5

In the old days it was possible to beat computers basically by playing slow, closed positions with weaknesses that can be exploited in the long term, beyond the calculation depth of the engine.

As far as algorithms are concerned, what has been done to improve on this weakness of computers? Or is it just the increased calculating power/depth that makes beating computers so much harder nowadays?

  • 2
    I remember once I created simple position for human, plenty of checks and captures along the way. I could calculate quickly it's mate in 22. I gave it to Fritz, after 15 minutes he evaluated it as perpetual check. I gave it to Rybka, it started with evaluation #22 and changed it in second to #17 improving on my line... – hoacin Nov 8 '17 at 11:38
  • 2
    Evaluation functions have also become enormously better since the 1990s. I don't know the exact technical reason for that, though. – Dag Oskar Madsen Nov 8 '17 at 22:16
2

Advances in the tactical acuity of the computer programs coupled with hardware improvements to CPU, memory, and disk storage, plus the availability of robust opening book and endgame tables has made them all but unbeatable.

I don't know if computers have strategic "thinking", but their tactical prowess is such that it doesn't really matter.

2

The development of Monte-Carlo tree-search has been a key element to overcome "horizon-effect" issues for chess softwares in the mid-2000.

0
[Event "Blitz 2m+2s"]
[Site "Microsoft"]
[Date "2018.02.06"]
[Round "?"]
[White "Lyudmil Tsvetkov, owner"]
[Black "Stockfish 9 64 POPCNT"]
[Result "1-0"]
[Annotator "owner"]
[SetUp "1"]
[FEN "rnbqkbnr/pp1p1ppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[PlyCount "77"]
[EventDate "2018.??.??"]
[TimeControl "120+2"]

{512MB, OWNER-PC} 1. e4 {0} Nf6 {2.40/22 12} 2. Nc3 {3} Nc6 {1.75/21 7} 3. Nf3
{2} d5 {1.97/20 3} 4. exd5 {2} Nxd5 {2.15/22 5} 5. Nxd5 {1} Qxd5 {2.09/22 3} 6.
d4 {2} Bf5 {1.92/23 3} 7. c3 {2} Be7 {2.01/20 4} 8. Bd3 {2} O-O {2.02/21 3} 9.
Bxf5 {3} Qxf5 {2.17/19 2} 10. O-O {2} Rfe8 {2.09/22 3} 11. Be3 {2} Bf6 {2.23/
23 22} 12. Qb1 {2} Qd5 {2.07/22 0} 13. Qd3 {3} Rad8 {2.19/20 4} 14. Rfe1 {6} g6
{2.32/21 7} 15. Bd2 {5} Ne5 {2.30/19 3} 16. Nxe5 {3} Bxe5 {2.25/22 2} 17. b3 {
16} b5 {2.18/21 6} 18. Qf1 {0} h5 {2.42/22 19} 19. Rad1 {0} Bf6 {2.70/24 11}
20. Rxe8+ {2} Rxe8 {2.89/23 2} 21. Be3 {0} Rc8 {2.86/24 5} 22. Qd3 {6} Qf5 {3.
01/25 0} 23. Qxf5 {0} gxf5 {3.17/23 3} 24. Rc1 {5} Kf8 {3.24/25 9} 25. c4 {14}
bxc4 {2.46/22 1} 26. Rxc4 {4} Rxc4 {2.79/25 0} 27. bxc4 {1} f4 {3.07/24 4} 28.
Bxf4 {2} Bxd4 {3.31/24 4} 29. Be3 {2} Bb2 {4.06/29 9} 30. Bxa7 {3} Ke7 {5.02/
30 12} 31. Kf1 {9} Kd7 {4.44/28 4} 32. Ke2 {1} Be5 {5.18/28 7} 33. h3 {3} Bd6 {
4.98/23 2} 34. g4 {7} hxg4 {6.03/26 6} 35. hxg4 {1} Ke6 {6.35/26 2} 36. Kd3 {2}
f6 {7.18/25 3} 37. Be3 {8} f5 {7.21/21 0} 38. gxf5+ {4} Kxf5 {10.49/27 0} 39.
c5 {3} 1-0

[Event "Blitz 2m+2s"]
[Site "Microsoft"]
[Date "2018.02.05"]
[Round "?"]
[White "Stockfish 9 64 POPCNT"]
[Black "Lyudmil Tsvetkov, owner"]
[Result "1/2-1/2"]
[ECO "D94"]
[Annotator "owner"]
[PlyCount "111"]
[EventDate "2018.??.??"]
[TimeControl "120+2"]

