Optimizing ChessPhilology and Algorithmic Culture
This essay utilizes chess to ventilate and then re-inter a sustaining premise of Western thought: the belief that practically all phenomena are susceptible to rational calculation, and that even the most aleatory cultural processes can be captured within a matrix of formal procedures. Although this tendency toward mathematical operationalization—the re-inscription of culture into algorithms—is often depicted as emerging alongside computers, my study of chess demonstrates that algorithms have a much longer history within the European cultural imaginary, particularly where Europeans have imagined the culture of the Other. Computers and computation, I argue, must be considered as only one narrow part, albeit a vigorous part, of this older and broader historical project. To begin, I show how algorithms aided European philologists who traced chess back to its origins in India; next, I use this philological history to contextualize subsequent debates in the prominent Artificial Intelligence subfield of computer chess. For computer scientists as well as textual scientists, chess's internal mechanisms become a locus for validating judgments and actions that far exceed the confines of the chessboard itself. Reading AI within the shadow of European philology discloses a critical history of computation, broaching the impingements that computers place upon critical scholars in the present. If today's culture is indeed a digital culture, then it is incumbent upon us, as humanists, to investigate, study, and critique the digital aspects of that culture. But this means seeking digital problems and digital objects where they actually are, rather than imagining them where they are convenient. A postcolonial account of chess forges one—but only one—new trailhead for this spatio-temporal re-mapping of our so-called digital culture.
[End Page 30]
At the end of his classic 1950 paper on computing machinery and intelligence, Alan Turing recommended two possible starting points for simulating human thought: mathematicians could either program a computer to play chess, or they could program a computer to speak and understand English. "I do not know what the right answer is," Turing admitted, "but I think both approaches should be tried."1 Seventy years and several billion research dollars later, it would appear that the latter approach—the linguistic one—still leaves much to be accomplished. After the falsely prophetic chatbots ELIZA and SHRDLU, after the 1966 ALPAC Report and its abrupt defunding of Machine Translation, after the "Chinese Room" and other refutations of strong AI, after Amazon Alexa and the specter of "psychoacoustic hiding," experts continue to question whether a computer will ever speak, much less understand natural language in a manner equivalent to a human.
Chess, on the other hand, succumbed rather quickly. Any doubts whether a computer could play chess were summarily quashed on May 11, 1997, when the world's top-ranked chess player lost in six games to IBM's Deep Blue supercomputer. The following morning the front page of the New York Times hailed Deep Blue as an "inscrutable conqueror." Unlike its human opponent, Deep Blue "never agonized, was never tired, never showed joy or disappointment."2 With mechanical tenacity, it evaluated 100 million positions every second. It exploited differentials in tempo, positional strength, and king safety. It formulated long-term strategies and mobilized short-term tactics. Yet one could still wonder: at any point during the match, did Deep Blue manage to think? Did its behavior qualify as "intelligent," in some human sense of the word? It came as a surprise to some observers when IBM answered, "no." IBM officials acknowledged that "Deep Blue is stunningly effective at solving chess problems, but it is less 'intelligent' than even the stupidest human."3 IBM's cautious modesty reflected a broader shift in the priorities of artificial intelligence research, which had largely abandoned chess a decade earlier. By the late 1980s, contributors to the Journal of the International Computer Chess Association were already mourning their field's regression into an engineering gimmick, or a business venture, rather than a science.4 Deep Blue's victory therefore offered little to celebrate. Computers had grown powerful enough to defeat humanity's chess champion, but, as Noam Chomsky put it, "Who cares?" The fact that a computer played better chess than the world champion was "about as interesting as the fact that a bulldozer can lift more than some weight lifter."5
Although chess may no longer have much to tell us about computers or cognition, it can still tell us quite a bit about culture. The following essay reconsiders chess as what Erich Auerbach called an Ansatzpunkt: a point of departure, or a "handle" for seizing and synthesizing an otherwise nebulous subject.6 Unlike AI researchers, my intended subject is not intelligence. Instead, I use chess to grasp the far more diffuse subject of computationalism: the steadily intensifying process whereby all social phenomena—intelligence, but also language, literature, and other objects of humanistic inquiry—are inscribed, or are imagined to be inscribable, within a discrete set of algorithms.7 While experts in natural language processing continue to probe English with the latest neurally [End Page 31] networked methods of algorithmic inscription, chess represents one prominent cultural domain where computation has, without a doubt, achieved functional preeminence. Chess can therefore show literary scholars how, why, and what happens when culture becomes computable.
