Match Kasparov - Deep Blue, New York 1997: The Deciding Party and Conclusions
In the summer and autumn of 2013 in the newspapers, The Guardian, The New York Times and The New York Post, I read about the play by Matt Charman “The Machine”, which tells about the match Kasparov - Deep Blue. Now Matt Charman became famous as the author (along with the Cohen brothers) nominated for the 2016 Oscar for the original script of Steven Spielberg's film Spy Bridge. The play “The Machine” - about the dramatic epic battle of a man with a giant computer - was staged in New York at Park Avenue Armory. The performances were held from 4 to 18 September 2013.
By 2013, I had already forgotten the score of the match, and how it went: I myself did not analyze the games, I had a different opening repertoire, as a theorist I specialized in other principles. But he remembered exactly the victory of the IBM computer. Journalists trumpeted: The victory of artificial intelligence over the world chess champion! The car beats the man! Computer - superstar! A milestone in the history of AI! Looking into Wikipedia, I found that the match ended with a computer overweight by only one point 3.5-2.5, and the outcome of the match was decided in the last sixth game. This party and the Wikipedia notes seemed strange to me. Kasparov quickly lost, Vicki commentators put three questions to his moves 7 ... h6? 11 ... b5? and 16 ... CC6 ?, and some experts said that instead of 8 ... Fe7, it was necessary to immediately take a horse. Did the world champion make four mistakes in 18 moves? I decided to deal with this and find out the truth, using both modern chess programs, and my own, human understanding of positions using heuristics or general strategic principles of the game. Mostly analyzed using Houdini, sometimes, in some controversial situations, used Fritz, Rybka or Stockfish.
So, before the 6th game, the score was 2.5-2.5, and if the game ended in a draw, Man would stand against the Machine. In the article "The Truth About the Decisive Game Between the Deep Blue and Garry Kasparov, New York 1997, Game 6" 10/20/2013 I answer the questions: How many mistakes did Kasparov make? Could he equalize the game?
Deep Blue - Garry Kasparov (2785) [B17]
1. e4 c6 According to the Mega Database, before this game with Deep Blue, Garry Kasparov played Karo-Kann defense with blacks 19 times (9 wins, 2 losses) from 1977 to 1982, white pieces 28 times (15 wins, 2 losses) from 1976 to 1996. He remembered youth. 2. d4 d5 3. c3 dxe4 4. xe4 d7 5. g5 gf6 6. d3 e6 7. 1f3 h6 Risky, but quite normal move for a player who knows all the nuances. I do not like such options, because I do not like to defend, it is more pleasant to attack. Although, if you study a month, you can play. Simply 7 ... d6.
8. e6! The sacrifice of a horse is the strongest answer. Modern programs also choose this move. I suspect that Kasparov was expecting from Deep Blue to retreat the knight on e4, since in the 1990s computers rarely sacrificed figures in obscure positions. Now, in February 2016, I looked at the party logs at www.research.ibm.com/deepblue/watch/html/c.html IBM The Match. In the 6th game logs of www.research.ibm.com/deepblue/watch/html/game6.log from 1 to 11, “book moves” is written, which means that these moves are taken from the debut library. Kasparov did not suspect that the aggressive move 8. Kxe6 was simply recorded in the memory of Deep Blue.
8 ... Fe7? Mistake. A decisive mistake in a decisive game. It is correct to immediately beat the knight 8 ... fxe6, and as the analysis shows, Black has counter chances in all variations. Although White has good compensation for the piece, if the game of both sides is correct, the game should end in a draw. In one of the last grandmaster games of H. Stefansson - H. Granda Zuniga, Reykjavik 2015, black won in 18 moves.
Heuristics: In the debut, quickly develop shapes. Based on the general principles, the move 8 ... Fe7? also worse than 8 ... fxe6, after 9. g6 + 7 10. OO c7 11. c4 d8 Black can develop the kingside, leading the bishop and the rook to the open, after 8 ... Fe7? it's impossible. Why did Kasparov prefer 8 ... Fe7? Most likely, during home preparation, he did not seriously study the sacrifice of a horse, thinking that Deep Blue would go 8. Ke4.
