Author: Robert Hyatt
Date: 15:13:22 11/26/98
Go up one level in this thread
On November 26, 1998 at 04:41:15, Graham Laight wrote: > >On November 24, 1998 at 18:19:02, Amir Ban wrote: > >>On November 24, 1998 at 14:16:16, Ernst A. Heinz wrote: >> >> >>>"DarkThought" finished iteration #21 after 210 hours of computation and I >>>finally had to stop it there. Qb6 remained the best move and the PV took about >>>156 hours and exactly 120,180,459,947 nodes to resolve (3x more than the PV of >>>the previous iteration). The expected best reply switched back to Rd8 and the >>>score sank to a new low of +1.13. >>> >>>21.01 Qb6 Rd8 Be4 a5 axb5 axb4 Rxa8 Rxa8 Ra6 Rd8 ... (1.13) #120180459947 >>> >>>=Ernst= >> >> >>Hey ! You are back on the "drawing" line and you have a novelty ! >> >>Dark Thought now wants to decline the bishop sacrifice and play Ra6 (rather than >>Rxa8 Qxa8 Qxd6 Qa1+ etc.). We completely overlooked this possibility when we >>analyzed this line last year. It seems to hang on to a +1 or so advantage, with >>reasonable winning chances, I think. >> >>In view of this, I need to change the conclusion of the analysis, which was that >>36.Qb6 Rd8 draws. >> >>Great effort, Ernst. >> >>Amir > >Given this claim that a big new discovery has taken place, may we just recap >what we're discussing here, please? > >Correct me if I'm wrong (I probably am, because I haven't been following the >thread in detail), but my understanding is as follows: > >In DB V GK 97 game 2, DB played 36. axb5. > >This is a move one would not expect a computer to play, because it misses an >opportunity to win material with 36. Qb6, which is what one would expect a chess >computer to do. > >Those who believe that DB found the move for itself, without help, have, in >support of their case, stated that DB would have been able to see that 36. Qb6 >leads to a forced draw. > >To discover if this is true, Bruce Moreland has asked that as many people as >possible should run the position for as long as possible on their own computers, >to see whether, at some depth of search, the eval for 36. axb5 jumps ahead of >the eval for 36. Qb6. > >In doing this, Ernst Heinz has discovered that, at ply 21, a line of >continuation that includes move 40. Ra6, appears to push the draw too far into >the distance for any computer to have a realistic chance of seeing it. > >This adds to the case that 36. axb5 is not a move one would expect a computer to >make, and therefore supports Gary Kasparov's view that the move requires further >explanation from IBM. > >Right or wrong? neither. I searched both moves to great depth... Qb6 after 22 plies was right at +1.0... axb5 after 20 plies was +.7. axb5 is harder for crafty to search for reasons I haven't tried to examine.. but notice that the two moves are *very* close. I tried an older version that was more paranoid on king safety (the infamous 13.x series used in Paris) and it likes axb5 from early on, then decides that Qb6 goes to +2 later, but after quite a while when Qb6 drops back into the +1.5 range it gets replaced by axb5 again. So there's nothing remarkable when you consider that axb5 and Qb6 are not separated by even a whole pawn... and it is easy to believe that either DB found Qb6 led to a drawn (or drawish) position and axb5 didn't, or else it evaluated the position for axb5 as better when it got deep enough. I think that it liked axb5 for some positional reasons and when Qb6 dropped low enough it was replaced... nothing magic, nothing sinister, nothing unethical, and nothing immoral at all...
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.