Intro

You know that chess program on your computer, the one you only use if you’re at the airport and you’ve played so many rounds of solitaire that the opening cards are repeating? There was a time when it seemed possible if not inevitable that chess programs less sophisticated than that were harbingers of the end of humanity. We all thought The Matrix would kick off once a machine beat the best chess player in the world. It didn’t quite happen that way, but that doesn’t mean that there was no lasting impact. The history of computer chess runs parallel to the recent history of the mind and this is the first installment of a three part series which will explore these parallel histories. In part 1 we’re going to go back in time: briefly back in to the late 1700s, then the mid 1900s, then we’ll settle into the early days of computer chess.

For a while, from about the mid 1970s to the end of the ‘90s, chess represented a major frontier in the race to develop artificial intelligence. This was partly due to rapid developments in processing speeds, programing sophistication, and the relative ease with which computers were able to duplicate and exceed the human brain’s capacity for rational thought. But there was also a much older idea at work, a lingering Cartesianism de-essentialized by European secularism, Freudian psychology, the germinal neurosciences, and posthumanism, among other things, but which we’ve never quite relinquished. If you take the ghost out of the machine, what’s left? The relationship between chess and thinking is very much an active one, and you need look no further than the latest Hollywood portrayal of Sherlock Holmes which culminates in a series of chess games, one literal and the others virtual. Look at the floor in the ballroom scene at the climax. Black and white marble squares, 8×8, a chessboard. In fact, both of Guy Ritchie’s depictions of Holmes deploy what could be called chess-thinking, which I’ll spend some time discussing later on.

But before we get into the history of chess computing, a short detour.

The Gambit

I’ll start with some weird stuff from the ‘60s. Nothing inspired by Dr. Leary – though he may be lurking behind this too – but two fictionalizations of Artificial Intelligence. These asides may not seem immediately intelligible, but I’ll come back to them occasionally over the course of these articles, use them as a keisaku – a crack over the head when things seem too familiar or intelligible.

John Barth prefaced his 1966 novel Giles Goat Boy with a bizarre recusal which, given the publisher’s disclaimer that appears just before it, he must have actually sent to Doubleday. Barth claims that he didn’t write the manuscript which he was submitting in lieu of a promised novel and that said manuscript was produced by a collaboration between a man raised as a goat, who dictated the text, and an artificially intelligent computer, who ‘read out and corrected the mistaken passages.’ This text is then delivered to Barth by one Stoker Giles, himself a bit goatish, not as any kind of memoir or personal history but as The Revised New Syllabus , which is the title under which Barth submitted the manuscript to his publishers. This all probably begs a lot more explanation, but for my purposes its only important that you’re thinking about the fact that the ensuing novel is the product of two non-human intelligences, one anthropomorphic and one artificial. But not really, because it’s pretty obvious that Barth did in fact write the novel and that this concealment is ironic. So, then, these two non-human forms of intelligence are still human, or the product of human intelligence. Imagined intelligences, but also perhaps aspects of human intelligence (itself as we imagine it) recapitulated in these non-human forms.

Now, let’s move from the page to the screen. The close of Stanley Kubrick’s 2001: A Space Odyssey . An iconic and nearly incomprehensible scene. Nearly. There is a lot of doubling in the movie, but maybe none more poignant than the point at which Dr. Bowman, having rapidly aged and watched himself doing so, is laying in what must be his death bed at the foot of which stands the black monolith. The long progression of evolution which launches the film culminates in the creation of the HAL 9000 computer is doubled in Bowman’s own aging. Even at this stage of development, the monolith represents an unknowable, maybe infinitely unknowable. But the monolith is also a symbolic inverted cinema screen and the incursion of its significance into the cinematic space as symbol of itself starts to unravel our hopes for representation – that same representation on which my claim about Bowman’s aging as a symbol for the other evolutionary progressions would depend. The screen reveals nothing, not just because its symbolic double reveals nothing. That would be too simple. It would mean that the symbolic representation of non-representability would actually be a functioning representation (and at the same time, not). It’s a kind of trap, a trap which is doubled again in the act of its own conception. And this is why the monolith and the screen are such appropriate doubles; both represent puzzles which are incomprehensible or which point out the limits of logic and so are doubled again in the re-presentation of human thinking, HAL. In other words, everything presented on the screen is a reflection presented as a projection (in this case, into the future). Cinema, as a technology, is a contingent re-presentation. And let’s not forget that writing is also a technology…

