGo fare against a serious Korean or Chinese pro playing at tournament strength. Fan Hui became 1 dan pro at 16, achieved only 2 dan in 3 years and moved to France, effectively terminating his professional career. There are hundreds, if not thousands, of such rejects from the extremely rigorous and competitive Korean, Chinese and Japanese professional ladders, occupying a gray area between amateurs and professionals. If, say, Lee Sedol plays seriously instead of as a PR feature for the AI team (as such matches often seem to be), I bet he will crush AlphaGo.

]]> By: Ross https://www.isegoria.net/2016/01/alphago/comment-page-1/#comment-2455451 Ross Fri, 29 Jan 2016 22:37:09 +0000 http://www.isegoria.net/?p=39733#comment-2455451 Cool Story. Didn’t think Go AI would move quite this quickly. That said….

Gary Marcus wrote that this Euro-Go chap is ranked something like 600. Given DeepMind’s play, it might be as good as 275 ranking, roughly. So, there’s a bit further to go (NPI) before “Skynet”.

FWIW this is not undifferentiated pattern matching à la the current rage, “deep neural nets”. I really look forward to to reading the background but IIRC there’s a bit of tree-search blended with the hidden layers as well. That’s cool.

]]>