Author: Uri Blass
Date: 00:54:35 01/09/01
Go up one level in this thread
On January 08, 2001 at 16:02:57, Dann Corbit wrote:
>On January 07, 2001 at 23:08:45, Jon Dart wrote:
>
>>Here is an interesting example of a successful attack by a human
>>against a computer:
>>
>>[Event "?"]
>>[Site "?"]
>>[Date "2001.01.07"]
>>[Round "?"]
>>[White "realpravo"]
>>[Black "Arasan 6.0"]
>>[Result "1-0"]
>>[ECO "A08"]
>>[WhiteElo "2164"]
>>[BlackElo "2561"]
>>[TimeControl "300+8"]
>>
>>1. e4 e6 2. d3 d5 3. Nd2 c5 4. Ngf3 Nc6 5. g3 g6 6. Bg2 Bg7 7. O-O
>>Nge7 8. Re1 b6 9. e5 Qc7 10. Qe2 O-O 11. c3 Bb7 12. Nf1 Qb8 13. h4
>>Nf5 14. g4 Nh6 15. Bh3 a5 16. Bg5 Kh8 17. Ng3 a4 18. h5 Ng8 19. h6
>>Nxh6 20. d4 Ng8 21. Be3 cxd4 22. cxd4 Qc7 23. Kg2 a3 24. b3 Rfc8
>>25. Rac1 Ba6 26. Qd2 Qe7 27. Ng5 Qd7 28. f4 Bf8 29. f5 Bb4 30. Qf2
>>Bxe1 31. Rxe1 Rc7 32. f6 Nb4 33. Ne2 Nd3 34. Qh4 Nxe1+ 35. Kg3 h6
>>36. Ne4 g5 37. Bxg5 Nf3 38. Kxf3 dxe4+ 39. Kg2 Kh7
>>40. Bxh6 Nxf6 41. Bg5+ Nh5 1-0 {ArasanX resigns}
>>
>>The crucial position seems to be at Black's move 40:
>>
>>[D] r5nk/2rq1p2/bp2pP1p/4P1B1/3Pp1PQ/pP5B/P3N1K1/8 b - -
>>
>>Here any move but Qb5 leads to a fairly quick mate. Qb5 appears
>>to hold, although Black is still worse and may go under
>>eventually.
>>
>>The interesting thing here is that Black is way up material:
>>any computer would probably prefer Black in the moves
>>leading up to this one.
>>
>>Arasan takes a long time to find the Qb5 defense, and even longer
>>to have the score go negative:
>>
>>ply 9. Kh7 11 seconds. score: 5.02 2181239 nodes.
>>ply 9. Kh7 11 seconds. score: 5.02 2181239 nodes.
>>ply 10. Kh7 15 seconds. score: 5.02 3106545 nodes.
>>ply 10. Qc6 25 seconds. score: -156.03 6032366 nodes.
>>ply 10. Qb5 239 seconds. score: 0.39 70340167 nodes.
>>ply 10. Qb5 241 seconds. score: 0.39 70815106 nodes.
>>ply 11. Qb5 273 seconds. score: 0.39 77059646 nodes.
>>ply 12. Qb5 280 seconds. score: 1.02 78577842 nodes.
>>ply 12. Qb5 335 seconds. score: 1.02 89195295 nodes.
>>ply 13. Qb5 456 seconds. score: 1.02 112609559 nodes.
>>ply 13. Qb5 600 seconds. score: -0.23 141594783 nodes.
>>235987 nodes/second.
>>28414464 regular nodes, 113180319 quiescence nodes.
>>27712078 searches of hash table, 6431604 successful (23 percent).
>>500000 hash entries inserted, 1738345 entries replaced, 19925639 inserts failed.
>>
>>search: Qb5 600 seconds. score: -0.23 141594783 nodes. ++
>>correct
>>Qb5 Bxh6
>
>Qb5 does NOT help. It's all over but the shouting by this time. The real
>mistake is earlier, but I am not sure exactly where.
I agree and it proves that chess programs have stupid scores.
You can try Crafty on the following position from the same game.
[D]r5nk/2rq1p2/bp2pPpp/3pP3/3PN1PQ/pP2B1KB/P3N3/4n3 b - - 0 1
Crafty17.14 suggests the game move(g5) with a score of +4.56 for black at depth
14.
g5 is a losing move and the only question is if black can save the game(black
cannot win).
Humans are going to prefer 2 pawns with no compensation and not this position
because they have not linear evaluation but programs seem to use the simple sum
of positional score and material score.
I did not analyze enough to find if black can save the game but it seems that if
black can save the game then h5 is the right move.
Uri
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.