# Computers finding computer moves

There are positions in chess where the only good move (e.g. only move that preserves equality) is very unnatural for a human player. For instance if this move goes against common principles or if it opens the king to a potential attack, a human might be less likely to play this move.

Have there been attempts to find such positions with a computer?

I am not interested in using this to identify cheaters (=humans using illegally computer help in chess).

I am more interested in whether this could be used to win equal positions against humans. For example with such analysis, one could improve opening preparation, steering towards positions where the opponent has to find a "computer move". Also in endgame studies it could be used for the same reason.

• You'd have to teach a computer how humans evaluate positions for it to be able to judge how difficult it would be for humans to find a certain move. Teaching computers to think like humans is very difficult. Commented Nov 27, 2016 at 21:56
• I did not say it would be easy ;-), however I don't think you really need to teach them to think like humans. As an example one could for instance search for positions where the only decent move (which preserves equality) sacrifices material. Or where the only decent move creates an isolated doubled pawn or.... Probably the output from the computer would still need to be checked by a human, whether it satisfies the requirement (being a non-human computer move), but the first selection could be done by a computer. Commented Nov 27, 2016 at 22:31
• In your comment you named some elements which would make a move less likely to be played by a human. I would say that there are positions in which sacrifices are obvious and if your isolated pawn would be a passer, that would make it obvious as well. I think it would be very hard to reduce human evaluation of positions to such simple criteria (which do not have lots of "ifs" and "buts" attached) in such a way that a computer can make a workable selection. If you can make such a reduction accurately, however, it would be relatively easy to run such an attempt yourself. Commented Nov 27, 2016 at 22:39
• I guess what I mean to say is that this question comes down to if you can formulate a set of reasonably narrow criteria which are formal enough for a computer or not. Commented Nov 27, 2016 at 22:41
• Yes, I am aware of that. That's why I asked whether anybody has made an "attempt" (...to find such criteria and run them through an engine).. Commented Nov 27, 2016 at 23:29

One of the most common "unnatural" moves for a human is an under-developing move. Computers find them easily, as it's just another move to check but human's have invested time and energy developing a piece - why then move it back again?

Computers just crunch and then evaluate. Humans are much more emotive and have certain biases for development so it's very easy as a human to overlook a move that a computer would find in a heartbeat.

(one example might be re-routing a queen from one side of the board to the other, via the home-square - e.g. Qa5-d8-h4)

I realise this is a general answer but might help :)

Have you seen the Computer Chess Rating Lists competitions? They pit chess engines against one another for the purpose of assigning chess engines consistent ratings, and for determining which chess engines are the strongest using fair methodologies (such as by using controlled conditions and hardware).

I ask because in the process of doing this you can find games of chess engines (with ratings much higher than humans) playing chess engines (with ratings much higher than humans). Some of these games are very strange to watch.

These chess engines can see so many moves beyond what a normal human can (even GMs), that the moves they make are very nuanced and strange to us (and GMs).

Have there been attempts to find such positions with a computer? Indirectly, yes but as a consequence of pitting chess engines against one another. These games, and their odd moves can be downloaded and witnessed.

