Man vs machine - how to beat chess engine

[overlord]

You are right,it is enough to increase engine aggressivness (contempt in Houdini for example).And even average engines will be deadly for humans.The second possibility is engine that uses gambit book...

[overlord]

Don, I have only Komodo 3 (I don´t if there is higher version for Android), but I wanted to know how will perform in the same opening (Frech advanced). Here is the result...

[Event "Blitz:5'"]
[Date "2013.06.13"]
[White "Kvicala, Miroslav"]
[Black "Komodo 3"]
[Result "1-0"]
[TimeControl "300"]

1. e4 e6 2. d4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Qb6 6. a3 c4 7. Qe2 Na5 8. Nbd2 Bd7 9. g3 Be7 10. Bg2 Nh6 11. O-O O-O 12. h3 Nf5 13. Nh2 Ba4 14. g4 Nh4 15. f4 Nxg2 16. Kxg2 Qc7 17. g5 Bc2 18. Rg1 b5 19. Ng4 Bd3 20. Qf2 Rfd8 21. Kh2 Rab8 22. Nf6+ Kh8 23. Qh4 gxf6 24. gxf6 Bf8 25. Nf3 Nb3 26. Ng5 Nxa1 27. Nxh7 Bxh7 28. Rg5 Qxe5 29. fxe5 Nc2 30. Rh5 Kg8 31. Rxh7 Bg7 32. Rxg7+ Kf8 33. Qh8# 1-0

[Lyudmil Tsvetkov]

I think their are 3 ways of drawing strong engines:

- the first one is to block the position as much as possible, as in this case the proportion of variations requiring calculation decreases sharply
- the second one is to keep on exchanging as much material as possible, which will also result in simplification of the position and significant decrease in the number of variations requiring computation; proceeding in this way is simple - wherever you notice the engine has developed a piece, you attack it with you own and exchange it, even at the cost of some (but only reasonable) compromising on the pawn structure. When you echtange a couple of minor pieces and a pair of rooks, the position will already look very simple to you to hold. It will be difficult to skip some tactics, while engines are far superior in complex positions with many pieces being open to a large degree.

- and the third one is, of course, to play like an engine ; it is difficult to learn to do that, but not impossible - I have set myself a target of learning to do that in a couple of lifetimes

Realistically the third one is inaccessible to the albeit limited human brain: we don't compute millions of moves per second, especially the awkward and counter-intuitive ones. It would take Mikhail Tal multiplied by Bobby Fisher, elevated to the power Garry Kasparov to reach that level, at least...

But the first two, illustrated by the draws you posted, are indeed a threat to poor computers

The problem is that engine developpers tune their engine to increase their ELO against other engines. That increases strength against a GM to a certain extent, but eventually you reach a glass ceiling, and you can gain lots of elo in comp vs comp, you're still dumb in positions where there is nothing to calculate (like blockades or lots of pieces exchanged and symmetric pawn structures).

It is possible however, to compromise and make an anti-human setting of an engine, that may be weaker against other computers, but where the engine will seek to avoid:
* exchanging pieces
* symmetric pawn structures
* blocked pawns
I can easily code that up in DiscoCheck, and I'm quite sure it will be weaker in computer vs computer rating lists, but it will give you a hard time to draw against it. If the computer refuses to play it your way (ie. cowardly draw) and forces you to play complex and sharp positions (and refuses any unfought 3-repetition draw), you are dead meat

Hi Lucas.
I will be eagerly waiting for DiscoCheck_Anti-human to play some games against. Then I will post the results for entertainment purposes. I think the more anti-human the engine will be, the more you will be compromising on quality. But you never know, you might be able to come up with something worth testing. It is very likely that a version tested against engines and against humans will perform better than a version tested exclusively against engines. Adding a human factor might certainly improve the quality of play.

