WEAKNESS OF COMPUTER CHESS

The First Story: A Night in Riga

Prologue

It was late evening. The capital of Latvia was burried in the darkfall. The two of us: me Marijus Kulvietis, the president of Lithuanian chess club "Baltijos Lyga" and my good friend International master Oleg Krivonosov were sitting in an ordinary hotel room. He was going to start in Latvian men`s championship next day. Oleg felt tiredness after the journey from his native town Daugavpils and slight anxiety telling upon him my naturally stoic friend, either. It was because of experienced mistakes from the part of some officials and elementary bureaucracy that led to the situation when IM Oleg Krivonosov quite undeservedly was not enrolled in the last FIDE rating list.

It was not clear whether he would be allowed to play in Latvian championship or not up to the very last moment. That was the evening when a green light opened Oleg the way to the mentioned tournament. The first battle was ahead. I knew that the chessplayer had to rest well however I dared to disturb him.

Our club had sponsored the computer programs tournament "Odyssey 2001". Organizator Thorsten Czub from Germany and supporter Ed Schroder/boss of the Rebel company/ from Holland were waiting for the commentaries about the game of the programs. We hadn`t got much time for discussion that evening, but before leaving Oleg for a week`s tournament I was eager to receive at least some information.

Oleg Krivonosov distinquished himself in perfect technique, firm schemes in opening and endgames. No wonder, in the past he was taught by legendary people: Viktor Kortchnoj and Tigran Petrosjan. I was sure he was the man who had what to say to the participants and organizers of the "Odyssey".

"Oleg",- I said,- "we won`t have computer programs today. We can write only what we have left in our heads".

"Well",- answered my friend,- " I remember my both games against Rebel-Century from our official match. Let`s use what we have got. Your corrchess games! I have enough ideas and have only to pick out the most proper ones for illustrations".

I`m really playing correspondence chess. Everybody knows that using computers here for analysis is a permissible standard. That`s why any victory in corchess means the triumph of human understanding against schemetism of computer analysis.

"That`s true ",- I confirmed,-" I`m sure the weakness of computer chess will cause interest to the creaters of the programs, fans and people associated with this field."

Oleg started writing from time to time asking me to remember my corchess game which could illustrate his thoughts. So immediately I had got an interesting manuscript, corrected a little after returning to Lithuania and to be presented to the reader`s judgement. It`s the proffesional`s analysis for professionals and those studying chess seriously. So I had to overcome my modesty to show my games in the illustrations as I am not a great player. I expect the reader to understand the situation. The purpose is not to demonstrate a high level but to analyse typical weakness of computer chess on the basis of officially played games. I do believe many of my opponents were not blindly following the computer, but the picked illustrations are those were computer style and analysis effected my opponents greatly. Please, estimate how it has worked.

ANALYSIS

IM Oleg Krivonosov:

I had two programs to check the given results:

• "Fritz-5"
• "Rebel-Century"

All the analysis I had divided in such a way: 6 problems of computer programs:

• Knight on a1;
• Bishop on h1;
• King is bravely walking to the centrum;
• No understanding of the endgame (1);
• No understanding of the endgame (2);
• No understanding of the endgame (3).

1.. Knight on a1

Move

B92 Marks,Timothy Kulvietis,M IECG CUP 5 Event CP-1999-P-00027 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be2 e5 7. Nb3 Be7 8. Bg5 Be6 9. O-O Nbd7 10. Qd3 O-O 11. Rad1 Rc8 12. Rfe1 Qc7 13. Qg3 Rfe8 14. a3 Kh8 15. Qf3 Nb6 16. h3 h6 17. Be3 Nc4 18. Bxc4 Qxc4 19. Rd2 Kg8 20. Red1 Qc6 21. Nd5 Bxd5 22. exd5 Qa4 23. Re1 Rc4 24. Na1 and here we have it!!!! When the computer doesn't know what to do-it proposes such moves like this. Till this moment all game was going according to computer`s reccomendation, but black are step by step creating something new and black following only computer-FRITZ 5 plays the same all moves and proposes the same 24.Na1. You can check it. Seems nothing important-but if computer proposes such moves it is already clear it doesn't understand the position already and step by step going into worse and worse game. 24... Qb5 25. b3 Rc3 26. a4 Qd7 And now white is totally lost possitionally. White has no game while black has comfortable game and can create concrete plans. Less and less space for white. But the main thing that fatal 24.Na1 -that seemed so innocent first- now has simply eliminated the white knight from the game! I can`t see how to return this knight to the game. Principally it is the same as white has one knight less. White is lost. But I`m not interested in a showing all the games till the end and how it happened. I want to find out where and why the computer advices bad! And such moves like Na1 are advised by programs very often in a silent positions, when computer simply has no good plans. 0-1