9. OO fxe6 10. g6 + d8 11. f4!
The winning move, as shown by my research. He first applied E. Geller in 1986 at the Chigorin Memorial in Sochi, beating E. Medun. White has a clear developmental advantage, powerful elephants control the fields around the black king.
After seeing the games of this variant, I discovered that it was played by my Hungarian acquaintances, international master Zoltan Siklosi and grandmaster Peter Leko. Shikloshi defeated M. Nemeth in Lenk, Switzerland in 1990 after 11. f4. Leko chose another strong move 11. c4 and defeated G. Bakhtadze in 22 turns, Las Palmas, 1995.
11 ... b5 !? Kasparov is trying to activate the forces on the queenside and plans to transfer the knight to the center. However, this and all other moves lead to defeat. 12. a4! Deep Blue attacks! 12 ... b7 13. e1 d5 14. g3 c8 15. axb5 cxb5 16. d3 c6 The black king is in danger.
17. f5! exf5 The computer takes the queen. 18. Lxe7 Cxe7 19. c4! Again the best move: Deep Blue reveals the vertical for the queen. Kasparov surrendered.
Answers on questions. Kasparov made only one mistake 8 ... Fe7. That was enough for a supercomputer, Deep Blue played well and won the champion. But if Kasparov beat the knight 8 ... fxe6 at once, he could have equalized and made a draw in the game and the match.
')
A bit of criticism. In Geektimes there is a post geektimes.ru/post/246086 about the match. When a non-expert raises the topic, he rewrites the main thing with others, adding from his own stupidity. It is evident that the person did not play chess seriously, never prepared for tournaments, did not analyze opening options. Almost from the first lines of this post causes laughter. Reti's “extravagant” debut was 100 years ago. The “bizarre” system with double fianchetto until 1997 was used by Kasparov at least 11 times. In 1978 against G. Govashelishvili in Baku (1-0, 36 moves). In 1980 against B. Toro Sanchez in Dortmund (a draw, 22 moves). In 1987 against V. Korchnoi in Brussels (a draw, 51 move), ibid against G. Sosonko (1-0, 34). In 1987 against A. Karpov in Seville (1-0, 64 moves, the match for the title of world champion). In simul sessions against B. Hund in 1988 in Basel (1-0, 63), in Cannes against M. Ulybin (draw, 40). In 1990 against N. Short in Paris (1-0, 59). In a simultaneous game in 1991 against A. Núñez in Galicia, Spain (1-0, 30 moves). In 1993 against F. Taylor Bouda in London (1-0, 22 moves). In 1994 against Anand in New York (draw, 24 moves). Double fianchetto is outlandish only for neophytes who know one move - 1. e2-e4. As Vladimir Vysotsky sang: "Chto something familiar to me ... So, so!". The debut of the 4th game is not the defense of Pirc-Ufimtsev, the move there is obligatory ... g6, fianchetto.
Anti-computer strategy is the creation of a closed position with pawn chains, so that long-term planning can play the main role. If the program counts 15 moves ahead, then the implementation of the plan for the transfer of figures can last 20 moves. A good example of anti-computer strategy is the 3rd game of the match Kasparov - X3D Fritz, New York 2003, is on the site Chess Siberia. About the evaluation function and the differences I write in the last chapter of my science fiction novel "The Shining Path" (from the words "Man and the computer play chess in different ways").
Garry Kasparov, 13th World Chess Champion. Photo © Boris Schipkov, Novosibirsk 2005
1st party. Kasparov - Deep Blue 1-0 . Reti's debut. Kasparov played hard, it was difficult for the computer to evaluate such positions, especially with two passed pawns.
2nd party. Deep Blue - Kasparov 1-0 . Spanish party. After a weak game in the opening and a strategic mistake, Kasparov lost his counterplay. In the end, he made a mistake and Deep Blue, but Harry did not use the gift of fate.
3rd party. Kasparov - Deep Blue 1/2 . English beginning. Kasparov attempted to drive the car away from the theory of the “extraordinary” 1. d3, but it took moves from its debut base until the 8th move. The horse in the center provided Deep Blue approximate equality.
