Digital Historiography: Kasparov vs. Deep Blue

According to many observers, the chess match between Garry Kasparov and the IBM computer Deep Blue in May 1997 was an "historic event." If true, then this match would seem to be the sort of event that would be of special concern to historians interested in the marriage of computers and history. For historians of computing, Deep Blue's defeat of Kasparov—arguably the greatest chess player ever—has the potential to be like Gettysburg or the storming of the Bastille. Like these events, Kasparov vs. Deep Blue is a significant moment in time that encapsulates a much larger historical reality, the implications of which historians may well wrestle with again and again. Each new generation of historians always returns to notable events seeking new insights, new meanings, and fresh interpretations. Similarly, since Kasparov vs. Deep Blue is an historic event, then succeeding generations will more than likely revisit it. Like any historic event, Kasparov vs. Deep Blue is also an historiographic event.

Some historiographic boundaries of this historic event are already forming, and I imagine these will continue to shape interpretations into the foreseeable future. At the center of these debates are several important questions. What is the significance of Kasparov losing to a computer? Does this signal the beginning of the end of human intellectual superiority, thus consigning humans to a subservient role behind computers? Or is the victory of the computer as inconsequential as an automobile outracing a human runner or a bulldozer moving more earth than a human shoveler? That is, have we simply created a technology that outperforms humans but is ultimately designed and controlled by humans? Is chess itself a legitimate gauge to measure the intellectual capacity of computers (or humans), or are there more complex activities that would serve as a better litmus test? To answer these questions one must determine if Deep Blue was consciously playing chess or merely carrying out commands made by human programmers.

IBM's victory did not come out of the blue: designing a program to defeat a human world champion has been an explicit goal of computer science and artificial intelligence research since the 1950's. This history is suggestively sketched by Monty Newborn, a leading figure in computer chess. Claude Shannon, the father of information theory, first proposed the idea of programming a computer to play chess in 1950. In general, Shannon's ideas worked like this: the program would take a given board position, then calculate out the possible moves that might follow from that position. A picture of these multiple possibilities resembles the branches of a tree, hence they are called "search trees." The program would have to include a final level or cut-off point beyond which the computer would stop calculating; to calculate all possible variations would require too much time and computing power. One solution would be to construct computers with increasing computational power. Another solution involves writing programs in such a way as to "prune" the number of variations. The program would then assign a "score" to each position as a way to gauge its effectiveness as a move. The program would weigh such factors as material, pawn structure and the mobility of the pieces, just as a human chess player might. While the specifics of Shannon's theory have been surpassed by subsequent generations of programmers, the basic idea behind computer chess programs is essentially the same.

Shannon's theory was entirely speculative; he wrote no actual program. Alan Turing, another pioneer in computer research, created an actual chess playing program, albeit by hand. Turing's program was based on similar principles to Shannon's theory, in that the program assigned scores to various positions based on certain criteria, such as mobility, piece safety, and king mobility. All chess players evaluate these and other similar considerations; the difference is that the programmer must assign a numerical value to what are often very subjective, intuitive qualities. Turing's program was tested in 1951 against a weak human opponent. The program—playing White—resigned after 29 moves.

Interestingly, Turing noted that the program was weak because "It was in fact based on an introspective analysis of my thought processes when playing"(28) which proves to be significant in understanding the history of computer chess. From this early date it is clear that the intention of the programmer is to model the human thought process. Indeed, this explains why computer scientists are so serious about chess-playing programs. If a computer can be programed to play great chess—any program can merely play chess—then perhaps a computer could be programed to carry out other complex cognitive tasks. And the only way to determine how well a program is playing chess is to pit it against the best human competition.

What would anyone prove by creating such a program? As Newborn observes, computer chess is analogous to automobile racing. "Many advances in the auto world," he notes, "were first tried on racing models and then after refinement incorporated into commercial vehicles. This may be the pattern in the computer field, too, where techniques used by computers to play chess are on the cutting edge of developments in complex problem-solving." (279) Chess represents a domain of human skill that is simple enough to model yet complex enough to reflect deep levels of cognition. If a computer can play good chess, then perhaps it can be programed to predict the weather or sort through scheduling problems, or even write poetry or music.

The rest of Newborn's book is a chronicle of computer chess from these early beginnings up to the first official Kasparov-Deep Blue match in 1996, which Kasparov won 4-2. This chronicle includes results from laboratories and universities in the United States, the former Soviet Union, the Association for Computer Machinery Championships, the World Computer Chess Championships and various exhibitions. While much of the book is reportage—"this program defeated that program," or "this program defeated that human player"—there is a detailed record of the moves for each match with some limited commentary. Chess enthusiasts will find these sections most interesting, but those lacking an understanding of chess can easily pass them over.

