Black hole

lech · Post by **lech** » Sat Aug 06, 2011 1:50 pm

This text is a try to show why machines (computers) are not able to solve many positions.
Machines will do all. Only wrong software can interfere with theirs good work.
Diagram 1:
[d]3N2r1/2K1p3/4Pk2/8/Bp5b/8/2P5/8 b - - 0 1
It is a simple endgame position with 10 pieces.
We are waiting that top engines are able to find the easy way to win:
1… Bg3+ 2. Kd7 Rxd8 3.Kxd8 Kxe6 (diagram 2 below).
Diagram 2:
[d]3K4/4p3/4k3/8/Bp6/6b1/2P5/8 w - - 0 1
It is a win for black
I tried to use (diagram 1) the best engine and after 1 hour (PIV, 3GHz, 2 threads) I resigned.

Code: Select all

Houdini_15a_w32&#58;
  4/10	00&#58;00	         929	58.000	+2,42	Bh4g3+ Kc7d7 Bg3e5 Nd8c6
  5/13	00&#58;00	       1.784	89.000	+2,25	Bh4g3+ Kc7d7 Rg8g5 Nd8c6 Rg5d5+ Kd7e8
  5/13	00&#58;00	       2.114	100.000	+2,51	Rg8g2 Nd8c6 Bh4g3+ Kc7b6 Bg3e5 Nc6xb4 Be5d4+ Kb6c7 Kf6xe6
  5/13	00&#58;00	       3.168	99.000	+2,58	Rg8g4 Nd8c6 Kf6xe6 Ba4b3+ Ke6f5 Kc7d7
  6/15	00&#58;00	       4.217	113.000	+2,41	Rg8g4 Nd8c6 Kf6xe6 Ba4b3+ Ke6f5 Kc7d7 Rg4e4
  7/15+	00&#58;00	       7.868	135.000	+2,54	Rg8g4
  7/15	00&#58;00	      11.941	178.000	+2,30	Bh4f2 Nd8c6 Bf2c5 Ba4b3 Rg8g2 Kc7d7 Rg2d2+ Kd7c7
  7/15	00&#58;00	      13.485	192.000	+2,33	Bh4e1 Nd8c6 Be1c3 Ba4b3 Rg8g2 Kc7d7 Rg2d2+ Kd7e8
  8/15	00&#58;00	      14.320	168.000	+2,33	Bh4e1 Nd8c6 Be1c3 Ba4b3 Rg8g2 Kc7d7 Rg2d2+ Kd7e8
  8/15	00&#58;00	      15.558	174.000	+2,35	Rg8g4 Ba4b3 Bh4g3+ Kc7d7 Rg4d4+ Kd7e8 Rd4d2 Nd8c6 Bg3c7
  9/17	00&#58;00	      30.492	260.000	+2,38	Bh4g3+ Kc7d7 Bg3d6 Nd8c6 Rg8g2 Ba4b3 Rg2d2 Kd7e8 Bd6c5 Ke8f8
 10/17	00&#58;00	      55.705	357.000	+2,30	Bh4g3+ Kc7d7 Bg3d6 Nd8f7 Bd6b8 Ba4b3 Rg8g3 Kd7c6 Bb8e5 Kc6c5 Be5c3
 10/19	00&#58;00	      69.451	394.000	+2,37	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Nd8c6 Bd4c5 Kd7c7 Bc5d6+ Kc7d7 Rg2d2 Kd7e8 Bd6c5 Ke8f8 Rd2g2
 11/19	00&#58;00	      84.148	414.000	+2,37	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Nd8c6 Bd4c5 Kd7c7 Bc5d6+ Kc7d7 Rg2d2 Kd7e8 Bd6c5 Ke8f8 Rd2g2