This essay has two goals. First, I argue that the game of chess has long been analyzed in algorithmic terms, dating back several centuries before the appearance of modern computers. These pre-computer analyses were not performed by engineers or mathematicians; rather, they were the work of European philologists, who studied the game's internal mechanisms in order to trace its Indian origins. Second, I show how philologists and, eventually, computer scientists used the game's internal mechanisms to make authoritative judgments about external phenomena: judgments regarding what is intelligent and what is not, what is human and what is not, what is European and what is not. I argue that this long-established sympathy between literary research and AI research makes chess a valuable Ansatzpunkt for apprehending the deep historical tensions circumscribing cultural knowledge at a moment when all cultural objects confront the widening gyre of computational expertise. My intention is not to escape but to inhabit this gyre, and, through historical and historicizing analysis—the obverse lamina of philological method—to build what Norbert Wiener called a "local and temporary island of decreasing entropy."8 Rather than an algorithmic study of culture, then: a cultural study of algorithms.9
From Chaturanga to Chess
H. J. R. Murray's A History of Chess is best understood as the work of his father's son. This is because his father, James Murray, was the renowned editor of the Oxford English Dictionary, overseeing its inaugural publication in 1888. Half rationalist and half romanticist, his father rejected the notion that a language's origin lay somewhere behind it, or that English is only the still-settling dust of a prior Indo-European explosion. Languages, rather, are perpetually reborn, every day, and although this process is beyond our control, we can glimpse it through the scientific histories of words.10 Chaos yields to knowledge in the OED—but also, as we will see, in A History of Chess.
Murray published his History in 1913, and it remains one of the few obligatory touchstones for a historical field that is otherwise diffuse, methodologically inarticulate, and unevenly hagiographic and pedagogical. Seventy years after its publication historian Richard Eales could lament that "the very excellence of [Murray's] work has had a dampening effect on the subject, for whenever historians or critics come across a mention of chess in their researches they almost invariably put in a reference to Murray's History and pursue the matter no further."11 In the words of Marilyn Yalom, a more recent chess historian, Murray was "one of those late Victorian giants":12 a giant, because he aspired to assemble chess in its totality, tracing chess's broader cultural and political meanings from its Indian genesis to its European maturation, grounded by manuscript readings in Sanskrit, Arabic, Latin, French, German, and Spanish, all condensed into his 900-page [End Page 32] tome; and late, because he worked within the twilight of an ambitious and imperial epoch, which began in 1694 with Thomas Hyde's theory of chess's Indian origin, and continued through the research of Sir William Jones, Duncan Forbes, Antonius van der Linde, and drew to a hesitant conclusion with Murray. In 1900 Willard Fiske summarized the interminable search for chess's origins in The Nation, writing that chess's history was unknowable beyond the seventh century: "down to that date it is all impenetrable darkness, nor can any ingenuity evoke, from the misty fables of later time, a solitary gleam of light."13 Hyde, Jones, Forbes, van der Linde, and Murray all agreed that chess began as the Indian game chaturanga, and each attempted to penetrate this "impenetrable darkness" with more and more precision.14 But Murray's History marks a point of exhaustion, a philological limit where the unanswerable question of chess's origins would be displaced by something different: in addition to asking when, where, and how chaturanga was invented, one must also ask when, where, and how it became European.
