Author: Allen Lake
Date: 11:34:21 11/15/02
Go up one level in this thread
On November 13, 2002 at 15:09:14, Bob Durrett wrote: >To be completely fair about it, you must admit that there is much in the >present-day chess engine algorithms which is the same for every engine. For >example, there is all this lip service paid to "alpha-beta." > >My guess is that MOST, if not all, of the basic algorithm content of present day >chess engines is the same from one engine to the next. Correct me if I'm wrong. > >If you buy that, then you must admit that it's unfair talking about "using bad >algorithms" versus "using good algorithms." If all the chess engines are using >essentially the same algorithms, then ALL chess engines will be using "good >algorithms" [giving present-day algorithms the benefit of the doubt.] > >Differences between engines discussed here at CCC seem to me to differ in only >two important ways: >(1) Coding skills of the programmers [some are experienced professionals], and >(2) Maturity of the programs. [Crafty mature. Russell's still unnamed.] > >My perception is that the basic algorithms used do not change significantly from >one modern engine to the next, except as noted above. > It's probably more accurate to say that the basic _design_ of modern chess engines do not change significantly from one to the next, rather than that the basic algorithms don't change. Engines may differ in their use of various types of brute force search techniques or in the concrete implementation of such techniques, and those differences, slight as they may be in some cases, account for most of the variability in the results that those engines get in their play. For example, given that elapsed time is an important factor in the game of chess, an engine that finds ways to save time without compromising its quality of play will probably have better results (match/game scores) than an engine that does not contain those methods. Remove elapsed time as a factor in the game, and many of the techniques used today might lose some value to the engine programmer. Additionally, you can think of techniques such as opening books and endgame tablebases as "pre-digested" searches. The engine sees that the current board position matches a position in either the opening book or the endgame tablebase and it plays the recommended next move. That recommendation is based on some kind of predetermined evaluation (search), such as master analysis or master play (for openings) or examination of all subsequent positions (endgame tablebases). Engine programmers can then take this information, convert it to a format that the engine can read quickly, and program the engine to access this information in real-time. Once again, the ability (or lack thereof) to use this information causes differences in the play of different engines. For the past couple of decades, the pace of increase in computing power (i.e. faster processors, more and faster memory, multi-processing, etc.) has rewarded the developer of the brute-force engine design. As machines get faster and have more memory to use, searches go deeper and wider as needed in the same length of time, which improves the apparent level of play. At the risk of opening another large can of worms, what I think you are asking this community to think about (in this post and others you've posted here recently) is whether the current state-of-the-art in engine programming is on an optimal track. In short, are we beginning to reach the point of diminishing returns in trying to improve brute-force search techiques and should we lessen these improvement efforts in favor of a new programming paradigm for chess? For what it's worth, I think that the next great leap forward in chess programming (or game programming in general) is probably going to come from someone who finds a new way to think about the game. Once that new paradigm is demonstrated and understood, the cycle will begin again -- incremental improvements on known techniques, new derivative ideas, better and faster hardware, etc. And there won't be any doubt whether computers are stronger than humans -- not that there is much doubt for 99 percent of us now ;) Maybe not in my lifetime, but it will probably happen.
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.