For example:

``````

[Event "CCRL 40/4"]
[Site "CCRL"]
[Date "2016.11.25"]
[Round "351.6.759"]
[White "Fire 5 64-bit 4CPU"]
[Black "Stockfish 8 64-bit 4CPU"]
[Result "0-1"]
[BlackElo "3481"]
[ECO "B06"]
[Opening "Modern"]
[Variation "1.e4 g6 2.d4 d6"]
[WhiteElo "3324"]
[Termination "normal"]
[PlyCount "119"]
[WhiteType "human"]
[BlackType "human"]
[FEN ""]

1. e4 g6 2. d4 d6 3. c3 Nf6 4. Bd3 Bg7 5. f4 O-O 6. Nf3 c5 7. dxc5 Nbd7 8.
cxd6 exd6 9. Be3 Nxe4 10. Bxe4 Re8 11. Qd3 Nf6 12. Bxg6 hxg6 13. O-O Bf5
14. Qd2 Qa5 15. Re1 Rac8 16. a4 a6 17. Bd4 Bg4 18. Rf1 Ne4 19. Qd1 Rc4 20.
Bxg7 Kxg7 21. Qd3 Qc5+ 22. Nd4 Be2 23. Qxe2 Rxd4 24. Qe3 Rxa4 25. Qxc5 Nxc5
26. Rxa4 Nxa4 27. b3 Nc5 28. Nd2 Re3 29. Rf3 Rxf3 30. gxf3 Kf6 31. Kf2 d5
32. Ke3 Ne6 33. c4 dxc4 34. Nxc4 b5 35. Nd6 Ke7 36. Ne4 f5 37. Nc3 Kd6 38.
b4 Nd8 39. Na2 Nc6 40. Ke2 Nd4+ 41. Ke3 Kd5 42. Kd3 Nxf3 43. h3 Ng1 44. h4
Nf3 45. Nc3+ Ke6 46. Ke3 Ne1 47. Kd2 Ng2 48. Ne2 Kd6 49. Kd3 Nxh4 50. Ng1
Ng2 51. Ne2 Kc6 52. Nd4+ Kc7 53. Ne2 Kb6 54. Kd4 a5 55. bxa5+ Kxa5 56. Kc3
Ka4 57. Kd2 Ka3 58. Kc2 b4 59. Kb1 Kb3 60. Nd4+ 0-1

```
```

And in this game (NOTE) Watch WHITE's Knight in this match between move 12-15. WHITE spends 3 moves to relocate his Knight to an adjacent square (because he can see far enough ahead to want it there, and can afford 3 moves to get him there):

``````

[Event "CCRL 40/4"]
[Site "CCRL"]
[Date "2016.11.25"]
[Round "351.6.716"]
[White "Stockfish 8 64-bit 4CPU"]
[Black "Fire 5 64-bit 4CPU"]
[Result "1-0"]
[ECO "A00"]
[Opening "Benko's opening"]
[PlyCount "208"]
[WhiteElo "3481"]
[BlackElo "3324"]
[FEN ""]

1. g3 e5 2. c4 Nf6 3. Bg2 Nc6 4. Nc3 Bb4 5. Nd5 O-O 6. a3 Bc5 7. Nc3 Bd4 8. Nf3
Bxc3 9. bxc3 d6 10. O-O Bd7 11. d3 h6 12. e4 Re8 13. Ne1 Qc8 14. Nc2 Bh3 15.
Ne3 Ne7 16. f4 Bxg2 17. Kxg2 exf4 18. Rxf4 Nd7 19. Rf1 f6 20. a4 Ne5 21. a5 c5
22. Qe2 Qd7 23. Bd2 Kh8 24. Ra2 Rac8 25. h4 Qc7 26. g4 Rg8 27. Be1 Nf7 28. Bg3
Nc6 29. Rb1 Rcd8 30. Rb5 Qc8 31. Nf5 Rd7 32. Rb1 Re8 33. Rba1 Qd8 34. Qf3 Nfe5
35. Qd1 Nf7 36. Kg1 Kh7 37. Kh2 Kg8 38. Ne3 Nfe5 39. Qe2 Rc7 40. Bf2 Rc8 41.
Nf5 Qc7 42. Bg3 Nf7 43. Rf1 Rcd8 44. Qf3 Nfe5 45. Qd1 Ne7 46. Ne3 N7c6 47. Kg1
Rf8 48. a6 b6 49. Raf2 Qd7 50. d4 Nf7 51. Nd5 Qc8 52. Qd3 Rde8 53. Rxf6 Qxg4
54. R1f4 Qd7 55. Bf2 Ne7 56. Qg3 Nxd5 57. cxd5 cxd4 58. cxd4 Nh8 59. e5 Rxf6
60. exf6 h5 61. Kh2 Re2 62. Rf5 Re8 63. Qg5 Rf8 64. Bg3 Rf7 65. Rf4 Qc8 66.
Qxh5 Rxf6 67. Rxf6 gxf6 68. Bxd6 Qc2+ 69. Kg3 Nf7 70. Ba3 Qc7+ 71. d6 Qc2 72.
Qf3 Qa4 73. Qd3 Kg7 74. h5 Kh8 75. Kf4 Nh6 76. Kf3 Qd7 77. Ke2 Nf7 78. Qf3 Kg7
79. Qe4 Kh8 80. Qe7 Qb5+ 81. Ke3 Qg5+ 82. Kd3 Qf5+ 83. Kc3 Qa5+ 84. Bb4 Qa1+
85. Kb3 Qb1+ 86. Kc4 Qa2+ 87. Kd3 Qb1+ 88. Ke2 Qb2+ 89. Bd2 Qb5+ 90. Ke3 Qb3+
91. Ke4 Kg7 92. d7 Qb1+ 93. Ke3 Qg1+ 94. Kd3 Qf1+ 95. Kc2 Qc4+ 96. Kb2 Qd5 97.
h6+ Kg6 98. h7 Qb5+ 99. Bb4 Qh5 100. Qxf7+ Kxf7 101. d8=Q Qe2+ 102. Kc3 Qf3+
103. Kc2 Qe2+ 104. Bd2 Qc4+ 1-0

```
```

