Author: Bob Durrett
Date: 05:57:35 11/11/00
Go up one level in this thread
On November 11, 2000 at 00:45:46, Bruce Moreland wrote: <snip> > >You can't prove this, in theory, because every position would resolve to a win, >loss, or draw, given enough time. > >What this means is that the evaluation of a position is subject to change, and >therefore it is likely that the program will change its mind eventually. > >Perhaps you can find a series of positions where a program won't change its mind >in any practical period of time, but you'd always have to wonder if it will >change its mind in another five minutes. Yes, that makes sense. A little background: This inquiry was originally prompted by the planned matches between top GMs and the two programs, Fritz and Junior. After seeing the announcement of the matches, I began wondering whether or not the top GMs could effectively prepare an "anti-Fritz" or "anti-Junior" opening repertoire which would give the GMs and advantage, based on the assumed deterministic nature of the programs. A thread resulted on rec.games.chess.computer in which an idea occurred [an idea of Bob Hyatt's] which, it seemed to me, might hold out some hope of finding effective anti-computer lines. The idea was that it might be possible for a GM to find an opening line leading to a position having the special property that the computer's response to that position could be predicted, at least in some practical sense. If the GM knew in advance what the computer would play from that position, then he/she could select a move in advance, essentially extending the GMs opening repertoire line by one move for each side. If, then, the next position were to also have that same special property that the computer's move could be predicted [in some practical sense], then the line could be extended another move for each side. On then must wonder whether or not it might be possible for this to be continued until the GM found a line leading to a win. [The GM would have the flexibility of selecting his moves in coming up with the line.] There are some relatively trivial special cases, such as where there is only one reasonable move for the computer. Another, more exotic, case occurred in one of my games I played against a friend, and later post-mortem analyzed using Fritz. At a certain point, Fritz said the evaluation was 0.00 and suggested a move. It would never change it's mind, as later explained to me by Fritz's programmer [I forgot his name, I'm ashamed to admit]: He said that when Fritz sees that repeating a position would lead to a draw [by repetition of the position] but not repeating the position would lead to a bad game, then Fritz will always opt to play the move that leads to the repetition of the position. > >If you want the looser case of a line that will win at most time controls, you >can start by finding a line that wins at blitz, and seeing if you can still win >with the line at longer time controls. > >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.