The fall of the last intellectual bastion: is a computer chess player really stronger than a man?
At writing this review pushed the post “The secret of the ancient game of go. Why has the computer still not beat the man? ” , Published on 25 May. In the post itself, and, especially, in the comments, a lot was said about computer chess in general and the Deep Blue - Kasparov (1997) match in particular. It is clear that now, after nearly twenty years, very few people are interested in all the details of that match: computers are developing at a tremendous speed, modern smartphones will easily give odds to computers of that time, and perhaps chess itself has lost some popularity lately - for some reason - this is a topic for another conversation.
However, some details, apparently, are really unknown, and these details are such that the headlines about the “fall of the last intellectual bastion” are nothing more than a newspaper device, for the result of the match, in fact, a scandal, due to its chess specificity would hardly be of interest to the general public. No, I, despite the fact that I have always been a fan of Garry Kasparov (exclusively in chess terms), I am not going to justify him for that defeat and try to prove that everything was not at all the way it is now widely known. And even more so the goal is not the refutation of some comments on the chess theme to the alizar habraouser post . The only goal is to tell some details of what exactly happened in New York in early May 1997, and why, according to the author, the result of this confrontation has really proved nothing to anyone.
Immediately make a reservation: the author is not a highly qualified chess player. Perhaps some purely chess moments in this material may be criticized by people involved in chess professionally. Also, alas, the author is not strong either in game theory or in chess programming, and will only try to express the opinion of a partially chess player who is partially familiar with computers, and not create a Worldwide and All-Encyclopaedia of Chess Programming.
It is not enough to be a good player; you should also play well
The history of chess programming began, in fact, immediately with the usual programming. Artificial intelligence, the possibility of its creation, has been worrying people for a long time, and programmable computer technology turned out to be a more suitable means for finding such an opportunity than the person who beat up Napoleon sitting in the nightstand . The first chess programs were, by the way, only rivals to each other: due to the huge number of options for chess games , there was no question of complete enumeration, and there is not, by the way, to this day. (And it is not foreseen (c) Shvonder.) This did not prevent chess programs from periodically surprising professional chess players with unexpectedly best moves (which is primarily due to the fact that the computer considers forced versions still faster and more accurately) and sudden views of the mat in a couple -troiku moves to save the queen. However, if in tactics even relatively weak computers were quite strong, then in the positional game everything was very sad - if you could somehow teach the program that the queen is more expensive than a pawn (which in some cases led to the aforementioned views of the mat) It was not possible to teach an assessment of the position on which, in fact, the whole party holds. This is not surprising - the correct assessment of the position is not an easy task, even for chess players, there is no single recipe, even two almost identical positions, differing only in the position of one pawn, can have opposite evaluations.
')
Over time, however, progress began to take its own. Debut libraries allowed programs not to “float” in the initial stage of the party - such “voyages” often led to the end of the party in the opening - nevertheless, several centuries of experience and the growth of opening theory from the beginning of the twentieth century gave a person a considerable head start. Debut libraries this handicap is not just eliminated - unlike the person, the computer now could play absolutely any debut, get away from any debut trap, and he, respectively, could catch any trap! There are no chess players who know all the openings - there is a concept “debut repertoire” - this is a certain number of beginnings used by a chess player. A person can expand, replenish this repertoire, prepare, for example, new debuts for a new tournament, but with the advent of computers in the memory of the debut encyclopedia, people began to lose in this component. By the way, a double handicap came out here: unlike a computer, which, in fact, plays in the opening library, a person cannot spy on the opening encyclopedia during the game. The only way to equalize chances here is incorrect beginnings that are not included in the encyclopedia, and the maximum chances are to equalize exactly - a person, after all, also has no opportunity to apply the baggage of his debut knowledge.
It must be said that the problem of opening knowledge and their excessive prevalence and accessibility was also of concern to the players themselves. So, situations in which players made 30–40 moves, and there was nothing new in them, became completely normal in professional chess - all these moves were already in another game. (Here, of course, the matter concerns not only the opening, but, as a rule, such situations were associated primarily with debut disputes, and some option could be defended in a deep middle game, and even in the endgame!) supergross-masters - a party with up to 20 moves, usually - 16-18, all “out of the book” and an agreement for a draw - because the position is theoretically drawn. Salvation saw Chess Fisher, they are Chess-960, where the figures in the initial position can be placed arbitrarily. Alas (or fortunately?), This option could not be ousted by ordinary chess. Why it is difficult to say, but I would venture to suggest that he was not quite interesting for professional chess players, they lacked something to do in the stock debuts, did not give any advantages to the rest, see Kasparov’s comment. In fact, the opening principles did not change - “develop your pieces as quickly and safely as possible and prevent them from doing the same to your opponent”, and those who understood these principles in standard chess didn’t stop understanding them in Fisher, and who moved the pieces exclusively from memory from the reference book and with any deviation fell into a panic - began to fall into this very panic much faster, at 2-3 moves.