 12/21	00&#58;00	     130.681	491.000	+2,34	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Kd7c6 Bd4c3 Kc6c5 Rg2g5+ Kc5c4 Rg5g4+ Kc4b5 Rg4h4 Nd8f7 Rh4h5+ Kb5c6
 13/21	00&#58;00	     212.766	573.000	+2,34	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Kd7c6 Bd4c3 Kc6c5 Rg2g5+ Kc5c4 Rg5g4+ Kc4b5 Rg4h4 Nd8f7 Rh4h5+ Kb5c6
 14/21	00&#58;00	     317.682	621.000	+2,34	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Kd7c6 Bd4c3 Kc6c5 Rg2g5+ Kc5c4 Rg5g4+ Kc4b5 Rg4h4 Nd8f7 Rh4h5+ Kb5c6
 15/24	00&#58;01	     548.784	683.000	+2,34	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Kc6d7 Rg8g2 Kd7c6 Bd4c3 Kc6c5 Rg2g5+ Kc5c4 Rg5g4+ Kc4b5 Rg4h4 Nd8f7 Rh4h5+ Kb5c6
 16/25	00&#58;01	   1.107.215	750.000	+2,32	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Nd8f7 Bd4c3 Bb3d5 Rg8a8 Kc6b5 Ra8a1 Bd5b3 Kf6f5 Kb5c5 Ra1a5+ Kc5c6 Kf5f6 Kc6d7 Bc3d4 Kd7c6 Ra5a6+ Kc6b5
 17/28	00&#58;02	   2.061.996	782.000	+2,32	Bh4f2 Ba4b3 Bf2c5 Kc7c6 Bc5d4 Nd8b7 Rg8a8 Kc6b5 Ra8b8 Kb5c6 Rb8h8 Kc6b5 Kf6e5 Nb7a5 Bd4c3 Kb5c5 Ke5e4 Na5c6 Rh8h5+ Kc5c4
 18/29-	00&#58;04	   3.351.510	803.000	+2,21	Bh4f2 Ba4b3
 18/31	00&#58;08	   7.077.865	814.000	+2,20	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Rg2g5 Kd7c7 Rg5g2 Kc7d7
 19/31	00&#58;11	   8.942.509	823.000	+2,20	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Nc6a5 Bd6e5 Na5c6 Rg2d2+ Kd7c8 Be5d6 Kc8b7 Rd2g2 Kb7b6 Rg2g5 Kb6a6 Rg5h5 Ka6b6 Rh5h2 Kb6b5 Rh2h4 Bb3d5 Rh4h5 Kb5c4
 20/33	00&#58;14	  11.947.177	834.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8d7 Rg5b5 Nc6d4 Rb5h5 Nd4f3 Rh5c5 Nf3d2 Rc5c7+ Kd7e8 Bd6e5 Nd2f3 Be5f4 Nf3d4 Rc7c3 Ke8d7 Bf4c7 Nd4c6
 21/34	00&#58;21	  17.717.071	842.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Bf4d6 Kb7b6 Rc5f5 Kb6a6 Rf5h5 Ka6b7 Rh5b5+ Kb7a6 Rb5h5 Ka6b7
 22/37	00&#58;30	  25.225.789	846.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Bf4d6 Kb7b6 Rc5f5 Kb6a6 Rf5h5 Ka6b7 Rh5b5+ Kb7a6 Rb5h5 Ka6b7
 23/41	00&#58;44	  38.196.018	855.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Kb7b6 Bf4e3+ Kb6c7 Be3c5 Kc7c8 Rg5h5 Kc8b7 Bc5e3 Kb7c7 Be3f4+ Kc7b7 Bf4d2 Kb7c7 Bd2f4+ Kc7b7
