Man vs machine - how to beat chess engine

[overlord]

Komodo 4 is much stronger than Komodo 3 from the human point of view...after maybe 30 loses here is the first draw...

[Event "5 min. game (+3 s/per move for man)"]
[Date "2012.06.13"]
[White "Kvicala, Miroslav"]
[Black "Komodo64 no-sse Version 4"]
[Result "1/2-1/2"]
[ECO "C02"]
[WhiteElo "2035"]

1. e4 e6 2. d4 d5 3. e5 c5 4. c3 b6 5. Nf3 Qd7 6. Bd3 Ba6 7. Qe2 Bxd3 8. Qxd3
Nc6 9. O-O Nge7 10. a3 c4 11. Qc2 Na5 12. Nbd2 Nec6 13. Rb1 Be7 14. Re1 h6 15.
Nf1 Nb3 16. Bf4 O-O 17. Ng3 a6 18. Nd2 Nca5 19. Nxb3 Nxb3 20. Qe2 Kh8 21. Nf1
Rae8 22. Nd2 Nxd2 23. Qxd2 g5 24. Bg3 b5 25. f4 Rg8 26. Rf1 Ref8 27. Rf2 Bd8
28. Rbf1 Rg7 29. f5 Rfg8 30. Rf3 Qc6 31. f6 Rh7 32. Bf2 g4 33. Rf4 Bb6 34. Be3
h5 35. g3 a5 36. R4f2 Rg6 37. Ra1 a4 38. Bg5 Qc8 39. Raf1 Qg8 40. Bh4 Qf8 41.
Kg2 Qh6 42. Qxh6 Rhxh6 43. Rf4 Kg8 44. h3 Kf8 45. hxg4 hxg4 46. Rh1 Ke8 47.
Rhf1 Kd7 48. R1f2 Kc6 49. Rf1 Bc7 50. Re1 Rg8 51. Re2 Bb6 52. Re1 Rh5 53. Re2
Rg6 54. Re1 Rh7 55. Re2 Rg8 56. Re1 Rh6 57. Re2 Kb7 58. Re1 Rhg6 59. Re2 Rh8
60. Re1 Kc6 61. Re2 Bc7 62. Re1 Rc8 63. Re2 Rgg8 64. Re1 Ra8 65. Re2 Kb7 66.
Re1 Kb6 67. Re2 Rac8 68. Re1 Rcd8 69. Re2 Kc6 70. Re1 Rde8 71. Re2 Ra8 72. Re1
Rad8 73. Re2 Ba5 74. Re1 Bb6 75. Re2 Ra8 76. Re1 Rae8 77. Re2 Rb8 78. Re1 Rbc8
79. Kf2 Rg6 80. Kg2 Ra8 81. Ra1 Bc7 82. Rf2 Rag8 83. Re2 Rh8 84. Re3 Rh5 85.
Re2 Bb6 86. Rf2 Rhh6 87. Re2 Kd7 88. Rd2 Bc7 89. Re2 Rh8 90. Re3 Rh5 91. Re2
Rgh6 92. Re3 Rf5 93. Re2 Kc6 94. Rf2 Rxf2+ 95. Kxf2 Rh5 96. Rf1 Bxe5 97. dxe5
Rxe5 98. Re1 Rxe1 99. Kxe1 Kd6 100. Ke2 e5 101. Ke3 Ke6 102. Kf2 d4 103. Ke2
Kd5 104. Kd2 d3 105. Ke3 Ke6 106. Bg5 Kf5 107. Bh4 e4 108. Kd2 Kg6 109. Ke3 Kf5
110. Kd2 Kg6 111. Ke3 Kf5 1/2-1/2

[overlord]

Maybe not so correct game...but clear fight

[Event "5 min. game"]
[Date "2012.06.13"]
[White "Kvicala, Miroslav"]
[Black "Cheng 1,07 ja"]
[Result "1/2-1/2"]
[WhiteElo "2035"]

