Author: Mark Young
Date: 02:52:11 10/19/04
Go up one level in this thread
On October 19, 2004 at 05:08:18, Uri Blass wrote: >On October 19, 2004 at 04:50:16, Tony Nichols wrote: > >>On October 19, 2004 at 03:52:10, Uri Blass wrote: >> >>>On October 19, 2004 at 02:56:31, Tony Nichols wrote: >>> >>>>On October 19, 2004 at 02:31:54, Roger D Davis wrote: >>>> >>>>>Several years ago, back before RGCC even existed (before Rec.games.chess split), >>>>>computers were lucky to beat human masters. Then the masters fell, then the >>>>>international masters, and now computers are as good as most GMs, maybe as good >>>>>as all but the top GMs, and maybe somewhat better than the top GMs. Who knows. >>>>>The point, however, is that progress is indeed being made, and it doesn't show >>>>>any sign of abating. >>>>> >>>>>My questions are these: Will computers ever become so strong that GMs will feel >>>>>lucky even to draw? Will the percentage of GM versus computer draws slowly >>>>>diminish, even among the top humans, so that computers will someday completely >>>>>and totally dominate? >>>>> >>>>>Remember...chess isn't a solved game. Perhaps white always win. So as computers >>>>>improve, they should begin to win more and more often as their strength comes to >>>>>approximate perfect play. But even if white doesn't always win, it may >>>>>nevertheless be that if the 2nd best move is made in any position, that side is >>>>>lost. Maybe perfect play can only draw and anything else loses. And just which >>>>>side do you think might make the 2nd best move...the human or some future >>>>>Quantum-computing beast? >>>>> >>>>>Another reason to believe that eventually even the strongest humans will be on >>>>>the losing side: Recently, it was posted that as computers have become faster, >>>>>programs authors have actually been REMOVING knowledge from their evaluation >>>>>function. In other words, deeper searches are better than explicit knowledge, >>>>>this presumably because chess has proven to "consist" more of combinatorial >>>>>tactics than of positional strategy. >>>>> >>>>>Accordingly, it would seem that the humans are the ones with the "horizon >>>>>effect" (Surprise!!), meaning that the combinatorial tactics that computers >>>>>handle quite nicely just doesn't reduce as much to positional rules as we might >>>>>like. Sure, humans might learn a few tricks from computers as computers continue >>>>>to improve, but once we've lost the lead, we won't ever regain it. What happens >>>>>when a computer regularly searchs to double the number of plies we see today. >>>>>Can a human GM even draw such a beast? >>>>> >>>>>Roger >>>> >>>> >>>> >>>> >>>> Hi Roger, I believe that most GM's can easily make a + score against the >>>>computers. >>> >>>In that case they could prove it in the israeli league when the result was >>>importnat for their teams and not only for themselves by beating humans >>>convincingly when the teams could choose the person to play against the computer >>>but they did not do it even there and score near 50%. >>> >>>I remember for example that Yona kossashvili lost against Fritz6 and we are >>>talking about human who did 6/6 in humans against machines in 1997. >>> >>>I remember that computers had bigger problems against weaker players and 3 chess >>>programs could only draw against arnold hasidovsky that has rating near 2200. >>> >>>Remember that computers today are clearly better than the level they were in the >>>time of the Israeli league so my guess is that most GM's cannot have positive >>>score against the machines. >>> >>>Uri >> >> Hi, Uri >> I'm not familiar with the Israeli league but I will accept your information. I >>think Human players understand chess programs better today than they did then. I >>would also say that if the engines had trouble with a 2200 player that helps my >>argument not yours. I agree that programs have gotten stronger but surley not >>500 elo. So if programs draw against master level players how can they be better >>than GMs? >> >>Regards, >> >>Tony > >I think that the reason is simple. > >The 2200 player played for a draw when the GM's wanted to win. >When you try to make a draw against computers your task is clearly easier. This is an area where progammers could do better. I think Crafty on ICC is the only program that knows the rating of the player it is playing. And can avoid many draws by playing a bit riskier to avoid draws or open the position. Humans do this all the time when playing weaker players. The stronger player will play a less then perfect move. Knowing that the weaker play can not exploit the move. This is done to avoid drawing, opening up the position, or going for a quick knock out by the stronger player. > >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.