Author: John Sidles
Date: 05:22:54 02/28/06
Dear CCCers I'm much more a "lurker" here than any kind of chess expert, but I would be interested in learning from CCC's programming experts mroe about how (if at all) chess programs evaluate, not the score of the board, but the score of the move tree. The point being that maybe Rybka's surprising strength comes not from its knowledge of chess, but from a superior assessment of the branching and topology of the move tree. E.g., suppose white and black both look 15 moves ahead, and they foresee that with optimal play, the board score will be roughly equal. But even if the score is equal, if black's play is essentially forced, while white's play has many strategic options, then white has a huge advantage. In military language white's advantage is called a "favorable strategic landscape". So it is clearly important for any chess program to steer the game so as to achieve a favorable strategic landscape. There are at least two ways to do this. The first way is a highly tuned evaluation function, i.e., knowledge that rooks have a higher weight than knights. The second, more subtle way, is achieved purely by examining the branching and topology of the search tree, i.e., determining whether the strategic landscape with a rook in it is more favorable than the strategic landscape with a knight in it. This latter technique, in principle, requires no chess knowledge. A good program will use both methods. So I would be very interested to learn more about how chess program authors assess their search tree. SIncerely ... John Sidles
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.