Finally I've got some progress!
"Samuel has long maintained that his ignorance of checkers was helpful in getting the program going, that he was never tempted to put in what he himself knew, because he knew nothing. But ignorance didn’t help the hopeful designers of chess machines, and in fact, one of the first really successful programs was done by Alex Bernstein, who knew a great deal about chess to begin with, and soon learned more. Bernstein had a flair for it. He had begun playing when he was nine or ten, and as I was waiting for him one afternoon in his pleasant Brooklyn Heights brownstone, I watched his own nineyear-old patiently teaching a friend some chess tactics. Bernstein was captain of the Bronx Science High School chess team, a school whose pasteboard playing fields have done as much for American science as Eton’s grassy ones did for the British Empire. He recounted the early history of his involvement with the game:
I started playing chess seriously, I guess, when I was in high school. I played chess so much that it affected my grades in college. One year I played chess to the exclusion of everything else and woke up at the end of the term and discovered I had failed two courses. I was going to City College at the time. I failed a physics course and a math course— theory of functions of real variables. It was quite a shock and I gave up chess after that term. I suppose I continued reading about it, but I stopped playing chess. Then in graduate school, although I’d given up math for medieval literature and poetry, I worked as an assistant in the civil engineering department. After that I went into the army, and because of my work at Columbia and what I was doing in the army—working in a special research and development outfit of the Signal Corps—I became acquainted with computers and what they could do.
From the summer of 1953 to the summer of 1955, then, Bernstein worked at the Bureau of Standards, his flagging interest in mathematics revived by a kind woman who had known his father, a well-known European mathematician, and who taught him about the operations of the bureau’s computers. Bernstein was given permission to try his hand at simulating the air-defense system protecting Washington—the first missile air-defense system—to see how it would work. This was at just about the same time that Allen Newell was involved in simulating an air-defense message center out at the RAND Corporation. Computer simulation was in the air. But one main difference between Newell’s and Bernstein’s models was that Bernstein’s was a mathematical computer model and Newell’s was nonmathematical.
After Bernstein got out of the army he returned to Columbia, but the academic atmosphere got him down, and when the opportunity came to work full-time for IBM he took it. He had been working part time for them, and had become friendly with a young man named Hal Judd. Judd knew Bernstein had played serious chess, and one day suggested they try to produce a chessplaying program. He was a very poor chess player but an avid one, as most Americans who play chess are [Bernstein recalls]. I guess he knew computers far better than I did at the time, but he really didn’t know chess, and he felt that with his expertise in computers and my understanding of chess, perhaps we could produce something. Well, almost immediately after I’d said that I thought it would be a good idea, he was transferred someplace out of New York City. Eventually he wound up working for IBM in Australia. Originally, he’d said he’d work with me at long distance, but it became impossible. Nevertheless, the idea of a chess-playing program really intrigued me, so I decided to see if I could come up with a scheme for producing this program and for getting some backing. It wasn’t impossible to work on this problem oneself, but one did need computer time and computer time is very expensive, although at the time there was a good deal of unused computer time during the third shift. Nevertheless, we had to have some permission at least to do it. So we went to Charlie DeCarlo, who was head of the so-called Applied Science Division at IBM. DeCarlo is a mathematician and came out of Carnegie, and he was very sympathetic to the idea, and said he’d support it on a limited basis. Thus I went to work full-time for IBM at what was then the Scientific Center, and later became part of the Service Bureau, in Manhattan. I was given other work to do, but essentially it was understood that half of my time I would be allowed to spend working on the chess program. Bernstein drew upon not only his own experience with chess, but began to study Modern Chess Openings, which came out then every two years, and spent six months going through some five hundred chess openings. He assigned scores to various positions, scores that depended not only on the pieces retained, but also on area control of the board and mobility. He also developed a fourth measure, what he called a “greens area” around the king, meaning that the more squares outward from the king controlled by his own side the better. But after six months of this he gave it up. He couldn’t make any sense out of it.
