Author: Robin Smith
Date: 22:06:39 12/13/00
Go up one level in this thread
On December 13, 2000 at 23:07:25, Michael Neish wrote: >On December 13, 2000 at 19:26:20, Robin Smith wrote: > >>The likelyhood of chess being a win for white, with perfect play from both >>sides, is very low. This is born out by computer-computer games, where the draw >>percentage increases as search depth/time increases and also by the fact that >>super GM vs. super GM games have a much higher draw rate than games by lower >>rated players. In both cases as playing strength increases the percentage of >>draws also increases. Strong evidence that chess is a draw. Also, in decisive >>games one almost invariably finds that one of the players either made a mistake >>or mistakes, or at the very least took unneccesary risks. And most strong >>chesspayers believe a perfectly played game should end in a draw. When Kasparov >>was once asked why he did not win a particular game he replied "Chess is a draw, >>no?". >>So, there will NEVER be any computer opening database, no matter how big, where >>computers (playing white) think the 1st position out of book is always at least >>+2.50 for the computer .... unless the computer has a seriously flawed >>evaluation, in which case it will hardly mean chess is solved. >> > >With all due respect, the points you make in your post, if correct (and some I >think are debatable), merely suggest that Chess might be a draw, and do not >prove it. They do not justify the strong conclusion you make at the end. Although I will be the first to agree it is not proof, the evidence is strong, and I believe it supports a strong conclusion .... chess with perfect play IS a draw. This is not proven, but it is the most logical conclusion from the data. Here is some data from Chessbase's Mega2000 database rating of BOTH players # games in mega2000 % of games drawn >2600 ~12,000 56% 2400-2600 ~169,000 52% 2200-2400 ~176,000 39% 2000-2200 ~36,000 33% Similar results have been posted here for comp-comp games, where a program would play at progressively deeper search depths, as search time/depth increased (for BOTH sides equaly) the % of drawn games increased and the percentage of white wins decreased. >The draw rate is reflected in a player's rating. If two player's ratings are >close, then of course they are going to draw more often than not. See above for what happens when two closely rated players at ~2100 are compared to two players closely rated at >2600. >At any rate, >super-GMs might not be seeing anywhere near far enough over the board for the >outcome of a position to be proved beyond a doubt. I'm not claiming that super-GMs are perfect. Only that they are CLOSER to perfect than 2000 rated players. And that as one gets closer to perfect, the draw rate goes up .... for BOTH humans AND computers. >It was my understanding that there is no evidence of a decline in the rate of >improvement of a computers' play with increasing ply depth. I.e., a 7-ply >searcher is expected to have the same rating difference compared to a 6-ply >searcher as a 13 to a 12, etc. I think this is the established view, although >I've also heard of (but never directly read) an opposing view. There was a very good study posted here by a fellow whos name escapes me right now (starts with letter H .... Heinz perhaps?). It showed some evidence for a decline in the rate of improvement in computer play with increased depth ... but even STRONGER evidence for an increasing number of draws. >Maybe someone >who knows more about this can confirm or deny it. I think the reason why the >draw rate increases with increasing search time is that the search time per ply >increases exponentially. So if you are playing at 40/120, chances are that both >programs will be searching roughly to the same depth, and will not be outdone >tactically. If you are searching 3 seconds per move, then the probability of >missing tactics that your opponent misses, or vice-versa, increases. Yes! My point exactly. The fewer tactics that get missed the more likely the game ends in a draw. Please note that what there DOESN'T seem to be any evidence for is: as depth increases, and mistakes correspondingly decrease, more wins by white are seen. In fact the opposite occurs, fewer white wins. What does this strongly imply?? To me it implies that black losing games comes from black making mistakes. And when black makes a mistake it is often fatal. >Just my opinion. I also tend to think Chess is a draw, but piling up anecdotal >evidence doesn't prove it, whatever Kasparov might say. I'll agree, Kasparovs statement is anecdotal. See above for hard eveidence. One other piece of subjective/anecdotal data, I am soon to be a 2 time winner of the US correspondence chess championship. And I have never played a game with a decisive result where it was not possible to find at least one mistake by the losing side. In fact even many drawn games have mistakes, they just aren't (for whatever reason) enough to be fatal. Chess is a draw. Robin Smith
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.