DeepMind has released a new set of AlphaGo self play games. Up to fifty in total will be released eventually. It is very interesting to see how it all unfolds
Share your thoughts, and try some new moves!
Given how 8 out of 10 games were won by white, it may suggest the komi is a little too high. Then again, the statistician side of me screams there’s not enough sample to make that conclusion (yet).
David Silver from the Deepmind team said that AG thinks the game is pretty balanced with 7.5 komi, although AG slightly prefers White.
I see that these games are all Chinese rules (at least, the ones I looked at).
Does AG ever play Japanese?
Here are the SGF files uploaded on OGS, for everyone interested https://online-go.com/library/784/1830
White won 38 of all the 50 games (76%).
I have never seen AlphaGo play Japanese. The Lee Sedol series of games were conducted under Korea Baduk Association in Korea, but they used Chinese ruleset. And of course Master games on WildFox and Ke Jie / Pair / Team games are all chinese.
The UEC Cup (http://senseis.xmp.net/?UECCup), the world cup equivalent of Go AI’s, has been held in Japan using Japanese ruleset, but AlphaGo has never participated in this tournament, except for Aja Huang’s Erica in 2011 before Aja joined DeepMind. Computer Go Server (http://senseis.xmp.net/?CGOS) uses a modified version of New Zealand ruleset called the Tromp-Taylor rules (https://www.cs.cmu.edu/~wjh/go/tmp/rules/TrompTaylor.html), which is the easiest for programmers to implement on Go bots. But again, AlphaGo never played on it.
Japanese ruleset is the most difficult for programmers to implement in my opinion.