Likewise, the sections dealing with the technical issues of programming and computer hardware can be left to programmers. Yet within these sections, Newborn notes some of the theoretical problems chess programming opens up. For example, human players do more than simply calculate positions and assign numerical scores; human players have the ability to use the "method of analogy."(43) This refers to the ability of a human player to see similarities in two board positions without having to go through the redundancy of searching both. Such analogous thinking is important to real human thinking, but as Newborn shows has proven very difficult to program without a great deal of computation. This suggests the difficulties in modelling and programming human thinking. Many computer scientists, in fact, downplay Deep Blue's victory over Kasparov, for they understand the limitations of computers and recognize all the things computers still cannot do.

Overcoming the technical and programming hurdles is an important plotline of Newborn's book, which is clearly about the programmers. Only in a very few cases does he use language that would ascribe to the computer any consciousness or independence of thought. It is clear that Newborn does not believe the computer is capable of independent thought. Thus, when Kasparov plays Deep Blue in 1996, he is actually playing a group of IBM technicians, programmers and their program, not a sentient electronic being. Newborn writes that the IBM team "didn't expect their computer to defeat Kasparov," (174) but pressed on anyway. Feng-Hsiung Hsu, for example, works long hours and sleeps on the floor at IBM preparing the computer for the match, like a heroic figure fighting all odds. While Kasparov won that first encounter, Newborn writes of the programmers as heros returning from an epic battle. Newborn's book was written prior to the May 1997 match; one can only speculate as to how the IBM team would be lionized had this book been written after IBM's victory.

In the mass media, that battle was described as "man vs. machine," but Newborn sensibly notes that the real challengers have always been the computer scientists. The "machine" is the hardware (constructed by humans) that executes programming commands (authored by humans). Nevertheless, these programmers seem to misanthropically pit themselves against the rest of humanity. Newborn describes those pulling for Kasparov as "all of mankind's supporters," (173) implying that Deep Blue's creators are somehow "against" mankind. Newborn concludes his book by pronouncing that "the day is not too far off when the best players will no longer be serious competition. Computers will simply consider too many possibilities and set up positions that are too complex for mere mortals to cope with." (280) I assume "mere mortals" here means people who cannot program, suggesting that programmers look at the whole of humanity with disdain (or perhaps it is self-loathing?). After such a time, virtuosity in chess will no longer involve artfully moving pieces on the board. "Perhaps," speculates Newborn, "eventually the game will become programming chess." (281)

Bruce Pandolfini, himself a national chess master, does not see Kasparov playing a team of programmers. He has no doubt that Deep Blue was "playing chess," and seems both fascinated and appalled by the prospect. At the beginning of this very interesting book, Pandolfini describes Deep Blue as "A supercomputer, developed and fine-tuned by ingenious programmers, technicians and grandmasters, with the megabacking of a giant, multinational corporation," (7-8) at least acknowledging that Deep Blue is not a sentient being. Yet the commentary in the rest of the book treats Deep Blue as if it were sentient, and the programmers, technicians and grandmasters fade into the background. The result is a book which amplifies the popular "man vs. machine" motif.

Pandolfini's book is a careful commentary on each of the six matches played in 1997. Chess books usually include detailed analysis of matches, yet Pandolfini's commentary is very accessible to non-chess players. He carefully points out nuances of the game lost on the uninitiated. The effect is to demonstrate why Deep Blue won the match by pointing to the tactical and strategic choices the computer made.

The key word there is "choices," for Pandolfini seems convinced that Deep Blue was making decisions and choices. Throughout the book, he anthropomorphizes Deep Blue. Deep Blue has "the proclivity to retain Bishops," (in fact, "loves Bishops") will never "surrender material lightly," "wants to insure" a sound position, "has demonstrated strategic understanding" and always "decides" at critical moments. When Deep Blue plays a particularly curious move, Pandolfini remarks that "There must be some deep—very deep—tactical reason for this retreat, hidden in the recesses of Deep Blue's scheming mind." (58) Clearly, Pandolfini does not care for Deep Blue, referring to it at several points as the "metal mind," "a chunk of metal," and "a Machiavellian machine." He likens Deep Blue to Frankenstein, which is interesting, for that fictional character was an artificial creation of a human. The creation later develops sentience, which terrifies the creator. These statements say much about how Pandolfini understands the significance of this match.

It might be easy to dismiss these descriptions as poorly chosen words; the effect, however, is to treat Deep Blue as if it were capable of independent thought. I do not believe Pandolfini actually believes this; sprinkled throughout the commentary are words and phrases which remind the reader that Deep Blue is in fact programed by humans. Pandolfini constructs these phrases in the passive voice, however, thus removing the human subject from these sentences, and thereby making it seem, at least in the eyes of this National Master, as if the computer has passed the Turing's test.

