Author: blass uri
Date: 03:58:10 01/31/00
Go up one level in this thread
On January 30, 2000 at 09:47:23, Robert Hyatt wrote: >On January 29, 2000 at 20:26:33, Christophe Theron wrote: > >>On January 28, 2000 at 08:51:14, Alvaro Polo wrote: >> >>>On January 27, 2000 at 13:51:01, Christophe Theron wrote: >>> >>>>On January 26, 2000 at 18:28:22, Dann Corbit wrote: >>>> >>>>>On January 26, 2000 at 18:23:50, Bruce Moreland wrote: >>>>> >>>>>>On January 26, 2000 at 18:10:10, Dann Corbit wrote: >>>>>> >>>>>>>IOW, more horsepower is a tough way to make chess programs play better. There >>>>>>>is also evidence (according to some) that the increase in speed has >>>>>>>*diminishing* returns. Hence, it may take a terahertz to get there. Don't know >>>>>>>of any material that could do that, not even a Josephson Junction. >>>>>> >>>>>>I think it's a great way. You just take a vacation, preferably a long one, and >>>>>>when you come back you make one call to Gateway and poof, free Elo points. >>>>>> >>>>>>Got an article that shows that the Elo curve flattens out with increased depth? >>>>> >>>>>Darn. I knew someone would ask that! I just read it somewhere, but I will have >>>>>to go and look for it now. >>>>>:-( >>>> >>>> >>>>Just my 2 cents: this "dimishing returns" theory is an urban legend. >>>> >>>>You are almost certain to have seen this demonstrated, you'll find people that >>>>will tell you they have seen this demonstrated, but you'll eventually find no >>>>proof of this. >>>> >>>>But everybody wants to believe it because it fits so well our common sense. When >>>>everybody in a group believes in something, it eventually because "real" for >>>>this group. >>>> >>>>Computer chess is CROWDED with legends like this one. >>>> >>>>The programmers that do better than their peers are those who do not believe >>>>these legends. >>>> >>>> >>>> Christophe >>> >>>I have a question for you. Do you believe that the "diminishing return for each >>>extra ply" theory is false for comp-comp only, or also for comp-human? >>> >>>Alvaro >> >> >>I make no difference (well almost) between comp-comp and comp-human. >> >>I don't believe that "dimishing returns" is of any pratical use for us. In >>theory there must be dimishing returns if you can search until the end of the >>game, in pratice it will have no influence on the way we program a computer to >>play chess. >> >> >> Christophe > >I think diminishing returns might be an issue vs humans. Because obviously >deeper searches never hurt a computer, but they can and do hurt humans, as the >STM memory of the human brain is not infinite in size. IE I think that as the >computer goes deeper and deeper, it will eventually reach a depth beyond which >the human just can't see. I don't think we can tell the difference between >outsearching our opponent by 5 plies or 10 plies. Either is a crushing >advantage. I see that sometimes humans simply outsearch computers. A master outsearched Rebel Century in the second round of the Israeli league and won a pawn. Rebel blundered and played in the following position Nf5 and the master found Rxf5 that is winning a pawn. Rebel fails low on Nf5 if it has one more iteration. [D]2r3k1/p1r2pq1/1p2pR1p/3bP1pQ/3P3n/2P5/P1BB1R1P/6K1 b - - 0 1 Rebel won the game but it could lose if it played against a better player. I found at home that Rebel could avoid the tactical mistake if it searched one more iteration. Another problem is that sometime you cannot see a tactical win if you outsearch the opponent by 5 plies but you can see a tactical win if you outsearch the opponent by more plies. > >Against computers, that extra ply will always be worth something. Not always. There are cases when one program wins not because of outsearching the opponent but because of better evaluation function. I agree that the difference from faster hardware can be smaller against humans because of the fact that humans try to play anti-computer style in order to go to positions when one ply of the computer is not going to help the computer but positions when one ply cannot help can be also in games between computers if you choose the right opening book and programmers can do it when they have to play with opponents with faster hardware. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.