Author: Ratko V Tomic
Date: 19:27:59 10/17/00
Go up one level in this thread
> I don't think you should presume Bob is ignorant (does >not know/has never heard of) of what you layed out though. I don't, we had gone over this before. He just doesn't think there is much to Botvinnik's approach (or anything based/inspired by it). Bob is right that most new ideas, trends, fashions, are worthless nonsense, flash in the pan. But, it is also true that old ways always do get displaced by the new ones. And the guardians of the old ways, the "old guard," are the most uniquely qualified to miss the new idea which will take off (and especially right after it has taken off). > I am not certain you can simply present certain patterns that >are precise enough to be useful but necessarily vague to cover > enough situations (so as to be practical), and proclaim it is > worth a piece. But neither is the "worth of a piece," say 3 pawns, in itself more than an empirically supported construct. If, for example Gambit Tiger, with its "+3 pawns" evaluations in positions where Crafty counts wood as equal, manages to perform as well or better than Crafty, that is as good a support as any other commonly accepted belief has (such as "worth of a piece"). While GT's computation of a "worth of the king-side attack" may be more complex (and I am sure it is) than computing the "worth of a Bishop" (a trivial lookup), this computational complexity alone doesn't make it less tangible or less reproducable or less valuable. They're all humanly constructed models, a guesswork, however widely believed, whose sole justification are the results. So, adding numbers attached to the pieces of wood isn't inherently any more accurate decision making mechanism than any other conceivable computation which over the board performs equally or better. The simplicity and the long lasting habits only give illusion of counting something tangible, accurate. In fact, the conventional programs are still merely adding the "faith" points (other than when they stumble into a checkmate or a table-base position). > As for non-symmetric weighting, >isn't that rather double-edged? I mean, do you want your program to understand >(with the proper weights) about attacking, but nothing about being attacked? In >a double-edged position, such a program would always be wildly optimistic as it >would only take into account its possibilities but not its opponent's. Or am I >missing something? If you observe small children looking at the books, they're perfectly happy looking at them upside down. Same with left and right shoes. Or drawing with left or right hand. The gradual crystallization of asymetry in their cognition is a sign of more advanced processing. Virtually all brain functions are asymmetric (with left and right hemisphere specializing in different types of processing). Similar process of this type (advancing through creating asymmetries) is social specialization. Here again, even though "all men are (supposedly) born equal," they don't all pick the same profession or do little bit of everything. The more advanced society is the more specialized its members are. Or look another complex system, the embryo. Its initial symmetries are repeatedly broken as it unfolds toward more advanced state. Or you can go as far back as creation and evolution of physical universe -- again you see that even at the level of purely physical laws, the present day richness of their manifestations unfolded via successions of symmetry-breaking (phase) transitions. Abstracting for one more step, symmetry is less efficient, more constrained, more information impoverished modus operandi than asymmetry. Asymmetry, and its special cases, such as functional specialization, allow for more efficient utilization of resources, allow system to absorb more information, thus to model better its environment, since they allow for more distinct internal states of a system. Humans and animals find symmetry pleasing or pretty since it taxes nervous system less, it is simpler, less surprising, cheaper to model i.e. everything you wouldn't want your chess program to be. Returning back to the evaluation functions, in the absence of other indicators, the symmetric evaluation is an a priori sign of a more primitive chess program. Of course, that doesn't mean it is impossible to write poorly performing asymmetric evaluator. But if one can pick between the best each kind could be, the assymetric one will be more advanced.
This page took 0.01 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.