{256MB, Mirekbook.ctg} 1. d4 g6 2. e4 Bg7 3. Nc3 d6 4. Be3 Nf6 5. Be2 Nc6 6.
Qd2 Ng4 7. Bxg4 Bxg4 8. f3 Bd7 9. O-O-O e5 10. Nge2 Qh4 11. g4 O-O 12. d5 Nb4
13. Ng3 Bf6 14. Kb1 Rfe8 15. Rdg1 a5 16. Nf5 gxf5 17. gxf5+ Kh8 18. Rg4 Qh5 19.
Rhg1 Rg8 20. Bh6 Qh3 21. a3 Rxg4 22. fxg4 Na6 23. g5 Be7 24. Qf2 Nc5 25. Rg3
Qh4 26. Qg2 b5 27. Rh3 Qe1+ 28. Ka2 Na6 29. g6 fxg6 30. Bd2 Nb4+ 31. axb4 axb4+
32. Kb3 Qa1 33. Rxh7+ Kxh7 34. Qxg6+ Kh8 35. Qh6+ Kg8 36. Qg6+ Kh8 37. Qh6+ Kg8
38. Qg6+ Kh8 1/2-1/2

[tmokonen]

You are using the 64 bit SSE4 version of Critter, which may give different results than the 32 bit version that Miroslav is using. With one thread, using the 32 bit version of Critter, on slower hardware than yours, this is what I get:

