On Tue, 2006-12-05 at 09:51 -0800, steve uurtamo wrote: > > I'll bet Mogo would give a dan level player fits at > > 9x9 if 1 week of > > thinking time per move could be compressed enough to > > play a 30 minute > > game. > > you could always get a dan player to volunteer for > such a game. he would promise not to spend more > than 1/2 hour on the game, and mogo would play > postally. > > i'd be very impressed if mogo could "give him fits".
I don't think such an experiment could be fairly constructed. You cannot prevent a strong player from dreaming about the position at night. The brain is a wonderful thing and solves problems even when you are thinking about other things. What do you mean by dan level? I don't mean "high dan", I mean 1 dan. Mogo would also have a memory problem. The UCT programs build trees in memory. My own program cannot think more than a few minutes without running out of memory - so even the experiment you propose cannot be done. But I'm quite sure Mogo would win such a match were it possible with time-compression. I think you are falling for the standard misconception that the computer must be superior in every aspect of the game to have a chance. I speak from experience. I know exactly how these things work. The match would begin, the human would probably be outplaying the computer and then make some error. The computer would win and everyone would cry it shouldn't have happened. The computer just got lucky this time. In a long match everyone would cry that the human outplayed the computer, but the computer kept getting "lucky." The human might win a few games and observers would believe these were the only relevant games, proof the human is better. The others games were just "dumb" errors by the humans that shouldn't have happened. (The human was unfairly thrown off balance - yeah, that's it.) Games like chess and go are all about statistics. We think it's about brilliant play but it's much more about who makes the errors and if they get punished. Both the computer and the human will make errors. The humans will tend, on average, to make the more serious errors. The computers will make more minor errors. Based on what I've seen it's pretty much all about the more serious errors. This is why Mogo would win. It would not play as brilliantly as the one dan player or as purposefully, but it would be solid like a big dumb rock. Eventually the human would do something stupid enough that the computer would actually respond correctly and punish it. It probably wouldn't have to be horribly stupid for a 1 week thinking time Mogo to see. When I started playing tournament chess, I was quite weak. One day it occurred to me that I was blundering too many pawns and pieces. I "set my mind" to not make these trivial blunders - after than I never made a move without first looking around to see if the piece I just moved was directly hanging or I was moving a defender of something else. I had a very simple and trivial checklist I always followed that dealt only with the most stupid kinds of mistakes, nothing strategic or deep that required any calculation. My rating shot up to over 1700 from about 1300 even though I was still playing horrible chess. I was even still blundering and making errors, just not the most trivial kind. I had discovered this very basic principle, it's not what you do well, it's what you don't do badly. To this days, humans are still far above computers in chess skill - but computers have continually raised the level of their "worst moves." Their very worst moves are well above the level of the humans "worst moves" and this is pretty much what really counts. In fact, in honor of Chrilly's laws I will call this "Don's law". "What really counts is how bad your bad moves are." A perfect example of this happened recently. One of the top Grandmasters missed a mate in 1 against a computer in chess. Unless there is a bug in a program, certain kinds of errors are impossible for computers to make, but not humans. For the longest time it was the "brute force" searching programs that did so well. This is the exact reason why this did so well. The selective programs saw tactics deeper and could find better moves - but all the brute force programs had to do was wait for a serious mistake due to the selectivity. Eventually we figured out how to do selectivity safely and by doing so we enabled the computer to make even less serious errors. - Don > also, 30 minutes / 50 moves (guess) == .6 min / move. > 7 days = 10080 minutes, so it's just a factor of > 16800 speed increase that's needed. is mogo > parallelizable? if so, i could probably get 10 > machines on the job, so we'd just need another > factor of 1700 increase or so. :) > > s. > > > > ____________________________________________________________________________________ > Any questions? Get answers on any topic at www.Answers.yahoo.com. Try it now. > _______________________________________________ > 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/