From The Turk to the Turing Machine

In 1769 or 1770 Wolfgang von Kempelen built and first exhibited his chess automaton, The Turk, in Vienna. The Turk consisted of a large cabinet, about the size of an executive desk, which housed the mechanical components that allowed the automaton to play chess and, on top of that, The Turk himself, looking a bit like Zoltar from Big . In 1809 Napoleon played The Turk, by then in the possession of Johann Mälzel. In the first game, he made an illegal move three times which caused The Turk to first replace the piece, then remove it from the board altogether, and finally to swipe its arm across the board. Tetchy machine. One assumes the temper was part of its design. Sufficiently amused, Napoleon played his second game fairly and resigned after 19 moves.

There were several attempts to debunk this chess playing machine over the years, but this didn’t seem to hinder its success much. It was exhibited all over Europe while in von Kempelen’s possession. Mälzel brought it to America and toured everywhere from Canada to Cuba before selling the machine to Edgar Allan Poe’s physician in 1840. Poe himself was among The Turk’s debunkers and published a paper entitled ‘Maelzel’s Chess Player’ in which he claimed that the machine had to be a hoax because it would have been much harder to design a machine that won only occasionally, as The Turk did, rather than one that won all of its games. Just about every theory about The Turk, including Poe’s, came to the right conclusion by the wrong path. A vital piece of technological development was still yet to occur, so detractors who knew that the machine with a hoax were forced to speculate. And many of the theories hinged on an assumption about the perfectibility of the machine and the imperfection of man.

It was eventually revealed that the furtive purpose of the cabinet was to conceal the man that operated the automaton. But the popularity of The Turk makes for a fascinating case study and its enigma would eventually find a double in the machine that renewed the popularity of chess machines and which would participate in what was probably the most important chess match between man and machine. The Turk was destroyed in a fire in 1854. Nearly 150 years would pass before its natural successor appeared.

From this historical vantage point, it’s clear that von Kempelen’s invention preceded the scientific concept that would have made its legitimate functioning possible. Though mechanical computing devices predate The Turk, its most sophisticated contemporary was Charles Babbage’s ‘difference engine’. Babbage first publicly described his machine in 1822, over 40 years after The Turk was originally exhibited. He was never able to build his difference engine, and even if he had it would have been a long way from anything capable of playing chess. It wasn’t until Alan Turing came up with a theory of computation that mechanical chess became a real possibility. A Spanish engineer, Torres y Quevedo, successfully designed and built a chess playing device in 1914. But this device was only capable of playing king and rook versus king endgames. Ultimately it wasn’t until Turing’s contribution that a more complex chess playing machine could be designed.

Turing thought that certain human processes could be mechanized and conceived of a theoretical machine, what became known as the ‘Turing Machine’, that was programmable. The program – sets of symbols representing discrete segments of an operation (the theoretical basis of programing language as we now know it) – would determine the operations the machine could perform. As an abstract concept, and with infinite storage capacity, the Turing Machine would be limited only by the imagination of its human programmer. It’s not hard to recognize in the Turing Machine a theoretical version of the modern computer. And consequent with the theory of the computer is a theory of human as computer.

The Birth of Computer Chess  

Turing’s idea of a stored program is what made computer chess possible, and before his death in 1954 he published a paper theorizing the use of a stored program specifically for chess. But perhaps the most important early contribution to the theory of chess programing came from Claude Shannon. It was Shannon that had earlier made Turing’s idea a reality by quantifying information in a mathematical theory of information; representing it with binary symbols that now allow anything from simple text to high resolution video to be stored as a series of 1s and 0s. When Shannon turned his attention to the problem of computerized chess, his theories established the basis on which many of the early programs which began appearing in the ‘60s.