Murray's inquiry centers on one unambiguous element: the problem of time. In predecessors such as chaturanga (Indian) and shatranj (Muslim), the game's pieces were far less mobile than modern chess pieces.15 Early pieces known as the counsellor and the elephant functioned somewhat like the modern queen and bishop, but unlike queens or bishops, which can travel across the entire board whenever their paths are unobstructed, counsellors and elephants could only move one or two squares at a time. "Early chess undoubtedly suffered from the serious drawback that the game was slow in opening," Murray writes. "How to obviate or do away with this drawback confronted chess players everywhere, and is the key to all the modifications and improvements which have taken place in the game."16 An early technique for accelerating gameplay was the ta'bīya, or "battle array," a device developed among medieval Muslim players. Since shatranj pieces travel so slowly, competing players could exchange several turns before either could threaten to attack. Opponents paid little attention to one another, working independently instead to construct a favorable opening position, or a ta'bīya, and only afterwards did proper gameplay commence.17 Using the ta'bīya as a heuristic guide, shatranj openings became well-drilled performances that reflected "no necessity for thought."18 Thus the problem of time was resolved through attrition. Gameplay was expedited by eliminating the opening phase altogether, and "hurried, simultaneous, and unconsidered play gradually became the rule for the earlier moves of the game."19 [End Page 33]
The ta'bīya was a local solution but a historical dead end, accelerating the game exclusively by prompting players themselves to play faster. Muslims might experiment by adding more pieces or enlarging the board, but, in effect, the game played in Damascus in 850 CE was the same game being played there 700 years later: shatranj, a game without history.20 "It is plain that there were no forces making for change from within," Murray writes, "and, if changes were to come, the motive forces must be looked for from without." During the tenth century shatranj made its initial contact with Europeans, who "never exhibited the reluctance to make changes in the game of chess which was shown by Muslims."21 Shatranj's newest players were not content merely to accelerate the old game. The "European player, unlike the Muslim, felt some disappointment with chess," such that the game itself had to change.22 By the fifteenth century the problem of time penetrated directly into the game's pieces: the sedentary counsellor and elephant were abandoned for the far-ranging queen and bishop. The game grew faster, the ta'bīyāt were forgotten, and shatranj disappeared from Europe:
The changes in the move of the Queen and Bishop completely altered the method of play at chess. The initial stage in the Muslim or mediaeval game, which lasted until the superior forces came into contact [i.e., until the ta'bīyāt were completed], practically ceased to exist; the new Queen and Bishop could exert pressure on the opponent's forces in the first half-dozen moves. . . . The player no longer could reckon upon time to develop his forces in his own way; he was compelled to have regard to his opponent's play from the very first. It became necessary to examine into the validity of the different possible ways of commencing the game. Thus analysis came into being, and the game was played in a more scientific way.23
The queen and bishop did not simply add or subtract time; they invested and modulated it. Tempo itself became a commodity, something to be won or lost each turn. Games no longer began at some arbitrary moment surrounding the ta'bīyāt; suddenly every move, including the very first move, became absorbed into the meaningful time of gameplay A player's earliest choices incurred consequences deep into the endgame, prompting detailed analyses of the unexplored opening phase. Meanwhile, new systems of writing allowed these analyses to proliferate, as "algebraic" and "descriptive" notations provided a common script for recording, studying, and discussing past games.24 These three factors—faster pieces, consistent notation, and opening analysis—coalesced around an object bearing little resemblance to shatranj. European chess was "almost a new game," and by 1600, after only a generation or two, "practically the whole of the science, the literature, and the problem lore of the old game . . . became obsolete."25
But if European chess so completely broke with shatranj, then how can we be sure they were related at all? The Arabic word shatranj, for instance, derives from its Eastern predecessors, chatrang (Persian) and chaturanga (Sanskrit), but is unrelated to later terms like échecs (French), schach (German), or chess (English). Chess's European names derive instead from the Arabic word for a specific game piece—shah, the king—and its Latin adaptation, scacus, which, in its plural form, scaci, acquired the more general meaning "chessmen," from which échecs, schach, and chess ultimately derive.26 [End Page 34] Unfortunately, however, etymological analysis fails to answer the essential question. Shatranj and chess became completely different games during the fifteenth century, but the word shah became scacus as early as the tenth century, which means that a fourteenth-century manuscript mentioning "chess" still refers to the Muslim game, shatranj, whereas the modern Arabic word (shatranj) refers to the European game, modern chess.27 When did the word chess cease to mean shatranj?28
