The crowd at Philadelphia’s Masonic Hall may have heard rumors of the mysterious machine that had the ability to play chess, but the day after Christmas of 1826 was, for most, the first time they would see the “Chess Playing Automaton” in the flesh. Johann Maelzel, a showman par excellence, took the stage and directed the attention of the audience to the machine by his side: a box about the size of an executive desk, with a mannequin emerging from the top, dressed in robes and a turban in the style of “an oriental sorcerer.”
With theatrical flourish, Maelzel opened the side door to “The Turk,” revealing gears and gadgetry. By the time the machine had checkmated its first opponent, the crowd was astounded. Silas Weir Mitchell, the eminent physician and writer, was moved to observer that “the Turk—even he, with his oriental silence and rolling eyes, would haunt your nightly visions for many an evening after. Since then we have known him better, and we confess to this day a certain mysterious awe of his eternal cross-leggedness, his turbaned front, and left-handed activity.” In that face of that, sorcery seemed a plausible explanation.
The only sorcery, though, was how Maelzel managed to get away with it. The Turk was a hoax: inside the elaborately designed device, behind the gears and pulleys, was a human chess player, navigating each game like a puppeteer. Some of the era’s greatest chess players would power the Turk, and yet its secret remained hidden from the public for several decades. Edgar Allan Poe, among others, was prescient enough to investigate the phenomenon, fixing his suspicion on one of the Turk’s handlers “who is never to be seen during the exhibition of the Chess Player, although frequently visible just before and just after.”
But Poe’s skepticism was a minority opinion, and for much of the century it was believed that the machine was just as good—and terrifying—as advertised. The Turk tapped into an anxiety as persistent as it was powerful. Before the legend of John Henry and the fear of a machine to surpass man’s might, before science fiction imagined artificial intelligence or the singularity, there was the Turk, a machine that claimed to surpass its creators. Of course, the Turk was a hoax; but the hoax was only a reprieve.
If Shannon was cheerfully optimistic about the possibilities of thinking machines, it wasn’t only because he had built a mechanized mouse that could find its way through a maze to a piece of steel cheese—and remember its path. It was also because, in the late 1940s and early 1950s, he turned his curiosity to the question of how and whether a computer might be programmed to compete against a human being in a chess match. It didn’t matter that the recent history of such machines was a story of hucksters; Shannon believed it was possible for a computer to play honestly, and to play better than a human. What this research gave Shannon was yet more assurance that a properly programmed machine could do more than mimic a human brain—it could best it.
In a life of pursuits adopted and discarded with the ebb and flow of Shannon’s promiscuous curiosity, chess remained one of his few lifelong pastimes. One story has it that Shannon played so much chess at Bell Labs that “at least one supervisor became somewhat worried.” He had a gift for the game, and as word of his talent spread throughout the Labs, many would try their hand at beating him. “Most of us didn’t play more than once against him,” recalled Brockway McMillan.
On a trip to Russia in 1965, Shannon offered a friendly game to Soviet international grandmaster and three-time world champion Mikhail Botvinnik. Botvinnik, having presumably endured countless games of show for various dignitaries, agreed to the match but played without paying much attention and nursed a cigarette throughout, his uninterest apparent to all in the room. Then, suddenly, Shannon managed to win the favorable exchange of his knight and a pawn for Botvinnik’s rook early in the contest. Botvinnik’s attention was instantly yanked back to the board, and the atmosphere of the room shifted as the Russian champion realized that his challenger was more than just another hapless dignitary. “Botvinnik was worried,” Betty would remember years later.
The game went on far longer than anyone, including the surprised champion, could have predicted. But there was still no real doubt about the outcome. After forty-two moves, Shannon tipped his king over, conceding the match. Still, lasting dozens of moves against Botvinnik, considered among the most gifted chess players of all time, earned Shannon lifelong bragging rights.
(Another incident from the same trip to Russia spoke to his and Betty’s sense of humor. When Shannon complained aloud that the lock on their hotel room’s door was broken, a locksmith instantly appeared—leading them to suspect that their room had been bugged by the Soviet authorities. Their next move was to complain aloud that they had never received the royalties for the Russian edition of his book—and a check materialized the next day.)
His work on computerized chess would, in time, be recognized as another instance of Shannon dropping into a field and, in one stroke, defining its limits and unearthing many of its central possibilities. Decades after the publication of his paper “Programming a Computer for Playing Chess,” Byte magazine would put it succinctly: “There have been few new ideas in computer chess since Claude Shannon.” The paper that would bring the world a significant step closer to an actual, working Turk attracted none of the hoax’s audience or attention. Shannon introduced his idea for a chess-playing computer with characteristic modesty: “Although perhaps of no practical importance, the question is of theoretical interest, and it is hoped that a satisfactory solution of this problem will act as a wedge in attacking other problems of a similar nature and of greater significance.”
Shannon imagined some future applications of a chess-playing artificial intelligence: routing phone calls, translating text, composing melodies. As he reminded his readers, these machines were right around the technological corner, and no one doubted their economic utility. As diverse as these applications were, they had an important quality in common: they didn’t operate according to a “strict, unalterable computing process.” Rather, “solutions of these problems are not merely right or wrong but have a continuous range of ‘quality.’ ” In this way, chess was a valuable test case for the emerging generation of artificial intelligence.