4th party. Deep Blue - Kasparov 1/2 . Wrong start, defending Caro-Kann with a loss of pace. The debut library, before the 7th move, after which Deep Blue recklessly closed the position with 9. e5 (humor: this is an anti-computer move), and Kasparov got a pleasant game. True, I would prefer 12 ... fxg6 !, immediately opening the f line. Deep Blue weakened his king with 26. b5 ?, but Kasparov changed queens, then more precisely 35 ... Luff2! .. The endgame is drawn.
5th party. Kasparov - Deep Blue 1/2 . Reti's debut. Deep Blue changed the bishop for knight 4 ... xf3 simply because this move was in the debut library. The iron monster in vain moved into the endgame, and Kasparov could have won with 44. Rg7 + with a further 45. g6. Then Deep Blue, having calculated the four-way option, confidently made a draw with a clear 47 ... Krb6. Kasparov lost composure.
6th party. Deep Blue - Kasparov 1-0 . Protection Caro-Kann. Kasparov stumbled in the opening with 8 ... Fe7, and Deep Blue quickly knocked out his opponent with heavy blows.
Deep Blue prevailed in the match against Garry Kasparov with a score of 3.5-2.5.
Deep Blue played no better than Kasparov, so there was no point in fighting the world champion in the IBM rematch, the result could be anything.
Kasparov did not know that Deep Blue has a huge debut library and base of parties. Bad debut preparation of Kasparov is noticeable: 2nd, 6th games.
Kasparov did not understand why Deep Blue played with drops, made a strong move, then a weak one. But it is explained very simply. Deep Blue for the time given to it can calculate the position, for example, 7 moves ahead. However, on the 8th move the opponent has a powerful answer that refutes the computer variant (for example, on the 6th move he takes the rook, and on the 8th loses the queen). After the first move, Deep Blue notices this powerful response and deviates to the side. But late, since the previous move was not the best, and perhaps even a mistake. In 7 or 8 moves, Deep Blue could not recognize the eternal check, so he chose 45.Ra6? in the 2nd installment.
Kasparov lost the match due to the fact that he didn’t know how the chess program works, didn’t know that the opening library in the computer’s memory could be very large, and also because of excessive emotionality and poor preparation.
By 2013, I had already forgotten the score of the match, and how it went: I myself did not analyze the games, I had a different opening repertoire, as a theorist I specialized in other principles. But he remembered exactly the victory of the IBM computer. Journalists trumpeted: The victory of artificial intelligence over the world chess champion! The car beats the man! Computer - superstar! A milestone in the history of AI! Looking into Wikipedia, I found that the match ended with a computer overweight by only one point 3.5-2.5, and the outcome of the match was decided in the last sixth game. This party and the Wikipedia notes seemed strange to me. Kasparov quickly lost, Vicki commentators put three questions to his moves 7 ... h6? 11 ... b5? and 16 ... CC6 ?, and some experts said that instead of 8 ... Fe7, it was necessary to immediately take a horse. Did the world champion make four mistakes in 18 moves? I decided to deal with this and find out the truth, using both modern chess programs, and my own, human understanding of positions using heuristics or general strategic principles of the game. Mostly analyzed using Houdini, sometimes, in some controversial situations, used Fritz, Rybka or Stockfish.
So, before the 6th game, the score was 2.5-2.5, and if the game ended in a draw, Man would stand against the Machine. In the article "The Truth About the Decisive Game Between the Deep Blue and Garry Kasparov, New York 1997, Game 6" 10/20/2013 I answer the questions: How many mistakes did Kasparov make? Could he equalize the game?
Decisive match match Kasparov - Deep Blue, New York 1997
Deep Blue - Garry Kasparov (2785) [B17]
1. e4 c6 According to the Mega Database, before this game with Deep Blue, Garry Kasparov played Karo-Kann defense with blacks 19 times (9 wins, 2 losses) from 1977 to 1982, white pieces 28 times (15 wins, 2 losses) from 1976 to 1996. He remembered youth. 2. d4 d5 3. c3 dxe4 4. xe4 d7 5. g5 gf6 6. d3 e6 7. 1f3 h6 Risky, but quite normal move for a player who knows all the nuances. I do not like such options, because I do not like to defend, it is more pleasant to attack. Although, if you study a month, you can play. Simply 7 ... d6.