This question cannot be answered through etymology alone, but where language fails to grasp chess's catachrestic impasse, one can consult chess games themselves, and to this extent chess is a rare object for the philologist. Although Perso-Arabic manuscripts lack a science of openings, they abound with endgame compilations called manṣūbāt, or what modern players simply call problems. Each manṣūba is a pedagogical puzzle presenting players with a game in medias res, prompting them to discover a combination of moves to ensure victory. To use a modern example, Thomas Taverner's famous 1881 problem instructs the player to win in two moves, or to "mate in two." The surprising solution is revealed through a sequence of steps, or an algorithm, written as follows: 1. Rh1 Bxc7 2. Rh4#.29
Today it would not be unusual to find a similar problem in a newspaper, printed alongside a crossword puzzle, but for Murray this is the nearest approximation to a Muslim science of play. The 553 extant manṣūbāt, available only through their encoded solutions, form a narrow vista for reconstructing the style and substance of Muslim chess in its entirety. Thus Murray concludes that the majority of chess problems in a specific European manuscript are "unmistakably Muslim" because they are composed in the singular fashion of the manṣūbāt, which means these problems are populated with erroneous pieces; black and white are evenly matched; the winner's king is under immediate threat of check; and the number of moves to secure checkmate is unspecified.30 The true genealogy of chess is revealed through the manṣūbāt and through their recurrence in medieval European manuscripts.31 By pinpointing repeated arrays and solution steps, Murray traces chess's evolution not through mutations of language, but through different styles of provoking, limiting, and producing the strings of code that solve various chess problems.
This philology is not linguistic, but algorithmic.32 Although modern chess is wholly distinct from shatranj, we can pursue it backwards into the earliest European problems, which are algorithmically identical to what must have been their Muslim progenitors. Around 1300, there was no appreciable difference between a manṣūba and a "problem," at least in the mute mechanics of their step-by-step unfolding. But once these Euro-Arabic chess problems began eliminating unnecessary pieces, tipping the balance of forces in white's favor, specifying timelines and mate conditions, and exploring alternatives [End Page 35] beyond an incessant mate-drive, then, suddenly, "we are in a different world, the creation of the European problemist."33 And the European creator abhors excess: we can authenticate European problems because they contain, on average, three fewer moves and ten fewer pieces than the cluttered manṣūbāt, proving that "economy of force was a principle of composition that was not yet dreamed of, and the presence of inactive and superfluous men was no blemish in Muslim eyes."34 Eventually, but only eventually, the game divulged its science, as the foggy manṣūba gave way to the clear-eyed problem, and then, after the influential Göttingen Manuscript and the chess masters Lucena and Philidor, as endgame compilations altogether succumbed the systematic analysis of openings.35 In the meantime chess became faster, scientific, better, which is to say, European.
By now it should be clear that Murray partakes in the ready-to-hand machinery of orientalism. Reading his History, one begins to picture that same "single shelf of a good European library" imagined by Thomas Babington Macaulay, a shelf no doubt stocked with Philidor's Analyse du jeu des échecs. Modern chess is a troublesome contradiction, both "an advanced variety of Muslim chess" and, at the same time, "European from start to finish." But this contradiction can be resolved by the philologist, who separates Europeans from Muslims until, out of one game, emerge two. The first "stood still," while the second entered Europe and "became subject to those laws of development and progress which were working in all other branches of human activity."36
If this maneuver sounds at all familiar, then perhaps this is because it was preceded by a century of Europe's most preeminent philologists. When Wilhelm von Humboldt hypothesized that each nation's mental power could be measured through its primordial language-form, he was in the midst of disentangling, like Murray, the troublesome genealogy of the Kawi language, separating its Sanskritic (inflectional) features from its Malayan (agglutinating) features. Kawi speakers, von Humboldt concluded, exiled Sanskrit to the fallow fields of agglutination, constricting an otherwise dynamic language within a moribund language-form.37 Modern chess, by contrast, resembles those "staunch and sturdy" daughter-languages of Latin: the queen and bishop melted shatranj's rigid syntax and infused it with simultaneity, so that an agglutinative game became inflected. In chess as in language, mental power surfaces wherever time submits to form.