 24/41	01&#58;03	  54.783.060	866.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Bf4d6 Kb7b6 Rc5f5 Kb6a6 Rf5h5 Ka6b7 Kf6f5 Kb7b6 Kf5e4 Bb3c4 Rh5h2 Kb6b5 Rh2xc2 Nc6xb4 Rc2c3 Bc4d5+ Ke4d4
 25/41	01&#58;37	  85.099.561	873.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Kb7a6 Bf4d2 Ka6b7 Rg5h5 Kb7c8 Rh5h4 Kc8b7 Bd2c3 Kb7b6 Rh4h5 Kb6c7 Bc3e5+ Kc7b6 Be5d6 Nc6d8 Rh5f5 Nd8c6 Rf5f2 Kb6b5
 26/45	02&#58;32	 134.162.166	879.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Rc5h5 Kb7c8 Rh5b5 Kc8d7 Bf4d6 Kd7c8 Rb5f5 Kc8b7 Rf5g5 Kb7b6 Rg5h5 Nc6d8 Rh5f5 Nd8c6 Rf5f2 Kb6b5
 27/47	04&#58;25	 232.912.976	877.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Kd7c8 Rg2g5 Kc8b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Rc5h5 Kb7c8 Rh5b5 Kc8d7 Bf4d6 Kd7c8 Rb5h5 Kc8b7 Rh5f5 Kb7b6 Kf6g6 Bb3c4 Rf5f2 Bc4b3
 28/47	08&#58;34	 418.836.500	814.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Nc6d8 Bd6e5 Nd8c6 Rg2d2+ Kd7c8 Be5d6 Kc8b7 Rd2f2 Kb7a6 Rf2f5 Ka6b6 Rf5h5 Kb6a6 Rh5g5 Ka6b6 Rg5f5 Kb6a6 Rf5h5 Ka6b6 Rh5g5 Kb6b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Rc5h5 Kb7c8 Rh5b5 Kc8d7 Bf4d6 Kd7c8 Rb5h5 Kc8b7 Rh5f5 Kb7b6 Rf5h5
 29/51	18&#58;08	 844.917.265	776.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Nc6d8 Bd6e5 Nd8c6 Rg2d2+ Kd7c8 Be5d6 Kc8b7 Rd2f2 Kb7a6 Rf2f5 Ka6b6 Rf5h5 Kb6a6 Rh5g5 Ka6b6 Rg5f5 Kb6a6 Rf5h5 Ka6b6 Rh5f5 Kb6a6
 30/53	32&#58;14	1.422.643.813	735.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Bc5d6 Nc6d8 Bd6e5 Nd8c6 Rg2d2+ Kd7c8 Be5d6 Kc8b7 Rd2f2 Kb7a6 Rf2f5 Ka6b6 Rf5h5 Kb6a6 Kf6g7 Nc6a5 Rh5f5 Na5c6 Rf5g5 Ka6b6 Kg7f6 Kb6b7 Bd6f4 Bb3c4 Rg5c5 Bc4b3 Rc5h5 Kb7c8 Rh5b5 Kc8d7 Bf4d6 Kd7c8 Rb5h5 Kc8b7 Rh5c5 Kb7b6 Rc5f5 Kb6a6 Rf5g5 Ka6b7 Bd6f4
 31/55	58&#58;04	2.456.449 659	705.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Rg2d2+ Kd7c8 Rd2h2 Kc8b7 Rh2h5 Kb7a6 Bc5d6 Ka6b6 Rh5f5 Nc6d8 Rf5g5 Nd8c6 Rg5h5 Kb6a6 Rh5f5 Ka6b6 Rf5g5 Kb6a6 Rg5h5 Ka6b6
 31/55	1&#58;10&#58;19	2.973.738 488	704.000	+2,19	Bh4f2 Ba4b3 Bf2c5 Nd8c6 Rg8g2 Kc7d7 Rg2d2+ Kd7c8 Rd2h2 Kc8b7 Rh2h5 Kb7a6 Bc5d6 Ka6b6 Rh5f5 Nc6d8 Rf5g5 Nd8c6 Rg5h5 Kb6a6 Rh5f5 Ka6b6 Rf5g5 Kb6a6 Rg5h5 Ka6b6