8. e6! The sacrifice of a horse is the strongest answer. Modern programs also choose this move. I suspect that Kasparov was expecting from Deep Blue to retreat the knight on e4, since in the 1990s computers rarely sacrificed figures in obscure positions. Now, in February 2016, I looked at the party logs at www.research.ibm.com/deepblue/watch/html/c.html IBM The Match. In the 6th game logs of www.research.ibm.com/deepblue/watch/html/game6.log from 1 to 11, “book moves” is written, which means that these moves are taken from the debut library. Kasparov did not suspect that the aggressive move 8. Kxe6 was simply recorded in the memory of Deep Blue.
8 ... Fe7? Mistake. A decisive mistake in a decisive game. It is correct to immediately beat the knight 8 ... fxe6, and as the analysis shows, Black has counter chances in all variations. Although White has good compensation for the piece, if the game of both sides is correct, the game should end in a draw. In one of the last grandmaster games of H. Stefansson - H. Granda Zuniga, Reykjavik 2015, black won in 18 moves.
Heuristics: In the debut, quickly develop shapes. Based on the general principles, the move 8 ... Fe7? also worse than 8 ... fxe6, after 9. g6 + 7 10. OO c7 11. c4 d8 Black can develop the kingside, leading the bishop and the rook to the open, after 8 ... Fe7? it's impossible. Why did Kasparov prefer 8 ... Fe7? Most likely, during home preparation, he did not seriously study the sacrifice of a horse, thinking that Deep Blue would go 8. Ke4.
9. OO fxe6 10. g6 + d8 11. f4!
The winning move, as shown by my research. He first applied E. Geller in 1986 at the Chigorin Memorial in Sochi, beating E. Medun. White has a clear developmental advantage, powerful elephants control the fields around the black king.
After seeing the games of this variant, I discovered that it was played by my Hungarian acquaintances, international master Zoltan Siklosi and grandmaster Peter Leko. Shikloshi defeated M. Nemeth in Lenk, Switzerland in 1990 after 11. f4. Leko chose another strong move 11. c4 and defeated G. Bakhtadze in 22 turns, Las Palmas, 1995.
11 ... b5 !? Kasparov is trying to activate the forces on the queenside and plans to transfer the knight to the center. However, this and all other moves lead to defeat. 12. a4! Deep Blue attacks! 12 ... b7 13. e1 d5 14. g3 c8 15. axb5 cxb5 16. d3 c6 The black king is in danger.
17. f5! exf5 The computer takes the queen. 18. Lxe7 Cxe7 19. c4! Again the best move: Deep Blue reveals the vertical for the queen. Kasparov surrendered.
Answers on questions. Kasparov made only one mistake 8 ... Fe7. That was enough for a supercomputer, Deep Blue played well and won the champion. But if Kasparov beat the knight 8 ... fxe6 at once, he could have equalized and made a draw in the game and the match.
')
A bit of criticism. In Geektimes there is a post geektimes.ru/post/246086 about the match. When a non-expert raises the topic, he rewrites the main thing with others, adding from his own stupidity. It is evident that the person did not play chess seriously, never prepared for tournaments, did not analyze opening options. Almost from the first lines of this post causes laughter. Reti's “extravagant” debut was 100 years ago. The “bizarre” system with double fianchetto until 1997 was used by Kasparov at least 11 times. In 1978 against G. Govashelishvili in Baku (1-0, 36 moves). In 1980 against B. Toro Sanchez in Dortmund (a draw, 22 moves). In 1987 against V. Korchnoi in Brussels (a draw, 51 move), ibid against G. Sosonko (1-0, 34). In 1987 against A. Karpov in Seville (1-0, 64 moves, the match for the title of world champion). In simul sessions against B. Hund in 1988 in Basel (1-0, 63), in Cannes against M. Ulybin (draw, 40). In 1990 against N. Short in Paris (1-0, 59). In a simultaneous game in 1991 against A. Núñez in Galicia, Spain (1-0, 30 moves). In 1993 against F. Taylor Bouda in London (1-0, 22 moves). In 1994 against Anand in New York (draw, 24 moves). Double fianchetto is outlandish only for neophytes who know one move - 1. e2-e4. As Vladimir Vysotsky sang: "Chto something familiar to me ... So, so!". The debut of the 4th game is not the defense of Pirc-Ufimtsev, the move there is obligatory ... g6, fianchetto.
