I disagree; I don't think that the draw factor has to be corrected for, because it *is* legitimately part of the notion of "deepness" that the method tries to measure.
If a "perfect player" will still draw 80% of the time with the top 200 players in the world, by what reasonable measure are those top players not very close to perfection? Take tic-tac-toe; there's really only room for three levels there. You have a totally random player, a beginner who knows the rules and avoids immediate traps, but doesn't really get the game, and then you have somebody who's played enough to read ahead the three or four moves needed to always secure a draw. The game simply doesn't have room to fit any more categories; once you have made two or three steps in improvement, you're done. Is there any rational sense in which one player on that third level can be said to be "better" than the other? He may be slightly brighter and see the draw sooner than his opponent, but the game is not "deep" enough to be able to distinguish subtle differences like that. It is an intrinsic property of the game, not the players. The same it is with chess, although obviously the scale is much larger than 3. We may say that Anand is stronger than Kramnik, etc., but the fact is that they are both so close to each other (and to the top limits of human capability, I suspect) that chess as a game is not able to distinguish between them except when one of them makes an uncharacteristic error. This is a property of chess, and I think *does* say a lot about its deepness. To say that you can get a similar result in Go by requiring a 50-point gap (which is a ridiculous value, by the way. Maybe even 5 would be enough to achieve chess-like draw rates in top pro games) is pointless. Of course you can make Go unable to make fine distinctions if you blur the line enough; so what? That's true about any game. Go as it exists *is* able to distinguish even very small differences in skill between players, which makes it "deep" (at least by the definition used in this discussion, which I happen to think is a very good one). On Tue, Oct 26, 2010 at 4:45 PM, <dave.de...@planet.nl> wrote: > I think draws play a big factor in the length of an Elo-scale. > Chess has a large draw margin, go has a small one. A perfect chess game > probably ends in a draw, so the stronger the chess players, the harder it is > for the stronger player to secure a win. This effect compresses the upper > part of the chess Elo-scale, reducing the complexity number. Go would have > less levels if we enlarged the draw margin by adding a rule that a game is a > draw when the score difference is less than 50 points. But then again, that > kind of rule would make the game a lot less easier to play. > > Dave de Vos > > ------------------------------ > *Van:* computer-go-boun...@dvandva.org namens Ashley Griffiths > *Verzonden:* di 26-10-2010 23:43 > *Aan:* computer-go@dvandva.org > *Onderwerp:* Re: [Computer-go] human complexity measure of games > > I am pretty sure the original article came to the conclusion that Poker > was > a 1 or 2, and backgammon was a 4. Been a while since i saw it, but I think > those were the numbers. With checkers at 8, chess at 16 and go at > approximately 40. So its not like the authors had a problem with poker > having a low complexity. > > The 40 rating for go is representative of a pro player versus an absolute > begginner and gives the beginner something like 8E-23% chance to beat a pro > (thats probably less than the chance of the pro dropping dead mid game :p) > > If poker had a 1 rating it says an absolute beginner has a 25% chance to > beat Dolly (if its 2 then its a 6.25% chance, and from that I am inclined > to > think its probably actually somewhere between the two, which based on the > definition would mean it is a complexity of 1) > > Wow that was a bit rambly, sorry about that > > --Ash > > ----- Original Message ----- > From: "Christoph Birk" <b...@obs.carnegiescience.edu> > To: <computer-go@dvandva.org> > Sent: Tuesday, October 26, 2010 10:25 PM > Subject: Re: [Computer-go] human complexity measure of games > > > > On Tue, 26 Oct 2010, Nick Wedd wrote: > >> I believe that I am much stronger than IdiotBot. IdiotBot makes its > >> moves at random. But it is not impossible that IdiotBot will beat me, > by > >> luckily happening to make good moves. > > > > You are arguing using a real edge case. More realistically, > > if I (3 kuy) play a pro I will not win a game in my lifetime, > > even if we play every day. > > If I (_not_ a poker pro) play 100 sessions of poker > > (say 4 hrs) against Doyle Brunson, then I am confident to > > finish ahead a few times. > > > > Christoph > > _______________________________________________ > > Computer-go mailing list > > Computer-go@dvandva.org > > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > > _______________________________________________ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > > _______________________________________________ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
