Author: Steven Edwards
Date: 14:10:25 03/15/04
Go up one level in this thread
On March 15, 2004 at 13:33:57, Andrew Wagner wrote: >Steven? I saw your post after I posted mine this morning. Would you care to >comment on how you think my ideas compare to your own with symbolic? My guess is >that my ideas are somewhere in between yours and the classic alphabeta searcher. >Comments? My first comment is that, if you haven't already done so, is to read the papers by de Groot and others on the psychometrics of chessplayer memory (piece position recall), perception (eye tracking), and move selection (anaylsis of audiotaped stream of consiousness) processes. It's from works like these that we have some clues at to what happens inside a chesspaler's head and how behavior varies by playing strength. My second comment is about the prospect of evolving smarter programs by evolving the traditional A/B paradigm. First, a parphrase from Slate and Atkin' from thirty years ago: "How to improve Chess 4.x? ... One idea is to gradually introduce more sophisticated evaluation terms while retaining the overall search structure. As the evaluation becomes more reliable, one could then allow the search to be more trusting of its values." And: "A major problem today is that the tools we have are woefully inadequate for the task of describing and adding chess specific knowledge to a program." So, they suggested an evolutionary approach long ago, but it really hasn't produced the hoped for results. And the tools are still inadequate. My take on this is that an evolutionary approach could work, but it would require more effort than a start from scratch approach.
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.