FEN: r1bqk2r/pp2bppp/2n1pn2/2pp4/3P4/2PBPN2/PP1N1PPP/R1BQ1RK1 b kq - 2 7

Critter 1.4 32bit:
2/2 00:00 163 0 -0.09 O-O Nf3-g5 Bc8-d7 Nd2-f3
3/4 00:00 824 412,000 +0.07 O-O Nf3-e5 Nc6xe5 d4xe5
4/6 00:00 3,195 399,375 +0.12 O-O b2-b3 c5xd4 e3xd4 Qd8-c7
5/6 00:00 3,976 397,600 +0.10 O-O b2-b3 c5xd4 e3xd4 Be7-d6 h2-h3
6/8 00:00 5,203 371,642 +0.07 O-O b2-b3 c5xd4 e3xd4 Be7-d6 Nf3-g5 h7-h6
7/10 00:00 9,673 420,565 +0.19 O-O b2-b3 Bc8-d7 Bc1-b2 c5xd4 e3xd4 Qd8-c7 Nf3-g5 h7-h6
8/11 00:00 20,876 509,170 +0.19 O-O b2-b3 Bc8-d7 Bc1-b2 c5xd4 e3xd4 Qd8-c7 Nf3-g5 h7-h6
9/13 00:00 27,928 537,076 +0.21 O-O b2-b3 Bc8-d7 Bc1-b2 c5xd4 e3xd4 Qd8-c7 h2-h3 h7-h6 Qd1-e2 Ra8-c8
10/15 00:00 51,555 579,269 +0.19 O-O b2-b3 Bc8-d7 Bc1-b2 c5xd4 e3xd4 Qd8-c7 Qd1-e2 h7-h6 Nf3-e5 Nc6xe5 Qe2xe5 Qc7xe5 d4xe5
11/18 00:00 137,117 652,938 +0.24 c5-c4 Bd3-c2 O-O b2-b3 b7-b5 b3xc4 b5xc4 Ra1-b1 Qd8-c7 h2-h3 h7-h6 Qd1-e2 Ra8-b8 Rb1xb8 Qc7xb8
12/20- 00:00 294,105 682,378 +0.15 c5-c4 Bd3-c2 O-O b2-b3 b7-b5 b3xc4 b5xc4 Ra1-b1 Qd8-c7 h2-h3 h7-h6 Qd1-e2 Ra8-b8 Rb1xb8 Qc7xb8
12/20 00:00 362,488 678,816 +0.09 c5-c4 Bd3-c2 O-O e3-e4 b7-b5 Nf3-e5 Nc6xe5 d4xe5 Nf6-d7 Nd2-f3 Bc8-b7 e4xd5 Bb7xd5 Bc1-e3 Bd5xf3 Qd1xf3 Nd7xe5
13/20 00:01 528,279 682,531 +0.09 c5-c4 Bd3-c2 O-O e3-e4 b7-b5 Nf3-e5 Nc6xe5 d4xe5 Nf6-d7 Nd2-f3 Bc8-b7 e4xd5 Bb7xd5 Bc1-e3 Bd5xf3 Qd1xf3 Nd7xe5
14/29- 00:04 2,743,412 720,244 +0.01 c5-c4 Bd3-c2 O-O e3-e4 b7-b5 Nf3-e5 Nc6xe5 d4xe5 Nf6-d7 Nd2-f3 Bc8-b7 e4xd5 Bb7xd5 Bc1-e3 Bd5xf3 Qd1xf3 Nd7xe5
14/29 00:04 3,141,347 714,104 +0.04 c5-c4 Bd3-c2 b7-b5 e3-e4 Qd8-c7 e4-e5 Nf6-d7 b2-b3 O-O b3xc4 b5xc4 Ra1-b1 Bc8-b7 Qd1-e2 Ra8-b8
15/34- 00:09 6,736,043 723,916 -0.04 c5-c4 Bd3-c2 b7-b5 e3-e4 Qd8-c7 e4-e5 Nf6-d7 b2-b3 O-O b3xc4 b5xc4 Ra1-b1 Bc8-b7 Qd1-e2 Ra8-b8
15/34 00:09 6,982,827 722,785 -0.04 c5-c4 Bd3-c2 b7-b5 e3-e4 Qd8-c7 e4-e5 Nf6-d7 b2-b3 O-O Bc1-b2 Bc8-a6 a2-a4 c4xb3 Bc2-d3 Ra8-b8 Qd1-e2 Qc7-b7 Nd2xb3 b5xa4 Bd3xa6 Qb7xb3
15/34 00:10 7,393,879 724,038 +0.01 O-O d4xc5 Be7xc5 b2-b4 Bc5-d6 b4-b5 Nc6-e5 Nf3xe5 Bd6xe5 Bc1-b2 a7-a6 b5xa6 b7xa6 Qd1-c2 Ra8-b8 Ra1-b1 Qd8-c7 f2-f4 Be5-d6 c3-c4 Bd6-c5 Rb1-e1 h7-h6 Bb2xf6 g7xf6 c4xd5 e6xd5
16/34 00:11 8,156,612 730,224 +0.01 O-O d4xc5 Be7xc5 b2-b4 Bc5-d6 b4-b5 Nc6-e5 Nf3xe5 Bd6xe5 Bc1-b2 a7-a6 b5xa6 b7xa6 Qd1-c2 Ra8-b8 Ra1-b1 Qd8-c7 f2-f4 Be5-d6 c3-c4 Bd6-c5 Rb1-e1 h7-h6 Bb2xf6 g7xf6 c4xd5 e6xd5
17/34 00:15 11,497,601 744,663 +0.01 O-O d4xc5 Be7xc5 b2-b4 Bc5-d6 b4-b5 Nc6-e5 Nf3xe5 Bd6xe5 Bc1-b2 a7-a6 b5xa6 b7xa6 Qd1-c2 Ra8-b8 Ra1-b1 Qd8-c7 f2-f4 Be5-d6 c3-c4 Bd6-c5 Rb1-e1 h7-h6 Bb2xf6 g7xf6 c4xd5 e6xd5

I think it's entirely possible that Critter played c4 here in a 5/0 game, depending on the hardware and the number of threads being used.

[overlord]

Thanks for analysis. It is as you said. It was 32 bit Critter running on smartphone Samsung Samsung (N7000 - 2x1,4 GHz).