Much better in theory were things with an endgame. Especially with low-figured endings - the smaller the figures - the better. Such situations, unlike debuts, were much easier to figure out: there are far fewer figures! Yes, and restrictions on the number of moves contributed (see. Rule 50 moves ). One way or another, computers were able to use the Endgame Tables of Nalimov, and this was probably an even bigger victory than debut libraries. However, the person here had his trump cards - for example, the fact that many variants of chess endings could be played, practically not counting the options, but using well-known algorithms, the most famous example (the first study, the probability of such motives in a real game is quite high).
But in the middlegame, man remained the master of the situation. Yes, the algorithms were improved to assess the position, yes, computers considered farther and deeper, but one thing remained, which is why computers have so far not become either writers or composers: the plan. If in the opening the computer could move the pieces exclusively “according to the book”, and in the endgame it could follow certain algorithms, then there were difficulties with the plan. Yes, obviously, if, when evaluating a position, a car found weak points for an opponent, then it could make moves aimed at using them. If the weaknesses were at the car - then she could take action to protect. However, a live chess player will not be guided only by his plan - he will try to determine the opponent’s plans. Remarkable task for AI. Is it just doable?
In 1996, IBM representatives offered Garry Kasparov a match against their Deep Blue chess program with a prize pool of $ 500,000. Deep Blue is a supercomputer based on the RS6000 system, consisting of 32 nodes, each of which consists of 512 processors. hardware optimized for chess program. The performance of Deep Blue corresponded to 11.38 GFLOPS, and the computer could estimate up to 200 million positions per second ( Wikipedia ).
The thirteenth world champion Garry Kasparov has always considered his occupation primarily to be creative. That is, the fact that the machine is not available and will never be available. Kasparov said:
Ironically, it was Kasparov who became the first world champion to lose a computer. The very first game of the Deep Blue - Kasparov match (Philadelphia, February 1996) brought a sensation - the World Champion was hit by a car. Alas for fans of AI - in the second game the world champion was rematchable, and after two more games that ended in a draw, he won twice, thus proving that the world champion is a world champion. Nevertheless, it was already clear that computer chess is a real power, and that in the second match, which IBM offered a little over a year later, the champion-man will have to try to protect the human race.
Kasparov's second match with “Dark Blue” began, probably, even with a disappointing result for fans of intrigue - Kasparov won with White easily, choosing the wrong start (more precisely, this debut can be classified as Reti Debut). It seemed that everything was clear, the idea of Kasparov was simple - to deprive the opponent of the first trump card, the debut library, and simply play chess, which the machine was not given. Of course, it is much more difficult for Black to move away from theory when playing, white will set the tone, and here Kasparov probably hoped for the Spanish game, the same one that Ostap Bender himself played in the amount of eighteen pieces, where the old debut was not enough , you must be able to play the middle game. And here it began.
He is not a chess player who, having lost the game, does not declare that he had a winning position
So, at first - only game facts. The games were played with a control of 2 hours for 40 moves (classic time control). Playing black, Kasparov tried to revive the position with the help of the victims, the car rejected these victims, and as a result, the world champion, who was in the most difficult position, gave up. Everything is simple and clear. Further, part of the description and quotes borrowed from here (there you can see the whole batch number 2), part - described by the author.
According to the champion, the first bell rang on the 35th move - a supercomputer, which previously spent no more than three minutes on the course, thought for a quarter of an hour. Another six minutes - over the next turn. The result was even more unexpected - the computer rejects the pawn sacrifice with black.
Under the terms, after each batch, IBM provided printouts of the analyzes that the computer produced during each batch. According to Kasparov, the variant chosen by the car was assessed as unclear to her, unlike acceptance of the sacrifice, which the car regarded as advantageous for itself. "Very cute. We are dealing with a unique event, the Machine refuses to win three pawns, because it is allegedly “unclear” to her - G.Kasparov.
The party continued, and moved to a logical ending, which finally came - Kasparov surrendered. However, the miracles did not end there. As it turned out almost immediately - the world champion surrendered in a drawn position. It would seem that there is no one to blame here, except for the victim himself. But everything is not so simple. Kasparov surrendered, simply believing that the car is infallible. And wrong. With his last move in the game, the machine made a gross tactical mistake, after which Black could end the game in a draw. Kasparov's surprise after the game is completely understandable: the machine that found the strongest positional move, which even a lot of protein chess players could not do, that is, bypassed people on their own territory - in a positional game, immediately lost on its territory, where it cannot be mistaken - allowing an elementary tactical blunder?
