Author: KarinsDad
Date: 10:32:50 01/18/99
Go up one level in this thread
On January 18, 1999 at 12:32:31, John Coffey wrote: >On January 18, 1999 at 01:59:39, Laurence Chen wrote: > >>I'm sorry, but I have to disagree with your statements. It's like asking a GM >>rated 2500+ to play at 1400 ELO which is impossible to accomplish. > >Of course you can make a 2500+ program have an option to play at 1400 >elo. There are many ways to do it. (Some programs don't even try.) If it >can play at one second per move then why couldn't it play at a tenth of a second >per move or even less? Another way to do it is to increase the randomness. >(i.e. don't differentiate between moves below a certain threshhold. That >threshhold could be made pretty large if necessary.) > >I still like my idea of having levels where level 1 looks at 30 possibilities, >level 2 at 60, level 3 at 120, level 4 at 240 etc. > >Best wishes, > >john coffey John, Although I agree that any of the methods you describe above will force a computer to play weaker moves, I do not think that simplistic approaches such as these will enable a program to play human-like 1200 level chess. The approaches above result in a more randomized level of play, not a weaker and more flawed level of play. I've analyzed quite a few amateur games (both my own and from others) and have noticed that weaker players do not play random moves. They play what to them appears to be the strongest move and quite often, these moves are identical to what a strong computer would play. However, quite often, these moves are not identical. One mechanism that a chess program could emulate human play is to not make one move hangs of pieces, but rather 3 move or 5 move hang of pieces. I've noticed that in CM6000, you can be playing a 1400 rated player and the program will do a 1 or 2 move hang of a piece. Although this occassionally happens in 1400 level chess (and I'm talking slower speeds here, not G5), it is more rare than you would think. A computer analysis of lower rated games could probably result in a statistical evaluation of the range and frequency of errors made by weaker players at various portions of the game and the program could attempt to emulate that frequency and range of errors. Remember that on a good day, a 1400 player can beat an 1800 player and on a bad day, can lose to a 1000 player. The program could also be designed to have "good" and "bad" games as well. Something to chew on :) KarinsDad
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.