2.. Bishop on h1

In the game I will show -White had nice pressing. Both players had played rather "human" party. Nevertheless the game had ended as a draw and I see some influence of computer chess experience as 38.Bh1? and Bishop trip 21.Bf3, 22.Bh5, 23.Bg4 -rather doubtful and empty bishop moves at a important moment and FRITZ-5 analysis recommends such moves really in a position I will show. This recommendations I would name just empty shots and time wasting. Here is a game:

Move

B92 Kurylo,Ryshard Kulvietis,Marijus Lithuanian e-mail chess championship 1. Nf3 d5 2. g3 Bg4 3. Bg2 c6 4. O-O Nd7 5. d4 e6 6. Nbd2 Ngf6 7. Re1 Be7 8. e4 dxe4 9. Nxe4 O-O 10. c3 Nxe4 11. Rxe4 Bf5 12. Re1 Qb6 13. a4 a5 14. Nd2 Qa6 popular field for a queen. Liked by my childhood friend GM A. Shabalov currently playing in USA 15. b3 Qd3 16. Bb2 Qc2 17. Re2 Qxd1+ 18. Rxd1 Nf6 19. Nc4 Nd5 20. Ne5 Rfd8 21. Bf3 Rac8 22. Bh5 Bg6 23. Bg4 Here it is - all 3 last white moves are recommended by FRITZ5. I doubt if a human master would do them. Maybe in some live game when time trouble is coming, just to save time but here we have a correspondence game and what those moves can bring. I still can't understand why FRITZ recommends those moves and what is the reason of such walking with a bishop? 23... Nf6 24. Nxg6 hxg6 25. Bf3 Nd5 26. c4 Nb4 27. h4 Bf6 28. Bc3 Rc7 29. Kf1 Rcd7 30. Red2 Kf8 And all main battle is going to be on a queenside-so black has to walk with this king g8-f8-e7-d6-c7. It is typical human idea. This King`s trip was loved and used by famous GM V.Bagirov. 31. Ke2 Ke7 32. Ke3 Kd6 33. Be4 Kc7 34. Bxb4 axb4 35. f4 b6 36. Bg2 Rd6 37. Rd3 R6d7 38. Bh1 Once again this empty move is recommended by FRITZ. Black can't win here. And white has some chances. Thus strange look such recommendations by program analysis when it is time to create some win plans, but not empty moves. 38... Be7 39. d5 exd5 40. cxd5 c5 41. Kd2 Bf6 42. Re3 Kd6 43. Kd3 Bc3 44. Kc4 Ra7 45. Re2 Rb8 46. Bf3 Raa8 47. Rd3 Bd4 48. Rdd2 Re8 49. Ra2 1/2-1/2

3.. King is bravely walking to the centrum..... and loosing

The game I will show now -is interesting to watch-it is short, tactical and nice win. Also after one side made a mistake another side has not listened to computer`s recommendations to take the material back that was sacrificed. But human moves made instead have won and showed human victory vs computer`s analysis.

Move

Kulvietis,M. Pljusnin, Victor Event IECG cup-5 CP-1999-P-00027 1. Nf3 d5 2. c4 dxc4 3. Na3 Nf6 4. Nxc4 e6 5. g3 b5 6. Nce5 Bb7 7. Bg2 Nbd7 8. Nxf7 Nice sac. The computer sees it-but programs make this only to return the material back after several moves. While the human idea is to make weak king and to attack it. In short computer`s idea is- a tactical sac- while human`s- a positional sac. 8... Kxf7 9. Ng5+ Kg6 10. Bxb7 Kxg5 11. Qc2 A human move - computers want to take back the material and recommend Bxa8- but then the game returns to +-. While Qc2 is a typicall human`s move and no computer recommends it. And this move prepares a killing attack on a totally weak king. 11... Kh5 This position reminds me a game from 18-19 century. I haven't seen such in a nowadays. King in a middle of a desk walking bravely after 10 moves. 12. Bf3+ Kh6 13. d4+ g5 14. h4 Bb4+ 15. Kf1 Kg7 Computer still shows only a little advantage for whites while the game is totally lost by black. Now it is principally forced win. 16. hxg5 Nd5 17. Qe4 Qe7 18. Rh6 N5b6 19. Rxe6 Qf7 20. d5 a5 21. Be3 Rhe8 22. Bd4+ Kf8 23. Qh4 Rxe6 24. dxe6 Qxe6 25. Qxh7 Rb8 26. g6 Qg8 27. Rd1 Bd6 28. g7+ Ke7 29. Qh4+ Pay attention how late the programs give more points for white. It shows they don't see clear positional win and count only material results when it is funny clear that the game was lost after 11 move! Any master would laugh at a king walking after 10 move to the centrum of the board. Really-humans victory vs computer`s material counting! 1-0

