Games Analysis of Engine Positional Flaws

This is my analysis of today's chessgames.com Game of The Day between GM Anthony Miles and IM Arkhipov. This game clearly demonstrates how a GM's positional understanding of chess OTB is superior of that of Fritz, or that of the IM in this game.

The critical inaccuracies are 8...a4 and especially 12...0-0, which loses. Enjoy!

[Event "Munster"]
[Site "Munster"]
[Date "1993.??.??"]
[Round "?"]
[White "Miles Anthony J (ENG)"]
[Black "Arkhipov Sergey"]
[Result "1-0"]
[ECO "E21"]
[Annotator "Yuga"]
[PlyCount "107"]
[EventDate "1993.??.??"]

1. d4 Nf6 2. Nf3 b6 3. c4 e6 4. Nc3 Bb4 5. Qb3 a5 {I recall Ray Keene in one
of his annotated games saying that this move was a bit of a sideline.
(Whatever.)} 6. Bg5 Bb7 7. e3 Qe7 (7... Bxf3 $2 8. gxf3 d6 9. Be2 h6 10. Bh4
Nbd7 11. O-O-O Bxc3 12. Qxc3 Ne4 13. fxe4 (13. Qc2 {Fritz 1st choice} Qxh4 14.
Qxe4 Qxe4 15. fxe4 $14) 13... Qxh4 14. f4 a4 15. a3 O-O-O {The a pawn is very
weak so the initial Fritz recommendation of castling Qside is certainly not
best, and Black is lost.} 16. Kb1 Kb8 (16... e5 17. Qc2) (16... g5 17. Rdf1 Qh3
18. Bd1 f5 19. e5 gxf4 20. exf4 Qxc3 21. bxc3 c6) 17. Rdf1 e5 18. Qc2 Qh3 19.
Rf3 Qg2 20. Qd1) 8. Qc2 a4 {
Dubious. d6 is then necessary, followed by a later Bc6 is then best to hold.} (
8... Be4 9. Bd3 Bxd3 10. Qxd3 Bxc3+ 11. bxc3 d6 12. e4 h6 13. Bh4) 9. a3 Bxc3+
10. bxc3 Be4 {Perhaps castling is best} (10... Ra5 {Crikey! Fritz 1st choice}
11. Be2 O-O (11... d6 {Fritz 1st choice} 12. Rb1 Nbd7 13. Rb5 Rxb5 14. cxb5 h6
15. Bxf6 Qxf6 16. Qxa4 O-O $16) 12. Bh4 (12. O-O {Less strong.} Bxf3 13. Bxf6
Qxf6 14. Bxf3 Rg5 $11) 12... d6 13. O-O Rh5 {Fritz 1st choice} 14. Qxa4 Nbd7 {
Fritz says only +.39 but where is Black's compensation for the pawn?}) 11. Qb2
(11. Qxe4 {Fritz 1st choice. Indeed if Miles had played it, there would have
been high draw chances for Black.} Nxe4 12. Bxe7 Kxe7 13. Bd3 f5 14. Nd2 Nxd2 (
14... Nxc3 $2 15. Rc1 Ne4) 15. Kxd2 Nc6 16. Rab1 Kf6) 11... h6 (11... d6 12.