[overlord]

After many bad loses against new Critter 1.6 32-bit (on smartphone) here is the first draw

[Event "10 min. game"]
[Date "2012.06.16"]
[White "Kvicala, Miroslav"]
[Black "Critter 1.6 32-bit"]
[Result "1/2-1/2"]
[ECO "D35"]
[WhiteElo "2035"]

1. d4 e6 2. c4 d5 3. Nf3 Be7 4. Nc3 Nf6 5. cxd5 exd5 6.
Bg5 O-O 7. Qc2 Nc6 8. e3 h6 9. Bh4 Re8 10. a3 Ne4 11. Bxe7 Rxe7 12. Bd3 Nxc3
13. Qxc3 Bg4 14. Be2 Qd6 15. b4 Nd8 16. O-O Ne6 17. Rfe1 Bxf3 18. Bxf3 Ng5 19.
Be2 Ne4 20. Qc2 c6 21. Bd3 Rd8 22. Bxe4 Rxe4 23. Qc5 a6 24. Qxd6 Rxd6 25. a4
Re8 26. Rec1 Ra8 27. Rc5 g5 28. Rac1 Kg7 29. Kf1 Kg6 30. b5 cxb5 31. axb5 a5
32. Ra1 b6 33. Rc6 Rxc6 34. bxc6 Rc8 35. Rc1 a4 36. c7 Kf6 37. Rc6+ Ke7 38.
Rxb6 Rxc7 39. Ra6 Rc4 40. Ke2 Rb4 41. g4 Rb2+ 42. Kf3 Rb4 43. Kg3 Kd7 44. Rxh6
Kc7 45. Ra6 Kb7 46. Ra5 Kb6 47. Ra8 Kb7 48. Ra5 Kb6 49. Ra8 Kb7 50. Ra5 1/2-1/2

[overlord]

One more draw against Critter 1.6 32-bit on smartphone

[Event "5 min. game"]
[Site "Microsoft"]
[Date "2012.06.18"]
[Round "?"]
[White "Kvicala, Miroslav"]
[Black "Critter 1.6 32-bit"]
[Result "1/2-1/2"]
[WhiteElo "2035"]
[PlyCount "85"]
[TimeControl "300"]

{256MB, Mirekbook.ctg} 1. d4 d5 2. Nf3 Nf6 3. c3 Nc6 4. Bf4 h6 5. Nbd2 Bf5 6.
e3 e6 7. Ne5 Nxe5 8. Bxe5 Bd6 9. Nf3 O-O 10. Bd3 Bg4 11. Bxd6 Qxd6 12. h3 Bxf3
13. Qxf3 e5 14. dxe5 Qxe5 15. Qf5 Qxf5 16. Bxf5 g6 17. Bd3 Kg7 18. c4 Rad8 19.
cxd5 Rxd5 20. O-O-O Re8 21. Rd2 Rg5 22. g3 Rd8 23. Bc2 Rxd2 24. Kxd2 Rb5 25.
Kc1 Rd5 26. Rd1 Rh5 27. h4 Ra5 28. a3 Rc5 29. Rd4 Re5 30. Kd2 Rb5 31. Kc1 a5
32. Be4 h5 33. Bf3 c6 34. Kc2 Rf5 35. Rf4 Re5 36. Kc3 Ne8 37. Re4 Rc5+ 38. Rc4
Rf5 39. Rf4 Rb5 40. Rd4 Rf5 41. Rf4 Rc5+ 42. Rc4 Rf5 43. Rf4 1/2-1/2

[overlord]

[Event "5 min. game"]
[Date "2012.06.19"]
[White "Kvicala, Miroslav"]
[Black "Critter 1.6 32-bit"]
[Result "1/2-1/2"]
[WhiteElo "2035"]


1. e4 d6 2. d4 Nf6 3. Nc3 g6 4. f4 d5 5. e5 Ne4 6. Nxe4
dxe4 7. Be3 Nd7 8. Ne2 f5 9. c4 e6 10. Nc3 Nb6 11. Be2 c6 12. a3 a5 13. h4 Be7
14. g3 h5 15. Qd2 O-O 16. Rd1 Kh7 17. Kf2 Kg7 18. Rh2 Bd7 19. Rhh1 Rc8 20. Rh2
Qc7 21. Rhh1 Be8 22. Rh2 Bf7 23. Rhh1 Rcd8 24. Qc2 Rd7 25. Rd2 Rfd8 26. Rhd1
Ra8 27. Kg2 a4 28. Kh3 Kh7 29. Bf1 Kg8 30. Be2 Rdd8 31. Bf1 Kg7 32. Be2 Bg8 33.
Bf1 Bf8 34. Be2 Kh8 35. Bf1 Kg7 36. Be2 Rd7 37. Bf1 1/2-1/2

