33

I have heard many people saying, “Hah! I beat Stockfish,” and one saying, “I am the best chess player ever! I beat Stockfish.” So I wonder if it is possible, just to know whether I should try to beat it. I tried to play it once; I barely played 25 moves.

3
  • 14
    +1 I have heard people claim that too. Computers have kicked my butt since the early 1990s when I used to have MChess Pro running on a 486/50. It would kick my IM friend's butt too. – PhishMaster Feb 19 '20 at 19:07
  • 5
    P.S. I just looked up MChess Pro, and there was a game where it beat GM Zsuzsa Polgar, os no wonder it kicked my butt. en.wikipedia.org/wiki/MChess_Pro – PhishMaster Feb 19 '20 at 19:27
  • 1
    Your question is almost like asking whether the Carlsen of Carlsen is beatable. Sure it's technically possible, but your odds of doing it are incredibly low. – Inertial Ignorance Feb 20 '20 at 21:33
61

The answer is that they either played it on some very handicapped mode, they took back A LOT of moves, or they are probably lying.

Stockfish 11 is currently rated about 3607, and that is in standard chess. Computer programs are typically even HARDER to beat for humans at even faster controls.

There is a reason elite humans have not played matches against computers for a long time: even the Carlsens of the world cannot beat them any more. According to the ELO winning expectancy calculator, there is currently a 745 point difference between Stockfish 11 and Carlsen, and per the chart, that gives the computer a 99.5439696% chance of winning.

Note: The ratings used above are from different rating pools, and thus, are not truly directly comparable, but used as a rough estimate of what would happen between the computer and Carlsen.

16
  • 34
    The 3607 rating of Stockfish 11 are not directly comparable to Carlsen's rating because they are from different pools (computer engines vs. human chess players). Of course the tendency is correct and even Carlsen would have a very low chance of winning a game, but giving the probability to 9 decimal places is misleading (because there is a larger error margin). – user1583209 Feb 19 '20 at 19:10
  • 5
    @user1583209 I did realize that, but I did not really want to get into a discussion about rating pools, and frankly, was not expecting anyone to bring it up, especially so quickly. Nevertheless, it is still pretty indicative of what you really could expect, even against Carlsen...a bloodbath. – PhishMaster Feb 19 '20 at 19:14
  • 38
    I am unable to find my favorite Carlsen quote about playing the computer. Something like, "They play like they don't understand anything, and then they win" – Michael West Feb 19 '20 at 22:30
  • 37
    @MichaelWest "It’s like playing someone who is extremely stupid but who beats you anyway" newyorker.com/magazine/2011/03/21/the-princes-gambit – Martheen Feb 20 '20 at 6:39
  • 11
    Yeah, I'd suggest editing "0.995439696/1.00" to simply read 99.54%. Removes the decimal places that definitely aren't accurate, and makes it easier for a human to read. – Kevin Feb 20 '20 at 14:13
41

It's absolutely beatable, but not by an unassisted human. Anyone who claims to have done so is either lying or stacked the deck super heavily in their favor (e.g., by having Stockfish search only to depth 2). "Slow computer" isn't good enough - Stockfish 11 running on 1999 hardware would still have handily beaten Kasparov. You will need help from another engine to win.

Here's a recent game Stockfish lost to Leela Chess Zero.

[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[Event "Trillion-Node Throwdown II Tiebreaker (30|5)"]
[Site "?"]
[Date "2020.02.18"]
[Round "1"]
[White "Lc0"]
[Black "Stockfish"]
[Result "1-0"]
[ECO "A41"]
[GameDuration "01:13:46"]
[GameEndTime "2020-02-18T11:44:09.048 PST"]
[GameStartTime "2020-02-18T10:30:22.171 PST"]
[Opening "Queen's Pawn"]
[PlyCount "199"]
[TimeControl "1800+5"]