Shannon’s paper – simply titled ‘Programming a Computer for Playing Chess’ and well worth a read even 60 years after its initial publication – lays out the major factors for teaching a computer to play chess. One is the undecidability of categorization of a particular position. There are three possible categorizations: 1. white has won the position, 2. black has won the position, 3. the position is a draw. So, in practical terms, this means that if white makes an opening move – say pawn to e4 – there is no way to tell for sure which category this position belongs to. This is important because if it weren’t true it would mean that chess would be solvable in such a way as to make the end result of any game calculable from the opening move. Even knowing this, it is still possible (though impractical) to conceptualize computing a perfect game of chess because chess is a system of perfect knowledge – in other words, all the data about the system is knowable. However, the possible configurations are vast. Going by the conservative assumption that the average chess game lasts about 40 moves (the typical point of resignation at high level play), Shannon makes the following striking statement:

However, even at this figure there will be 10 ¹²⁰ variations to be calculated from the initial position. A machine operating at the rate of one variation per micro-second would require over 10 ⁹⁰ years to calculate the first move!

The trick, obviously, is to give the computer a way to bypass the necessity of calculating all the possible board configurations in order to decide on a move. Just to give you an idea of how vast the number of configurations is, the estimated number of atoms in the known universe is 10 ⁸⁰ . So there are significantly fewer atoms than possible chess positions. The trick, then, is to teach a computer to strategize and to focus only on relevant configurations.

The principle things that a chess program must consider are the relative value of pieces, strategy (particularly dependent on game stage), and evaluation, the latter of which can either be accomplished by positional knowledge or by tree search. Piece value helps develop strategy. For example, say you’ve forked a pawn and a bishop. It would generally be better to take the bishop because of its greater value. This isn’t always the case, though, so strategy has to be determined by multiple factors. The stage of the game is important here because the objectives in chess change throughout the game. Tree search is a particularly problematic aspect of early computer chess because of the restrictions of processing speeds. The tree to which ‘tree search’ refers is an expanding tree of possible moves starting from any one position. This tree extends with exponentially increasing possibilities with each projected move forward. So imagine that every move has three possible outcomes and the corresponding moves. This would mean that the second projected move would have nine possible outcomes and corresponding moves, the third 27, and so on. Calculating a few moves forward isn’t terribly hard, but imagine calculating 10, 15, or 20 moves ahead.

The first interest in computer chess from the artificial intelligence community came as early as 1968. MIT had developed a program called Mack Hack  which was the first program capable of outplaying the average human chess player. The idea was that from experimentations with chess, researchers might discover how to imbue a computer with any human creative ability. Almost immediately two divergent schools of thought formed around performance and competence as the respective speculative foci, and the field has split many times since. Mack Hack Six , the most advance iteration of MIT’s program, was successful because it had an advanced ability to store positions and so could recognize them if they repeated. This was a significant development because the early advantage of a human over a computer was that no human player would ever have to consider the range of possible configurations that a computer did. There are, let’s say, a much smaller number of probable games, and many positions which are easily recognizable to anyone who studies chess. Many researchers believe the human brain’s ability to recognize patterns is another distinct advantage.

The first World Computer Chess Championship was held in 1974 and  Kassia ² , developed by Moscow’s Institute for Theoretical and Experimental Physics, emerged as the winner. It would be a bit of an exaggeration to say that part of the cold war was being staged on a digital chess board, but the weight of the institutions backing program development in the U.S. and the Soviet Union gives some idea of the anticipated payoff. This became even more evident in the 90s when interest reached a fever pitch. But with this steadily increasing interest came a hermetic approach to development. In 1975 the program Chess 4.7 became the first to break the expert level. By then programs were already significantly more sophisticated in terms of employing chess theory, concepts, and were gradually improving at closed game positional play (which always seemed to give chess programs problems), but the rising interest and competitiveness in these programs led their creators to be much more protective about how the programs worked.

