2014 Game 2
White: Crazy Stone
Black: Franz-Josef Dickhut, 6 Dan
Black wins by resignation
Total Score after this game: Crazy Stone – Franz-Josef Dickhut: 1-1
At the very start of this game it seemed as if Franz-Josef Dickhut wanted to counter a possible big moyo of Crazy Stone by playing thickly, emphasizing influence. Thus, Crazy Stone had separate small groups in all four corners. But maybe this was just the consequence of Crazy Stones playing White, because around move 70 the typical pattern of the first game seemed to be re-established: Dickhut grabbed territory in the corners and Crazy Stone tried to create large territories on the outside. Crazy Stone in its own estimation and many spectators seemed confident that Dickhut would get another drubbing.
What happened thereafter is somewhat hard to explain. Beginning with the invasion of Dickhut’s upper left corner at move 86, Crazy Stone began a series of doubtful maneuvers. In these the program switched seemingly without focus from one area of the board to another, losing aji and also points in the process. The worst of these maneuvers seemed to be the reviving of the white corner in the upper right in the sequence of moves starting with 132.
At some point around this time of the match Crazy Stone must have realized that it was no longer ahead but actually behind. In the first game, one could witness how a go program defends a lead: by playing super-thick but objectively useless security moves until the lead is brought down to nearly nothing. Now, in the last phase of game 2 one could observe how a Monte Carlo bot tries to catch up in a virtually lost position: It starts to play absurd attacking moves, such as 202. The reason, I suppose, is that in some of its zillions of Monte Carlo playouts of this move the opponent reacts wrongly and so manages to lose the won game: A sequence of events that is never going to be realized in the actual game being played.
Some commentators remarked that Dickhut did not win this game – Crazy Stone lost it. This may be so. What seems to have been Crazy Stone’s undoing is that (in contrast to the first game) this time there were many hot, unsettled positions on the go board in the middle of the game. It is objectively difficult to decide in which area of the board it is most important to make the next move – thus presumably diverting the attention of the Monte Carlo engine to hopping between these areas and thus having too few useful playouts for a valid evaluation of each move. If this speculation of mine is somewhat correct, Dickhut may not have won this game by clever maneuvers, but by posing insoluble strategic questions to Crazy Stone. This is, what one of the godfathers of combinatorial game theory, Prof. Elwyn Berlekamp of Berkeley, called the “enough rope principle” – a strategy to employ when one is behind in a game of no chance: Give your enemy enough rope so that he can hang himself. Dickhut seems to have done just that. He gave Crazy Stone a lot of rope – and the bot hanged itself.
If you want to download game in SGF format, please click here.