1. d4 d6 2. Nc3 Nf6 3. e4 e5 4. Nf3 Nbd7 5. Bc4 Be7 6. O-O O-O 7. a4 c6 8. Re1 Re8 9. Ng5 Rf8 10. Nf3 h6 11. h3 Re8 12. a5 Qc7 13. Be3 exd4 14. Nxd4 Bf8 15. Bf4 Rb8 16. Bg3 b5 17. axb6 axb6 18. b4 b5 19. Bb3 Bb7 20. Re3 Ne5 21. Bh4 Nfd7 22. f4 Nc4 23. Rg3 g6 24. f5 Kh7 25. e5 Ndxe5 26. Ne4 Be7 27. Nf6+ Bxf6 28. Bxf6 Qd7 29. Ne6 Rxe6 30. fxe6 Qxe6 31. Qf1 Re8 32. Bxe5 dxe5 33. Rc3 e4 34. Bxc4 bxc4 35. Qxc4 Qd7 36. Qc5 h5 37. Re3 Qd2 38. Qc3 Qd8 39. Ree1 Qc7 40. Qd4 f5 41. Rad1 Kh6 42. Qd7 Re7 43. Qxc7 Rxc7 44. h4 Ba6 45. Rd8 Bb5 46. Kf2 Rf7 47. Rd4 Ba4 48. Rc4 Bb5 49. Rc5 Ba6 50. g3 Bb7 51. Ke3 Rd7 52. Ra1 Kg7 53. Ra7 Kf7 54. Rc4 Bc8 55. Rxd7+ Bxd7 56. Rd4 Ke7 57. Rd1 Be6 58. Ra1 Kd7 59. Ra5 Kd6 60. c3 Bb3 61. Ra8 Be6 62. Ra7 Kd5 63. Rg7 f4+ 64. gxf4 Bf5 65. Rc7 Kd6 66. Ra7 Kd5 67. Ra6 Bd7 68. b5 cxb5 69. Rxg6 Bc8 70. Rg5+ Ke6 71. Rxh5 Kf6 72. Rxb5 Bh3 73. f5 Bg4 74. Kxe4 Bh5 75. Ke3 Be8 76. Rb4 Kxf5 77. h5 Kg5 78. h6 Kxh6 79. c4 Kg5 80. c5 Kf5 81. Rd4 Bb5 82. Rd5+ Ke6 83. Kd4 Bf1 84. c6 Ba6 85. c7 Bc8 86. Rd8 Bd7 87. c8=Q Bxc8 88. Rxc8 Kd7 89. Ra8 Ke6 90. Ra5 Kd6 91. Re5 Kc6 92. Rd5 Kb6 93. Rc5 Kb7 94. Kc4 Ka6 95. Rc6+ Ka7 96. Kb5 Kb7 97. Rc5 Ka7 98. Kc6 Ka6 99. Kc7 Ka7 100. Ra5# 1-0

If these people insist they can still beat Stockfish 11 unaided, challenge them to prove it live. You can probably offer them odds of pawn and move and they'll still lose (they won't even draw). If they actually manage to win, it'll be big news. Here's an example of a claimed win from 2011, at a time when computers were already unbeatable by humans. As ChessBase put it, "Believe it or not? Not, of course. Definitely not." Pointedly, the guy making the claim was later convicted of fraud.

9
  • The link is for a draw? – D. Ben Knoble Feb 20 '20 at 7:16
  • @D.BenKnoble weird. Looks like CCC's website is being wonky right now. I'll remove the link, the information needed to dig up the game from CCC's website is still in the answer. – Allure Feb 20 '20 at 7:54
  • 3
    @mbrig Komodo played a bunch of handicap games against GMs and did quite well. en.chessbase.com/post/komodo-9-odds-matches-against-gms – James Hollis Feb 20 '20 at 19:14
  • 1
    @mbrig based on what I saw following computer chess (no source however), Stockfish will crush a GM at pawn odds, is about even at knight odds, and is disfavored at rook odds. – Allure Feb 20 '20 at 21:32
  • 1
    @mbrig Turns out I might be wrong. GM David Smerdon gave Komodo knight odds and won 5-1. davidsmerdon.com/?p=2133 Komodo is significantly weaker than Stockfish, but still: it seems like humans can steer the game towards positions which engines cannot compensate for the knight odds. – Allure May 22 '20 at 3:27
8

People who say they beat Stockfish did it at a low level, with a fast time control so Stockfish could not do its best, and perhaps also on slower computers.