Pandolfini becomes quite grave in describing the fateful game six. "It will undoubtedly go down as one of the most famous games in the annals of chess," he writes, not only because Kasparov lost "but also for its import in the history of civilization." (144) Kasparov played particularly poorly in that last game; in fact, most chess experts understand that Kasparov lost the series because he did not play his usual game. Rather than using the tactics and strategies that vaulted him to the level of World Champion, Kasparov adopted an "anti-computer" strategy, meaning he played deliberately obscure, odd moves and positions in an effort to jar the computer out of its program. The tactic failed, and Pandolfini laments that Kasparov is "the thirteenth and last human world champion." (149) As Kasparov's position deteriorates very quickly, Pandolfini observes that "What's at stake is the coordination of [Kasparov's] forces, the safety of [his] King, and, perhaps, the dignity and supremacy of humankind." When Deep Blue plays the coup de grace, "The future seems bleak." (159)

For Michael Khodarkovsky and Leonid Shamkovich, Kasparov was the victim of the invisible programmers, technicians and grandmasters hidden in Pandolfini's account. The authors were members of Kasparov's team, thus they offer a "behind the scenes" look at Kasparov's preparations for both of the Deep Blue matches, as well as an explanation for Kasparov's defeat. This book is valuable in aiding our understanding of this historic event, for it provides a detailed description of Kasparov's psychology and impressions, the sorts of thoughts missing from the other accounts, which seem to focus on Deep Blue's "psychology."

Kasparov did not write this account—he does write the forward, however, and includes his own chess-notational analysis of the first five games—and as such does not directly reflect Kasparov's thoughts. These are, instead, filtered through Khodarkovsky and Shamkovich. The book contains several long quotes attributed to Kasparov and snippets of dialog between the grandmaster and his advisors. We are never quite certain, however, whether these quotations are Kasparov's own words or the recollections of the authors. Reading this book is like reading Thucydides' account of the Peloponnesian War: one is never certain whether one is reading the actual speech or the gist of that speech as filtered through the authors.

Thus, the book should be read with caution, for it sounds more like a hagiography of St. Garry than an objective study of the event. Consider this description of Kasparov's arrival for Game One:

In move 30 of that first game, after Kasparov makes a particularly good move, the authors exclaim "This was a heroic counter by Kasparov!" (189) A good portion of this book reads like Soviet-style hero worship.

Kasparov's forward contains very little hint as to why he lost the match (Deep Blue certainly didn't defeat him). Instead, the prologue contains a detailed description of Kasparov's activities on behalf of the game of chess. Kasparov writes of his run-ins with chess officials of FIDE, the governing body of the sport, mostly dealing with sponsorship contracts and big money prizes intended to attract the best players. Kasparov comes across as a shrewd businessman in these opening pages. He wants to make chess a "major league" sport with multi-million dollar prizes, a strategy that runs him afoul of FIDE. The IBM matches should be viewed against this backdrop; the implication is that Kasparov agreed to the match for the sponsorship deals he could strike with IBM.

In recounting the matches, however, the authors emphasize Kasparov the explorer, the scientist and the popularizer of chess more than Kasparov the businessman. Kasparov plays the computer as an experiment; the matches are "research into human abilities," (2) for he wishes "to be one of the men who contributed to the progress of artificial intelligence." (180) "Until the machine's been tested against humans," says Kasparov, "we don't know at what point it will have 'graduated' to a new understanding of chess." (177) Kasparov himself uses computers in his preparations as something like a sparring partner, or as an elaborate calculator that can calculate several variations. Computers are useful tools, and playing against this elaborate tool teaches us about the power of that tool, which ultimately serves our needs. Computers and humans will inevitably work closer and closer together, believes Kasparov, and any experiment that explores that relationship is ultimately beneficial. Thus, Kasparov willingly served as a guinea pig in a scientific experiment; the implication is that the matches with IBM were not real chess competitions of the type in which Kasparov has been so wildly successful. At no time does Kasparov or any of his team admit that they are playing a real chess match against a sentient opponent.

Given these caveats, then, Deep Blue's victory seems less significant. Furthermore, Kasparov's team asserts that IBM's victory was the result of human duplicity and "psychological warfare" more than computer skill at chess. Before the rematch, Kasparov had requested the results of Deep Blue's recent matches against human players. It is standard practice for chess players to study their opponent's matches in order to discern patterns in their play that might prove of value. IBM never provided these results. IBM did not provide a room or other such facilities for Kasparov's team, and when accommodations were finally made, only two seats were secured, not nearly enough for Kasparov's advisors. Kasparov was provided a rest area between moves—standard in such matches—that was a minute and a half from the playing area. Since Deep Blue's average move time was about 3 minutes, Kasparov would not have been able to take a break or have a refreshment (the rest area was later moved closer). These might seem like picayune details, but Kasparov has long been noted for his fastidiousness about playing conditions (he once halted a match because he did not like the type of pieces being used). "As every professional chess player knows," observes Khodarkovsky, "you can defeat an opponent even before you sit down at the board if you can destroy his emotional equilibrium." (185) IBM's strategy for the second match, it seems, was to destroy Kasparov's emotional equilibrium.