These questions still remain questions, and the match, nevertheless, continued. Three games ended in a draw. Was the sixth game for Kasparov a sort of final? The final in the match for the honor and mind of the human race.
In chess, the one who makes the last but one mistake wins
Perhaps, from the point of view of computer chess, nothing interesting happened in the sixth game. The questions in this game are mainly for the world champion - how could he make a move 7. ... h6 ? The computer immediately sacrificed the knight and the position of Black rolled down the slope. After the 19th move, White Kasparov surrendered. His only achievement in this game is that officially it did not become the fastest defeat in his career - chess statistics do not take into account the party of a person with a computer. Although formally, as you understand, this is exactly such a party.
Then it was less interesting - allegedly, IBM refused to provide logs of analyzes of this game, which, apparently, is not true (see links below), and Kasparov’s proposal to play another match was greeted by the corporation in an original way - the hero of the match (Deep Blue) was dismantled and put in ascrap museum. However, this is just easy to explain by the fact that one way or another, the world champion was defeated, the goal was achieved, and once again they didn’t really want to lay out a round sum for Kasparov’s consolation.
And yet, what proved these matches? What man is defeated? If we take both matches of Kasparov with DB, then the score remained in favor of Kasparov - 6.5-5.5. Even if we take only the second - firstly, probably, statistically the result of 2-1 in favor of one side (without taking into account any draws) does not prove anything. Secondly, the defeat of Kasparov in the sixth game - due to the blatant yawn - is unlikely on the basis of such a gross error (I am silent about the second game in the draw position by Kasparov) can prove the strength of the chess program.
After moving Deep Blue to the museum, people, of course, did not stop playing chess with a computer. However, such matches are not audible at all, grandmasters now prefer more exotic uses of the computer . Maybe the leading grandmasters are just afraid of defeats from the car? Why, then, the owners of supercomputers, the authors of powerful chess programs, do not declare, they say, that we have proposed a match, but the world champion refused? Most likely it’s really a matter of falling interest in chess, it’s more interesting now to set other tasks for AI, and defeat from a chess machine, even a world champion, to whom it is interesting, because “the car won the world champion a long time ago”.
References:
IBM's 1997 match report
Sixth installment of the 1997 match on Wikipedia
However, some details, apparently, are really unknown, and these details are such that the headlines about the “fall of the last intellectual bastion” are nothing more than a newspaper device, for the result of the match, in fact, a scandal, due to its chess specificity would hardly be of interest to the general public. No, I, despite the fact that I have always been a fan of Garry Kasparov (exclusively in chess terms), I am not going to justify him for that defeat and try to prove that everything was not at all the way it is now widely known. And even more so the goal is not the refutation of some comments on the chess theme to the alizar habraouser post . The only goal is to tell some details of what exactly happened in New York in early May 1997, and why, according to the author, the result of this confrontation has really proved nothing to anyone.
Immediately make a reservation: the author is not a highly qualified chess player. Perhaps some purely chess moments in this material may be criticized by people involved in chess professionally. Also, alas, the author is not strong either in game theory or in chess programming, and will only try to express the opinion of a partially chess player who is partially familiar with computers, and not create a Worldwide and All-Encyclopaedia of Chess Programming.
It is not enough to be a good player; you should also play well
(Z. Tarrasch)
The history of chess programming began, in fact, immediately with the usual programming. Artificial intelligence, the possibility of its creation, has been worrying people for a long time, and programmable computer technology turned out to be a more suitable means for finding such an opportunity than the person who beat up Napoleon sitting in the nightstand . The first chess programs were, by the way, only rivals to each other: due to the huge number of options for chess games , there was no question of complete enumeration, and there is not, by the way, to this day. (And it is not foreseen (c) Shvonder.) This did not prevent chess programs from periodically surprising professional chess players with unexpectedly best moves (which is primarily due to the fact that the computer considers forced versions still faster and more accurately) and sudden views of the mat in a couple -troiku moves to save the queen. However, if in tactics even relatively weak computers were quite strong, then in the positional game everything was very sad - if you could somehow teach the program that the queen is more expensive than a pawn (which in some cases led to the aforementioned views of the mat) It was not possible to teach an assessment of the position on which, in fact, the whole party holds. This is not surprising - the correct assessment of the position is not an easy task, even for chess players, there is no single recipe, even two almost identical positions, differing only in the position of one pawn, can have opposite evaluations.
