Computers have problems with certain positions

[Uri]

Computers are good but they still have problems with certain positions. For example go to the game Kasparov vs Deep Blue 1996. I analyzed this game with Toga II 1.3.4 and with Shredder 10 and they played crazy with black. In this game there were no tactical chances and Kasparov gained a huge space advantage.

http://www.chessgames.com/perl/chessgame?gid=1070879

The Stonewall Attack is another opening strategy which computers don't play well. Traditionally, chess computers have been vulnerable to the Stonewall because the positions are usually without clear tactical lines. White simply prepares for an assault by bringing pieces to aggressive posts, without making immediate tactical threats. By the time the computer realizes that its king is under attack it is often too late.

http://chess.eusa.ed.ac.uk/Chess/Trivia/stonewall.html

Eduard Nemeth also demonstrated how computers sometimes can't see far away dangers.

http://www.xs4all.nl/~timkr/chess2/honor.htm

[Uri]

[d]8/6p1/2p4p/1p1k4/1P1pq3/6PP/P4K2/2Q5 b - - 0 50

Here in this position where black is to move, Toga II 1.3x4b and Shredder 10 can't find 50...Qe3+ 51.Qxe3 dxe3+ 52.Kxe3 Kc4 and black wins.

I think humans are still better than computers when it comes to finding the best move.

[Allard Siemelink]

[d]8/6p1/2p4p/1p1k4/1P1pq3/6PP/P4K2/2Q5 b - - 0 50

Here in this position where black is to move, Toga II 1.3x4b and Shredder 10 can't find 50...Qe3+ 51.Qxe3 dxe3+ 52.Kxe3 Kc4 and black wins.

I think humans are still better than computers when it comes to finding the best move.

The current development version of bright, which includes some experimental PP evaluation rules, finds it in ~30 seconds:
[14/41] 0:00:26 38533k 669 Qe4e3+ Qc1xe3 d4xe3+ Kf2xe3 Kd5c4 a2a3 Kc4b3 Ke3d4 Kb3a4 Kd4c5 Ka4xa3 h3h4 h6h5 Kc5xc6 Ka3xb4 Kc6d5 Kb4c3
[14/41] 0:00:26 38619k 669 ok/1.8

Here's the final position:
[d] 8/6p1/8/1p1K3p/7P/2k3P1/8/8 w - - 2 17

[jdart]

In my experience it is also hard for computers to handle the Classical King's Indian as White - since the standard plan for Black is a long buildup followed by a breakthrough on the kingside (a Kasparov specialty in his early career). (I see the Stonewall and other systems mostly used by human players who want to close the position and aim for a draw).

--Jon

[Mike S.]

D945/3.4 GHz, 256 MB hash

Engine: Rybka 2.3.2a 32-bit
von Vasik Rajlich

11.00 0:01 +2.09 1...Df5+ 2.Kg2 De6 3.Kf2 Df6+ 4.Kg2 De7 5.Dd2 De3 6.Dxe3 dxe3 7.Kf3 Kc4 (78.216) 72.1
(...)
17.01 0:37 +2.73 1...Df5+ 2.Kg2 De6 3.Kf3 Df6+ 4.Kg2 De7 5.Df1 De4+ 6.Kg1 De3+ 7.Kg2 d3 8.Df7+ (2.987.525) 82.1
17.15 0:44 +5.17 1...De3+ 2.Dxe3 dxe3+ 3.Kxe3 Kc4 4.a3 Kb3 5.Kd4 Kxa3 6.Kc5 Kb3 7.h4 h5 (3.868.319) 89.5
(...)
21.01 85:07 +10.61 1...De3+ 2.Dxe3 dxe3+ 3.Kxe3 Kc4 4.a3 Kb3 5.Ke4 Kxa3 6.Kf5 Kxb4 7.Kg6 c5 8.Kxg7 (548.268.421) 109.9
Bester Zug: De4-e3 Zeit: 98:23.031 min K/s: 108.844 Knoten: 627.451.026 TB: 649

(D=Q)

I let it run for so long only because I forgot that it was running in the background

[George Tsavdaris]

[d]8/6p1/2p4p/1p1k4/1P1pq3/6PP/P4K2/2Q5 b - - 0 50

Here in this position where black is to move, Toga II 1.3x4b and Shredder 10 can't find 50...Qe3+ 51.Qxe3 dxe3+ 52.Kxe3 Kc4 and black wins.

My Shredder 10 can find it very quickly....

(S10 with 3,4,5 Shredderbases and the Nalimov 6 piece KPPKPP loaded, on a AMD3500+ 2.2MHz)

Analysis by Shredder 10 UCI:

1...Qe4-e6 2.a2-a3 Qe6-f6+ 3.Kf2-e2 Qf6-e7+ 4.Ke2-d1 d4-d3 5.Qc1-b2
-+ (-2.49) Depth: 8/27 00:00:02 182kN
1...Qe4-e6 2.a2-a3 Qe6-f5+ 3.Kf2-e1 Qf5-e4+ 4.Ke1-f2 d4-d3 5.Qc1-c5+
-+ (-2.50) Depth: 9/28 00:00:03 318kN
1...Qe4-f5+
-+ (-2.51) Depth: 9/28 00:00:03 379kN
1...Qe4-f5+ 2.Kf2-e1 Qf5-e6+ 3.Ke1-d1 Qe6-f6 4.Qc1-c5+ Kd5-e4 5.Kd1-e1 d4-d3 6.Qc5-c1 g7-g5 7.g3-g4 Qf6-e6
-+ (-2.68) Depth: 9/28 00:00:03 460kN
1...Qe4-f5+ 2.Kf2-e1 Qf5-e6+ 3.Ke1-d1 Qe6-f6 4.Qc1-c5+ Kd5-e4 5.Kd1-e1 d4-d3 6.Qc5-c1 g7-g5 7.g3-g4 Qf6-e6
-+ (-2.68) Depth: 10/30 00:00:04 613kN
1...Qe4-f5+ 2.Kf2-e1 Qf5-e6+ 3.Ke1-d1 Qe6-f6 4.Qc1-c5+ Kd5-e4 5.Kd1-e1 Ke4-f3
-+ (-2.82) Depth: 11/32 00:00:05 1511kN
1...Qe4-f5+ 2.Kf2-e1 Qf5-e6+ 3.Ke1-d1 Qe6-f6 4.Qc1-c5+ Kd5-e4 5.Kd1-e1 Ke4-f3 6.Qc5-c2
-+ (-3.02) Depth: 12/34 00:00:06 3227kN
1...Qe4-f5+ 2.Kf2-g2 Qf5-e6
-+ (-3.26) Depth: 13/36 00:00:12 6903kN
1...Qe4-f5+ 2.Kf2-g2 Qf5-e6 3.Qc1-c5+ Kd5-e4 4.Qc5-c2+ Ke4-e3 5.Qc2-c1+ Ke3-d3 6.Qc1-b1+ Kd3-c3 7.Qb1-a1+ Kc3-c4 8.Qa1-b2
-+ (-3.31) Depth: 14/39 00:00:22 13576kN
1...Qe4-e3+
-+ (-3.32) Depth: 14/39 00:00:24 14440kN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.a2-a3 Kc4-b3 5.Ke3-d4 g7-g5 6.Kd4-c5 Kb3xa3 7.g3-g4
-+ (-3.44) Depth: 14/39 00:00:24 14467kN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.a2-a3 Kc4-b3 5.Ke3-d4 g7-g5 6.Kd4-c5 Kb3xa3 7.g3-g4 Ka3-b3 8.Kc5xc6 Kb3xb4 9.Kc6-d5 Kb4-c3
-+ (-3.44) Depth: 15/26 00:00:24 14504kN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.a2-a3 Kc4-b3 5.Ke3-d4 g7-g5 6.Kd4-c5 Kb3xa3 7.g3-g4 Ka3-b3 8.Kc5xc6 Kb3xb4 9.Kc6-d5 Kb4-c3 10.Kd5-c5
-+ (-3.52) Depth: 16/29 00:00:34 20768kN
1...Qe4-e3+ 2.Qc1xe3
-+ (-3.77) Depth: 17/32 00:00:50 30944kN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.a2-a3 Kc4-b3 5.Ke3-d4 g7-g5 6.Kd4-c5 Kb3xa3 7.g3-g4 Ka3-a4 8.Kc5xc6
-+ (-4.22) Depth: 17/32 00:00:51 31016kN
1...Qe4-e3+ 2.Qc1xe3
-+ (-4.47) Depth: 18/31 00:01:06 40431kN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.a2-a3 Kc4-b3 5.Ke3-d4 g7-g5 6.Kd4-c5 Kb3xa3 7.g3-g4 Ka3-a4 8.Kc5xc6 Ka4xb4 9.Kc6-d5 Kb4-c3 10.Kd5-e4 b5-b4 11.Ke4-e3
-+ (-4.61) Depth: 18/31 00:01:06 40524kN
1...Qe4-e3+ 2.Qc1xe3
-+ (-4.86) Depth: 19/35 00:01:35 58362kN
1...Qe4-e3+ 2.Qc1xe3
-+ (-4.86) Depth: 19/35 00:01:35 58491kN
1...Qe4-e3+ 2.Qc1xe3
-+ (-5.11) Depth: 20/39 00:03:19 124mN
1...Qe4-e3+ 2.Qc1xe3 d4xe3+ 3.Kf2xe3 Kd5-c4 4.Ke3-d2 Kc4xb4 5.Kd2-c1 Kb4-a3 6.Kc1-b1 g7-g5 7.Kb1-a1 b5-b4 8.Ka1-b1 h6-h5 9.Kb1-a1 c6-c5 10.Ka1-b1 g5-g4
-+ (-5.29) Depth: 20/39 00:03:19 125mN

[Uri]

How do you put the evaluations into the forum?

[Mike S.]

How do you put the evaluations into the forum?

This depends on which interface you have. Usually (like with Fritz), you would right-click into the engine window where you see the engine output, and select something like "copy to clipboard". Then, you press Ctrl-V in the posting to paste the computer analysis into the posting.

Also, in Fritz it is useful to select "Scroll new moves only" which will skip useless in-between lines where the first move was not changed. The first occurance with the decision time will be kept. That helps to keep the postings smaller and easier to read (sometimes, engines will produce an output of dozens of lines in the first few seconds, always with the same pm).

[George Tsavdaris]

How do you put the evaluations into the forum?

This depends on which interface you have. Usually (like with Fritz), you would right-click into the engine window where you see the engine output, and select something like "copy to clipboard". Then, you press Ctrl-V in the posting to paste the computer analysis into the posting.

"Clip Analysis" it is called in Fritz or similar GUIs.....

[Uri]

How do you put the evaluations into the forum?

This depends on which interface you have. Usually (like with Fritz), you would right-click into the engine window where you see the engine output, and select something like "copy to clipboard". Then, you press Ctrl-V in the posting to paste the computer analysis into the posting.

Also, in Fritz it is useful to select "Scroll new moves only" which will skip useless in-between lines where the first move was not changed. The first occurance with the decision time will be kept. That helps to keep the postings smaller and easier to read (sometimes, engines will produce an output of dozens of lines in the first few seconds, always with the same pm).

Ok, thanks Mike. By pm you mean program memory?