Mind Games

The possible implications of successfully developing a chess program were already evident to Claude Shannon, though not to the extent that they would later be (in all likelihood over-) estimated. By the 1980s computer chess had become the Drosophilia of artificial intelligence. A $100,000 dollar prize was offered to anyone who could develop a program capable of beating a world champion human chess player. Computer chess was also part of the broader development of game theory, but maintained a privileged role because of its particular challenge to computation and because it was possible to demonstrate a machine’s proficiency at the game in direct competition with man’s.

The link back from all this activity to a more general way of thinking about thinking comes from an earlier contribution by Claude Shannon. Mathematical information theory didn’t just make computation possible, it signaled a significant change in the material of thought. Quantifying information separates it from any meaning; information theory is only concerned with the transmission of information, not interpretation. What differentiates one data set from another, prior to their decoding, is just quantity and therefore the amount of space needed to store the message. This simple principle of information storage and retrieval gave rise to a whole new idiom for the philosophy of mind. Gilbert Ryle’s The Concept of Mind appeared a year after Shannon published his work on information theory and may have bore some marks of Shannon’s theories, but more so those of Wittgenstein and the misuse of language. More importantly, Ryle’s book eviscerated Cartesian mind-body dualism, what he called Descartes’ ‘Ghost in the Machine’. That Shannon’s digitalization of information and Ryle’s recombination of the mind with the physical influences of the body appeared at about the same time gave subsequent thinkers access to a completely new paradigm of the mind and the technical processes of the mind began to take on a patently computational lexicon. Chess quickly became a preferred analogy between the human and computer mind.

Triumph or terror?

In the early fervor of computer chess development there were mixed sentiments about the possibility of creating a program that could beat the best human chess players. By the mid-80s speculation ran from a few years to never. But the fact remained that this represented a possible realm of computer superiority in a time when machine paranoia was in full force, which somewhat ironically constituted a significant source of inspiration in American arts and letters.

Near the close of David Mamet’s 1984 play Glengarry Glen Ross , Richard Roma, conning his own coworker who unbeknownst to him is about to get busted for breaking into the office and stealing the sales leads, delivers an uncanny line:

I’ve wanted to talk to you for some time. For a long time, actually. I said, “The Machine, there’s a man I would work with. There’s a man…” You know? I never said a thing. I should have, don’t know why I didn’t.

“The Machine” is an immediate reference to the character Shelley Levene, but also an implicit reference to an absence in the play. With the exception of the telephone (the original sin of technological dependency?), there are no machines in Glengarry Glen Ross . At least none that would be easily recognizable as machines. The play was written at the height of U.S.-Soviet antagonism following the 1979 Soviet invasion of Afghanistan; the US had put a missile base in West Germany; NATO was simulating nuclear attacks. But the machines of war, the same that were responsible for the kind of Western pessimism that pervades the play, are absent. There is another absent machine which I’ll get to shortly, but there is another, more pressing absence in my own argument which probably seems obvious if you’re familiar at all with the play or the movie. The extortionary tactics of the salesmen are mandated by the absurd demands of the job which most of the characters seem locked into despite moral objections. This is the machine of late capitalism for which Mitch and Murray’s scamming real estate agency and Shelley Levene are synecdochic. Such a mechanistic view of the play makes the entropic conclusion all the more cynical. When Shelley “The Machine” Levene turns out to be the unlikely burglar, his mechanical cannibalism is only a minor setback. The parts, the phones and the leads, are replaced and the machine’s functioning is barely impeded.

Andrew Bujalski’s recent film Computer Chess is an interesting take on the weird world of chess programing in the 1980s. The idea that a bunch of socially awkward grad students were being scooped up by big corporations and courted by the Pentagon as a result of these computer versus computer tournaments in which glorified fax machines played a very awkward brand of chess is perplexing today. But at the time the implications were, on the one hand that the human brain was being reduced to the status of glorified fax machine, and on the other that machines were competing with humans in serious ways. Chess was a symbolic field for that battle.

The next article in this series will look at the most important match in the history of computer chess and the events that led up to and away from it.

Cray Blitz is a former chess champion from Hattiesburg, Mississippi.