Kasparov won game one of the rematch. He lost the second game, and was suspicious about the way the computer played the game. Something was fundamentally different about its tactics, and Kasparov wanted to study the explanatory printouts of Deep Blue's moves. While IBM promised these after the conclusion of the match, they never arrived. In fact, after repeated efforts to secure the printouts, IBM never released the information, much to the consternation of the Kasparov team. In this context, the manager of the Deep Blue project Chung-Jen (C.J.) Tan is portrayed as the chief villain, the man who refuses Kasparov his repeated requests. A good deal of commentary on the rest of the match is given over to this "war of the printouts."

Conspiracy run throughout the rest of the book. Kasparov's team learned that IBM had hired grandmasters Nick De Firmian and John Fedorowicz, even though IBM kept their involvement a secret. "This additional information," observes Khodarkovsky, "added fuel to all our suspicions that, unknowingly, we were facing the efforts of an untold number of grandmasters. Why was their participation kept secret? Was it also possible that this news had been leaked out at an opportune time to cause Garry more emotional turmoil?" (204) Kasparov's mother observed that IBM's treatment of her son reminded her "of 1984 when Garry played the world championship match against [Kremlin favorite Anatoly] Karpov. This led to a fight not only against Karpov but against the well-organized Soviet bureaucratic system. Now, thirteen years later, Garry is fighting not only against a supercomputer but against a well-organized capitalist system employing psychological warfare." (212) Kasparov is described as increasingly moody, temperamental and dispondant, very much the genius afflicted with melancholy. IBM's duplicitous maneuvering robs Kasparov of his ability to concentrate. "Although everyone thought that they were watching a contest of man against machine," Khodarkovsky somberly observes, "what they were really seeing was a very tired man pitted against relentless electronic calculation aided by psychological warfare—it was no contest." (228)

Kasparov has developed a reputation as a master of psychology at the chess board. He will often stare at opponents and move pieces with exaggerated confidence, all in an effort to intimidate his opponents. This psychological dimension to chess is never noted in the official analysis of a match, but is as important to a championship-caliber chess player as a straight face is to a poker player. Thus, if Kasparov lost because he was psychologically battered, it is difficult to feel sorry for him. At the same time, if psychology was the determining factor, then Deep Blue's victory seems hollow, or at least not as meaningful as someone like Pandolfini might believe. If IBM manipulated the situation, then Deep Blue did not win because it outperformed Kasparov. The victory, then, was not a programming triumph, and Deep Blue is exactly as Kasparov sees it: a tool which serves human ends.

Moreover, it suggests that people defeated Kasparov, not an inanimate machine. Human beings—serving as programmers, technicians and grandmasters—teamed together, the machine simply a tool toward their ends, an observation supported by Newborn's account. In the latest edition of the venerable Modern Chess Openings (New York: David McKay Co., 1999), editor Nick de Firmian notes that

De Firmian all but admits that he defeated Kasparov, since he designed the critical opening strategy that gave the computer its advantage. If true, then Kasparov was fighting the wrong opponent.

This event—a computer defeating a human world champion—has been a long time in coming, ever since Shannon first proposed computer chess. For some, this is a disaster, the beginning of the end of human intellectual superiority. Others see this event as evidence of superior human programming and engineering ability. Some computer scientists claim that other human domains are next; researchers at Carnegie Mellon, for example, assert that they hope to field a robotic soccer team that can defeat the human world champion by the middle of this century. Others might claim that this is a non-event and a ruse since chess is far too limited a domain of human activity to claim it as evidence of artificial intelligence. Players of go—the ancient Chinese strategy game—note that no computer has yet been developed that can defeat even the best amateur players, the game being far more complex than chess and thus less amenable to algorithms and brute-force calculation.

Others might point to the fact that while Deep Blue possesses an understanding of chess, it lacks any understanding of "meta-chess." That is, Deep Blue cannot carry out a conversation about the activity in which it is engaged, something humans do all the time with relative ease. Deep Blue was not programed to justify its moves, or to comment on the value of the moves so chosen, or even to analyze the significance of the Kasparov/Deep Blue match, something Kasparov and other humans have done at length. This meta-level thinking is not incidental: if computers are to be programed to be intelligent, they must possess this recursive, self-reflexive mode of thought. Until Deep Blue writes a book about the chess match, it will remain simply a very powerful chess-playing tool.

David J. Staley