Author: Robert Hyatt
Date: 11:04:23 06/02/04
Go up one level in this thread
On June 02, 2004 at 07:34:16, Uri Blass wrote: >On June 02, 2004 at 06:52:25, Vasik Rajlich wrote: > >>On June 01, 2004 at 07:26:14, Uri Blass wrote: >> >>>On June 01, 2004 at 07:02:11, Vasik Rajlich wrote: >>> >>>>On May 31, 2004 at 07:22:55, Uri Blass wrote: >>>> >>>>>On May 31, 2004 at 07:15:31, Vasik Rajlich wrote: >>>>> >>>>>>On May 30, 2004 at 14:58:45, Jorge Pichard wrote: >>>>>> >>>>>>> >>>>>>>Kasparov-Deep Blue >>>>>>>Philadelphia (6) 1996 >>>>>>> >>>>>>> >>>>>>>The Opening has been a sucess for Kasparov. He has good central control, and >>>>>>>prospects of a gradual queenside advance. More importantly, there is no direct >>>>>>>plan for Black, so Deep Blue drifts for a few moves with disastrous >>>>>>>consequences. The bishop is already a little clumpsy on d7; I suspect a strong >>>>>>>human player would have sunk into thought, and devised a plan for deliverating >>>>>>>his game. >>>>>>> >>>>>>>[D]r2q1rk1/pp1bbppp/2n1pn2/3p4/2PP4/1P1B1N2/PB1N1PPP/2RQ1RK1 b - - 0 1 >>>>>>> >>>>>>>11...Nh5? >>>>>>>This over-ambitious idea met with strong disapproval from most strong human >>>>>>>commentators. However, Yasser Seirawan said "oddly enough, one well-known chess >>>>>>>computer scientist suggested that the move may well be OK, but it might need a >>>>>>>highly advandce program and computer in a few years' time to justify this move". >>>>>>>I suspect that this is a case in point of someone believing that a strong >>>>>>>chess-playing program is doing something profound, when in fact is just >>>>>>>crunching numbers, Few GMs back in 1996 felt that 11....Nh5 was anything other >>>>>>>than a bad move. >>>>>> >>>>>>This type of position is very difficult for any chess program. >>>>>> >>>>>>Sometimes, the engines will find some way to make a very strange move work, like >>>>>>11. .. Nh5. Even in this case, it's not very good if a person is using an engine >>>>>>to help him understand the position. >>>>>> >>>>>>Note that search depth is not important here. For another example of this, see >>>>>>Kasparov-Fritz, X3D, game 3, where Fritz was doing 18-19 ply in the middlegame. >>>>> >>>>>I disagree that search depth is not important. >>>>> >>>>>The fact that 18-19 plies of Fritz was not enough does not mean that search >>>>>depth is not important. >>>>> >>>>>Uri >>>> >>>>Rrrrrr. :) >>>> >>>>See how the scores for the top four or so moves here change with depth. (Or >>>>trust me: not much.) >>> >>>It proves nothing. >>> >>>The fact that the scores do not change much does not mean that the moves that >>>the program suggest at bigger depth are not better. >>> >>>If you want to convince me you need to show me that depth X+2 scores less than >>>55% against depth X from the relevant position >>> >>>[D]r2q1rk1/pp1bbppp/2n1pn2/3p4/2PP4/1P1B1N2/PB1N1PPP/2RQ1RK1 b - - 0 1 >>> >>> >>> >>>You can use some commercial programs against themself at depth X+2 against depth >>>X for X=14-18 so you may get enough games(for me 50 games are enough). >>> >>>A better solution may be to use unequal time when you give 27 hours per game for >>>one side and 3 hours per game for the second side. >>> >>> >>>If after 50 games you do not get at least 55% for the deeper searcher then you >>>convince me that the position is position that programs do not earn much from >>>deeper depth. >>> >>>Of course you can choose dead drawn position and get exactly 50% but I doubt if >>>you can find a position when both sides have chances based on the games and >>>still get less than 55% after 50 games. >>> >>>Uri >> >>No. >> >>:) >> >>The deeper searcher will score better, because it will play better later, when >>there are either tactics, or positional decisions which are searchable. >> >>For the purely strategic moves, extra depth will give you pretty much nothing. >> >>Vas > >In this case the quality of the purely startegic moves is not very important >because they do not decide the game but what happens later. > >Uri This is short-sighted. In 1984 Cray Blitz was black against Nuchess at the Los Angeles ACM event. We were playing a poor opening and were getting strangled. Nuchess reached a point in the game were it could win a pawn, and all it had to decide was whether to take an already weak isolated a-pawn or take a strong center pawn and thoroughly weaken our pawn structure. It chose the center pawn, which was a strategic decision. In doing so, it left CB with a passed/isolated A pawn that was not a problem, until CB later found a way to force the trade of _all_ pieces and the a-pawn became the deciding feature of the game. That "strategic" choice to rip our pawn structure apart rather than remove that weak a-pawn lost the game. strategic moves _can_ (and often do) decide the game... Otherwise the moves would not be strategic and the "strategy" would be unimportant for them. BTW CB thought NuChess should have removed the passed a-pawn, just as Crafty does today, because both understood that a distant passer is of strategic value later in the game as pieces come off, even if the pawn is a weakness when there are lots of pieces to attack it. I can think of dozens of similar decisions my program (and others) have made. "Mobilize the majority or let it stay where it is safer?" In some endgames a single tempi decides. push an a-pawn to create an a6/b5 chain with b5 weak on an open file? But where b6 later forces a quick-promoting passer that wins the game? Another strategic decision that can win or lose the game.
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.