Houdini can’t solve this simple position, immediately.
Why?
To explain it we must test diagram 2, after 3 moves of solution.
Diagram 2:
[d]3K4/4p3/4k3/8/Bp6/6b1/2P5/8 w - - 0 1

Code: Select all

 Houdini_15a_w32&#58;
  4/8	00&#58;00	         227	227.000	-0,25	Ba4b3+ Ke6d6 Kd8e8 e7e5 Ke8f7
  4/8	00&#58;00	         354	29.000	-0,21	Ba4c6 Ke6d6 Bc6e4 e7e5
  5/9	00&#58;00	         554	34.000	-0,26	Ba4c6 Ke6d6 Bc6e4 e7e5 Kd8e8
  6/9	00&#58;00	       1.080	56.000	-0,23	Ba4c6 Ke6d6 Bc6e4 e7e5 Kd8e8 Bg3f4
  7/11	00&#58;00	       2.163	90.000	-0,26	Ba4c6 Ke6d6 Bc6e4 e7e5 Kd8e8 Kd6e6 Ke8d8
  8/11	00&#58;00	       3.770	145.000	-0,23	Ba4c6 Ke6d6 Bc6e4 e7e5 Kd8e8 Kd6e6 Ke8d8 Bg3f4
  9/20	00&#58;00	      35.410	472.000	-0,45	Ba4c6 Ke6d6 Bc6e4 e7e5 Be4f3 Kd6c5 Kd8e7 Kc5d4 Bf3e2 e5e4 Ke7f6 e4e3 Kf6f5
  9/20	00&#58;00	      47.188	462.000	-0,38	Kd8c8 Ke6d6 Kc8b7 Kd6c5 Ba4d7 e7e5 Bd7f5 Kc5d4 Kb7b6 e5e4 Kb6b5
 10/20-	00&#58;00	      53.297	455.000	-0,43	Kd8c8 Ke6d6
 10/20	00&#58;00	      80.089	530.000	-0,48	Kd8c8 Ke6d6 Kc8b7 Kd6c5 Ba4c6 e7e5 Bc6g2 Kc5d4 Kb7b6 e5e4 Kb6b5 e4e3 Kb5xb4 e3e2 c2c3+ Kd4d3
 10/20	00&#58;00	      86.078	531.000	-0,45	Ba4c6 Ke6d6 Bc6e4 e7e5 Be4f3 Kd6c5 Kd8e7 Kc5d4 Bf3e2 e5e4 Ke7f6 e4e3 Kf6f5
 11/20	00&#58;00	     114.822	574.000	-0,49	Ba4c6 Ke6d6 Bc6e4 e7e5 Be4f3 Kd6c5 Kd8e7 Kc5d4 Bf3e2 e5e4 Ke7f6 e4e3 Kf6f5 Bg3c7 Kf5g4
 12/24+	00&#58;00	     222.806	673.000	-0,43	Ba4c6
 12/24	00&#58;00	     226.442	675.000	-0,43	Ba4c6 Ke6d6 Bc6g2 e7e5 Kd8e8 Kd6c5 Ke8f7 Kc5d4 Kf7g6 e5e4 Kg6f5 e4e3 Bg2f1 Kd4c3 Kf5e4 Bg3f2 Bf1d3
 13/24	00&#58;00	     294.929	710.000	-0,40	Ba4c6 Ke6d6 Bc6g2 e7e5 Kd8e8 Kd6c5 Ke8f7 Kc5d4 Kf7g6 e5e4 Kg6f5 e4e3 Bg2f1 Bg3e5 Kf5g4 Kd4c3 Kg4f3 Kc3d2 Bf1e2
 13/24	00&#58;00	     458.519	733.000	-0,39	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 Kd6d5 Kf5g4 Bg3c7 Ba4b3+ Kd5d4 Bb3f7 Bc7b8 Bf7g8 Kd4c3 Bg8b3
 14/25	00&#58;00	     526.906	742.000	-0,39	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 Kd6d5 Kf5g4 Bg3c7 Ba4b3+ Kd5d4 Bb3f7 Bc7b8 Bf7g8 Kd4c3 Bg8b3
 15/25	00&#58;01	     752.661	767.000	-0,39	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 Kd6d5 Kf5g4 Bg3e5 Ba4b3+ Kd5d4 Bb3a4 Be5c7 Ba4c6 Bc7b8 Bc6d7 Kd4c3
 16/25	00&#58;01	   1.176.799	796.000	-0,39	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 Kd6d5 Kf5g4 Bg3e5 Ba4b3+ Kd5d4 Bb3a4 Be5c7 Ba4c6 Bc7b8 Bc6a4 Bb8e5 Ba4b5 Kd4c3
 17/28	00&#58;02	   2.013.568	828.000	-0,41	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 e4e3 Ba4b5 Kd6c5 Bb5e2 Kc5d4 Be2b5 Kd4c3 Kf5e4 Kc3d2 Ke4f3 Bg3e5 Bb5f1 Be5d4 Bf1d3
 18/28	00&#58;04	   3.350.386	853.000	-0,41	Kd8e8 Ke6d6 Ke8f7 e7e5 Kf7g6 e5e4 Kg6f5 e4e3 Ba4b5 Kd6c5 Bb5e2 Kc5d4 Be2b5 Kd4c3 Kf5e4 Kc3d2 Ke4f3 Bg3e5 Bb5f1 Be5d4 Bf1d3
 19/30-	00&#58;04	   3.936.505	860.000	-0,47	Kd8e8 Ke6d6
 19/30-	00&#58;06	   5.341.053	867.000	-0,59	Kd8e8 Ke6d6
 19/45-	00&#58;16	  14.813.456	921.000	-1,79	Kd8e8 Ke6d6
 19/57	00&#58;27	  25.953.183	957.000	-2,25	Kd8e8 Ke6d6 Ke8f7 e7e5 Ba4b5 e5e4 Bb5f1 Kd6c5 Kf7e8 e4e3 Ke8d7 Kc5d4 Kd7c6 Bg3h4 Kc6b6 Kd4c3 Kb6b5 Bh4e1 Bf1e2 Kc3xc2 Kb5c4 b4b3 Be2d3+ Kc2d2 Bd3e4 b3b2 Kc4b3 Kd2c1 Kb3a2 e3e2 Be4b1 Be1c3

Houdini needs 19 depths to get result less than -2.19. Why is it important ?
Engines use “null move pruning” to detect and eliminate weak moves.
In fact it is a cascade of threats with a gigantic reduction, for me “black hole”.
Houdini starts the first position (diagram 1) and gets the second position (diagram 2) after 3 moves and next tests a threat (null move pruning) which must give less than –2,19 to be a good threat.
Since the engine needs 19 depths to solve diagram 2 directly, it seems to be impossible to pass the threat.

