First of all congratulations to Lee Sedol for his wonderful win in round 4 against AlphaGo!
The last part of the game was a bit disappointing for spectators: AlphaGo, in the expectation of losing, started making 15-kyu threads to avoid the unavoidable. One of the leading German players (FJ Dickhut) was so embarrased that he made a comment on this in the German computer go forum. The operators in the AlphaGo team had mercy and resigned on behalf of their bot after move 180. But this created another problem: My wife, who has just started to play Go, wanted to see the final position. I showed her, but she was not able to count territory/area, because the game was still far from complete in the traidtional sense. So, "we" had a double-edged sword: either resigning early or to see terrible bot moves for another 20-50 moves. ************************************ Dynamic komi proposal: Assume a position where the bot is hopelessly behind (say at 20 % or 25 % winrate). Then find out what the expected score is (for instance the central peak in the playout histogram). Select an artificial komi value in this region and let the bot complete the game with respect to this value. (Perhaps adjust this value once or twice when it turns out that the bot is losing even higher.) Disclaimer: Of course, this would not result in higher playing strength. But it would be good for the Go community and their/our PR work. The final phase of the game would look much more natural. Ingo. _______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