[styx]

why does stockfish play 10. ... h6 here? on my 4-core desktop computer it's not even considered (SF thinks Qa5 ist the best move). stockfish 2.2.2 on my mobile phone evaluates h6 for less than a second in the beginning, then it sticks with Nc6 as the best move. i cannot reproduce 12. .. Bf5 either. i don't know. maybe it's the difference between our hardware. but my phone (LG P990) should be slower than yours.

one thing for sure: a mobile phone is much weaker than a fast desktop machine. especially on short time controls like 5 minutes. but i am pretty sure, that only your phone had 5 minutes to think and you've taken much more time. it's still hard to win, though maybe not impossible for a 2000+ elo player.

but i still think you have absolutely no chance if:
- you both use equal time controls (a loss on time is a loss)
- you use a PC with modern multicore hardware
- you are not allowed to take back a move

you don't even have to try it with houdini, komodo, critter or stockfish. show me a game against deep shredder or programs of similar strength. that would be challenging enough.

[...]

[Event "5 min. game"]
[Site "Microsoft"]
[Date "2012.05.28"]
[Round "?"]
[White "Kvicala, Miroslav"]
[Black "Stockfish 2.2.2"]
[Result "1-0"]
[ECO "C42"]
[WhiteElo "2035"]
[PlyCount "123"]

1. e4 e5 2. Nf3 Nf6 3. Nxe5 d6 4. Nf3 Nxe4 5. Nc3 Nxc3 6. dxc3 Be7 7. Be3 O-O
8. Qd2 c5 9. O-O-O Be6 10. Kb1 h6 11. Bxh6 gxh6 12. Qxh6 Bf5 13. Rd5 Bg6 14.
Bd3 Bxd3 15. Rh5 f6 16. cxd3 Kf7 17. Nh4 Ke8 18. Nf5 Nc6 19. Re1 Kd7 20. Qe3
Re8 21. Rh7 Kc7 22. Nxe7 Rxe7 23. Rxe7+ Nxe7 24. Qxe7+ Qxe7 25. Rxe7+ Kb6 26.
h4 Rf8 27. g4 Rh8 28. h5 f5 29. f3 fxg4 30. fxg4 Rg8 31. Re4 Kc6 32. c4 Kd7 33.
Kc2 Rh8 34. Kd2 Rg8 35. Ke3 Rh8 36. Kf4 a5 37. Kg5 d5 38. cxd5 Kd6 39. h6 b5
40. Kg6 Kxd5 41. Rf4 Ke5 42. Rf5+ Kd4 43. Kg7 Rd8 44. h7 a4 45. h8=Q Rxh8 46.
Kxh8 a3 47. bxa3 c4 48. dxc4 bxc4 49. Rf4+ Kd3 50. Rxc4 Kxc4 51. g5 Kd5 52. g6
Kc4 53. g7 Kb5 54. g8=Q Kb6 55. Qc4 Kb7 56. Kg7 Kb6 57. Kf7 Kb7 58. Ke7 Kb6 59.
Kd7 Ka5 60. Qb4+ Ka6 61. Kc7 Ka7 62. Qb7# 1-0

[...]

[Kyodai]

Andreas

"but i still think you have absolutely no chance if.."

sigh... that's not the point at all - not even vaguely interesting...
The central issue here - the kern of the pudel if you like - is that
weaknesses of the chess programs are reveiled

And *that* is interesting - might even contribute to even stronger programs...

[styx]

you are right. and i appreciate miroslavs effort in revealing and exploiting the engines weaknesses. but playing on limited hardware on short time controls is a bit like showing games against engines with limited depth or uci-elo-limit. it is still kind of impressive (i personally could never beat my cellphone stockfish/critter with full strength at 1 sec/move), but then it's more a hardware "weakness" than an engine issue.