Handicap 29 Prize

offered by Ingo Althofer; January 26, 2012


In the American Go Congress 1998, a demonstration game on 19x19-board was played between human Dr. Martin Müller and Many Faces of Go which was the leading bot in that year. Martin gave 29 stones handicap and nevertheless won the game by a margin of 6 points. sgf of the game can be downloaded from Nick Wedd's list of human vs computer games. The starting position - with the handicap stones - can be seen in the following figure.

Due to the concept of dynamic komi, during the last 18 months Monte Carlo bots have "learned" to deal with handicap situations. This in mind time is ripe for the following challenge: Try to achieve bot wins against the 1998 version of Many Faces of Go at 29 handicap stones!

I want to encourage people by announcing the following prize: I give 1,000 Euro to the programmer of the the first bot that achieves such a win at least three times in a five games match.

Details:
* Matches have to be played on KGS. The five games of a match have to be played within 24 hours.
* I have to be informed about such a match two weeks ahead. Use my email address ingo.althoefer bummbumm uni-jena.de, where bummbumm is substituted by the wellknown symbol. After a match the contender has to send me sgf of the games and in case of a win also his/her banking connection data.
* Time for the challenger is 60 minutes for a game in total.
* The opponent will be the 1998 version of Many Faces of Go that is currently running under the KGS-account mfgo1998. This version is playing rather quickly. The option "random" is switched on.
* Rules are Japanese, with komi = 0.5.
* Each game has to be started in the following way: handicap 9 and only pass moves by White in her first 20 moves. During this phase mfgo1998 will place 20 stones on the board, not in each game in the very same places.
* After a non-successful attempt, the corresponding programmer(s) have to wait at least 12 months for a new attempt with their bot.
* Contenders are allowed to play sparring games against mfgo1998. When mfgo1998 should become available on CGOS, contenders will also be allowed to play sparring games against it.
* David Fotland has announced that he himself will not apply for the prize.
* It will be reported on this site when the prize task has been solved.
* The offer ends on December 31, 2020.


Many thanks to David Fotland for providing the service with mfgo1998 on KGS!

In case of disputes I am willing to take advice from experts like David Fotland, Martin Müller, Nick Wedd, and others.

Legal measures are excluded. (In German: "Der Rechtsweg ist ausgeschlossen.")


PS 1. Once the prize is paid I would be interested to see at which handicaps above h-29 bots will be able to beat mfgo1998.

PS 2. In August 2009, I was able to beat an old ManyFaces version, similar to mfgo1998, semi-automatically by a dynamic komi procedure ("Rule42"), at handicap 25 on 19x19-board.
Here you can download the sgf with some comments.
And this figure shows that Rule42-reduction of (dynamic) komi happened mainly in two avalanches in that game.




Back to the main site of Ingo Althofer