Anti-computer strategy is the creation of a closed position with pawn chains, so that long-term planning can play the main role. If the program counts 15 moves ahead, then the implementation of the plan for the transfer of figures can last 20 moves. A good example of anti-computer strategy is the 3rd game of the match Kasparov - X3D Fritz, New York 2003, is on the site Chess Siberia. About the evaluation function and the differences I write in the last chapter of my science fiction novel "The Shining Path" (from the words "Man and the computer play chess in different ways").
Match Party Overviews
Garry Kasparov, 13th World Chess Champion. Photo © Boris Schipkov, Novosibirsk 2005
1st party. Kasparov - Deep Blue 1-0 . Reti's debut. Kasparov played hard, it was difficult for the computer to evaluate such positions, especially with two passed pawns.
2nd party. Deep Blue - Kasparov 1-0 . Spanish party. After a weak game in the opening and a strategic mistake, Kasparov lost his counterplay. In the end, he made a mistake and Deep Blue, but Harry did not use the gift of fate.
3rd party. Kasparov - Deep Blue 1/2 . English beginning. Kasparov attempted to drive the car away from the theory of the “extraordinary” 1. d3, but it took moves from its debut base until the 8th move. The horse in the center provided Deep Blue approximate equality.
4th party. Deep Blue - Kasparov 1/2 . Wrong start, defending Caro-Kann with a loss of pace. The debut library, before the 7th move, after which Deep Blue recklessly closed the position with 9. e5 (humor: this is an anti-computer move), and Kasparov got a pleasant game. True, I would prefer 12 ... fxg6 !, immediately opening the f line. Deep Blue weakened his king with 26. b5 ?, but Kasparov changed queens, then more precisely 35 ... Luff2! .. The endgame is drawn.
5th party. Kasparov - Deep Blue 1/2 . Reti's debut. Deep Blue changed the bishop for knight 4 ... xf3 simply because this move was in the debut library. The iron monster in vain moved into the endgame, and Kasparov could have won with 44. Rg7 + with a further 45. g6. Then Deep Blue, having calculated the four-way option, confidently made a draw with a clear 47 ... Krb6. Kasparov lost composure.
6th party. Deep Blue - Kasparov 1-0 . Protection Caro-Kann. Kasparov stumbled in the opening with 8 ... Fe7, and Deep Blue quickly knocked out his opponent with heavy blows.
Deep Blue prevailed in the match against Garry Kasparov with a score of 3.5-2.5.
Let's sum up
Deep Blue played no better than Kasparov, so there was no point in fighting the world champion in the IBM rematch, the result could be anything.
Kasparov did not know that Deep Blue has a huge debut library and base of parties. Bad debut preparation of Kasparov is noticeable: 2nd, 6th games.
Kasparov did not understand why Deep Blue played with drops, made a strong move, then a weak one. But it is explained very simply. Deep Blue for the time given to it can calculate the position, for example, 7 moves ahead. However, on the 8th move the opponent has a powerful answer that refutes the computer variant (for example, on the 6th move he takes the rook, and on the 8th loses the queen). After the first move, Deep Blue notices this powerful response and deviates to the side. But late, since the previous move was not the best, and perhaps even a mistake. In 7 or 8 moves, Deep Blue could not recognize the eternal check, so he chose 45.Ra6? in the 2nd installment.
Kasparov lost the match due to the fact that he didn’t know how the chess program works, didn’t know that the opening library in the computer’s memory could be very large, and also because of excessive emotionality and poor preparation.
Source: https://habr.com/ru/post/390713/
All Articles