You rightly say the 'albeit' limited human brain - it is a well-known fact that humans use only about 10% of their mental energy and the other 90% are wasted to no avail. I think that it is possible to use a higher amount of your potential mental energy by trying to concentrate better. When I play chess against engines I try to do that: I shut myself into a private room, switch off all available phones, do not think about how the game is going to end, i.e. I am psychologically unburdened, I play during the night whenever possible and sometimes even use some additional hearing protectors, so that I do not get disturbed even by the humming of the fans. And consequently, I am able to play quite successfully against even top engines. I am able to concentrate deeply in this way. I know I can never replicate my performance in real-life tournaments, because there the level of noise is sometimes too high, your opponent and other people might constantly distract you, you lose a fair amount of mental energy while thinking how the game is going to end, and some people like myself might sometimes even get too nervous to play decently.

I think concentration is the key to performing optimally in chess. But that is true also for engines. You can compare an ill-concentrated player to a well-concetrated player to what a Pentium 200 Mhz would provide an engine in respect of a computer with multiple cores. Modern computers are much more 'concentrated' than older ones used to be

[Lyudmil Tsvetkov]

You are right,it is enough to increase engine aggressivness (contempt in Houdini for example).And even average engines will be deadly for humans.The second possibility is engine that uses gambit book...

But then, it is not my problem that engines do not understand closed positions, as it is not the engines' problem that humans have considerable difficulties in complex positions, especially when open.

[Laskos]

You are right,it is enough to increase engine aggressivness (contempt in Houdini for example).And even average engines will be deadly for humans.The second possibility is engine that uses gambit book...

But then, it is not my problem that engines do not understand closed positions, as it is not the engines' problem that humans have considerable difficulties in complex positions, especially when open.

Just use contempt -3 on any top engine, they will deblock positions. Then they will crush any <2000 elo player even having -3.00 evaluation in the middlegame.

[overlord]

If the man has a good day, engine like Shredder may be massacred in miniature game

[Event "Blitz 5m"]
[Date "2013.06.26"]
[White "Kvicala, Miroslav"]
[Black "Shredder 12"]
[Result "1-0"]
[ECO "D43"]
[PlyCount "41"]

1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. Bg5 h6 6. Bh4 dxc4 7. e4 g5 8. Bg3 Bb4 9. Bxc4 Nxe4 10. O-O Nxg3 11. fxg3 Nd7 12. Qe2 O-O 13. Ne5 Nxe5 14. dxe5 Be7 15. Rad1 Qc7 16. Qe4 a6 17. Rf6 Rd8 18. Rxd8+ Qxd8 19. Rxh6 Bc5+ 20. Kh1 Kf8 21. Rh8+ 1-0

Main line of Caro-Kann is not so innocent as it may look...

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Texel 1.02"]
[Result "1-0"]
[ECO "B19"]
[PlyCount "79"]

1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6 6. Nf3 Nd7 7. h4 h6 8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 e6 11. Bd2 Ngf6 12. O O-O Bd6 13. Ne4 Nxe4 14. Qxe4 Nf6 15. Qe2 Qb6 16. Kb1 O-O 17. c4 Qa6 18. Ne5 Rfd8 19. g4 Nh7 20. g5 hxg5 21. h6 g6 22. Nxg6 Rab8 23. Ne5 Kf8 24. f4 Ke7 25. fxg5 Bxe5 26. dxe5 Qb6 27. Qh5 Rxd2 28. Rxd2 Rd8 29. Rxd8 Qxd8 30. Rd1 Qg8 31. Rf1 Qg6+ 32. Qxg6 fxg6 33. Rf6 Nf8 34. c5 b5 35. b4 a6 36. Kc2 Nh7 37. Rxg6 Nf8 38. Rg7+ Ke8 39. Rg8 Kf7 40. Rxf8+ 1-0

Well, Anti-Moscow variation is really challange for engines. Without good book they may become man´s victim very easily

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Cuckoo chess"]
[Result "1-0"]
[ECO "D43"]
[PlyCount "77"]


