On Tue, 2007-07-10 at 14:06 +0200, Sylvain Gelly wrote: > Hi David, > > > >> (...) I cannot imagine that progress will be > > >> made without a great deal of domain knowledge. > > > Depending on what you exactly mean I disagree. > > I mean progress by the standard usually applied to computer Go: > > programs that can beat 1D humans on a full board, and then get > > better. > > For me progress meant an improvement, not a goal :-).
I think this is actually a very important concept. Computer go has traditionally set this long term goal but treated it as a short term goal. You get statements that someone tried something "but it didn't work" meaning that it didn't result in a 1 Dan player. David really hit the nail on the head when he said "progress by the standard usually applied to computer go" and then explained essentially that NO progress has been made by this yardstick (because the scale of progress doesn't even start until computers are 1 Dan strength.) In my opinion, to speed up progress enormously, you must set up some system of easily measurable progress or goals and give it some teeth, a way to "certify" this progress in the minds of everyone. A specific measurable goal and one that can be achieved without waiting years. > Anyway, very few months ago almost everyone in this list said that > UCT/MC was only suitable for 9x9 Go, which was said not representing > the Real Game Of Go, and will never (in near future, or even in far > future) do well in 19x19. Today, some UCT/MC based programs, e.g. > MoGo, can give 5 stones to gnugo on 19x19 in 30 minutes game, without > any additional go knowledge. For me it is an evidence that progress > can be made very quickly without "great deal of domain knowledge". Why > we (by "we" I mean the all community) could not imagine other > improvements? 1D humans on a full board is not so far, contrary as you > seem to say... This apparently is just human nature. We don't always adjust very quickly to change once we have anchored ourselves in the present. After years of very little progress, it's hard to imagine that it actually has occurred in a big way. Along these lines, when the highly arbitrary 1D level is actually achieved, it will take a good bit of time for people to recognize or acknowledge that it has happened. Even if it's directly shown to someone, they can pick apart the games of the new 1D player and will just say, "that's not 1D play" and will rationalize it away as a fluke. When 2 or 3 Dan is achieved, they will grudgingly admit that "maybe it's playing 1 Dan level", because their brain hasn't caught up yet. When it is 1 Dan it will have different strengths and weaknesses from humans but the humans who feel threatened will discount the strengths and only see the weaknesses. > > > I am less sure that knowledge representation in the > > classical programs is the right expertise for its representation in the MC > > playouts. > Yes, maybe you are right, I don't know. But I think it is not a bad > expertise :-). And also it is a very "expensive" expertise (in the > sense that it takes a lot of time to get). > > So many thing yet to do! > To me it sounds like a good news, don't you think? :-) > > Cheers, > Sylvain > _______________________________________________ > 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/
