Deep blue

from Wikipedia, the free encyclopedia
Deep Blue (similar model, 2007)

Deep Blue was a chess computer developed by IBM . In 1996, Deep Blue became the first computer to beat the then reigning world chess champion Garry Kasparov in a game with regular time controls . In 1997, Deep Blue won against Kasparov an entire six-game match under tournament conditions.

development

The project was invented by Feng-hsiung Hsu . He started it in 1985 with the development of a train generator integrated on a chip as a chip test at Carnegie Mellon University and named the finished system Deep Thought , after the computer of the same name in the novel The Hitchhiker's Guide to the Galaxy by Douglas Adams . In 1989 Hsu joined the IBM team and did research with Murray Campbell on parallel computing problems . Deep Blue emerged from this work. This name was derived from the American nickname for the US group IBM, which was called "Big Blue" because of its large market capitalization and its blue logo .

The system got its playing strength mainly from its enormous computing power. Deep Blue was a massively parallel, SP -based RS / 6000 computer. The 1996 version consisted of 36 nodes and 216 special VLSI chess processors, the 1997 version of 30 nodes with 480 chips. Each node had 1 GB of RAM and 4 GB of hard disk space. The chess software was written in C and ran under the AIX 4.2 operating system . Depending on the type of position, it calculated between 100 and 200 million, an average of 126 million positions per second.

Its evaluation function consisted of the extensive parameter evaluation carried out in hardware and the weighting of these parameters carried out in software (e.g. how important is the ultimate security compared to a space advantage in the center). The system itself determined the optimum values ​​of the parameters by analyzing thousands of master games. Before the second match, Grandmaster Joel Benjamin's chess knowledge was optimized. The opening book came from the Grand Masters Miguel Illescas Córdoba , John Fedorowicz and Nick de Firmian .

Competitions against Kasparov

Kasparov was able to win the first match, which took place with six games from February 10 to 17, 1996 in Philadelphia . He won three games, made two draws and lost one game, beating Deep Blue 4-2. The first game of the match that Deep Blue won became famous . The competition was for a prize fund of $ 500,000 and was broadcast live on the Internet.

Subsequently, IBM equipped its machine with more powerful hardware and in May 1997 again faced Kasparov. Deep Blue, who could now calculate 200 million positions per second, won the revenge 3.5: 2.5. This made the system the first computer that could win a competition against a reigning world chess champion under “tournament conditions”.

Deep Blue - Kasparov
  a b c d e f G H  
8th Chess --t45.svg Chess rdt45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg 8th
7th Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess kdt45.svg Chess pdt45.svg Chess --t45.svg 7th
6th Chess rlt45.svg Chess qdt45.svg Chess qlt45.svg Chess bdt45.svg Chess --t45.svg Chess pdt45.svg Chess --t45.svg Chess pdt45.svg 6th
5 Chess --t45.svg Chess pdt45.svg Chess --t45.svg Chess plt45.svg Chess pdt45.svg Chess plt45.svg Chess --t45.svg Chess --t45.svg 5
4th Chess --t45.svg Chess plt45.svg Chess pdt45.svg Chess --t45.svg Chess blt45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg 4th
3 Chess --t45.svg Chess --t45.svg Chess plt45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess plt45.svg 3
2 Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess plt45.svg Chess --t45.svg 2
1 Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess --t45.svg Chess klt45.svg Chess --t45.svg Chess --t45.svg 1
  a b c d e f G H  

Second match game, position after 45.Ra8 – a6, Kasparov resigned

Template: checkerboard / maintenance / new

After Kasparov had won the first match game, the second game came to a remarkable conclusion: Kasparov gave up in a draw position. In the position in the diagram he assumed that he would have to swap queens and that his position would be hopeless after 45 ... Qb6xc6 46. d5xc6. It was analyzed after the game that after 45 ... Qb6 – e3 46. Qc6xd6 Rb8 – e8 47. h4 he could have achieved a draw through continuous chess, but this analysis was rejected in favor of 47. Qd7 + Re7 48. Qc6, which with easier-to-play position for White leads to an unclear endgame. Kasparov, however, had not considered these options and was mentally ill after this defeat.

In the sixth and last game on May 11, 1997, Kasparov completely collapsed with Black and suffered one of the shortest defeats of his career:

1. e2 – e4 c7 – c6 2. d2 – d4 d7 – d5 3. Nb1 – c3 d5xe4 4. Nc3xe4 Nb8 – d7 5. Ne4 – g5 Ng8 – f6 6. Bf1 – d3 e7 – e6 7. Ng1 – f3 h7 –H6 ?! 8.Ng5xe6! Qd8 – e7 9. 0–0 f7xe6 10. Bd3 – g6 + Ke8 – d8 11. Bc1 – f4 b7 – b5 12. a2 – a4 Bc8 – b7 13. Rf1 – e1 Nf6 – d5 14. Bf4 – g3 Kd8 – c8 15 a4xb5 c6xb5 16. Qd1 – d3 Bb7 – c6 17. Bg6 – f5 e6xf5 18. Re1xe7 Bf8xe7 19. c2 – c4 1: 0

