Author: Ratko V Tomic
Date: 11:00:39 04/15/02
Go up one level in this thread
>But that information is completly subjective, hence, >worthless for any solid objective conclusion. If a GM plays a single game against a program and provides an in-depth analysis of the program's flaws, that info is much more valuable than a bare result, say, 72:28 you may get from hundred games. It is objective in the sense that most strong players would agree with it (they may add some more observations, of course, so it is not a complete information, but it is objective). Also, the info provided by a GM is in the form which is much more useful to the humans. It uses conceptual models which allow for generalizations, useful for other players (as instruction how to play against the program) and for programmers on how to fix/improve the program. Playing hundred games between two programs may tell you one is likely stronger than the other, but not exactly what is the problem or how to improve the program. To get anything barely comparable to the GM's evaluation you need to play two small variations of the same program and attribute the resulting score difference to the minor code change. Such process is vastly slower and still not as useful as a GM advisor giving an insight of the good and bad aspects of the change (from the game results you get only the net results of the sum of pluses and minuses of a given change combined with other unknowns, such as effects of choice of openings, opponent style, etc). Note also that subjective/objective distinction is itself subjective. The "objective" merely takes for granted some 'postulates' (which by themselves are 'subjective') and then deduces 'objective' conclusions. In particular, the ELO system assumption of memoryless stochastic process yielding the results of the games, while valid for coin tosses, is certainly invalid for learning programs and humans. Whether conclusion is 'solid' or 'soft'/'subjective'/'worthless' is decided not by the lesser or greater appearance of rigour (as when you leave the necessarily subjective 'postulates' unchallenged). It is decided by subsequent results. If GM makes certain recomandations and an opponent taking advice starts winning more, then the model he is proposing is as valuable and as solid as anything else which may have an appearance of mathematical conclusion and results in the same degree of improvement.
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.