Chess differs from language, however, in its criteria for speaking philologically, to borrow an expression from Edward Said.38 I have already suggested that these criteria are algorithmic, not linguistic, and I hope this carries us beyond the somewhat obvious argument that Murray was an orientalist. If Murray's Eurocentrism seems at all obvious or even predictable to us, today, then I would argue this is because scholars of language in the era of decolonization have spent several decades working through their ethno-philological inheritance. We already know, and are still learning, about language [End Page 36] in the world. We know, for instance, that von Humboldt's Indo-European obsessions were formative for the Western nation-state, the modern research university, literature departments, scientific racism, and the Eurocentric conception of "Man."39 As software and other automatically executable instructions come to occupy a central position in the inscription and re-circulation of our contemporary digital culture, it is worth asking whether there is a comparable inheritance for algorithms. What would it mean to posit a postcolonial algorithmic philology?40
Chess offers a potential case study, as it is one of those rare, early domains where analyses of algorithms and culture converge. Resisting the temptation either to elevate chess to the status of literature or to reduce chess to fit within computation's primal scene, Murray's History interposes chess in the middle, between culture and a system.41 Europeans, Muslims, history, thought: they all crystallize in and through the chess game, where they can be sorted, categorized, ranked, and even, someday, automated. We could learn quite a bit about the historical position of algorithms if we placed Murray's History in alignment with the discourse that likewise took chess apart while searching for evidence of "thought"—namely, the post–World War II discourse on artificial intelligence. If Murray shows us how an Indian game became a symbol of Western power, then AI researchers, and especially the AI pioneer Herbert Simon, will show us how this symbol became a fully automated instrument. Although Murray is a philologist and Simon is a computer scientist, each used chess's internal mechanisms as a discrete script for rendering intelligence intelligible. Each therefore belongs within a cultural history of reading culture, algorithmically.
From Minimaxing to Satisficing
Murray published his History in 1913. That same year a German mathematician named Ernst Zermelo published a short paper applying point-set theory to the game of chess, a paper which would become foundational for the military-economic science of game theory.42 In 1913, Murray understood improvement in chess to mean an economical restructuring of the game's internal temporality; in Muslim society, the game stood still, whereas in Europe it grew faster and more strategic, which means it became better. But by the time Murray and Zermelo were writing, chess had once again been standing still for almost four hundred years. After Zermelo, improvement would gradually cease referring to the judgments of players, problemists, or historians, and would increasingly fall under the scrutiny of mathematicians, economists, and computer scientists. The distinctly European science of analysis would reach a new threshold of abstraction, escaping the grasp of the game and its players, only to be re-imported by artificial intelligence researchers. Once again, evaluations regarding chess would require investigations of specific algorithms, the goal no longer being to disentangle Europeans from Muslims, but to separate generative thought from mere computation. This transition began as a thought experiment in 1913, but it would consume the attention of many early computer scientists after World War II. [End Page 37]
Zermelo's paper asked two questions: can a winning position be mathematically defined, and, if so, is it possible to know the number of moves required to force a win? This has already been proven, Zermelo noted, in the case of individual chess problems. But is it possible to treat the entire game as a single, fantastically elaborate chess problem? His interest was purely theoretical, as he did not care whether any particular position was a winning position, or which particular moves ensured victory. Yet he understood that if anyone managed to answer these more pragmatic questions, then "chess would of course lose the character of a game at all," and the appetite for scientific analysis that first made modern chess possible would be its undoing.43
Theory moved closer to praxis in 1928, when John von Neumann proved his groundbreaking minimax theorem. If for Murray the galvanizing element of chess was time, then for von Neumann this element was uncertainty. A chess player can calculate any number of hypothetical moves, at any point in the game, but no amount of time will allow the player to discern an opponent's thoughts. How should one strategize without knowing what will happen next? This question is no doubt central to all chess strategy, but von Neumann's minimax theorem wrested a baseline of mathematical certainty from the mind of the opponent. Specifically, he proved that in any two-player, zero-sum game of strategy, there is always an optimal choice, regardless of the opponent's counter-strategy. At every juncture, a player can optimize the outcome by choosing the move with the best worst-case result.44 The player succeeds by repeatedly minimizing the maximum possible loss, hence "minimax."