')
Over time, however, progress began to take its own. Debut libraries allowed programs not to “float” in the initial stage of the party - such “voyages” often led to the end of the party in the opening - nevertheless, several centuries of experience and the growth of opening theory from the beginning of the twentieth century gave a person a considerable head start. Debut libraries this handicap is not just eliminated - unlike the person, the computer now could play absolutely any debut, get away from any debut trap, and he, respectively, could catch any trap! There are no chess players who know all the openings - there is a concept “debut repertoire” - this is a certain number of beginnings used by a chess player. A person can expand, replenish this repertoire, prepare, for example, new debuts for a new tournament, but with the advent of computers in the memory of the debut encyclopedia, people began to lose in this component. By the way, a double handicap came out here: unlike a computer, which, in fact, plays in the opening library, a person cannot spy on the opening encyclopedia during the game. The only way to equalize chances here is incorrect beginnings that are not included in the encyclopedia, and the maximum chances are to equalize exactly - a person, after all, also has no opportunity to apply the baggage of his debut knowledge.
It must be said that the problem of opening knowledge and their excessive prevalence and accessibility was also of concern to the players themselves. So, situations in which players made 30–40 moves, and there was nothing new in them, became completely normal in professional chess - all these moves were already in another game. (Here, of course, the matter concerns not only the opening, but, as a rule, such situations were associated primarily with debut disputes, and some option could be defended in a deep middle game, and even in the endgame!) supergross-masters - a party with up to 20 moves, usually - 16-18, all “out of the book” and an agreement for a draw - because the position is theoretically drawn. Salvation saw Chess Fisher, they are Chess-960, where the figures in the initial position can be placed arbitrarily. Alas (or fortunately?), This option could not be ousted by ordinary chess. Why it is difficult to say, but I would venture to suggest that he was not quite interesting for professional chess players, they lacked something to do in the stock debuts, did not give any advantages to the rest, see Kasparov’s comment. In fact, the opening principles did not change - “develop your pieces as quickly and safely as possible and prevent them from doing the same to your opponent”, and those who understood these principles in standard chess didn’t stop understanding them in Fisher, and who moved the pieces exclusively from memory from the reference book and with any deviation fell into a panic - began to fall into this very panic much faster, at 2-3 moves.
Much better in theory were things with an endgame. Especially with low-figured endings - the smaller the figures - the better. Such situations, unlike debuts, were much easier to figure out: there are far fewer figures! Yes, and restrictions on the number of moves contributed (see. Rule 50 moves ). One way or another, computers were able to use the Endgame Tables of Nalimov, and this was probably an even bigger victory than debut libraries. However, the person here had his trump cards - for example, the fact that many variants of chess endings could be played, practically not counting the options, but using well-known algorithms, the most famous example (the first study, the probability of such motives in a real game is quite high).
But in the middlegame, man remained the master of the situation. Yes, the algorithms were improved to assess the position, yes, computers considered farther and deeper, but one thing remained, which is why computers have so far not become either writers or composers: the plan. If in the opening the computer could move the pieces exclusively “according to the book”, and in the endgame it could follow certain algorithms, then there were difficulties with the plan. Yes, obviously, if, when evaluating a position, a car found weak points for an opponent, then it could make moves aimed at using them. If the weaknesses were at the car - then she could take action to protect. However, a live chess player will not be guided only by his plan - he will try to determine the opponent’s plans. Remarkable task for AI. Is it just doable?
In 1996, IBM representatives offered Garry Kasparov a match against their Deep Blue chess program with a prize pool of $ 500,000. Deep Blue is a supercomputer based on the RS6000 system, consisting of 32 nodes, each of which consists of 512 processors. hardware optimized for chess program. The performance of Deep Blue corresponded to 11.38 GFLOPS, and the computer could estimate up to 200 million positions per second ( Wikipedia ).
The thirteenth world champion Garry Kasparov has always considered his occupation primarily to be creative. That is, the fact that the machine is not available and will never be available. Kasparov said:
“If a computer can beat the best of the best in chess, it will mean that the computer is able to compose the best music, write the best books. I can not believe it. If a computer with a rating of 2800, that is equal to mine, is created, I myself would consider it my duty to call him for a match in order to protect the human race. ”
Ironically, it was Kasparov who became the first world champion to lose a computer. The very first game of the Deep Blue - Kasparov match (Philadelphia, February 1996) brought a sensation - the World Champion was hit by a car. Alas for fans of AI - in the second game the world champion was rematchable, and after two more games that ended in a draw, he won twice, thus proving that the world champion is a world champion. Nevertheless, it was already clear that computer chess is a real power, and that in the second match, which IBM offered a little over a year later, the champion-man will have to try to protect the human race.
