Hello, > I would be more interested to see three strong programs voting for > moves and no human intervention. Perhaps add Patchi to the mix.
I did some work on "simple" voting systems in computer chess around 1990. The main result was: majority voting gives a SMALL improvement over the (equally strong) bots - indeed, much smaller than 200 rating points (the improvement was more in the range of 30-50 points). One problem in such systems is the synchronization of timing: you lose playing strength, when you brutally stop all three bots in a predefined moment. Much later (around 2004-2007), Helmut Weigel made very succesful experiments with three chess bots who together generated a small selective tree before deciding on a move. He called it Gecko. Gecko's crisis came when Rybka - being clearly stronger than all other bots in those days - entered the scene. Helmut did not find a good way to combine one very strong and two normally strong bots. Ingo. PS. Majority voting (with 3 bots) in 19x19 go might be problematic as you will have many situations with three different proposals. So, you would need clever tie-breaking. > > > On Wed, Apr 27, 2011 at 4:31 PM, "Ingo Althöfer" > <[email protected]>wrote: > > > Hello, > > > > now I found time to read several sgf protocols > > of KGS games played by BonoBot. > > > > Here is part of the chat from a game between > > BonoBot and Zen, which I find very interesting. > > > > PW[Zen19B]PB[bonobot] DT[2011-04-23] > > ************ Chat on ***************** > > ajahuang]: they are strong at different aspects > > [ajahuang]: Zen features more fighting and semeai > > [ajahuang]: Bonobot is better at direction > > > > [ThePok]: so they have to merge > > > > [ARPAnet]: What do you think, aja? How strong would > > be a mixed "program" where Zen and bonobot suggest > > a move in a current game situation but a human (for > > example you) decides which one of those moves should > > be played? > > > > [ajahuang]: it depends on fast or slow game > > > > [ARPAnet]: well, how about the current time setting > > > > [ajahuang]: 5d up, sure > > [ThePok]: than theres not much time to choose > > [ajahuang]: if not 5d, at least 4d > > [ARPAnet]: hmm, ok thanks > > ... > > [BOThater36]: so dont trust Aja can pick the best move :-) > > [ajahuang]: everything could happen in fast games...anyway :) > > ************ Chat off **************** > > > > From my chess experiences I would say that Aja is rather > > pessimistic. Between 1985 and 1997 I made several 3-Hirn > > experiments where I was the human having the final choice > > between candidate moves by two different chess bots. > > My rating was around 1900, and through a wide range of > > bot ratings (between 1500 and 2500) I was able to add > > 200 ratings points to the strengths of the bots (in these > > experiments the bots had very similar strengths - only my > > human strength differed a lot). > > > > If this would translate to "server go", 5d-Zen and 5d-BonoBot > > would be lifted to something like 7d by the final choices > > of Aja Huang[6d]. > > > > I am really eager to see, if this holds in reality. > > > > Ingo. > > -- > > NEU: FreePhone - kostenlos mobil telefonieren und surfen! > > Jetzt informieren: http://www.gmx.net/de/go/freephone > > _______________________________________________ > > Computer-go mailing list > > [email protected] > > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > > -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
