Analyze game and explain why bad moves break theory [closed]

Question

Is there an automated way to analyze a game and explain in human language why a particular opening move breaks theory?

Answer 1

You might be looking for something like decodechess.com/:

The first AI chess tutor, DecodeChess explains the why behind chess moves in rich, intuitive language.

This screenshot shows an example of a position of DecodeChess's natural-language explanation for why, for example, Qa1 is good:

• Escapes the black queen's threat (Qf6(b6)-b2)
• Supports the rook at a7
• Threatens to play Bxd5
• Supports the rook at d1

Although you ask for something to "explain in human language why a particular opening move breaks theory," in my limited exposure to DecodeChess it focuses on explaining why you should make a particular good move, rather than why you should not have made a bad move.

DecodeChess has a free trial available, which is pretty limited in capacity (two "decodes" per day), and monthly and annual subscriptions to get unlimited access.

Answer 2

Is there an automated way to analyze a game and explain in human language why a particular opening move broke theory?

The explanation part is very easy. Opening theory is the body of opening knowledge of what moves are considered to be good moves. When a move is made which isn't regarded by theory as good then it "breaks" opening theory and the reason it does so is because it isn't regarded as good.

The only difficult part of automating the process is building the database of all opening theory. Once you've done that comparing what is played in a game with "opening theory" is fairly straightforward. In effect this is what top level commentators are doing at events like the recent Altibox Norway event.

One additional, very obvious, problem is that as long as top level players are playing chess opening theory is changing and your database is out of date and needs to be updated.

Again, Altibox provides a very good example of this. In the first round game, Tari vs Caruana, Caruana's 13th move, Bxe3, was in the databases as a blunder which lost a pawn. It had only been played a handful of times. Caruana "broke opening theory" by playing this move.

After the game when he was asked about this he said that he had analysed the move in preparation and regarded it not as a pawn-losing blunder but as a pawn sacrifice which gave him an easy game and made life very difficult for his opponent. And so it proved. He went on to win the game. 13...Bxe3 has, for the moment, become part of opening theory in the Taimanov Sicilian.

Of course that is not the end of it. The move is bound to be analysed by other top players and if a way is found for white to hold on to the pawn with a good game then the move may again "break opening theory".

Here is the game:

[title "Tari - Caruana, Altibox 2020 Rd 1"]
[fen ""]

1. e4 c5 2. Nf3 e6 3. d4 cxd4 4. Nxd4 Nc6 5.Nc3 Qc7 6.g3 a6 7.Bg2 Nf6 8.O-O Nxd4 9.Qxd4 Bc5 10.Bf4 d6 11.Qd2 h6 12.Rad1 e5 13.Be3 Bxe3 14.fxe3 Be6 15.Qxd6 Rc8 {White is a pawn up but his pawns are doubled and the Black pieces are better than the white ones. Especially the bishop, so there is compensation} 16. a4 {this is a real novelty and the game is now original} h5 17. h3 Qxd6 18. Rxd6 Nd7 19. h4 Ke7 20. Rd2 Nb6  21. Nd5+ Bxd5 22. exd5 Nc4 23. Re2 Nd6  24. e4 a5 25. Ra1 Rc4 26. c3 f6 27. Re3 Rhc8 28. Kf2 g5 29. Bf3 g4 30. Bg2 R4c5 31. Bf1 f5 32. exf5 Rxd5 33. Rd3 Rcc5 34. Rad1 Rxd3 35. Bxd3 e4 36. Be2 Nxf5 37. Bb5 Nd6 38. Be2 Ke6 39. Rc1 Ke5 40. Rd1 Rc7 41. Kg2 b6 42. Ba6 Rf7 43. Be2 Rf8 44. Rb1 Nf5 45. b4 Ne3+ 46. Kg1 Nd5 47. Rb3 Rc8 48. bxa5 bxa5 49. c4 Nb4 50. Rb1 Kd4 51. Bf1 Nd3  52. Rb5 Rc5  53. Bxd3 exd3 54. Kf2 Kxc4 55. Rb1 Re5 0-1

Answer 3

It's possible to make an attempt to do this, although it would be inferior to the opinion of someone familiar with the opening.

A program can tell you when you've left its opening book, and what move(s) would have stayed in it. It can show you its evaluation of the standard move and its evaluation of the move played. It can show you the lines it expects in each. It would even be possible for it to tell you why it evaluates the position as it does, and this can be done in human language - something like "White's king is unsafe" or "Black's knight is on an outpost".

But the openings will sometimes result in greatly diverging lines without it being obvious why they diverge. Maybe it turns out that, in the non-theory line, the threat of a pin causes one side to have to play a move that exposes some other weakness, while in the theory line it's not a threat. But the pin doesn't actually happen in either line. It's pretty hard for an engine to list all the moves it didn't play.