Author: Zheng Zhixian
Date: 09:00:02 10/27/05
Go up one level in this thread
On October 27, 2005 at 10:10:39, Uri Blass wrote: >On October 27, 2005 at 09:41:46, Zheng Zhixian wrote: > >>On October 27, 2005 at 05:30:59, Uri Blass wrote: >> >>>On October 27, 2005 at 04:41:49, Alessandro Damiani wrote: >>> >>>>On October 27, 2005 at 04:39:22, Uri Blass wrote: >>>> >> >>>It will be interesting to know what rating can the best humans achieve against >>>humans when they are forced to play like a computer with definitive algorithm >>>(of course their oppoents should not know the algorithm because playing like a >>>computer is enough disadvantage) >> >>Sounds like you want a human to play a full game by acting as a computer? >> >>If the former such a test did happen (though the human knew it was playing a >>simulation of a chess program). In fact , I seen to recall reading that Alan >>Turing did it by hand simulating a simple chess program back before there werent >>real computers that could run the program. I think he did more than 1 ply of >>course. > >Yes but I think that he used more than average of 3 minutes per move. >I also think that it may be possible to improve his evaluation. I think evalution has to be simple, espically if he wants to do a deep enough search. If computers are slowed by complicated eval, humans are even penalised more! Of course better evalution doesn't mean more complicated to use, but I think for humans simulating computers, the problem is reaching decent search depths to avoid tactical blunders. Even with a simple alpha beta, material only evaluation, how deep can you get in 3 minutes? Hmm I just realised you as a human don't get transposition tables :). Would you do iterative deepening? Would it depend on what type of tools you had? What did turning use? Paper and pen only? I think the best chance for a human simulating a machine is to keep eval simple, like in the original turining program and hope search is quick enough to avoid stupid tactical errors. There's no point in a "accurate eval" that leads one to play proper positional pawn lever moves in different openings but fail to a fork because you didnt have time to calculate far enough. I guess that's the difference between human eval and comp eval, human eval is almost free but search is the very very expensive part. >I also do not claim that doing 1 ply search is the best strategy and maybe some >rules of selective search that allow extensions in some lines may be better(for >example a rule that say not to analyze king moves during the opening unless the >king is under attack). Hmm, isn't that exactly what a human does? You might as well ask the human to play chess normally. :) >>If you restrict it to one ply, it would simplify things a lot. Then the question >>would be if a computer is doing only one ply searches (no qsearch, no >>extensions, just one ply), can you beat it? I've never tried, because most programs that you can force to 1 ply, do q search+extensions also I think so it's much stronger than a plain 1 ply search. Strength improves quite a bit from a plain 1 ply search, picks up tactics i suspect. >I certainly can do it but a beginner who play his first games may fail in doing >it. > >I already saw weak correspondence games when the sides that play the games can >use 3 days per move when I am quiet sure that the players who played them are >going to lose against a simple algorithm. > >If you look at games of players with rating 1100-1200 that are not the weakest >players in the site in the link you may find a lot of stupid mistakes. Dropping pieces like pieces were going out of style? >http://gameknot.com/players.pl > > >Note that I do not suggest one ply with no qsearch and every algorithm that at >least one human can follow at tournament time control is accepted. My prediction if you force a human to follow strictly to a chess program algo, the result would be very very bad, unless he practised a lot before hand and/or had a lot of familarity with the program. But that's not surprising. >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.