AlphaGo vs AlphaGo

[John Fairbairn]

Anyone know anything about 3 recent internet games AlphaGo vs AlphaGo? Genuine?

If so, White won all three and two were finished early, which hints at a couple of things: Komi still too small? And even AlphaGo makes mistakes bad enough to lose mid-game.

Japanese pro Ohashi Hirofumi, who discovered the games, also says a lot of the moves don't look human. Created in Roswell?

These games are separate from the deepmind vs sky15 game. Two were played at 5 secs a move and were identical for 19 moves (so there is some sort of randomiser?). The other was at 1 min a move.

Not sure if Ohashi is deluding himself here, but he's very clued up about computer go and has a regular column about it in Go World.

There's another interesting case I found: Zhou Junxun played against the partnership of Xie Yimin and Dark Forest.

Slightly to my surprise, the Oriental pro go world seems to be embracing computer go with gusto from the word "go". This didn't seem to happen with shogi and xiangqi, although I think they've got over the hump now.

[yoyoma]

Are you talking about these AlphaGo vs AlphaGo games released by deepmind back on September 5?
https://deepmind.com/research/alphago/a ... es-english

[John Fairbairn]

Yes, they seem to be those. Thank you.

Have I missed any comments here on these games? If there's a thread I may get round to posting some of Ohashi's comments.

[hyperpape]

There was at least one thread some time ago, albeit with surprsisingly little commentary. http://lifein19x19.com/forum/viewtopic. ... nd#p209838