Can Stockfish be beat at high level on fast computer with long time control? Of course, but it would likely take another computer program to do it.

Remember that Stockfish was rated at 3388 as of July 2017 and would be even stronger now.

4

I would say it is practically unbeatable by a human player when running according to the TCEC hardware, but theoretically, it could be since chess has so many possibilities that we only completely understand it when only 7 pieces are on the board(Endgame Tablebases). Another fact is that these tablebases were created by supercomputers. I am certainly sure that it could be beaten by non-human players especially Neural Networks.

2

Gary Kasparov said this when Stockfish was only rated around 3100:

“There has been a steady, but slow, drive upwards. But the strongest chess [software] engines today -Stockfish, Komodo– are in the 3100-plus category, so much stronger that competition even with Magnus Carlsen would make no sense. Not because they understand chess, but because they make almost no mistakes,”

0

I played against Sockfish 5 2000. I had won the game, I was chasing the king down with three connected pawns, and my king. Driving him back until I forced Stockfish into a stalemate. But until the last move made, I was 1 second ahead of the computers time. I hadn't played the game seriously since the 70's. Seen Lichess, Gave it a go. I actually didn't know at the time Stockfish was a computer, I thought Man this guy is really fast, and good too! Can't let him think he can run over me. Ha, Ha, how funny! Game wasn't a fluke, won another game at 2000 rating, and beat computers time by 3 seconds.

enter image description here

1
  • 3
    AI level 5 is, however, a very weakened version of Stockfish, and not the 3300 version we are talking about here. Do you beat Stockfish at level 7? – null Sep 5 '20 at 0:59
-1

Well i usually play with stockfish on my mobile to sharpen my chess skills. I am now good at chess playing. My opening (as far as I think) is one of the best in the world. I never played officaly against any grandmaster but informally i have played against a state level champion (Bihar, India) i beat him easily in both the games (on a train journey). I tried several time to do something better against stockfish (10 and 11). In some of my best games against stockfish, I managed to pull the game upto 60 moves (but never won or made a draw). In my analysis, I was equally balanced upto 30 moves and then I made blunder (I was not able to guess the opponents move). So i can say it's not possible to beat stockfish 9 and above by any human now. I regularly watches international chess games specially Vishwanathan Anand's game, and many times i realized he made bad moves (some times blunders too). After his match when I analyzed, I found my guess better than him. Yet i cannot claim myself a master or grandmaster. My best chess rank on lichess was 2400+, after that I stopped playing on that. I usually play online anonymously. I am not serious anytime, so I make blunders. My normal chess rating lies between 1500-2500.

0
-1

Upd. Considering this question I did exercise of mating maxed out Chess.Com engine. During analysis it turned out that most like they are using some version of Stockfish or Komodo. It required a one restart of whole process, tries of different positions, sacrifices and total destruction on strategical level. It only understood that it has problems only when the position was already totally losing. It required nearly 60 moves, my guess it is not possible to do it in less amount of moves. I reduced required number of moves because I used moves from other engines, which were not critical for the win. Professional player can do the same without using chess engine and with less trials. The hardest part is to find key ideas against key weaknesses, essentially it is task for human not computer. Also you must find the good opening which would work. It is important that I do not know what engine I beat because essentially the required steps to make your own win would be the same nonetheless.

mating of maxed engine

Just note that game had 3 turnouts. I outtraded engine for a whole piece. Then I sack that bishop I had an advantage to get a grab on Queen which engine was thinking is OK because he was lured into a trap with overestimating his passed pawn. Engine really overestimated his ability to make checks and material advantage. Engines are so greedy that if you play by luring their risky attacks into clear trap they have zero brain to avoid it. And after you win once - you win forever. May be tweak some moves but you would repeat the whole idea.

