The supposed development of Artificial Intelligence during the Cold War was really about an expression of a voracious and destructive raison d’etat that left hundreds of thousands of victims across the planet. This is the story of how chess was destroyed by the shock wave of Hiroshima and how Go, thanks to its anti-essentialist logic more than its combinatorial logic, remains untouched until today.
This post is dedicated to my existentialist friend, Alan Furth.
We remember the Cold War as a time of polarization, but it was worse than that. It was the golden age of the raison d’etat. If I had to summarize its moral legacy, its message, it obviously wouldn’t be the value of freedom — remember McCarthy or the dictatorships supported by the United States – nor the unity of Europe against the totalitarian threat (Salazar, Franco or the Greek colonels had agreements of association with what was then the European Economic Community). No, the moral legacy of the Cold War, without a doubt, was “anything goes,” “the means don’t matter, only winning.” This left hundreds of thousands of dead bodies behind across the planet, and a good deal of cultural changes. The death of chess as a challenge and mystery was not the least of them.
From Torres Quevedo to Los Alamos
Even though all histories of chess played by machines begin with the famous imposture of “The Turk,” the prehistory of computational chess doesn’t begin until 1912. In that year, Torres Quevedo presented a true automaton that played exclusively by electromechanical procedures. The machine had the rules programmed in, could recognize when traps were set for it, and tried to checkmate the king and rook, whatever the game of the adversary was. Actually, it was little more than entertainment, a demonstration of power of automation and the very first electronics, as understood by the genius of Molledo.
In the middle of the Second World War, the first to think of a chess program as a software exercise was Alan Turing. His main problem was that he didn’t yet have a machine capable of running his program, so the first test was carried out with Turing himself doing the program’s calculations on a piece of paper. The truth is that Turing preferred Go to chess, and the program, which tried to evaluate the positions on board after seeing only one play, using the minimax theorem, reflected, according to critics, his poor conception of the game.
Meanwhile, in Los Alamos, the team that was working on the atomic bomb developed their own approximation, improving Turing’s focus: positionally evaluating every one of the possible movements and choosing the most strategically valuable. In this case they reduced the board to 6×6 and eliminated bishops. The result was stunning at the time: the program “beat” advanced players.
From reason to force
So far, the first computerized chess had followed the same logic: make algorithms that evaluated all possible positions after one play and compared them, recommending the most valuable. But it was like trying to reach the moon by climbing a tree — at first there’s rapid progress, but the path quickly disappears.
It would be another illustrious figure, the creator of the theory of information as a branch of mathmatics, Claude Shannon, who would figure it out. Shannon had worked with Turing and knew the Los Alamos team. He did not share the media fantasies of an “intelligent machine,” because he knew perfectly well that you can’t program what hasn’t been previously described as a process, which was very far from the current state of the discussion on intelligence and its nature. He defined two possible strategies this way: pure brute force and brute force based on heuristics.
The path of the pure brute force consisted of simply accumulating catalogs of board situations from real games that the computer read prior to beginning the game, the famous “Opening Books.” In each situation, understood in a more or less broad sense, a value was assigned on the basis of the real result of the game and the following movement was considered preferred or not on the basis of the results.
That was the path professional Soviet chess was following at the time, using a gigantic system of paper punchcards. Although, as the current world champion tells us, today that whole vast database fits on a CD that costs 150€, at the time it was kept in a special building and was considered a state secret.
Chess has been the Western metaphor for intelligence since Alfonso X, who defined it, in contrast to the others games of his court, as the only one “that depends on brains alone.” This association, as seen in the movies, was transformed, in the context of the Cold War, into a true “chess race” to discern the intellectual superiority of the socialist “new man” versus the American “homo economicus.” The war-propaganda machines’ exaggerations would culminate in the famous game between Fisher and Spassky in 1972.
So, in the Sixties, the world of chess was much more than a tournament. It was yet another battle line of the war, and as such, got military funds. And here is where Shannon’s heuristics strategy begins to make sense. Shannon’s proposal consisted of not using real games, but rather, incorporating programs that generated massive “plausible results” and then analyzing the decision trees that were generated.
The leap is important for the open path in Los Alamos, because it would end up basing the chess played by computers on a kind of calculation that, together with the bomb, is the most enduring legacy of the Manhattan Project: the Monte Carlo Method. In 1970, NASA presented its own chess software and organized the first “computer chess” championship in the US. In 1974, the first world championship was organized in Stockholm, which the Soviets won. More funds for this byproduct of the arms race. The famous Moore’s law, according to which computers double their power every two years, starts to take off. The idea of brute force takes on a new meaning: pure calculation capacity and processing speed. Everything is on the path towards Deep Blue (1997) and Deep Fritz (2002).
The last victim of Hiroshima
They arrived late. By the time computers were able to defeat the best players in what had been the USSR, it no longer existed. But nor did chess exist. The players were taught to play by imitating the evaluation system of the computers. The new champions of the world were no longer considered beings of vision and exceptional intelligence, but rather people with prodigious memories. And chess declined rapidly throughout the world.