1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. Bg5 h6 6. Bh4 dxc4 7. e4 g5 8. Bg3 Bb4 9. Bxc4 b5 10. Bd3 g4 11. Ne5 Qxd4 12. O-O Ba6 13. Qe2 Rg8 14. Rfd1 Bxc3 15. bxc3 Qxc3 16. Rac1 Qa3 17. Nxc6 Nxc6 18. Rxc6 Rg5 19. e5 Nh5 20. Rxa6 Nxg3 21. hxg3 Qc5 22. Bxb5+ Kf8 23. Re1 Rc8 24. Ra4 Rd8 25. Qc4 Rd5 26. Qxc5+ Rxc5 27. Bc4 h5 28. Bb3 Rgxe5 29. Rxe5 Rxe5 30. Rxa7 Re1+ 31. Kh2 Rf1 32. a4 Rxf2 33. a5 Rf5 34. a6 Ra5 35. Bc4 Rc5 36. Rb7 Rxc4 37. a7 Ra4 38. Rb8+ Kg7 39. a8=Q 1-0

Advanced French...what to say...this line is simply engines grave...

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Critter 1.6"]
[Result "1-0"]
[ECO "C02"]
[PlyCount "65"]


1. e4 e6 2. d4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Qb6 6. a3 c4 7. Qe2 Na5 8. Nbd2 Bd7 9. g3 Be7 10. Bg2 Nh6 11. O-O O-O 12. h3 Nf5 13. g4 Nh4 14. Nxh4 Bxh4 15. f4 Be7 16. Kh2 Rae8 17. Rg1 Qc7 18. g5 Ba4 19. Nf1 Nb3 20. Rb1 Bd8 21. Ne3 Qd7 22. Ng4 Bc6 23. Be3 a5 24. Bf3 b5 25. Nf6+ Bxf6 26. gxf6 g6 27. Qf2 Kh8 28. Rg4 g5 29. Rxg5 Rc8 30. Rg7 Nxd4 31. Rxh7+ Kxh7 32. Qh4+ Kg8 33. Rg1# 1-0

[Henk]

If the man has a good day, engine like Shredder may be massacred in miniature game

[Event "Blitz 5m"]
[Date "2013.06.26"]
[White "Kvicala, Miroslav"]
[Black "Shredder 12"]
[Result "1-0"]
[ECO "D43"]
[PlyCount "41"]

1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. Bg5 h6 6. Bh4 dxc4 7. e4 g5 8. Bg3 Bb4 9. Bxc4 Nxe4 10. O-O Nxg3 11. fxg3 Nd7 12. Qe2 O-O 13. Ne5 Nxe5 14. dxe5 Be7 15. Rad1 Qc7 16. Qe4 a6 17. Rf6 Rd8 18. Rxd8+ Qxd8 19. Rxh6 Bc5+ 20. Kh1 Kf8 21. Rh8+ 1-0

Main line of Caro-Kann is not so innocent as it may look...

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Texel 1.02"]
[Result "1-0"]
[ECO "B19"]
[PlyCount "79"]

1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6 6. Nf3 Nd7 7. h4 h6 8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 e6 11. Bd2 Ngf6 12. O O-O Bd6 13. Ne4 Nxe4 14. Qxe4 Nf6 15. Qe2 Qb6 16. Kb1 O-O 17. c4 Qa6 18. Ne5 Rfd8 19. g4 Nh7 20. g5 hxg5 21. h6 g6 22. Nxg6 Rab8 23. Ne5 Kf8 24. f4 Ke7 25. fxg5 Bxe5 26. dxe5 Qb6 27. Qh5 Rxd2 28. Rxd2 Rd8 29. Rxd8 Qxd8 30. Rd1 Qg8 31. Rf1 Qg6+ 32. Qxg6 fxg6 33. Rf6 Nf8 34. c5 b5 35. b4 a6 36. Kc2 Nh7 37. Rxg6 Nf8 38. Rg7+ Ke8 39. Rg8 Kf7 40. Rxf8+ 1-0