First off, this is a very valid question.

The short answer is that no - to my knowledge, there are no programs that can somehow "steer" the game into positions where "unnatural" moves are good. While it is certainly possible to create such an engine, remember that a natural move is subjective.

However, you don't need a computer program to do it for you! You can do it yourself. Select a complicated opening, especially something like the Najdorf or some other sharp line, and you'll notice that there will be many opportunities for you to lead your opponent into muddy water - complex positions where you know the theory sue to computer analysis, and your opponent might not.

Here's the thing though; to attempt such a thing, you'd need to be a VERY strong player already, at least FIDE master I would say. The reason is that while grandmasters play complicated positions on purpose all the time, they have to do it to beat equally strong players. But they don't need to if they're playing against even a 2000 rated player, as they can just gradually outplay them, without need for double edged positions.

For practical purposes, I will repeat that complicated openings, which you shouldn't be playing in the first place unless you're a master level player at least, will offer an incredible array of tough positions which are amazing fro computers, yet in which humans make plenty of mistakes.

Also, note that good players will play unnatural moves if they are good! That's part being a good player - considering all possible candidate moves.

To make myself even more explicit, Kasparov once said that even he can't calculate everything. In complex positions, there naturally tend to be no real natural moves. That's where computers dominate.

(I am an expert rated USCF player)

First of all I think that your question needs to center on how to discover unusual moves that hold or improve your position evaluation.

First of all, in equal positions sometimes any move is equally good. Therefore, is the frequency of moves what gives us an indication of the preferred variations that correspond aproximately to the same number of plans. In this circunstance an engine is almost uncapable to discover a move of value, due to the horizon effect. You should select by yourself a potential move.

This kind of position can be of advantage for your purpose. You can explore withing the several moves ordered by the engine the second or the third alternative and explore its continuation (different from chess database moves) and adjust the resulting line with your chess knowledge and your plan.

A different approach that sometimes I have experimented is to find positons in which there are only one best move according to an engine and the remaning moves have an inferior valuations of at least 1/4 of pawn. This is a kind of position that I consider risky for my oponent and if this variation is preceded by an unusual move as chosen with the method previously explained, you have build a potential surprise for your opponent that could be useful for your preparation.

• Whome voted negatively and why? Isn't right to say why? Why you hide? Commented Dec 20, 2016 at 4:31