[Computer-go] 29 stone game

Martin Mueller mmueller at ualberta.ca
Sat Feb 11 14:38:06 PST 2012


> The 29 stone game wasn't really a "demonstration game".  
> 
> My memory is that I had just finished igowin 9x9, and Martin played through
> all its levels up to professional, without losing a game.  I was surprised,
> and while we were talking the subject of a 29 stone game on 19x19 came up.  
> 
> I don?t remember if I challenged him or if he said he could beat 19x19 on
> that handicap.  So we made a very small bet (I don?t remember, maybe $5), on
> the game and he played.  I was very confident Many Faces would win and was
> really surprised when it didn't.  I don?t remember if we announced the game
> before it was played.  I think we didn?t, but there were a few people
> watching.
> 
> I don?t remember how the game record got out to the internet.
> 
> Martin, how do you remember it?  It was a long time ago...

I think we just agreed to try as an experiment. I think there were "side bets" with other people betting money, but not between you and me. Not sure why we chose 29, but it was related to Janice Kim's game vs Handtalk.
There is about one page of description in my survey article in AI journal.
M. Müller. Computer Go. Artificial Intelligence, 134(1-2):145-179, 2002.

If you have access to it (e.g. through your library), you can read it at http://dx.doi.org/10.1016/S0004-3702(01)00121-7

If not, a last author draft is on http://webdocs.cs.ualberta.ca/~mmueller/cgo/survey/index.html .
That page also has the sgf game records of this and other games from the article.

Here is a quote from the paper:

"In August 1998, at the US Go congress in Santa Fe, New Mexico, a 29 stone handicap game was played between David Fotland's program  The Many Faces of Go and the author, a 6 dan amateur player. Like Go4++ and Goemate, Many Faces is one of the strongest Go programs in the world, and a few months after this game it won the 1998 Ing Cup.  Many Faces is regarded as one of the best programs when it comes to tactical fighting. However, in this case its aggressiveness backfires. Despite the huge handicap, the game ends with a six point win for the human.

Figures \ref{mf1} to \ref{mf4} show the starting position and the game record. In the beginning, White sprinkles some stones around the board to probe for weaknesses, but Black defends well. In the bottom left corner, White uses a confused ko* fight to secure one group, then continues to create complications from this basis. By move 85, Black's group in that corner has been reduced to only one eye*. Black invests too many moves in a failed attempt to rescue this group and in a counterattack against the white stones floating in the center. With move 207 (7 in Figure \ref{mf4}), White isolates another black group in the lower right corner, and kills it a bit later by a combination exploiting a hidden dependency between two seemingly safe eye areas. This second big capture makes the game very close, and White easily overtakes Black in the remaining endgame. Throughout this game, most computer moves are quite reasonable, but there are just enough mistakes to allow White to grind out a win.

Conceptually, Black's main problem seems to be that the program tries to fight it out with a stronger opponent on even terms, instead of preserving some of its huge initial advantage by playing slow, ultra-safe moves. One might argue that a program playing a safer, more territorial style will be harder to overcome. However, a few weeks before this game, in an exhibition match held at the AAAI conference, professional player Janice Kim had already given Handtalk a similar huge handicap and won \cite{Mechner1998}."

--

If you compare this description with the behavior of current MCTS programs, I think we are very lucky that some of the exploitable behavior of old-style programs is "automatically" taken care of by the way the MCTS algorithms work.


> Martin once told/wrote me that the bet gave him three tries and he
> managed it already in the first one.

Yes, we all thought it would be hard, so we agreed that I would have three tries. But I won the first one by a small margin so we stopped there. I found it very difficult to play this game. I just tried very hard to not lose immediately at every single move, and imagine a path to winning by first isolating, then killing some groups. I got some chances when Many Faces played some unexpected tenuki and let me break through its positions. I was lucky :), but I think I was also playing well in giving it enough chances to confuse itself.

> 
> I saw a game between Janice Kim 1d (she is now 2d) and Handtalk (IIRC) in Providence at an AAAI event where Handtalk lost with a 25-stone handicap. That was 1996 or 1997.
> 
> Beating programs with very high handicaps was a sport in those days. It wouldn't surprise me if Martin had already engaged in this pastime prior to the bet. :-)

Janice is a 3 Dan pro.
http://en.wikipedia.org/wiki/Janice_Kim

At the time, I tried quite hard to get a game record for her game with Handtalk, but failed. As far as I know, it is lost.
I did not practice ultra-high handicap games. I played my own program at the time, Explorer, on 9 stones occasionally, but it was too depressing (for me as the programmer).

	Martin





More information about the Computer-go mailing list