After the game, it was discussed whether Black's 7th move was a finger mistake , because by moving 7 ... Bf8 – d6 Kasparov could have avoided tactical entanglements and achieved a solid position. After White's figure sacrifice , which the computer had saved in his opening book and was playing at tempo, Kasparov seemed surprised. Joel Benjamin suspects, however, that his reactions were about acting, because shortly before the variant appeared in a game by Gennadij Timoščenko against the chess program Fritz , which Kasparov was probably familiar with. On move 11 Kasparov made the game-deciding mistake, the 11th… Nf6 – d5 played by Timoščenko with an unclear position would have been better. Kasparov may have trusted Deep Blue to play similarly to Fritz, but it wasn't. In 2018, Kasparov stated that he consciously chose a variant he had never used before, in the expectation that it would not be in Deep Blue's opening book. He knew that 7… h7 – h6 was a bad move, but assumed that Deep Blue could not correctly calculate the potential of the piece sacrifice, since computers back then only sacrificed material if a concrete profit path could be calculated. The sacrifice, however, is speculative: White "only" receives a very strong attack for the piece. Kasparov therefore expected that Deep Blue would withdraw the attacked g5 knight to e4. Since the knight sacrifice was immediate, Kasparov knew that, contrary to his expectations, it was in the computer's opening book. The quick loss of the game, according to Kasparov, was less due to his bad position than to the fact that he was shocked to have made such a mistake, which threw him off his mind. Years later he learned that his assumptions were correct: Deep Blue could not correctly calculate the figure sacrifice and would have made another move himself, as the designers admitted. However, the morning before the game, they decided to add the victim to the opening book.

2003 appeared on the match documentary by Vikram Jayanti , entitled Game Over: Kasparov and the Machine.

Fairness of the competition

The Deep Blue team had a complete history of all Kasparov's public games, the analyzes of which were incorporated into the programming. In addition, Deep Blue's hardware and programming had been significantly improved compared to the first competition last year; Kasparov was thus de facto facing an unknown opponent.

The rules also gave programmers the ability to modify the program between games, which they did extensively. During the competition, errors in the source code were fixed and improvements made, so that Kasparov not only played against the machine, but also against the Deep Blue team, as they helped his system.

Hsu countered this objection by pointing out that a human opponent could learn from the games already played and avoid mistakes once made in further games as far as possible; However, changing the code by hand corresponds neither to machine learning in the sense of artificial intelligence nor to natural human learning . Even IBM contradicts Hsu's reasoning and agrees with Kasparov that Deep Blue is not a learning system, as the company states on its website for the Deep Blue project:

“Deep Blue, as it stands today, is not a 'learning system.' It is therefore not capable of utilizing artificial intelligence to either learn from its opponent or 'think' about the current position of the chessboard. "

And:

“Any changes in the way Deep Blue plays chess must be performed by the members of the development team between games. Garry Kasparov can alter the way he plays at any time before, during, and / or after each game. "

After the lost match, Kasparov said that he had observed high (human) intelligence and creativity in some of the machine's moves and assumed that the machine had been helped by people during the game. Kasparov demanded revenge, but IBM refused the world champion, among other things due to the allegations, a rematch and dismantled Deep Blue into its individual parts. The project cost IBM a total of about $ 5 million. Portions of Deep Blue can be seen today at the Smithsonian Institution in Washington, DC and the Computer History Museum in Silicon Valley . 20 years after the competition, Kasparov withdrew his allegations.

Denied rematch

In his book Behind Deep Blue , Feng-hsiung Hsu claims that he acquired the rights to the chess chips he created from IBM in order to build an even more powerful machine if necessary and to accept Kasparov's rematch offer, but Kasparov is now refusing one Rematch.

literature

  • Feng-hsiung Hsu: Behind Deep Blue. Building the computer that defeated the world chess champion . Princeton University Press, Princeton 2002, ISBN 0-691-09065-3 .
  • Martina Heßler: The success of "Stupidity". Deep Blues victory over the world chess champion Garri Kasparov and the dispute over its importance for artificial intelligence research . NTM Journal for the History of Science, Technology and Medicine 25 (1) 2017. pp. 1–33.
  • Michael Khodarkovsky and Leonid Shamkovich: A new era. How Garry Kasparov changed the world of chess. Ballantine Books, New York 1997, ISBN 0-345-40890-X .

Web links

Individual evidence

  1. Murray Campbell et al .: Deep Blue (2001, PDF; 349 kB)
  2. Kasparov - Deep Blue 1997: History gets it wrong! . Retrieved May 1, 2015.
  3. IBM website for the game, accessed on May 25, 2014
  4. Joel Benjamin: American grandmaster. Everyman Chess, London 2007, pp. 188-197.
  5. Kasparov about the game on Youtube (from 4:28)
  6. Interview with Vikram Jayanti at BBC Four ( Memento from February 10, 2010 in the Internet Archive )
  7. http://www.wired.com/playbook/2012/09/deep-blue-computer-bug/
  8. http://www.n-tv.de/technik/Software-Bug-besiegte-Kasparow-article7372096.html
  9. https://www.research.ibm.com/deepblue/meet/html/d.2.html
  10. Julie Moran Alterio: Deep Blue victory still a milestone 10 years later. In: The Journal News, May 6, 2007
  11. https://www.heise.de/tr/artikel/Deep-Blue-hat-nicht-betrogen-3770127.html
  12. Feng-hsiung Hsu: Behind Deep Blue. Building the computer that defeated the world chess champion . Princeton University Press, Princeton 2002, ISBN 0-691-09065-3 .