The minimax theorem, however, only exchanges one form of uncertainty for another. In their landmark text, Theory of Games and Economic Behavior, von Neumann and Oskar Morgenstern suggested that "if the theory of Chess were really fully known there would be nothing left to play. . . . But our proof, which guarantees the validity of one (and only one) [chess move], gives no practically usable method to determine the true one."45 A chess player can be mathematically certain that the optimal strategy entails minimizing the maximum possible loss, but he can almost never be certain about which move will actually achieve this optimization. To objectively determine the optimal choice, a player would have to construct a decision tree containing all possible moves, and then all possible moves following those moves, and so on, until every possibility had been exhausted, at which point each move's relative gains and losses could finally be calculated. But this quickly becomes impractical. Even today's most sophisticated computers would require well beyond a trillion years to evaluate each possible chess position. Again, but differently, time asserted itself as the central problem of chess: formerly, as a mismanaged commodity separating Muslims from Europeans; now, as a messianic promise separating humans from the para-secular timescales of computation.
This dilemma is not unique to chess. The gulf between everyday human behavior and ideally rational conduct harried all the mathematical and systems-oriented methods that seeped into the social and policy sciences following WWII. Compared to the peacetime applications of logistics and organization theory—minimizing highway congestion, adjudicating competing claims in state and federal budgets, optimizing inventory procedures [End Page 38] for huge warehouses; the list of potential examples is very long—playing a board game might seem oddly trivial. In 1950 Norbert Wiener put it plainly: "the reader may wonder why we are interested in chess-playing machines at all."46 Claude Shannon published the very first article about programming a computer to play chess that same year, and he made it clear that chess itself was in fact of little interest. Instead, Shannon hoped that a successful chess program might "act as a wedge in attacking other problems of a similar nature and of greater significance," including mathematical calculation, symbolic processing, telephone routing, logical deduction, language translation, and military strategizing.47 Regardless of whether human warfare or natural language translation really are "of a similar nature" to a board game, Wiener confirmed that the Theory of Games had made a considerable impression on government and military officials, so that "when Mr. Shannon speaks of the development of military tactics, he is not talking moonshine, but is discussing a most immanent and dangerous contingency."48 From the outset, chess was understood as a legitimate means to an end, an ersatz laboratory for intellectual automation at its most elementary level.
Shannon's initial paper outlined two approaches to computer chess, and these approaches defined the field for the rest of the century. On the one hand, chess programs could prioritize evaluation, heuristics, and domain-specific knowledge. In this case programs would implement or, better yet, autonomously generate short-term goals like center-control and king protection. This approach is potentially subjective and idiosyncratic, but it reflects the way that humans actually play chess. On the other hand, chess programs could prioritize a straightforward minimax strategy, which would be objectively certain but impossible to realize in a practical timeframe. This approach entails populating a decision tree as fully and efficiently as possible and then calculating the best move with brute computational force. The first approach prompts the computer to think like a grandmaster, while the second approach prompts the computer to think like a soothsayer. Shannon advocated for the former, heuristic approach, but, in a fateful turn for the field of artificial intelligence, he was summarily disregarded. As Nathan Ensmenger has shown, minimaxing spread through computer chess like an invasive species: minimax programs were simple to understand, they transferred easily between different operating systems, they improved with each jump in computing power, and, most importantly, they played good chess. Although there is no such thing as a pure minimax program devoid of an evaluation function, incredible human effort as well as RAND funding were dedicated to the wholly technical mission of economizing searches within decision trees. As computers grew faster and tree-search algorithms grew more robust, chess programs continued to beat higher-ranked players, and therefore seemed to grow smarter. Whereas other AI successes of the period performed esoteric tasks like proving [End Page 39] geometric theorems or solving "general problems," a chess game was easily intelligible, even exciting, particularly for non-specialists. Computer chess could demonstrate its improvement in a very public and measurable way, so that it came to represent the field more broadly, earning itself the moniker "the drosophila of AI."49
