Author: Mogens Larsen
Date: 12:05:42 02/07/03
Go up one level in this thread
On February 07, 2003 at 14:20:53, Chris Carson wrote: >I think you have one different question here, so I will re-state it and then >talk about it. > >Question: What is the Elo Rating of a Chess Engine (no book, tablebases, ...)? > or How many points do opening books, TB, ... add to a program? That wasn't really related to the original discussion, but I added the possibility because it's the only (IMO) relevant scientific question. This match (as well as others) have been a question of compensation from either perspective for the sake of fairness (and public interest). If we look at the Kasparov-DJ match it becomes apparent. The human preparation against a not too old previous version is compensation for not having a history of games to analyze. The engine opening book is compensation for the analytical baggage of human chess history. A similar reasoning can be made of endgame tables and any other compensating measure. However, science isn't about fairness. A human being and a chess program are two different objects. More or less random meddling compromises objectivity. The notion that a chess program requires a book to compete is in reality a derogatory concept. I imagine the initial idea was to avoid ugly games and easy human wins. We're beyond that. It can play chess on its own. The only real question is whether a measly engine (and a learning function) can defeat mankind without compensation and artificial aid. As I mentioned before some may disagree with that idea of science. That's okay with me. I doubt it'll attract commercial attention, but we'll end there eventually. >The spurious variables (motivation, conditions, ...) that you suggest may exist, >however, I have done some analysis on this and find no significant difference >(at a 95% confidence level) using ad-hoc analysis and SPSS (stats program) data >analysis. It could be there, but it is unlikely and funding for additional >tournaments, matches and analysis is unlikely. The events would have to be set >up so that a very large difference in motivation, conditions and the other >variables were present. Small and medium effects can be measured with existing >data and there was no significant difference. With all due respect, I doubt your capable of extracting any sensible data concerning motivation and conditions. At least not without a lot of assumptions. I don't think you're capable of forming those assumptions. It has no scientific basis in terms of objectivity, so the results are meaningless from my perspective. No offense intended. There are still the question of preparation and experience, which I deem more important. Parameters that have developed since the infancy of the human vs. chess program encounters. Of course the usual "humans almost never play against chess programs, so they're not optimally prepared" isn't credible in this computer age IMO. But claiming that they study intensively is also false IMO. There's room for improvement guided by the experience of others, ie. what general plans worked for Smirin, Kramnik and Kasparov, and what didn't. Regards, Mogens
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.