[ECO "A01"] 1.b3 d5 2.g3 e5 3.Bb2 Nc6 4.Bg2 Nf6 5.e3 h5 6.h3 Bf5 7.d3 Qd7 8.a3 O-O-O 9.Ne2 Be7 ( 9...Kb8 10.b4 a6 11.Nd2 Bd6 12.Nb3 e4 13.Qd2 Ne5 14.Nc5 Qc8 ( 14...Bxc5 15.Bxe5 Bb6 16.Bxf6 ) 15.Bxe5 Bxe5 16.d4 Bd6 17.b5 axb5 18.Rb1 Nd7 19.Rxb5 Nb6 20.a4 h4 21.g4 Bd7 22.Rb1 f5 23.Qc3 ( 23.a5 Nc4 24.Qc3 b5 25.axb6 cxb6 26.gxf5 Bxf5 27.Bf1 Bxc5 28.dxc5 Qxc5 ) ) 10.b4 a6 11.Nd2 g5 12.Nb3 Qe6 13.Nc3 Kb8 14.Qd2 h4 15.g4 Nxg4 16.hxg4 Bxg4 17.f3 Bh5 18.O-O-O f5 19.Rdg1 g4 20.fxg4 ( 20.f4 h3 21.Bf1 exf4 22.exf4 d4 23.Nd1 Bf7 ) 20...fxg4 21.Qe1 Bf7 ( 21...g3 22.Nc5 Qf7 ) 22.Kb1 Qd7 23.Ne2 Be6 24.Nc5 Bxc5 25.bxc5 h3 ( 25...g3 ) 26.Bf1 d4 ( 26...Rdf8 27.Ng3 Qe7 ) 27.e4 Rhg8 28.Ng3 Rg5 ( 28...Rg6 ) 29.Be2 Rdg8 30.a4 a5 31.Ba3 Ka7 32.Bd1 Nb4 33.Bxb4 Qxa4 34.Bc3 dxc3 35.Qxc3 b5 36.Qb2 b4 ( 36...Qd4 ) 37.Nf5 Qe8 38.Rh2 Kb8 39.Qa1 ( 39.Qxe5 $4 {very bad idea} ) 39...a4 40.c4 Bxf5 41.Bxa4 Bd7 42.Ra2 h2 43.Re1 Bxa4 44.Rxa4 Qc6 45.Ra6 Rh8 ( 45...Rh5 46.Kb2 h1=Q 47.Rxc6 Qh2+ 48.Kb3 ) 46.Rxc6 h1=Q 47.Kb2 Qg2+ ( 47...Rh2+ 48.Kb3 Qg2 ) 48.Kb3 Rh2 49.Rc1 Rh3 50.Rb1 Rh2 51.Rc1 Rh3 52.Rb1 Qd2 53.Ka4 Rh2 54.Kb5 Qa2 55.Ra6 Qxa1 56.Rbxa1 Kc8 57.Rf6 Rh8 ( 57...Rg8 58.Ra8+ ) 58.c6 Kb8 59.Rf2 b3 60.Rb2 Rh1 61.Rxh1 Rg8 62.Ra1 Rd8 63.Rxb3 Rd5+ 64.cxd5 g3 65.Kc5+ Kc8 66.Ra8# 1-0

============

Stockfish and other computer engines are beatable, considering these factors:

  1. Limited opening + end game databases. This is biggest factor. Computer engines do not know theory, so developers are strenghtening their weasknesses with providing tera-byte sized databases with sophisticated indexes for ultra-fast search operations.
  2. Weaknesses inside valuation algorithm. Every computer engine cannot get hold of trillions of possible combinations, so it is searching game tree, using valuation function somewhere at deepest levels. Latest engines are using Neural Network function at the end of tree, and balancing this neural network is also in fact another type of database humans can attack. If you analyze this neural function, then find out the position where it fails (it will fail somewhere), because of a limited time given to computer.
  3. Weaknesses at opening. Use highly non-standart opening, which you prepared by trial-and-error approach, hand-crafted. Every human-vs-computer match always goes prepared. Because humans cannot play with "black box", they should understand the opponent. Typically humans play with computer for months before they allow proper "match". Computer must do deterministic and recoverable moves. I.e. for every single run computer should not rely on some "random" variables. In historical Kasparov-DeepBlue match, computer chosen different move, which went to other direction, and did not repeat itself. So it was the question where humans corrected engine or database was different.

Attacking this weaknesses makes a lot of fun and creates different anti-computer strategies.