But as Ensmenger explains, the initial success of minimax algorithms actually stymied long-term progress, as AI researchers overinvested in a highly restricted method of problem solving. Teams of computer scientists could endlessly tinker with the boundary definitions of a tree-search algorithm, and they might succeed in improving the algorithm in the sense that it could win more games, but they came little closer to simulating intelligence or intelligent behavior. By the late 1980s, two prominent computer scientists could lament that AI research was too focused on "trivial optimizations" of common minimax algorithms, as opposed to interrogating intelligence from a computational perspective. In the end, brute-force minimaxing "may have done as much damage to the progress and credibility of computer-chess research as it has given chess programs high rankings." Chess had primarily become an "algorithmic problem, not an AI problem."50
The algorithmic stagnation of computer chess should help us understand the eccentric interventions of Herbert Simon, whose entire career was built in opposition to the minimax paradigm. As early as 1947, Simon complained that "the social sciences suffer from acute schizophrenia in their treatment of rationality," oscillating between a hyper-rational homo economicus, on the one hand, and a hypo-rational Freudian unconscious, on the other.51 Simon argued that humans are neither purely rational nor purely instinctual, but that reason and desire limit each other during every act of decision-making. Most importantly, humans are structurally incapable of maximizing. When playing chess, or when ordering lunch, or even when purchasing a house, humans do not search for the best choice, rather they search for whatever is "good enough." Like the optimal chess move, finding the best house or the best lunch could require years, so people seek instead to satisfy more immediate internal needs. In contrast to a "maximizing animal," "economic man is a satisficing animal whose problem solving is based on search activity to meet certain aspirational levels."52 Satisficing agents do in fact behave rationally, but not from the omniscient perspective of the game theorist. They act within what Simon calls "bounded rationality," or within a subjective, internalized model of objective, external experience. "He behaves rationally," Simon writes, but only "with respect to this model, and such behavior is not even approximately optimal with respect to the real world."53
Simon's work on satisficing and bounded rationality was at once influential and marginal. These concepts earned him the 1978 Nobel Prize in economics, yet, writing in the New York Times, Leonard Silk described this honor as "a surprise not only to him but also to the economics profession"; "from the standpoint of conventional economics, Mr. Simon is a heretic."54 There is no doubt that Simon was a major intellectual figure, and [End Page 40] he was instrumental in raising the status of the social and behavioral sciences within the National Academy of Sciences, but he was not exactly an economist.55 He identified at various moments as an economist, but also as a political scientist, an administrative theorist, a cognitive psychologist, a computer scientist, and, perhaps most accurately, as a scholar with a lifelong interest in decision-making.56 Bounded rationality, for instance, is less an economic or psychological theory than it is a theory of model-making, and of the habitual or second-nature models that validate certain behaviors as rational. If Simon was a heretic, then this is because he was willing to conflate long-established, discontinuous field knowledges within a set of unified, underlying models. The titles of his books confirm this: Models of Man (1957), Models of Discovery (1977), Models of Thought (1979), Models of Bounded Rationality (1982), and his autobiography, Models of My Life (1992). The unifying element of Simon's research is modeling and, notably, the way models are defined, simulated, and tested with computers.57
One of Simon's preferred model-making workshops was chess. He explains the rationale in a paper coauthored with Allan Newell and J. C. Shaw: "Chess is the intellectual game par excellence. Without a chance device to obscure the contest, it pits two intellects against each other in a situation so complex that neither can hope to understand it completely, but sufficiently amenable to analysis that each can hope to outthink his opponent."58 The only thing compelling a chess piece from one square to another is the player's earnest attempt to make sense out of a practical infinity of considerations. Whenever a player invents a strategy or a heuristic for managing this complexity, we are witnessing bounded rationality in its purest form. "Such characteristics mark chess as a natural arena for attempts at mechanization. If one could devise a successful chess machine, one would seem to have penetrated to the core of human intellectual endeavor."59
