Author: Bob Durrett
Date: 13:03:55 02/14/04
Go up one level in this thread
On February 14, 2004 at 12:13:06, Stephen A. Boak wrote: >On February 13, 2004 at 18:15:55, Bob Durrett wrote: > >>I want a "black box" which looks like, smells like, and feels like a human >>chessplayer. I don't care what's in the box. It could be a computer emulating >>a human. I wouldn't care. > >>Someday, robots may be indistinguishable from real humans. To make such a >>robot, it must be programmed to make the mistakes humans make as well as the > >>Bob D. > >You are looking for a program that passes the Turing Test, with the added >stipulation that it is indistinguishable from an 'average' or weak human player. Yes, but I want MORE. [never satisfied] My imagined ideal "black box" chess machine would be able to play chess at any human level from absolute beginner to World Champion level. The user would either give the "black box" a command which specified the desired level of play or the "black box" would infer the user's level of play from the user's games and then adapt to it [as Fritz tries to do today]. In this latter case, the user's games could either be games against the "black box" or a database of games played in the past by the user. There are different human styles of play from the aggressive sacrificial attacker to the quiet positional types. [Tal and Petrosian, for example.] In a conventional over-the-board tournament, one has no control over the type of opponent one must face. For this reason, if the "black box" is to be used as a training partner in preparation for an upcoming tournament [in which the user plans to participate] then, ideally, the black box would randomly select several different human styles of play to emulate. Similiarly, if the tournament is to be Swiss System, then the "black box" could emulate the variation in ratings to be expected. On the other hand, if the "black box" is to be used as a training partner for an upcoming serious rated match then the user should be able to input data to the "black box" permitting the "black box" to emulate the style of play of the upcoming match opponent. The user could either enter a command by selecting from several different available human personalities or the user could input a database of games played in the past by the opponent and let the "black box" determine the proper personality. Similarly for an opening repertoire. Keep in mind that at least 50% [estimated] of the chess-playing population worldwide performs at or below the 1500 level. For a commercial chess-playing program to be marketable, it must meet the needs of most of the likely future customers. For this reason, making the "black box" able to emulate 1500 level human play would be the right thing to do. > >Few strong programs could easily pass the Turing Test, in my opinion, if enough >example games & moves are reviewed by a strong human player. Eventually all or >most all programs would be detected as non-human. True today. As you may have already guessed, I'm not satisfied with that. I want ***better*** chess-playing programs, and that doesn't mean programs which ALWAYS play at the 3000 level. Maybe the day will come when it will be difficult to distinguish between human play and the play of a chess-playing maching designed to emulate human play [i.e. my desired "black box"]. > >A weaker program would, by subtrefuge, try to emulate a weaker player, but the >types of mistakes now made by human players is not identical to the types of >mistakes made by today's programs (even intentionally weakened). True today. But don't underestimate the resourcefulness of motivated programmers! [especially since some are supposed to be geniuses. : ) ] > >Suggested solution--a statistical study Agreed. However, the statistical study could be "computer-aided." There is no need to do things the way they were done in the past. These are modern times and computers can help a lot if allowed to do so. For example, a collection of human game scoresheets [preferably with time-used data] could be input into a computer and the computer would then perform the required statistical analyses. The chess servers, like Internet Chess Club, have many players playing all the time. These servers are the most obvious candidates for performance of the data collection task. PGN records could be stored with relevant data such as the time controls. We all know that people play speed chess differently than slow chess, so the time controls are a PARAMETER of the study. >of the probabilities & types of mistakes >made by weak players (opening, tactical, positional, sheer oversights/blunders) >could be used to create a program that makes similar probabilistic classes of >intentional machine errors in the attempt to emulate weaker human play. > >The probability of dropping a minor piece may be greater, for example, than the >probability of dropping the queen (weak players _may_ keep a better eye on their >queen than their minor piece; a queen may have more escape routes than a bishop >or knight and thus be less likely to be trapped). > >Pawn drops may be most common, whether due to accident or intentional but >improperly calculated in a tactic or gambit assessment. > >Possibly a piece drop is very unlikely in the early opening 10-12 moves, but >much more likely in the middle game, and not very likely in the ending, so an >even distribution of errors throughout a game would not appear very human. > >--Steve I think we are singing from the same sheet of music! : ) : ) Bob D.
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.