“It was essentially a chastening experience. I was sort of abashed because I had said to people, I can do it.” Bernstein was well aware that chess literature contained mentions of chess machines, and certainly science fiction—say, Ambrose Bierce’s “Moxon’s Master”—provided examples. Baron von Kempelen’s nineteenth-century chess machine might have been a fraud, but the idea was intriguing enough to even so serious a scientist as Charles Babbage, who hoped to raise money to complete his Analytical Engine by working up a little chess-playing machine for exhibition. And Torres y Quevedo had actually built two endgame players in the 1920s.
At this time, Bernstein was unaware of Shannon’s seminal papers, and did not know that chess had caught the interests of a group at Los Alamos, including J. Kister, P. Stein, S. Ulam, W. Walden, and M. Wells, who were working on a limited 6 x 6 board, rather than the regulation 8 x 8. Nor did he know that Allen Newell, J. C. Shaw, and Herbert Simon together, and John McCarthy independently, were also pondering chess-playing machines. Alex Bernstein only knew that the problem was hot, and though his confidence was slightly shaken by the experience of his first six months on the problem, he was all the more anxious to try again. That it was a classic problem made it even more tempting—the creation of a smart chess-playing machine had come to seem as elusive as solving the four-color problem, and the solution of either would have been a scientific coup. Bernstein went back to the chess-theory books. This time he found insight in a book called My System by Aron Nimzowitsch, an early twentieth-century Russian chess master who had revolutionized modern chess. Nimzowitsch had abandoned the traditional strategy of building up the center game and gradually grinding down one’s opponent. Instead, he stressed the notion of imbalance on the chess board—not necessarily immediate control of the center, but a delayed strategy, emphasizing what are called strong points and lead points. It was now that Bernstein became aware of Turing’s work and read at least one of Shannon’s papers. When he finally began to see how he might codify some of the principles he felt were essential, he telephoned Claude Shannon at MIT. “I went up to MIT and spent a day or two with him, telling him what I was planning to do, and he said he thought it was intelligent, and a good way of proceeding. Essentially I felt I’d received his blessings, which was pleasant.”
Bernstein also mentioned that he was working on the problem to Dr. Edward Lasker, a well-known chess writer, who introduced him to Stanislaw Ulam of the Los Alamos group. Bernstein had the advantage that the Los Alamos group didn’t have, of a machine with a large amount of memory, although the four thousand words of memory the IBM 704 had to begin with were insufficient for Bernstein’s program in the end. The 704’s memory was to have doubled by the time Bernstein finished his program, and he still came within two hundred words of overflowing memory. Bernstein’s program turned out to be a perceptive combination of his own chess intuitions, what he had learned from Nimzowitsch’s book, and happily, some of the things he had learned from his first six months with Modern Chess Openings. One of the program’s major features was that it eliminated a large portion of the legal possible moves from consideration, and concentrated upon those legal moves that were likely to prove fruitful.
As noted earlier, a popular misconception about a computer chess-playing program is that it can somehow consider all possible moves, and from this omniscient survey, pick the best move. As we’ve seen, in 1948 Claude Shannon had computed the possible moves in a chess game to be 10^120, which means that the heat death of the universe would terminate such a game before it could be fully played out. It had been clear to researchers from the start that some means of economizing was essential. Thus, for example, the Los Alamos group’s program considered all alternatives to a depth of two moves (that is, two for white and two for black), computing a score by evaluating mobility and material. To carry out computing on their machine, a MANIAC I, within reasonable times (if an average of twelve minutes per move can be considered reasonable), bishops were eliminated as pieces, as were all special moves such as castling and two-square pawn moves at the opening. Bernstein introduced some rules of thumb to his game that would direct it toward the most immediately fruitful moves, and Samuel had done the same for his checkers program. Such heuristics were that, and no more. They weren’t invariably the best strategy; sometimes they missed the unusual and the brilliant, and sometimes they didn’t work at all. But they were a reasonable means of pruning away the less likely possibilities of action. The notion of economizing searches for solutions among a host of possibilities would be central to artificial intelligence, with a good case to be made for its centrality in natural intelligence as well.
