On Sat, 2008-08-09 at 03:49 -0700, terry mcintyre wrote: > From: David Doshay <[EMAIL PROTECTED]> > > >One point not discussed much in this thread is the consistency issue. > >I think that if Kim were able to play a dozen games against mogo with > >this same handicap he would win the last 6 ... people manage to adapt > >and the computers do not. > > >But that much cluster time and Mr Kim time are probably not available. > >Perhaps with all of the interest this match is generating we will be > >able to get more of each (or some other pro) in the future. > > Unfortunately, it is not that easy to get a supercomputer. I hope that some > firm chooses to take this on as a challenge, the way IBM and Cray took the > game of Chess. As for the learning problem, now that we know how to generate > high-level plays which match the sophistication of high-level joseki, perhaps > it will be possible for computers to learn from pro-level games, including > those lost by the program to pros. Who's to say that a nice long post-game > analysis by the program could not lead to it returning with better moves? > > I should mention one point: Mogo was far enough ahead that Kim estimates an 8 > stone handicap would be tough for him to beat; perhaps a 7 stone handicap > would be appropriate. Interesting that Kim, though not familiar with > Montecarlo programs, quickly grasped that the weird-to-humans endgame play, > which blithely ignores the value of playing dame points under Chinese > scoring, provided that it preserves a 0.5 win, does not detract from the true > value of Mogo's lead. >
That does not surprise me at all. Although not being a strong go player myself, I've always assumed that all the fuss over not winning big would be something that really strong players would not think as odd. I would think that really strong players have a better global sense of what is really important. After all, it seems to be one of the things that has made this kind of major improvement possible. - Don > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
