Author: Walter Faxon
Date: 23:50:33 10/16/03
Go up one level in this thread
On October 16, 2003 at 15:25:43, Steven Edwards wrote: [snip] > >Computer chess was supposed to be the Drosephilia of AI. If so, CC theory is >still in the larval stage and I don't see wing buds popping out any time soon. >Where are the CC planning engines? Where are any general pattern recognition >algorithms in use? What program has real machine learning? Which programs are >adaptive and can re-write better versions of themselves? How many programs can >converse in natural language and answer the simplest of questions as to why a >particular move was made? Where are the programs that can improve based on >taking advice vs coding patches to the Evaluate() function? > >And the big question: What has CC done for AI in the past thirty years, and what >can it do for AI in the next thirty years? > >Hint: Any remotely correct answer does not include the phrase "nodes per >second". Hi, Steven. Computer chess is no longer the Drosophila of AI. That honor has passed to computer/robot soccer, especially as exemplified in the RoboCup tournament. The reason is simple: computer chess has long been dominated by ever-more-efficient search of huge game trees, a topic which is now understood pretty well. Fast searchers have no time for construction of dynamic patterns or learning or generalization, or any deeper understanding of causality than refutation by alpha-beta. In practice the best programs on good hardware consistently defeat all but the very best humans, so why go off on a tack that might take decades to come to fruition? Better to find a trick that gives you an extra ply in some critical positions -- do that and you'll have a world-beater. For a few weeks, anyway. I've posted on this subject here in CCC before. To bring computer chess as a topic back into the mainstream of AI will require more than a lone researcher writing a planning or pattern-matching chess program in order to support a master's thesis. No, not even a dozen of them. Rather, it will require an organized competition that will reward computer chess performance _without_ full-width search. I've called it "Limited-Search Computer Chess" (LSCC). Limit competitors to 100 nodes/sec or even 10 nodes/sec (nodes not knodes). Just an annual competition (online or in conjunction with some other AI conference) and a very modest award, maybe just $1,000 US, might do the trick. Of course the performance would initially be far below that of good programs today. There would be a temptation to cheat using a fast searcher. That's another reason to follow the RoboCup model of releasing all source for the competitors after every tourney. The exchange of ideas _and_ code has already been shown to substantially improve the average robot soccer performance. Competition is fun. Classical computer chess gained substantially from it. LSCC could, too. And its need for pattern-matching, planning, learning, generalization, etc., could well bring computer chess back into the mainstream of AI. -- Walter P.S. My answer did include the phrase "nodes per second". Sorry.
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.