Author: Robert Hyatt
Date: 07:08:51 06/13/01
Go up one level in this thread
On June 13, 2001 at 02:00:44, odell hall wrote: >On June 13, 2001 at 00:00:36, Robert Hyatt wrote: > >>On June 12, 2001 at 22:52:19, Fernando Villegas wrote: >> >>>Perhaps the debate could be clarified - well, I will try to do so- a little bit >>>if we discern between brute strenght showed by results and the quality of a >>>game, mainly grounded in positional understanding. In this last aspect I cannot >>>but to agree with you Bob; programs are not GM in quality of the game they >>>offer. That's clear. Nevertheless, in terms of results, maybe the other guy has >>>a point. Probably this is a tricky issue because chess games usually ends for a >>>tactical reason a lot of times, so the positional aspect is hidden. A computer >>>could be positionally busted and even so win the game because in the last minute >>>the human side missed a blow. It has happened to me -an to infinite human chess >>>player- thousands of times and I think it happens all the time to GM, although >>>in a lot more higher ground of finesse. So, in the paper, measured from a point >>>of view of "pure" chess, programs ar at most 2100 or 2200 players; in the realm >>>of facts, of rsults, they win enough times to be considered, I believe IM >>>strenght players, sometimes even more. >>>Fernando >> >> >>I don't disagree there at all. IM is believable in a way, because the >>difference between an IM and a GM is a _wide_ gulf of knowledge and experience. >> >>But you can't survive on tactics _all_ the time. The opponents simply stop >>letting the game get tactical... > > > Hi Bob > > Okay, i think i read a post of yours where you said Computers are not >Grandmasters, because a Real human Grandmaster would never make some of the >moves you have seen computers make, Am I correct?, is this your view? But can't >one use the same thinking to say the a Human Master would not make some of the >imbicile computer positional moves that computers make, so then we could say >computers are not even masters? Accoridng to your reasoning would not this >assumption be correct? For some programs, yes. However, for a "master" I am sure that at least a few programs know as much or more about strategy as the master does. Perhaps a bit less with respect to king attacks, perhaps a bit more with respect to endgames. For IM level players, things get more unstable. The IM players know more about strategy than the programs, but not necessarily by a wide margin. The programs are more accurate "calculators" than the IM, so this "fit" becomes very erratic and difficult to assess. For GM level players, they know more than the programs by a huge margin, and they are tactically as good or better in the right kinds of positions. IE a human GM can search 5X deeper than a program in positions that support long-term kingside attacks. The human GM has an incredible selective search that is hard to control there. But then the GM plays a computer, and finds himself in tactical positions unlike any he has seen before. In 1984 Cray Blitz (on a single processor fairly slow Cray 1S machine) entered the US Open speed chess championship in Pasadena, CA. We were invited to play an exhibition game against a USCF-rated 2400+ player, which we happily decided to do. We reached an incredible position and won pretty easily. But it was a position like I have never seen in a real game between two humans. Pieces were pinned. Down files, on diagonals, pieces were defending other pieces in very complex (and to a human very dangerous) ways. The 2400 player told me after the match "I simply had never played a game that got that 'wild'. I couldn't keep up with all the pins, overloaded pieces, xray attacks, desperados, and so forth..." That was a classic. We should _never_ have beaten this guy, but he let the program steer the game into a position where he was very poorly prepared to go. And he paid. He didn't _have_ to let the position get out of control, but he did. After the game he pointed out several places where he should have simplified and kept things "in a saner configuration" (his term). I believe that today, GMs play the computers all the time, and they notice all the positional weaknesses that they are sure they can exploit in real games. But when the rubber meets the road, they are thinking about exploiting the positional weaknesses without taking care of the "tactical sanity" of the position, and they let things spiral out of control because they _know_ "they can beat this thing easily." And they don't. With my baseball analogy, a good fastball pitcher is going to do his best to throw his "heat" past every batter. And even when he gets one slapped over the fence, he is _sure_ that his control was just a bit off, but the next time he faces that batter he will fan him by throwing a fastball by him. If he gets it slapped over the fence enough, he will finally decide that he has to throw something else to that batter, and then he _does_ strike him out. But until he realizes that what has worked in the past (against other humans) won't necessarily work in the present against this "electronic batter" he is going to watch the batter air-mail the ball over the center-field fence. Once he realizes that this batter can't hit fastballs, that batter can't hit a curve, another batter can't hit a slider, he becomes _much_ more in control of how he is going to do. Even against the electronic batter that can smash the 100mph fastball, but which can't touch a slow, looping pitch that drops in right over the plate. Ugly pitch, against everything the pitcher wants to do, but the _only_ way to win. If the humans had no "ego" the computers would have terrible problems. But the humans do, and the computers benefit.
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.