4.. No understanding of the endgame

Lets start from a position missing not the main interesting game-where white has played bad and went into troubles. The position:

Move

Kulvietis,M. Schulze,Michael Event-IECG master class tournament QM-2000-0-00024 Well, the endgame is clearly won for black. One pawn more and also very far and dangerous black pawn on d3. It maybe discussion by masters, but the programs show the victory for black. I believe them. And further game shows that black are following the recommendations of REBEL-CENTURY and FRITZ-5. Really according to them-black win seems very clear. Lets see what happened when black were playing according the programs: 27... f5 28. Rbd2 Red4 29. Kf2 g5 30. Rc1 Kf6 31. Rc6+ Ke5 32. g3 f4 33. Rxh6 Rc8 34. gxf4+ gxf4 35. Rh7 Rc2 36. Re7+ Kf6 37. Rxc2 dxc2 38. Rc7 Rd2+ 39. Kf3 Ke5 40. Rc4 Rxh2 41. Re4+ Kd5 42. Rc4 Kd6 A draw !!! And here you are the human builds a defence that wasn't recommended nor by Fritz, nor by REBEL. And result is a draw. Computer still shows some points, but endgame is a draw. 1/2-1/2

4.2 No understanding of the endgame (2)

One more exiting game where my friend had really to loose the endgame-and several times I had founded how but it is clearly seen the opponent was following computer`s recommendations and even when the programs where showing +3.00 for white, white didn't manage to realize such a dramatic endgame. As the programs also didn't manage to find the win. Let`s see the position:

Move

Wodniczak,Arkadiusz Kulvietis,M Event-IECG tournament L1-1999-0-00005 Again both programs FRITZ-5 and REBEL-Century show that white must win. Easy to believe them and whe are following their recommendations. 36. Rd4 a5 37. Rxc4 axb4 38. axb4 Rb5 39. Kf3 h5 40. e4 Kg6 41. exf5+ Rxf5+ 42. Rf4 Rd5 43. g4 h4 44. Rc4 Rd2 45. Rc6+ Kg5 46. Rc5+ Kg6 47. Re5 Rb2 48. Rb5 Kh6 49. Rb8 Kg6 50. b5 Rb4 51. b6 Kh7 52. g5 Kg7 53. Ke3 Rb2 54. Kf4 Rxg2 55. Rd8 Rb2 56. Rd7+ Kg6 57. Rd6+ Kg7 58. Kg4 Rb4+ 59. Kh3 Kh7 A draw was offered many times, but white couldn't believe they had missed a win in such a clear endgame. But it is really a draw now at the final position and no way to push the pawn. Simply black holds the defence - That wasn't seen again by any program. And REBEL and FRITZ had failed to break the human defence. More to say they have played bad and had got the real draw. Good position for chess pupils to study. 1/2-1/2

4.3 No understanding of the endgame (3)

We have studied some more endgames were computers didn't manage to win the endgame and human manage to draw a game. Now lets see the example how the computers didn't manage to find a win and a human finds a way to win the endgame.

The game is from the same event

Move

Kulvietis,M. DiTolla,Francesco Event-IECG tournament L1-1999-0-00005 Human will say the position is win for white. But the FRITZ and REBEL programs again are walking at the same place and can't find the win. For the human player this win is very simple. As here it is all up to positional understanding. We must eliminate the bishop which holds all the black defence and doesn't allow us to win. Here is the task. Computers don't understand it. The game was won in the following way: 35. h5 Rb8 36. Bc2 Rg8 37. Ba4 Here is the main way to win-trading the black bishop 37... Bxa4 38. Rxa4 f5 39. gxf5 Rg5 40. e4 dxe4 41. fxe4 Rg1 42. Rb4 Rd1+ 43. Ke3 Rd7 44. e5 Kf8 45. Ke4 Rc7 46. Rb5 1-0

EPILOGUE

Thanks to a great master Oleg Krivonosov for his comments. I guess advantaged chessplayers will understand our main idea. As to me -playing corrchess- it is clearly seen - computer programs are very great helpers.

They check our ideas, help to escape tactical mistakes and count the variations. And I do love such tournaments as Odyssey2001. As I`m interested to find VERY strong and interesting program here.

But our analysis and examples showed there are still spots in a computer`s analysis in very many fields and we can't trust them blindly! Until the perfect chessprogram is not find out the best game may be performed only in a cooperation between the chessprogram and a human player.

Marijus Kulvietis
President of the Baltijos Lyga Chess Club

2001.03.21
© All rights reserved by the Lithuanian Chessclub "Baltijos Lyga"

The first publication is kindly offered to our friends the organizers of a Odyssey 2001 computer tournament and personally to Mr. Ed Schroder.

The discussed games and positions can be downloaded here for private use and further study, the annotations remain the property of the Lithuanian Chessclub Baltijos Lyga.