Author: Robert Hyatt
Date: 07:05:21 01/18/00
Go up one level in this thread
On January 17, 2000 at 14:01:00, blass uri wrote: >On January 17, 2000 at 12:55:37, Jeremiah Penery wrote: > >>On January 17, 2000 at 06:32:48, blass uri wrote: >> >>>When Deeper blue made his last move in game 2 it did not expect the right >>>Qe3(see IBM logfiles). >>> >>>If I give crafty17.04 a long time it does expect the right Qe3. >>> >>>I know that no micro can find that Qe3 is a draw but some micro can at least >>>understand that Qe3 is the best move for black when deeper blue could not >>>understand it. >>> >>>Crafty is not the only micro that can expect Qe3. >>> >>>It increases the impression of most of the chess players that deeper blue was >>>better in tactics but had not better positional understanding than the >>>microcomputers. >> >>I was thinking about this last night. I'm not sure it is a positional thing. >>My guess is that DB saw that Qe3 led to a long series of checks, but couldn't >>find the quiet moves at that depth to find a draw. Since Qe3 doesn't lead to a >>draw (I.e., DB couldn't see it.), then white can have more attacking chances at >>black's king, because black's queen will be stuck on the other side of the >>board. (In most of the PVs, DB thought Kasparov would trade queens.) >>Of course this is just a guess, but it wouldn't seem completely inconsistent >>with the way DB seemed to evaluate certain things. > >I see it as a positional thing. >I think a good program should at least suspect that it is perpetual check. > >I think that if both sides have attacking chances it is illogical not to divide >the evaluation by a number bigger than 1 because it is clear that the position >is unclear and if you cannot do it clear by search then the best evaluation is >to admit that you are not sure by using smaller numbers for evaluation. > >I do not know if some of the programs that can expect Qe3 do it but it is clear >that deeper blue's evaluation was illogical. > >Uri I don't agree there. Computers have _always_ played "inhumanly". I can recall hundreds of games (mostly crafty, but also Cray Blitz and even Deep Blue) where a program will follow a line that appears to win quicker, but which leaves the program dead lost if it makes _one_ mistake. Where a human would follow the path of a sure (but longer) win, while leaving a lot more room for mistakes. The game from match 1 where DB was subject to a mate in 1 for several moves while winning was one example. I see Crafty follow these lines, which often are like balancing on a razor's edge... when it could easily trade queens and win the ending. And it _knows_ it could win, but the search finds a way to win one extra pawn down an incredibly sharp line.
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.