What is being “revived” today is really a new game of centaurs: half computer, half human. The chess that is characteristic of our time is the so-called “freestyle,” in which program and player form a pair. But regular chess, or what seems like it, is only differentiat in that humans must try to remember what his/her machine tutor recommended in a similar situation.
Chess, as it was known from Alfonso X to Kasparov, is dead. It was severely wounded in 1997 and died definitively in 2002. It was the last victim of the shock wave of Hiroshima.
Go against the bomb
The contemporary history of Go could be told, even literally, as the resistance to the logic that the bomb imposed on the world, and which, in good measure, we continue to live. It’s not just about the fact that there are 1.5×10768 positions (10110 more than chess); the combinations in Go demand much more processing capacity. It is the very essence of the game.
In 1989, Ing Chan-ki, the banker that challenged the world of Go, established the “Ing Prize,” a social challenge of $1,400,000 for the first program that defeated a professional Taiwanese Go player. Although smaller quantities were given out for the best programs of the times, in 2000, the challenge was retired, being considered useless.
The challenge had mobilized resources and programmers, above all, in the West. And it continued to do so after being declared cancelled. Victory in Go on the basis of brute force is a challenge in itself, but it has intrinsic problems. In 2006, in an interview with Wired, Remi Coulom, who was trying to apply Monte Carlo, commented on how the random results that the method provides are extremely difficult to evaluate, and how their program was then the most advanced, not by incorporating, but by throwing out a large number of them.
In 2009, the same magazine published a report on the first victory of a program based on Montecarlo over a professional player. Nobody seemed too happy. The magazine itself insisted that the person was a professional, but of a lower level, and made it clear that in any case, given the complexity of the game, it would not be like in chess: the players would not become memorizers of “solved games.” And in the words of the creator of Deep Blue, informatization of chess had come to “substitute [human] judgment with a search [in a database].”
So no one seemed to pay much attention to the event that had earned the Ing award: it wasn’t going to change the way of playing or the abilities demanded of players, and above all, it wasn’t going to teach us anything. As Bob Hearn, an artificial intelligence programmer at Dartmouth College, commented in the same report:
People hoped that if we had a strong Go program, it would teach us how our minds work. But that’s not the case. We just threw brute force at a program we thought required intellect.
One kind of reason that defeats force
Today, little remains of all that effort, of that last breath of the atomic bomb and the Cold War. Things like an XBox game or an Android app that, for 3.80€, let you connect as many times as you want to a supercomputer and play with a program equivalent to a 6d, a high-level amateur. It’s not, by any means, among the most sold or downloaded applications. Go players continue to prefer playing on online servers with other people.
But a lot can be learned from this story. The step from pure brute force to brute force accelerated through heuristics in chess seems like the perfect metaphor for going from the military theory of overwhelming superiority, created in the years of the Vietnam War, to the theory of overwhelming technological superiority that sank in the Gulf War. In both crises, part of the US strategists have looked at Go, intuiting a way of thinking in the adversary that eluded them, and returning over and over to the comparison between chess (metaphor for the West) and Go (the East).
Because chess and Go respond to completely different logics. Chess is the child of a caste society, India, in which each piece has a well defined role and a constant value, which is — barring extraordinary events — easily computable and invariable. To calculate the value of positions on the board is relatively simple. In Go, in contrast, all the pieces are, in principle, “equals,” that is, like in life itself, the value of each one is relational, does not depend on its origin, or on what they were prior to to beginning the game, but on their relationship to other pieces at every moment and to the general situation on the board.
Essentialism — emphasized by monotheist readings of the Greco-Latin classics — has defined a good part of Western thought. It values the why and the origins of things more than action, adopting a linear conception of time — another very “chess” thing — in which, as we move farther from the origin, subjects degrade, always seeking after a final redeeming victory (national plentitude, the classless society, paradise, etc.), in which they will recover their original purity.
In contrast, according to the authors, Go is the “playful expression” of a praxeology: a kind of non-essentialist and non-linear thought, oriented towards action, and which privileges “what can we do?” over “how did we get here?”. Existentialism told through a game, something similar to what Stephenson wanted to describe with his«Young Lady’s Illustrated Primer».
Indeed, all Go players know that to think linearly during a game turns out to be simply illusory, that the nature of a stone or a group changes throughout the game, and that with them, the board can “turn” even without mediating conflict, which is essential in chess but almost dysfunctional in Go. The issue is that this non-linear “building together” that the authors remark is characteristic of the game of Go, can learn very little from a program that works by “replacing judgment with search,” selecting movements in a database with a conception of game time that is necessarily linear, geared only towards the achievement of a final result.
That’s why Go has been relatively unaffected by pressure from programs based on Montecarlo. In contrast to chess, professional players do not study decision trees with a computer at their sides, nor does there exist a “freestyle” in which the human player selects between the branches suggested by a program. There must be something in this praxeology that, for once, enables human reason to overcome brute force.
Translation by Steve Herrick.