Unfortunately, such positions can be millions. It applies to endgames mainly, but not only.
Practically, all infinite searches are unreliable.
How to play against such “imperfect“ engines?
Opposite castling and sacrifices with a longer motivation.

zullil · Post by **zullil** » Sat Aug 06, 2011 3:47 pm

lech wrote:This text is a try to show why machines (computers) are not able to solve many positions.
Machines will do all. Only wrong software can interfere with theirs good work.

Diagram 2:
[d]3K4/4p3/4k3/8/Bp6/6b1/2P5/8 w - - 0 1
It is a win for black.

Yes, but to justify the sacrifice that leads to this position, we must be certain that White cannot draw. Are you certain, based on the extra pawn and the position of the kings? If so, can you convert your method of evaluation into code? Perhaps it is not so simple.

Rather than call the software "wrong," perhaps "incomplete" is a better word. Probably engine authors could add sufficient "endgame knowledge" to their programs so that this position is very quickly evaluated as an absolute win for Black. But writing that code may be tricky, and not much fun to do. It may be more practical to simply rely on 7-man tablebases.

mar · Post by **mar** » Sat Aug 06, 2011 4:07 pm

Practically, all infinite searches are unreliable.

Chess engines are nothing but stupid brute force searchers. No magic oracles. No real AI. Just brute force. But give them TIME and they will find the solution. In fact even the weakest engine ever written (in the case it has no bugs and having sufficient ply limit) should find the solution in a few (hundreds, thousands, millions, ...) years. Just sit and wait. Pruning, reductions: just turn them all off and the engine will see the solution at lower depth but it will take MUCH more time. These techniques simply sacrifice accuracy for speed. No magic either. And if you do it right (balance) it pays off in most cases. What I wanted to say is that search is actually always reliable. It simply works

Once the search hits terminal nodes/tablebase positions in all branches => problem solved. Otherwise it's nothing but a better or worse guess. Of course in theory if you had a perfect evaluation function you need no search at all (1 ply search will do). Once we have 32-man tablebases chess will be solved. But that's not going to happen. It would take insane amounts of time to calculate and insane amounts of space to store. So the best thing we can do is to search.

zullil · Post by **zullil** » Sat Aug 06, 2011 4:08 pm

zullil wrote:
lech wrote:This text is a try to show why machines (computers) are not able to solve many positions.
Machines will do all. Only wrong software can interfere with theirs good work.

Diagram 2:
[d]3K4/4p3/4k3/8/Bp6/6b1/2P5/8 w - - 0 1
It is a win for black.
Yes, but to justify the sacrifice that leads to this position, we must be certain that White cannot draw. Are you certain, based on the extra pawn and the position of the kings? If so, can you convert your method of evaluation into code? Perhaps it is not so simple.

Rather than call the software "wrong," perhaps "incomplete" is a better word. Probably engine authors could add sufficient "endgame knowledge" to their programs so that this position is very quickly evaluated as an absolute win for Black. But writing that code may be tricky, and not much fun to do. It may be more practical to simply rely on 7-man tablebases.

I wonder if anyone has written a "probabilistic static endgame evaluator?"

The input would be an "endgame position" (such as the one above) and the output would be three numbers pW, pD, pL in the interval [0,1] whose sum is 1. These would represent the (estimated) probabilities that the player to move wins, draws or loses, respectively. Obviously these could then be combined to yield the evaluation (ie, expected value) of the position.

lech · Post by **lech** » Sat Aug 06, 2011 8:32 pm

The main idea of this topic is to try to explain all who not programming, why top engines can’t solve such simple positions.
I hope I realized it.
All the rest is main personal opinion, and I have a right to do it.
I can’t see nodes like in “Matrix” yet

, but I hope that my knowledge in chess and programming give me the right.
I hope that this example positions will give a positive reflection for others.

