Author: Robert Hyatt
Date: 20:59:44 09/29/00
Go up one level in this thread
On September 29, 2000 at 15:32:13, Uri Blass wrote: >On September 29, 2000 at 14:40:05, Fernando Villegas wrote: > >>Hi Bob >>Well, let me give you a more detailed an idea of what I try to mean. >>Suppose the program is already in a losing track. From then on what I say is >>that he should try to put the opponent in the more tricky scenaries, not jus >>looking for the best thoeretical move to do. How to do it: maximizing the chance >>of the opponent to blunder. Example. Supose Crafty plays and has two moves and >>the adversary has three moves in answer for each of those two moves. This, of >>course, is just an example. >>Now, supose move A has the following answers: move x, score 5+; move y, score >>5,5+ and move z, score 5,9+ >>Then you have move B with the following possible answers: move x1, with score >>6,7+; move y2, with score 5,0+ and move z2, with score 1- >> >>Now, in the usual way, Crafty would choose move A, as much even the best >>opponent move there is just 5,9+, but with move B the opponent has the chance to >>play x1, with score 6,7+. >>What I say is that in this field of bad scores, that kind of reasonning has not >>too much sense as anyway, with 5.0+ or with 6,7+, anyway the program is lost. So >>the idea of a swindle comes, as in human games: you choose move B because there >>there is a chance the opponent will mistake and play z2, with score 1-. > >This is not so simple. >The question is if there is a practical chance that the opponent is going to >blunder. > >It is possible that move A is better from practical reasons because because >after move A there is a practical chance that the opponent is going to blunder >when after move B there is no practical chance that the opponent will miss the >+6.7 move. > >I think that it is not a good idea to invest time on swindle mode if you want to >win humans in regular games and it is better to invest time in preventing a bad >position in the first place. > >Ideas for swindle mode can be used only if they are good and simple to do and I >think that the idea that you suggest is not good and is not simple. > >Uri Probably the closest thing here was what Berliner did in Hitech: Assume that you search to depth=N, and for plies 1 thru N-1, move X is best. But suddenly, at depth=N, X fails low. If you can't recover the original score by playing another move, most programs play the best move they can find, which often just loses in a very obvious way. Hans would play move X, since it was best until the last iteration, which means it is _not_ obvious why it is bad. Against computers, that might be awful. Against humans? It might work. One classic example came up in a game Cray Blitz vs some 1800 player at the 1984 US Open speed chess championship. We had beaten several masters and IM players, and with Victor Korchnoi looking on, we played this 1800 player and out of the clear blue sky, CB gave up its queen for apparently nothing. Korchnoi was quite astonished as were the rest of us. He commented after the game, "computer make horrible blunder, just trade rooks and it wins easily." I tried his suggested move and CB responded instantly with a mate in 10 for the opponent. Korchnoi was even more astounded, and several of us had a long discussion about this. It gave up a queen to avoid a mate in 10 it saw. Would an 1800 player see it? Of course not. And Berliner's scheme would have worked quite well.
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.