Kasparov's second match with “Dark Blue” began, probably, even with a disappointing result for fans of intrigue - Kasparov won with White easily, choosing the wrong start (more precisely, this debut can be classified as Reti Debut). It seemed that everything was clear, the idea of Kasparov was simple - to deprive the opponent of the first trump card, the debut library, and simply play chess, which the machine was not given. Of course, it is much more difficult for Black to move away from theory when playing, white will set the tone, and here Kasparov probably hoped for the Spanish game, the same one that Ostap Bender himself played in the amount of eighteen pieces, where the old debut was not enough , you must be able to play the middle game. And here it began.
He is not a chess player who, having lost the game, does not declare that he had a winning position
(I. Ilf)
So, at first - only game facts. The games were played with a control of 2 hours for 40 moves (classic time control). Playing black, Kasparov tried to revive the position with the help of the victims, the car rejected these victims, and as a result, the world champion, who was in the most difficult position, gave up. Everything is simple and clear. Further, part of the description and quotes borrowed from here (there you can see the whole batch number 2), part - described by the author.
According to the champion, the first bell rang on the 35th move - a supercomputer, which previously spent no more than three minutes on the course, thought for a quarter of an hour. Another six minutes - over the next turn. The result was even more unexpected - the computer rejects the pawn sacrifice with black.
Under the terms, after each batch, IBM provided printouts of the analyzes that the computer produced during each batch. According to Kasparov, the variant chosen by the car was assessed as unclear to her, unlike acceptance of the sacrifice, which the car regarded as advantageous for itself. "Very cute. We are dealing with a unique event, the Machine refuses to win three pawns, because it is allegedly “unclear” to her - G.Kasparov.
The party continued, and moved to a logical ending, which finally came - Kasparov surrendered. However, the miracles did not end there. As it turned out almost immediately - the world champion surrendered in a drawn position. It would seem that there is no one to blame here, except for the victim himself. But everything is not so simple. Kasparov surrendered, simply believing that the car is infallible. And wrong. With his last move in the game, the machine made a gross tactical mistake, after which Black could end the game in a draw. Kasparov's surprise after the game is completely understandable: the machine that found the strongest positional move, which even a lot of protein chess players could not do, that is, bypassed people on their own territory - in a positional game, immediately lost on its territory, where it cannot be mistaken - allowing an elementary tactical blunder?
These questions still remain questions, and the match, nevertheless, continued. Three games ended in a draw. Was the sixth game for Kasparov a sort of final? The final in the match for the honor and mind of the human race.
In chess, the one who makes the last but one mistake wins
(S.Tartacover)
Perhaps, from the point of view of computer chess, nothing interesting happened in the sixth game. The questions in this game are mainly for the world champion - how could he make a move 7. ... h6 ? The computer immediately sacrificed the knight and the position of Black rolled down the slope. After the 19th move, White Kasparov surrendered. His only achievement in this game is that officially it did not become the fastest defeat in his career - chess statistics do not take into account the party of a person with a computer. Although formally, as you understand, this is exactly such a party.
Then it was less interesting - allegedly, IBM refused to provide logs of analyzes of this game, which, apparently, is not true (see links below), and Kasparov’s proposal to play another match was greeted by the corporation in an original way - the hero of the match (Deep Blue) was dismantled and put in a
And yet, what proved these matches? What man is defeated? If we take both matches of Kasparov with DB, then the score remained in favor of Kasparov - 6.5-5.5. Even if we take only the second - firstly, probably, statistically the result of 2-1 in favor of one side (without taking into account any draws) does not prove anything. Secondly, the defeat of Kasparov in the sixth game - due to the blatant yawn - is unlikely on the basis of such a gross error (I am silent about the second game in the draw position by Kasparov) can prove the strength of the chess program.
After moving Deep Blue to the museum, people, of course, did not stop playing chess with a computer. However, such matches are not audible at all, grandmasters now prefer more exotic uses of the computer . Maybe the leading grandmasters are just afraid of defeats from the car? Why, then, the owners of supercomputers, the authors of powerful chess programs, do not declare, they say, that we have proposed a match, but the world champion refused? Most likely it’s really a matter of falling interest in chess, it’s more interesting now to set other tasks for AI, and defeat from a chess machine, even a world champion, to whom it is interesting, because “the car won the world champion a long time ago”.
References:
IBM's 1997 match report
Sixth installment of the 1997 match on Wikipedia
Source: https://habr.com/ru/post/227679/
All Articles