Nd2) 12. Bh4 O-O {Very dubious! My later analysis indicates that this is the
losing move! d6 was vital.} (12... d6 13. Be2 (13. Nd2 Bc6)) 13. Nd2 Bh7 {
Initially, I thought this is a critical inaccuracy but I doubt that Bc6 holds.
Perhaps black hopes for a c6, d5 or c4 break!? But the b-pawn...and the a
pawn already looks very weak.} (13... Bc6 14. e4) 14. f3 e5 15. e4 Nc6 16. Be2
{No obvious mistakes by black, in fact Fritz at least mostly agrees with all
his moves, but white clearly has superior development, and proceeds to win.} d6
17. Nf1 {This type of move truly differentiates a GM from lesser players.} g5 {
Fritz 1st choice. Perhaps not best. But Black may be already losing, somehow.
} 18. Bf2 Nh5 19. Ne3 Nf4 {f5 is Black's best chance.} (19... f5 20. g4 {
Eventually Fritz settles on exf5 as best. But Fritz had this move had g4 as
best for some time; it overextends white's position and Black may have Bxe4
tactics in some lines.} (20. c5 {Fritz 1st choice (a very poor one) after 1
min; Fritz's recommended followup is amusing.} bxc5 21. dxc5 fxe4 22. Qb7 Na5
$17) (20. exf5 {Best.} exd4 (20... Nf4 21. Nd5 Qf7 22. Nxf4 exf4 23. g4 Na5 $18
) 21. Nd5 (21. cxd4 {Not as strong.} Nf4 22. Rd1 $16) 21... Qf7 22. g4 (22.
cxd4 {Fritz 1st choice, again probably not quite as strong.} Bxf5 23. O-O Nf4
24. Nxf4 gxf4 25. Qd2 Qg7 26. Kh1 Qg5) 22... dxc3 (22... Nf4 23. Nxf4 dxc3 24.
Qxc3 gxf4 25. O-O-O) 23. Qxc3 Nf4 24. Nxf4 gxf4 25. O-O-O Qg7) 20... fxg4 21.
Nxg4 exd4 22. cxd4 Nf4) 20. Nd5 Qd8 21. g4 Bg6 (21... Ng2+ 22. Kf1 Nh4 23. Rd1
Ng6 (23... Kg7 24. c5) 24. c5 bxc5 25. dxc5) 22. O-O-O f6 23. h4 Kg7 24. Rh2
Rh8 25. Rdh1 Na5 26. hxg5 fxg5 27. Kb1 Qd7 28. Rd1 Qf7 29. dxe5 dxe5 30. Nxf4
gxf4 31. Qb5 Qe6 32. Rd7+ Bf7 33. Rh5 c5 34. Rxe5 Qc6 (34... Qxe5 35. Bd4) 35.
Ree7 Rhf8 36. e5 Kg8 37. Rd6 Qxb5+ 38. cxb5 Rfe8 39. Bh4 Rxe7 40. Bxe7 Bc4 41.
Bd1 Bb3 42. Bxb3+ axb3 43. Rxb6 Nc4 44. Ra6 Re8 45. Bd6 Kf7 46. b6 h5 47. b7
hxg4 48. b8=Q Rxb8 49. Bxb8 g3 50. e6+ Kg7 51. Bxf4 g2 52. Bh2 Nd2+ 53. Kb2
Nxf3 54. e7 1-0

