Author: Ratko V Tomic
Date: 06:21:53 08/11/99
Go up one level in this thread
> You are misunderstanding.... it is _not_ a mobility issue. > Because at the next ply, the best move is to take the > promoted piece, no matter what it is. But if the promotion > is a check, that drives the entire line one ply deeper > than it if is not... No, I understood why a typical program makes such decision. You gave a good initial example. What I was saying is how a tiny bit of common sense reasoning about causes and effects, operating at the level above the tree searcher code, could help program realize that its choice =R instead of =Q cannot be right (that it is an artifact of the particular imperfect criteria for the search cutoff). What the chess programs could use is a hierarchical decision system, such that a higher level decision modules can guide and check on the lower level modules (which tend to get buried in the their trees, unable to see the forest). That's why human organizations are designed hierarchically, since the low level guy cannot be counted on to see well the forest while busy with his little details. Current programs, seem to throw in the so-called "knowledge" down at the low level, at the leafs of the search tree, which means it has to be primitive. The decisions are made by the hypertrophied tree traverser, who seems to forget that it is traversing only a tiny fraction of the full tree. It behaves as if it knows it all. In real world, this would be like some data processing nerd in a company making decisions for the whole company because computer says so. Botvinnik (among a few) had the right idea on structuring the decision making and at which level the chess knowledge and reasoning should go. Unfortunately, his hardware was much slower and had less RAM and disk than my kids' spelling checkers or "game boys", so they couldn't implement it well. (His dull exposition style didn' help either in bringing in others with resources.) Hopefully someone will eventually pick up where he left off.
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.