Nearly a half century before Deep Blue defeated the world’s human champion, Shannon anticipated the value of chess as a sort of training ground for intelligent machines and their makers:
The chess machine is an ideal one to start with, since: (1) the problem is sharply defined both in allowed operations (the moves) and in the ultimate goal (checkmate); (2) it is neither so simple as to be trivial nor too difficult for satisfactory solution; (3) chess is generally considered to require “thinking” for skillful play; a solution of this problem will force us either to admit the possibility of a mechanized thinking or to further restrict our concept of “thinking”; (4) the discrete structure of chess fits well into the digital nature of modern computers.
Shannon believed that, at least within the realm of chess, the inanimate had certain intrinsic advantages. The obvious ones were processing speeds well beyond the human brain and an endless capacity for computation. Further, an artificial intelligence wouldn’t be susceptible to boredom or exhaustion; it could continue to drill into a chess position well after its human counterpart had lost concentration. Computers were, in Shannon’s view, blessed with “freedom from errors,” their only mistakes “due to deficiencies of the program while human players are continually guilty of very simple and obvious blunders.” This extended to errors of the psyche: computers couldn’t suffer from a case of nerves or overconfidence, two deficits in human players that led to game-ending mistakes. A robot player could play emotionless, egoless chess: a clinical game in which each move was simply a new math problem.
But—and Shannon was emphatic about the “but”—“these must be balanced against the flexibility, imagination and inductive and learning capacities of the human mind.” The great downfall of a chess-playing machine, Shannon thought, was that it couldn’t learn on the fly, a capacity he believed was vital for victory at the elite levels. He cites Reuben Fine, an American chess master, on the misconceptions about top-ranked players and their approach to the game: “Very often people have the idea that masters foresee everything or nearly everything . . . that everything is mathematically calculated down to the smirk when the Queen’s Rook Pawn queens one move ahead of the opponent’s King’s Knight’s Pawn. All this is, of course, pure fantasy. The best course to follow is to note the major consequences for two moves, but try to work out forced variations as they go.”
In mastering the probabilities of each conceivable position, then, a chess computer would not simply be acting as a superpowered grandmaster, but as a fundamentally different kind of player. Essentially, human and computer would be playing two different games while seated across the same board.
So Shannon cautioned against programming computers to behave too much like human beings: “It is not being suggested that we should design the strategy in our own image. Rather it should be matched to the capacities and weaknesses of the computer. The computer is strong in speed and accuracy and weak in analytical ability and recognition.” Computers needed to be taken on their own merits and flaws, not as ersatz humans. What followed in the paper, and what Shannon would later popularize in a less technical article for Scientific American, was the range of strategies that could be programmed into a computer: a blueprint for turning a machine into a good, if not a great, player.
It is an admittedly broad survey: he studied each move’s possible outcomes, considered game-theoretic approaches, outlined how a machine might go about evaluating moves, and concluded that a computer could be programmed to play a perfect game of chess, but that such an outcome would be wildly impractical. This was, in a way, a limitation of the technology of the time: if a contemporary computer’s goal were to calculate all possible moves for itself and its opponent, it would not move its first pawn, Shannon calculated, for 1090 years.
Much as his information theory paper had done, Shannon’s chess paper acted as a road map for an emerging field. Shannon would live to see the fruits of these labors; he would purchase machine after chess-playing machine, leading his exasperated wife to remark that “Claude went hog wild.” But he took it one step further: Shannon’s answer to Maelzel, one might say, came in the form of a machine he built himself. Completed in 1949, the machine was referred to as both Endgame and Caissac (after the fictional “patron goddess of chess,” Caïssa). Shannon’s machine could only handle six pieces and focused on the final moves in a chess game. Over 150 relay switches were used to calculate a move, processing power that allowed the machine to decide within a respectable ten to fifteen seconds.
The machine has largely been absent from accounts of Shannon’s life. It is preserved at the MIT museum and in the memories of those closest to him. The box had the pattern of a chess board engraved on top of it; once the computer determined the correct move, a series of lights would indicate its choice to the user.
It was, by some accounts, the world’s first chess-playing computer. It is also, perhaps more importantly, another illustration of Shannon’s eagerness to build with his hands what he had dreamed up on paper.
For Shannon, both the chess paper and the chess machine addressed more enticingly ecumenical questions, as well. How should we think about “thinking machines”? Do machines think in the way we do? Do we want them to? What were an artificial brain’s strengths and weaknesses? Shannon gave a measured answer, one that surely reflected the fact that he himself hadn’t come to firm conclusions: “From a behavioristic point of view, the machine acts as though it were thinking. It has always been considered that skillful play requires the reasoning faculty. If we regard thinking as a property of external actions rather than internal method, the machine is surely thinking.”
Shannon would, over time, grow more positive that artificial brains would surpass organic brains. Decades would pass before programmers would build a grand-master-level chess computer on the foundations that Shannon helped lay, but he was certain that such an outcome was inevitable. The thought that a machine could never exceed its creator was “just foolish logic, wrong and incorrect logic.” He went on: “you can make a thing that is smarter than yourself. Smartness in this game is made partly of time and speed. I can build something which can operate much faster than my neurons.” There was nothing more mysterious to it:
I think man is a machine. No, I am not joking, I think man is a machine of a very complex sort, different from a computer, i.e., different in organization. But it could be easily reproduced—it has about ten billion nerve cells, i.e., 1010 neurons. And if you model each one of these with electronic equipment it will act like a human brain. If you take [Bobby] Fischer’s head and make a model of that, it would play like Fischer.