So Bernstein’s chess program selected what seemed to be the likeliest fruitful moves, and these it examined in considerable depth, comparing one to another among a number of dimensions. The program contained a large data base, which allowed it to examine any particular piece or square at any time. In descending order of importance, the program asked such questions as, Is the king in check? If the king is in check, there is nothing else to do. Is the king in double check? If he is, merely to capture one piece that threatens the king will be insufficient; the king must be moved. The next question had to do with material: is there any to be gained, or any in danger of capture? And clearly it is more important to rescue or capture a rook than to rescue or capture a pawn, and this was factored into the program. And so it went. By now Bernstein had transferred to IBM proper, and he and his group were sharing their machine with the research team that was designing FORTRAN, a computer language that would allow a human to write programs in a language more natural to humans, and that would then be translated by the machine itself into machine code. Each group needed the entire machine, and when one was on, the other had to be off. Since the groups needed huge amounts of computing time, both of them worked the third shift, so an extraordinary amount of computing was going on in the small hours at the corner of East 56th and Madison Avenue in Manhattan. Bernstein had begun working full-time on his project, and he had a number of assistants, some of whom were even officially his. The two groups watched and were cheered by each other’s progress.
There came a time to try out the chess machine. Bernstein had the sense to try it out first without an audience. “There was a bug. The very first move the machine ever made was to resign!” It had taken two years of work to get the chess-playing machine going, and for several years thereafter, bugs were still being discovered, not only by Bernstein and his group, but by outsiders who had requested copies of the program, and who uncovered more surprises. “It played, I think, a sort of respectable beginner’s game,” Bernstein says, “and every once in a while it made a move which was remarkably good.”
The success of the chess machine had some unexpected results. To be sure, Bernstein received all the publicity he could have hoped for—besides the usual scientific meetings he was invited to address, he found himself written up in the New York Times, and an article in Scientific American reached a wide international audience. Life magazine came asking for photographs of Bernstein sitting at the computer, and wondered if they could get Bobby Fischer to pose too. He was in his early teens at the time, but shrewd enough, and said he would for a fee of 2500, Bernstein recalls. They said forget it, and asked me did I know anyone else who might be willing to pose for a picture. And I said, I’m sure that Ed Lasker was a very respected name in chess and a charming man and a gentleman, and would not ask for 2500. He said he would be delighted, and they paid us each $1. The funny thing about it is they proceeded to go to some antique dealer on Madison Avenue and rented a chess set—an Indian chess set of the sixteenth century which cost all of 2500 and a chess board which cost 1200, and everybody was absolutely dying in case any of the pieces should fall over. They were extremely delicate filigree ivory, and they were insured.
But the pictures didn’t turn out and had to be retaken, meaning that the chess set and board had to be re-rented, the cost well exceeding, Bernstein calculates, what it would have cost to rent Bobby Fischer instead. As it turned out, Life never did use the pictures in its magazine, although six or seven years later Bernstein was surprised to discover a picture of himself and Ed Lasker standing in front of their 704 in the Time-Life series on mathematics. But T. J. Watson, the president of IBM, was not amused. IBM’s original, or at least official, justification for allowing Bernstein to use the first 704 for nothing more serious than game playing had been the hope that if he were successful, it would show the world—in particular, businesspeople—that computers could be used to solve problems even as difficult as ones that came up in business. But IBM’s stockholders had challenged Watson at the last meeting, wanting an explanation for the money being wasted on playing games."
McCorduck, P. (2004). Machines who think: a personal inquiry into the history and prospects of artificial intelligence (2. Edition). London: A.K. Peters. // https://books.google.ru/books?id=aH9QAAAAMAAJ