Author: Robert Hyatt
Date: 19:48:08 07/26/01
Go up one level in this thread
On July 26, 2001 at 12:41:34, Uri Blass wrote: >On July 26, 2001 at 10:48:47, Robert Hyatt wrote: > >>On July 26, 2001 at 08:40:33, Uri Blass wrote: >> >>>On July 26, 2001 at 08:18:42, Robert Hyatt wrote: >>> >>>>On July 26, 2001 at 00:43:48, Bruce Moreland wrote: >>>> >>>>>On July 25, 2001 at 21:26:18, Robert Hyatt wrote: >>>>> >>>>>>On July 25, 2001 at 18:36:08, James T. Walker wrote: >>>>>> >>>>>>>This position arose today in a game between Hiarcs 7.32 vs Century 3.2. >>>>>>>Although Hiarcs searched for 6 minutes on an Athlon 900 with 128M hash and it's >>>>>>>score dropped from the previous 0.87 to -0.56 it still could not resist the BxN >>>>>>>which I believe loses. I think almost any move which saves the Bishop keeps >>>>>>>white alive. Best is probably Kxb4 or Be3. Crafty is very fast to avoid BxN. >>>>>>>Junior7 is slow but finds it in a little over 1 minute. >>>>>>> >>>>>>>[D]8/p4k1p/6p1/8/1p2P1P1/1Kn3P1/3B3P/8 w >>>>>> >>>>>> >>>>>>This is all about knowledge. White's bishop is the only hope to restrain the >>>>>>black passed pawn and also help on the other side of the board. If it goes, >>>>>>black's distant passer makes this a normally won ending for black. >>>>>> >>>>>>I can't imagine a program trading the last piece and giving the opponent a >>>>>>distant passer, just on general principles, unless it sees some sort of tactical >>>>>>trick to sneak in its own pawn even quicker... >>>>>> >>>>>>If a program wants to aspire to be a GM, it _must_ know something about such >>>>>>endings... >>>>> >>>>>The reason a program would take the knight is that being up a pawn in a K+P >>>>>ending is usually better than being up a pawn in a B vs N ending. >>>>> >>>>>I don't think a program "has" to know about this kind of stuff in order to be a >>>>>GM, any more than a human has to be able to find a middlegame mate in 15 in 1/4 >>>>>second, has to be able to demonstrate a winning Fine 70 line in 1/2 second, or >>>>>has to be able to call mate in 95 in a KBN vs KN. >>>>> >>>>>Being a GM is not about being able to blend in perfectly with the GM community. >>>>>It's about being able to generate results comparable to those attained by GM's. >>>>>If the program is stupid in some circumstances, this is not necessarily fatal as >>>>>long as the weakness can be masked. >>>>> >>>>>bruce >>>> >>>> >>>>If a program doesn't know that being a pawn up, but with the opponent having >>>>an outside passed pawn, is most likely lost, then it is not going to be able to >>>>see that with search, >>> >>>Program can see it by search afer enough time. >>>Even Hiarcs that is relatively slow could see it after some minutes on pIII800. >>>Other programs are faster than hiarcs in seeing it but they do not see it at the >>>root like Crafty and they may need 1 minute on PIII800. >>> >>> and it is going to get hit on over and over once the >>>>opponent realizes it. >>> >>>The opponent needs to get into a pawn endgame and in most games the opponent >>>cannot go into a pawn endgame. >>> >>> It happened to me with cptnbluebear many times. Once, >>>>4 games in a row in fact, before Roman said "you _must_ fix this..." >>> >>>When the hardware gets better the program can see more by search and the program >>>may avoid the pawn ending by search. >>> >>>> >>>>I don't think you can mask a weakness that lets a program step into a totally >>>>lost endgame position thinking it is doing a good thing. >>> >>>I believe that Deep search can help here. >>>In the relevant example the program that I tried except Crafty do not know it by >>>chess knowledge but they are not going to fall into the trap at tournament time >>>control. >>> >>>Uri >> >> >>That doesn't help. back up 4 plies in the game. The program might still like >>this position because it thinks it can rip the knight and not lose any material. >>But once it gets here, it searches deeply enough to realize (a) if I take the >>knight, I don't lose material immediately but the passed pawn is going to win; >>(b) if I don't take the knight, I lose material in another way. I am screwed >>either way. But 4 plies ago I thought I could avoid the material loss by doing >>(a) but I just didn't understand the resulting position. > >In the relevant example not trading is a good alternative and there is no >problem of losing material. > >In theory there may be a position when the mistake was 4 plies before the >trading but practically it does not happen often. > >I agree that it is better to solve things by evaluation but it does not mean >that search does not help. > >Uri I totally disagree. Practically, it happens in _every_ game a computer plays and loses. It saw something that looked decent, then found that it wasn't decent later. Search helps, but only for tactics. an outside passed pawn or outside candidate simply isn't going to be understood by the search, period. I saw (and fixed) this problem in the early 70's for the same reason...
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.