Author: Bruce Moreland
Date: 17:13:03 07/28/99
Go up one level in this thread
On July 28, 1999 at 16:10:47, Peter Hegger wrote: >On July 28, 1999 at 12:27:56, Bruce Moreland wrote: > >> >>On July 28, 1999 at 11:18:19, Peter Hegger wrote: >> >>>The following is an excerpt from "How computers play chess" by Levy and Newborn. >>> >>>"In 1959 a Canadian program was demonstrated at the University of Toronto. It >>>was written by Frank Anderson, an International master and Bob Cody, and it ran >>>on an IBM 605 computer. The program dealt only with simple pawn endings (the >>>most complex was king and two pawns versus king and pawn).The programmers >>>devised a unique strategy that reportedly enabled their program to play these >>>endings perfectly. Their first version could cope with more than 180,000 >>>positions, a figure that was increased in later versions of the program. >>>When the program was demonstrated at the Canadian Conference of Scientists it >>>played against more than 50 different opponents, each of whom was allowed to >>>choose his own starting position, given the small number of pawns. In each case >>>the program apparently played perfectly. Unfortunately, the strategy that >>>enabled the endings to programmed successfully was never documented, and >>>Anderson even confessed to me in the early 1970's that he couldn't explain >>>why they worked!" >>> >>>Has anyone else ever looked into how this program worked? If a program could >>>handle KPPKP perfectly on a 1959 vintage computer then imagine how it would >>>perform on todays machines. Maybe 5 or 6 piece (or more) EGTB's would be not be >>>necessary if the program had the ability to figure all these positions out >>>perfectly. >> >>I am skeptical. >> >>bruce > >Skeptical that it actually happened, or skeptical that such an algorithm could >be duplicated? >Peter An important side-line in history of computer chess, particularly in the late 70's and early 80's, involves attempts to produce perfect, or at least extremely good, play in certain basic endings. I can't recall reading an article where someone expressed confidenced that they'd solved any of these domains via knowledge, other than Don Beal's KP vs K algorithm. So I think that if someone had produced a great KPP vs KP player, it would have been published or at least investigated thoroughly. Another issue involves apparent perfect play. Give 50 people who are not necessarily even chess people the opportunity to select a starting position in an elementary ending, and have them evaluate the play of their opponent, and who knows what they will say. So that's why I'm skeptical. As to whether it is practical to heuristically simulate perfect play in 5- and 6-man endings, the 5-man endings have been done via tables, so anyone wants to try has a perfect opponent that they can use to verify their heuristics. I think that this would be a great task for a student. I believe that if someone were to invent a perfect play heuristic even for some of the 4-man cases, notably KR vs KN, it would be a significant improvement over what went before, and would could probably be turned into a major paper. bruce
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.