Dirt · Post by **Dirt** » Sat Aug 06, 2011 8:40 pm

mar wrote:Chess engines are nothing but stupid brute force searchers. No magic oracles. No real AI. Just brute force. But give them TIME and they will find the solution. In fact even the weakest engine ever written (in the case it has no bugs and having sufficient ply limit) should find the solution in a few (hundreds, thousands, millions, ...) years.

It's hard to be sure there are no bugs, though. Even the strongest engine, Houdini 1.5a, fails that test.

Don · Post by **Don** » Sat Aug 06, 2011 9:05 pm

lech wrote:This text is a try to show why machines (computers) are not able to solve many positions.
Machines will do all. Only wrong software can interfere with theirs good work.
Diagram 1:
[d]3N2r1/2K1p3/4Pk2/8/Bp5b/8/2P5/8 b - - 0 1
It is a simple endgame position with 10 pieces.

Welcome to the wonderful world of selective search! The holy grail of computer chess is knowing which moves to prune and which moves not to prune so if you can figure that out, please let us know

Your position illustrates that computers still play far from perfect. Every once in a while someone asks how close are computer to playing perfect chess and the answer is "not very."

kgburcham · Post by **kgburcham** » Sat Aug 06, 2011 9:17 pm

I am going to play with this position a little more. Different gui and different settings. Also seems this is a mate in x.
kgburcham

[D] 3N2r1/2K1p3/4Pk2/8/Bp5b/8/2P5/8 b - -

Houdini 1.5ab-16 x64 [12 thr 14 spl] (4096 MB)

33/57 2:15 -2.24 1...Bg3+ 2.Kd7 Rg4 3.Nc6 Bd6 4.Bb3 Rg5 (4.086.089.061) 30114
34/63 3:16 -2.24 1...Bg3+ 2.Kd7 Rg4 3.Nc6 Bd6 4.Bb3 Rg5 (6.042.815.375) 30707
35/63 5:03 -2.24 1...Bg3+ 2.Kd7 Rg4 3.Nc6 Bd6 4.Bb3 (9.760.374.583) 32109
36/68 8:49 -2.36++ 1...Bg3+ (17.461.561.311) 33003
36/68 8:54 -2.60++ 1...Bg3+ (17.563.590.881) 32863
36/71 9:13 -3.17++ 1...Bg3+ (18.063.047.930) 32607
36/78 10:21 -4.71++ 1...Bg3+ (20.283.747.566) 32655
36/86 16:25 -8.86++ 1...Bg3+ (32.710.551.188) 33203
36/86 27:33 -11.12 1...Bg3+ 2.Kd7 Rxd8+ 3.Kxd8 Kxe6 4.Ke8 Kd6 5.Kf7 e5 (59.241.580.733) 35830
37/91 30:47 -11.12 1...Bg3+ 2.Kd7 Rxd8+ 3.Kxd8 Kxe6 4.Ke8 Kd6 5.Kf7 e5 (67.322.306.183) 36436

zullil · Post by **zullil** » Sat Aug 06, 2011 10:40 pm

kgburcham wrote: [D] 3N2r1/2K1p3/4Pk2/8/Bp5b/8/2P5/8 b - -

Houdini 1.5ab-16 x64 [12 thr 14 spl] (4096 MB)

36/86 27:33 -11.12 1...Bg3+ 2.Kd7 Rxd8+ 3.Kxd8 Kxe6 4.Ke8 Kd6 5.Kf7 e5 (59.241.580.733) 35830
37/91 30:47 -11.12 1...Bg3+ 2.Kd7 Rxd8+ 3.Kxd8 Kxe6 4.Ke8 Kd6 5.Kf7 e5 (67.322.306.183) 36436

Demonstrating that Houdini's search succeeds---just not quickly.

bob · Post by **bob** » Sat Aug 06, 2011 11:54 pm

Don wrote:
lech wrote:This text is a try to show why machines (computers) are not able to solve many positions.
Machines will do all. Only wrong software can interfere with theirs good work.
Diagram 1:
[d]3N2r1/2K1p3/4Pk2/8/Bp5b/8/2P5/8 b - - 0 1
It is a simple endgame position with 10 pieces.
Welcome to the wonderful world of selective search! The holy grail of computer chess is knowing which moves to prune and which moves not to prune so if you can figure that out, please let us know

Your position illustrates that computers still play far from perfect. Every once in a while someone asks how close are computer to playing perfect chess and the answer is "not very."

But don't forget to complete the statement...

"not very, but much closer to it than the best humans are."

Black hole

Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole

Re: Black hole