A handpicked example to illustrate what everybody knows? That computers are designed to be monsters at tactics but are weak positionally? The only thing missing from Fritz's Re3 against Kramnik was Fritz saying "Computer Thinking".

3 guesses as to who was playing in the following game. One was a silicon opponent and the other a famous pupil of Botvinnik.

Deep Blue - Kasparov


And the game I chose wasn't a handpicked example for the purpose of posting it here; it is normal for engine to make these positional mistakes, but it generally takes somebody of GM level to recognize them (or somebody who knows how to use a powerful engine for post-game analysis; i.e. somebody who knows a little bit about line pruning combined with some chess intuition).

Of course I am nowhere near the level of these two players, but even an average player can come to a very good understanding of what is going on in each position and even suggest improvements.

Originally posted by Yuga
And it wasn't a handpicked example. I saw that it was a game of a positional nature and I knew that engines are positionally weak immediately out of the opening.

I'm sure that one easily could find mistakes in engine evaluations of games of a positional nature. Especially when evaluations of various moves are quite similar.

Here is another example (one that I have posted before, but with some analysis). Fritz underestimates white's tactical resources, and black has no compensation for the pawn. Perhaps the best game of 2006.

[Event "Linares 2006"]
[Site "Morelia/Linares MEX/ESP"]
[Date "2006.02.25"]
[Round "6"]
[White "V Ivanchuk"]
[Black "P Svidler"]
[Result "1-0"]
[ECO "D80"]
[WhiteElo "2729"]
[BlackElo "2765"]
[Annotator "Yuga"]
[PlyCount "49"]
[EventDate "2006.02.18"]

1. d4 Nf6 2. c4 g6 3. Nc3 d5 4. Bg5 Ne4 5. Bh4 Nxc3 6. bxc3 dxc4 7. e3 Be6 8.
Qb1 c5 9. Qxb7 Bd5 (9... Qd5 10. Qxd5 Bxd5 11. f3 f5 12. dxc5 Nc6 13. Ne2 e5
14. Rd1 Bf7) 10. Qb5+ Nd7 11. Nf3 Rb8 (11... a6 12. Qa4) 12. Qa4 cxd4 13. cxd4
Qc8 (13... a5 14. a3 c3 15. Bb5 Bg7 16. O-O) 14. Rc1 e6 15. Bxc4 Rb4 16. Qa6
Bb7 17. Qa5 f6 (17... Bd5 18. Ne5) 18. Nd2 Bxg2 19. Rg1 Qc6 20. Rxg2 Qxg2 21.
Bxe6 Bd6 22. Rc8+ Ke7 23. Rxh8 Kxe6 24. Qd8 Qg1+ 25. Ke2 1-0

I posted the comments of a chessbase support employee on feedback he received regarding the following position. http://www.chessbase.com/support/support.asp?pid=20

1.e4 c5 2.d4 cxd4 3.c3 d5 4.exd5 Qxd5 5.Nf3 Nc6 6.cxd4 Bg4 7.Nc3 Bxf3



The problem, according to the caller, is that Fritz insists on playing 8.Nxd5 (which is an instant loser, according to the caller's friend who is able to beat Fritz game after game in this line). The idea for White is to then play 9.Nc7+ and then 10.Nxa8. The problem is that White's Knight gets trapped on the back rank and will eventually be copped off by the other Black Rook. (Remember the word "eventually" -- we'll come back to that).

Originally posted by z00t

I am sure a human could make the same mistake. But nice example.

Black to move and lose. A human could very easily make a mistake here too!

1.e4 c5 2.Nf3 g6 3.d4 Bg7 4.dxc5 Qa5+ 5.c3 Qxc5 6.Na3

Originally posted by Yuga
I am sure a human could make the same mistake. But nice example.

Black to move and lose. A human could very easily make a mistake here too!

1.e4 c5 2.Nf3 g6 3.d4 Bg7 4.dxc5 Qa5+ 5.c3 Qxc5 6.Na3


[fen]rnb1k1nr/pp1pppbp/6p1/2q5/4P3/N1P2N2/PP3PPP/R1BQKB1R b KQkq - 0 6[/fen]

I immediatley see Bxc3 which is a tactical maneuver winning at least the pawn... but then again, it is an anti-positional move because after Bxc3 bxc3 Qxc3 Bd2 Qxa3 white will have superiority over the long dark diagonal and the weak black kingside complex h6-g7-f6 and also a lead in development.

Originally posted by ChessJester
I immediatley see Bxc3 which is a tactical maneuver winning at least the pawn... but then again, it is an anti-positional move because after Bxc3 bxc3 Qxc3 Bd2 Qxa3 white will have superiority over the long dark diagonal and the weak black kingside complex h6-g7-f6 and also a lead in development.

Interesting. Perhaps white has sufficient compensation for the pawn in that line - I am not so sure; I will check later.

But instead of Bd2, what about Qd2?

Originally posted by Yuga
Interesting. Perhaps white has sufficient compensation for the pawn in that line - I am not so sure; I will check later.

But instead of Bd2, what about Qd2?

well then black can go for the rook and pin the bishop to the king... but maybe white can trap it... I can't tell without spending more time and I'm kinda busy right now.