Author: Robert Hyatt
Date: 07:42:23 11/21/00
Go up one level in this thread
On November 20, 2000 at 21:49:11, Bob Durrett wrote: >On November 20, 2000 at 13:06:26, Rafael Andrist wrote: > >>On November 20, 2000 at 12:26:24, Bob Durrett wrote: >> >>>On November 20, 2000 at 12:13:04, Severi Salminen wrote: >>> >>>> >>>>>Maybe a computer chess programmer would be willing to modify his program to add >>>>>the capability to measure the amount of time the program uses to find the >>>>>solution. That would not give a direct measurement of how difficult the >>>>>position would be for a human player, but might give some indication at least. >>>>> >>>>>Incidentally, you could try using a stopwatch. >>>> >>>>I don't think that is a good idea. Computers can find very deep combinations >>>>very fast and the same position might be allmost impossible for a human. I think >>>> better approach would be to check out the types of moves that lead to solution. >>>>If there are many non_captures, non_checks and non_promotions, it is probably >>>>more difficult for a human. Captures, checks and promotions are easier to find. >>>>Also the depth is an issue, mate in 10 is harder than mate in 2. And mate with 5 >>>> sequential captures is easier than a mate in 4 with "silent" moves. >>>> >>>>Severi >>> >>>The above hi-lights the issue of "What Makes a Position Difficult for a Human to >>>Solve?". This should be of great interest to computer chess programmers because >>>it would give an indication as to how to program chess-playing programs to put >>>up the greatest resistance to human opponents. >>> >>>In human vs human chess, one very successful strategy is to keep on posing very >>>difficult problems for the opponent to solve. Eventually, the opponent may >>>crack under the pressure! [Of course, if the opponent is following this same >>>strategy, then problems posed by the opponent must be solved also.] I see no >>>reason why this would not apply in human vs computer [or visa versa] games. >>>After all, the top GMs whip the chess programs by playing anti-machine >>>strategies. Why not let the engines turn the tables on the GMs and play >>>anti-human chess against humans? >> >>I would be nice to have algorithm that detect that. But: It's more difficult to >>find the silent moves for the computer than the human (at GM level). So, first >>the program need to find the silent moves in normal search and then detect that >>these are silent moves. I case of finding a mate this would be not very >>difficult to program, but if the gain is some positional advantage or a silent >>attack, a human can easier find the silent moves. > >Perhaps so. Perhaps it would be difficult to program. But when did that ever >keep a dedicated chess programmer from trying? > >If your life depended on it [obviously it does not], how would you tackle this >"difficult" programming task? Actually this is one contribution I would like to try to get the DB guys to expose. Most of us fail low and use more time. They did something better. They could tell when a position was getting "difficult" and use more time without a fail low. Hsu was trying to explain this at one point at an ACM event, but too much was going on and we never had time to finish the conversation. It would be interesting to ask them (again) to see how they were defining an "unstable" position. I'll ask...
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.