Well, Anti-Moscow variation is really challange for engines. Without good book they may become man´s victim very easily

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Cuckoo chess"]
[Result "1-0"]
[ECO "D43"]
[PlyCount "77"]


1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. Bg5 h6 6. Bh4 dxc4 7. e4 g5 8. Bg3 Bb4 9. Bxc4 b5 10. Bd3 g4 11. Ne5 Qxd4 12. O-O Ba6 13. Qe2 Rg8 14. Rfd1 Bxc3 15. bxc3 Qxc3 16. Rac1 Qa3 17. Nxc6 Nxc6 18. Rxc6 Rg5 19. e5 Nh5 20. Rxa6 Nxg3 21. hxg3 Qc5 22. Bxb5+ Kf8 23. Re1 Rc8 24. Ra4 Rd8 25. Qc4 Rd5 26. Qxc5+ Rxc5 27. Bc4 h5 28. Bb3 Rgxe5 29. Rxe5 Rxe5 30. Rxa7 Re1+ 31. Kh2 Rf1 32. a4 Rxf2 33. a5 Rf5 34. a6 Ra5 35. Bc4 Rc5 36. Rb7 Rxc4 37. a7 Ra4 38. Rb8+ Kg7 39. a8=Q 1-0

Advanced French...what to say...this line is simply engines grave...

[Event "Blitz:5'"]
[Date "2013.06.25"]
[White "Kvicala, Miroslav"]
[Black "Critter 1.6"]
[Result "1-0"]
[ECO "C02"]
[PlyCount "65"]


1. e4 e6 2. d4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Qb6 6. a3 c4 7. Qe2 Na5 8. Nbd2 Bd7 9. g3 Be7 10. Bg2 Nh6 11. O-O O-O 12. h3 Nf5 13. g4 Nh4 14. Nxh4 Bxh4 15. f4 Be7 16. Kh2 Rae8 17. Rg1 Qc7 18. g5 Ba4 19. Nf1 Nb3 20. Rb1 Bd8 21. Ne3 Qd7 22. Ng4 Bc6 23. Be3 a5 24. Bf3 b5 25. Nf6+ Bxf6 26. gxf6 g6 27. Qf2 Kh8 28. Rg4 g5 29. Rxg5 Rc8 30. Rg7 Nxd4 31. Rxh7+ Kxh7 32. Qh4+ Kg8 33. Rg1# 1-0

It would be more interesting to me if I knew the estimated ELO of both players. What are your winning chances. Do you win more than 1% of the time.

[overlord]

Well, I don´t have statistics because I don´t store lost games (it is too depressive). Anyway, against weaker engines like Fritz, Shredder, Cuckoo chess or Cheng I am able to achieve about 25% of points (playing white). In case of Stockfish it is about 5% points (not wins, distribution is something like 1 win and 10 draw), Critter is more unpleasant opponent (I guess 3% against him). Until now I didn´t try to play against Houdini 3 more games, but I assume that number of points will be close to zero

[Henk]

Well, I don´t have statistics because I don´t store lost games (it is too depressive). Anyway, against weaker engines like Fritz, Shredder, Cuckoo chess or Cheng I am able to achieve about 25% of points (playing white). In case of Stockfish it is about 5% points (not wins, distribution is something like 1 win and 10 draw), Critter is more unpleasant opponent (I guess 3% against him). Until now I didn´t try to play against Houdini 3 more games, but I assume that number of points will be close to zero

Sorry I forget, the hardware is also important. On what machine did you play against these chess program. How many cores ? etc.
If you played on a relative slow machine or on a fast machine that makes a difference.

[overlord]

Normally I use mobile phone Samsung Galaxy Note II (2 cores, ponder on, 64 MB hash) when I am playing against Stockfish, Critter or Komodo. If I play against older engines (Fritz, Shredder, Hiarcs...) I use PC (Q6600, but all versions are 1 core, hash 128 MB). I hope that Houdini for Android will be released soon