Kasparov vs. Deep Blue. Part III: Mezhmatche

The first serious chess match between a man and a car ended in favor of our species, but, as we already know, it was just a prelude to more exciting events.



Before continuing the fascinating reading about the battle of chess titans, representing different "life forms", I hasten to inform you that my co-author Roman Zhukov, who is also Rom77 , has appeared and is alive and well. Unfortunately, he does not have the time and interest to participate in the creation of new episodes, but he provided invaluable assistance that will have the most significant influence on the writing of this (and the following) article. At one time, Roman studied a lot of the vicissitudes of this confrontation, using, among other things, one source, very little known in the Russian-speaking environment.





When the chess passion around matches with Kasparov subsided, the father of the chess supercomputer Feng Xue , as is customary in America, wrote a book where he told a success story. The book, as far as is known, was not published in Russian. Rom77 did a tremendous amount of work in his spare time translating this work into Russian. The translation is not perfect, but it is easy to read and gives a good idea of ​​what happened in the depths of IBM , which managed in its secret laboratories to grow the most formidable artificial opponent for Kasparov.



Download here .

')

Well, we, relying on this book (and not only on it), will continue to recall “as it was” almost two decades ago.



Despite losing two points in the first match, the team of the creators of Deep Blue were optimistic. Programmers had a good understanding of the weaknesses of the supercomputer and, in general, saw ways to improve. There was a clear understanding that the matter was not in the gland - there were more than calculating abilities for the match of the highest level. The car was replayed by Kasparov primarily strategically, due to a more subtle understanding of positional nuances. The 1996 match gave a wealth of information for analyzing and further improving the quality of the game.



While IBM managers were negotiating with Kasparov about a new match, Fen Xu and his associates began to deduce the skill level of their combat robot to a new qualitative stage.



Grandmaster Joel Benjamin, famous for the Philadelphia match, remained the chess trainer of Deep Blue. Moreover, from a guest expert, he grew up as an official IBM employee. The task of Benjamin included endless sparring with the program on the principle of "taking a step back." If Joel got the worst (or just equal) position, he rolled the game back to a position that was more preferable for him, looking for new flaws in the game of his electronic student.





He played, Joel was not with the supercomputer itself, but with his “shadow” called Deep Blue Junior . The software used was almost the same as that intended for the Big Dip , but it was installed on a more modest (although powerful enough) hardware. So it was easier to create conditions for intensive testing.



During one of the merciless cyber-padawan training sessions, experienced Jedi grandmaster Benjamin drew attention to the underestimation of the importance of vertical lines and the inability to capture them with the help of rooks. As chess players know, if a free highway is laid between the opposing camps (or may be laid in the near future), the player who captured it sometimes gets a decisive advantage - the captured vertical can serve for a powerful invasion of heavy artillery (rooks and queen) into the adversary territory.



As it turned out, “dark blue” if he tried to fight for verticals, he did it wrong. He first opened the line, and only then tried to establish control over it, pulling up heavy figures. This gave Joel the opportunity to quickly transfer his rooks to the “road of life” too, easily nullifying the computer’s efforts to seize space. But sometimes you have to do the opposite - first set the rooks on the as-yet-not-open vertical, and only then open it — then domination on the board was achieved without much effort.



The computer had similar problems when fighting over the diagonals with the help of elephants. This, in particular, contributed to the defeat in the last game of the first match, where the black elephant did not find a comfortable place and was tightly locked in the corner.





Learning the computer to correctly capture verticals and diagonals has borne fruit in a future battle. Look at the position. This is a fragment of the 2nd meeting of the upcoming match. Deep Blue is not in a hurry to open the “a” line , but doubles her white rooks on it. Only through a dozen moves will he open for his tank column the passage to the opponent's camp. The invasion will be prepared by white elephants, who will clearly dominate their black counterparts. But it will be later, Deep Blue sample of 1996 for such maneuvers was not capable.



Were radically revised methods for assessing the safety of the king. In 1996, the computer tried to protect its monarch as much as possible by hiding it in bunkers of varying degrees of penetration. But as the parties won by Kasparov showed, the paranoid emphasis on royal security did not solve the problems. Programmers made a more flexible style of play in appropriate situations. If it was necessary, the king did not sit out in his "Wolf's Lair", but took an active part in the hostilities. Looking ahead, I will say that his Majesty’s brave demarche helped the computer to achieve a fantastic draw in the penultimate duel of the rematch. As well as the king's desire to stay closer to the thick of events almost missed the victory over Kasparov in the party number 2.



On the whole, tremendous work has been done to improve positional and strategic thinking in the game of electronic chess.









Grandmaster support was not limited to the help of Joel Benjamin. The famous Miguel Illescas took active part (in the photo). If Joel was just a master of high level, then Miguel at that time represented the elite, played in the same super tournaments as Kasparov. The Spaniard provided invaluable assistance by playing a series of control games with a computer. A number of grandmasters of smaller caliber also took part in the development of Deep Blue talent: Nick de Firmian, John F., Larry Christiansen, Michael Rode.



Despite the fact that "iron" was not a weak link, in parallel, work was underway to improve the old chips and create new ones. This insured the Deep Blue team from an unpleasant situation, if the new boards were not ready on time.





However, there was more than enough time, and fresh details for the computer brain arrived on schedule.



The fact that the next match will be much more interesting than the past showed the game between the old and the new versions of Deep Blue. The improved chess and computer genius easily outplayed his predecessor, which was hopelessly outdated in just a year.



The match scheduled for May 1997 was getting closer. Through the acquaintance Frederik Friedel, in common with Kasparov, the Deep Blue team received alarming rumors that Kasparov was preparing an insidious "anti-computer" strategy. The computer, first of all, is a rational player who relies on a distant calculation and a clear system of assessments of emerging positions. But is chess rational in its essence? Can a genius from among people create such positions on the board that are irrational and fundamentally defy cold counting?





The time when the answers to these questions should be received was inexorably approaching. The match was scheduled to be held from 3 to 11 May 1997. The prize fund was 1 million 100 thousand dollars, of which 700 thousand received the winner and 400 thousand - the loser. And so, on the 35th floor of the New York skyscraper Equitable Building began the final preparations for the competition. If during the battle in Philadelphia Deep Blue played remotely (the moves were transmitted using the Internet), this time the supercomputer was delivered directly to the building and only a few walls separated the rivals ...



Well, friends, we are very close to the apogee of our apotheosis. Initially, I was not going to spend so much personal time on creating a computer-chess saga, I just wanted to tell you about that epic match between Deep Blue and Kasparov, which took place in 1997. Immersed in the history of the issue, I realized that without half a liter here I could not figure out without the tension of the 8-year (!) Prehistory of the narrative would be incomplete. And in order for my graphomaniac efforts to be not only appreciated by the reading public, but also to bring creative satisfaction to me myself - or you will have to tell everything in order or not write anything.





Our serial thriller almost reached its ominous denouement. The climax series has almost passed the post-production stage and will soon be on the air.



( To be continued ... )