{512MB, OWNER-PC} 1. d4 {0.55/22 17} d5 {2} 2. c4 {0.51/22 8} c6 {3} 3. Nf3 {
0.54/23 8} Nf6 {2} 4. e3 {0.31/24 16} g6 {2} 5. Nc3 {0.59/21 3} Bg7 {2} 6. Be2
{0.45/23 6} O-O {2} 7. O-O {0.35/21 3} Bg4 {3} 8. a4 {0.53/21 5} a5 {0} 9. b3 {
0.47/22 7} e6 {11} 10. Ba3 {0.44/28 0} Re8 {2} 11. h3 {0.39/25 4} Bxf3 {3} 12.
Bxf3 {0.33/24 1} Bf8 {5} 13. Bb2 {0.42/22 2} Na6 {0} 14. Qc2 {0.34/21 3} Nb4 {3
} 15. Qb1 {0.35/22 13} Nd7 {23} 16. Na2 {0.24/28 0} f5 {11} 17. Nxb4 {0.31/24
11} Bxb4 {1} 18. g3 {0.66/19 2} Nf6 {3} 19. Kg2 {0.38/25 23} Ne4 {14} 20. Qd3 {
0.17/26 0} Qg5 {10} 21. Rh1 {0.48/22 10} h5 {2} 22. h4 {0.67/17 1} Qh6 {10} 23.
Rhc1 {0.57/22 3} Rf8 {0} 24. Qc2 {0.57/26 5} Rf7 {3} 25. Rh1 {0.49/25 13} Raf8
{2} 26. Be2 {0.40/20 3} Kh7 {0} 27. Bd3 {0.39/22 2} Qg7 {9} 28. Qd1 {0.39/25 3}
Qh6 {2} 29. Bc1 {0.39/26 6} Qg7 {2} 30. Qc2 {0.39/25 0} Rd7 {4} 31. Bb2 {0.39/
27 2} Rfd8 {4} 32. Rhc1 {0.39/23 3} Nf6 {4} 33. Be2 {0.39/20 4} Rg8 {9} 34. Bf3
{0.39/20 2} Qh6 {5} 35. Qe2 {0.39/20 2} Rdg7 {8} 36. Rf1 {0.39/22 2} Rd7 {9}
37. Rac1 {0.39/23 3} Kh8 {4} 38. Rh1 {0.39/23 2} Kh7 {4} 39. Bc3 {0.55/19 1}
Bxc3 {0} 40. Rxc3 {0.41/24 1} Ne4 {5} 41. Rd3 {0.41/21 2} Ra8 {6} 42. Rb1 {0.
41/22 1} Ra6 {3} 43. Qb2 {0.40/23 3} Rb6 {3} 44. Qc2 {0.36/21 3} Rb4 {5} 45.
Rf1 {0.12/23 2} Kg8 {9} 46. Rdd1 {0.12/21 1} Rf7 {0} 47. Be2 {0.12/25 2} Qf8 {
17} 48. Bd3 {0.11/35 0} Qd6 {3} 49. Ra1 {0.11/25 2} Rg7 {7} 50. Rab1 {0.11/24 2
} Kh8 {0} 51. Rfe1 {0.11/26 3} Qb8 {12} 52. Qb2 {0.11/21 2} Rf7 {2} 53. Rg1 {
0.11/27 2} Kh7 {3} 54. Qc2 {0.11/27 2} Qd6 {2} 55. Rgc1 {0.11/27 2} Rf8 {5} 56.
Qd1 {0.11/27 4} 1/2-1/2

Above the pgns for my first handicap win and my first draw with SF 9. Anticomputer chess still works, but you need more time to beat the very top without handicap.

Maybe someone is interested in such games. I tried to find the pgn tags to insert diagram, but don't see them, sorry, maybe someone could post the replayable game.

  • 4
    This would be a much better answer if you explained your strategy, instead of dumping games on us. (Not my downvotes, BTW, but I understand why people did that). – Glorfindel Feb 7 '18 at 12:20
  • I have explained my strategy in detail in the book. 1)Close the position, 2)Avoid immediate tactics, 3)Use concepts engines don't understand, like compactness of the pawn structure, unsheltered king with closed center, a very wide range of other sophisticated concepts, carefully described in what some people call my magnum opus, 'The Secret of Chess'. I have done my painstaking studies 16/7 in 5 years and the results are diligently described in my books, it is really difficult for me to do more. – Lyudmil Tsvetkov Apr 17 '18 at 8:02
0

Engines like Stockfish (Alphazero may be a different matter) are still not good at finding the best positional moves.

After doing an extensive study of the Fischer game collection, with the help of Stockfish, I just published a book on the theme:

While going through the positional test suite, including 112 test positions, I had to ascertain that Stockfish still fails to solve around one third, but maybe even close to half of the puzzles. With the tactical set, Stockfish has no problems at all, all solved.

Make the conclusions yourselves how weak actually Stockfish is and how strong Fischer.

So, you might just want to forget anything about alleged engine superiority in chess.

Edit: Here comes the FEN:

 [title "Exercice position"]
 [fen "r4rk1/2q1bpp1/pn3n1p/1pp1pP2/6P1/1BP4P/PP1N1P2/R1BQR1K1 w - - 0 17"]
  • Here one position from the positional test suite SF fails to see: r4rk1/2q1bpp1/pn3n1p/1pp1pP2/6P1/1BP4P/PP1N1P2/R1BQR1K1 w - - 0 17 (anyone knowing how to post fen diagrams here?) Why would SF not see 17. h4! here? – Lyudmil Tsvetkov Dec 30 '17 at 7:00
  • 4
    I have added the position. Now it is up to you to explain why 17.h4 is the best move, what SF thinks about it, and what conclusions you draw from that about the software's supposed shortcomings. – Evargalo Feb 7 '18 at 12:46
  • Hi Evargalo. Just saw that. Please check Talkchess forum on the matter. Analysis at LTC and 64 threads shows h4 is indeed the best move here. SF fails to see that at any reasonable TC smaller than couple of minutes and not a huge hardware. I have not checked the very latest version though. – Lyudmil Tsvetkov Apr 17 '18 at 7:58
  • 2
    SE's answers are supposed to be self-sufficient. If there's valuable information on a forum, you should sum up the most relevant parts in the body of your answer. And provide a link, of course. – Evargalo Apr 17 '18 at 8:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.