I would say, that we should define true computer engine, which would inherit these properties:

  1. No human intervention. At no time, engine variables would be arbitrarily fixed by humans.
  2. No neural networks + No peta-byte databases. It simply repeats 1 and in addition neural network basically require supervision by humans to learn, or some self-learn computing power. So engine is basically saving time it would require in real match by pre-computing databases.
  3. No random moves. Fixed behaviour for given CPU/GPU/time with deterministic formulae. Repeatable games.

True computer engines are beatable by humans, given there are competitive circumstances. Real game with stakes in money + time for human side for preparation at least few months + true computer engine which would be standing with sportsmanship. Beatable.

The difficulty associated with playing against computer engines comes from the time required for humans to come prepared to the battle. Because playing is radically different from playing against other humans. Playing against humans is all about strategy, while playing against computers is about making optimal moves.

Latest versions of Stockfish and Lc0 from beginning are not fair true computer engines, because they apply neural networks which were trained by supervision techniques - or by humans, or by other engines. So it's kind of a database which is built into engine to speed it up. So I don't think that "beating" such opponent would even count as "beating computer" because you are in fact beating other people who were training that specific neural nets.

PS. I did it. I started with observing how other engines deal with computer opponents and came up with moves which would counter the ideas which this particular engine were developing. Is there a guide to try to pull this off by yourself?

  1. Fix the opening moves. Typically you want first to play Ryu Lopez game because this is most like would be default for computers to play when they are not provided with opening book.
  2. Develop your pieces first, while not letting computer to exchange. Because less pieces -> computer algorithms are more effective. You want computer to suffer at grinding. Every time you make mistake of letting computer take exchange, roll moves back and analyze what's wrong.
  3. Develop your pawn chains and ideas. Key is to keep your idea working by trial-and-error. Idea moves + filling moves. When you just need to fill the space, make "filler moves" which are not making you risk but just increase complexity for computer.
  4. Catch computer when it screws its position. It will sooner or later. Computers are most likely screwing at passing pawns. Because when they notice pawn very close to passing it is too late already. Close your material advantage exchanging pawn for rook or something valuable -> win. Don't go for fancy endings, just do material things. It will take more moves, but it won't break your intuition.
  5. Once you found winning strategy, you have beaten this computer forever, because every time it runs, you can just repeat your moves and abuse same spots. It cannot do anything against. Because computer moves are computed by formulae, no matter how complicated those formulae are, they could be outplayed.

PPS. It's a game and computers are dumb at games! Just imagine you have bots against you in CSGO. Sure they can shoot, and their shooting could be improved up to milliseconds so no human player would be stronger at shooting. But there is more in CSGO then just pointing your mouse at specified directions.

4
  • -1 because screenshot indicates it's computer level 10, which is undoubtedly a handicapped version. – Allure Sep 5 '20 at 8:21
  • @Allure it corresponds to Komodo 20 lvl and is nearly the same as analytical engines on major platforms like Lichess with default 23 move depth. I'm pretty sure they use Komodo now because they acquired it and Komodo 20 depth is default setting for strongest – sanaris Sep 5 '20 at 19:02
  • @Allure also just read how entire game went, computer didn't think it made a missplay, up to the point where position was a mate already. If you think you put some better engine, I'm telling you it will tell the same. You know its just piece of math and for every math there is a value which goes beyond the scope. – sanaris Sep 5 '20 at 19:20
  • 1
    Level 10 on Chess.com engine is around 1400 rating. Try again at level 25, which is 3200+. On mobile or iPad, you'd play against Komodo, on the web, you'd play against Stockfish 10. Looking at your moves above, you clearly played against a handicapped version, chess programs in the 90s had such level easily. – Abel Feb 9 at 15:38
-1

Gary Kasparov has beaten Deep Blue several times. Carlsen and Kasparov haven’t played against Stockfish yet, as far as I know, so we can't comment on this issue. But in my opinion, world champions will he able to beat the best of the best computers for next 20 years or so. This is because, as we can easily see in TCEC games, no chess engine is unbeatable. If every single game they play against each other ends in a draw, then we can conclude that we have reached the stage of fool-proof chess engines.

1
  • 2
    Just because one engine can beat another, doesn't mean that the World Champion can beat Stockfish, which is much stronger than Deep Blue was. – Herb Apr 9 '20 at 20:37

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