Like Shannon, Simon saw chess as a valuable meta-model for understanding more complex systems, so studying chess was tantamount to studying economics, psychology, natural language, the stock market, etc. But unlike Shannon, Simon became a longtime contributor to computer chess. And unlike many other researchers, Simon was not content to construct a winning chess program. A successful program should win chess games, of course, but it should win by simulating a human chess-player's "elementary information processes," which should subsequently explain "other kinds of human thinking": logic, theory, math, language, etc.60 When Simon, Newell, and Shaw completed their pioneering Newell-Shaw-Simon (NSS) chess program in 1958, they boldly described it as a "deliberate attempt to simulate human thought processes."61
The NSS program did not maximize or optimize anything. Instead, it satisficed.62 Contemporary minimax programs explored a large number of potential positions, evaluated those positions, and then calculated an optimal move. The NSS program, by contrast, set an initial "acceptance or aspiration level" and then chose "the first acceptable move," or the satisficing move.63 Like all chess programs, NSS employed a tree-search algorithm, but while other contemporary programs investigated up to 800,000 positions, NSS investigated approximately fifteen, compensating for lack of depth with its heuristic functions.64 Instead of seeking future win-conditions at random, NSS prioritized [End Page 41] immediate goals such as material-balance, center-control, and piece development, and for this reason it appeared to know what move it was looking for. Simon emphasized the program's apparent intentionality, claiming that NSS could "build a bridge directly from a present situation to possible actions for transforming it in desired directions, instead of being required to consider large numbers of actions . . . with the hope that some may turn out to be relevant to the desired goals in the given situation."65 Minimax programs looked aimlessly into the future, hoping to discover certainty there. NSS embraced the uncertainty of each moment and, like a human, learned to live inside it.
The historical development of chess strategy will confirm Simon's bias toward heuristics. When Murray was writing his History, chess was reaching the end of its Romantic period, an era dominated by bold sacrifices and deep combinations. The quintessential exemplar of Romantic style is Adolf Anderssen's 1851 "Immortal Game," where Anderssen appears to make terrible decision upon terrible decision, only to force his opponent into a shocking checkmate at the very last moment—as if, like a computer, Anderssen could see the checkmate many moves in advance. The Romantic style of play was seductive and exciting, but it was supplanted in the early twentieth century by the modern and hypermodern schools, which eschewed daring combinations for the methodical, patient procedures of positional play. As Richard Réti writes in one of hypermodernism's founding texts:
The layman thinks that the superiority of the chess master lies in his ability to think out 3 or 4, or even 10 or 20, moves ahead. Those chess lovers who ask me how many moves I usually calculate in advance, when making a combination, are always astonished when I reply, quite truthfully, "as a rule not a single one." . . . The power of accurately calculating moves in advance has no greater place in chess than, perhaps, skillful calculation has in mathematics.66
Réti studied the games of an international chess legend like Paul Morphy not to imitate his end-games, but to distill the unwritten concepts of his positional style.67 Rather than memorize long strings of combinations, modern and hypermodern players established conventions, heuristics, and rules for strong development. Eight such heuristics are canonized in the opening sentence of Aron Nimzowitsch's paradigm-shifting My System: center-control, open files, the seventh and eighth ranks, passed pawns, pinned pieces, discovered check, material exchanges, and pawn chains.68 A Romantic player, by contrast, knows only two heuristics: attack and defend. According to Nimzowitsch, a strong positional move, such as overprotecting an unoccupied center square, "dumbfounds" the Romantic player, because this move "will not fit into either of these categories."69 Thus chess grows by inventing the categories that guide one's actions wherever no action is immediately obvious. For Simon, these categories are the history of chess, and his NSS program, like Réti and Nimzowitsch, sacrificed foresight for the heuristics that make decision-making meaningful.70
Unfortunately, however, NSS played bad chess. When the important considerations on the board were unrelated to its pre-programmed goals, NSS literally "ran out of things to do," and its play became erratic.71 Its strongest game was against Simon himself, [End Page 42] where it played admirably for the first nine moves, after which it began making decisions that were difficult to interpret, centralizing rooks for no apparent reason, and slowly "floundering in a lost position."72 Minimax programs went on to defeat the world's best players and inspire international fascination, while NSS was attacked by other scholars for failing in the one regard where it should have succeeded: cognitive verisimilitude. Specifically, a few years after Simon published his paper on NSS, two Soviet psychologists argued that computers in general and Simon in particular could not account for the fundamentally human capacity for insight. While studying the eye movements of chess players, O. K. Tikhomirov and E. D. Poznyanskaya noticed that before players choose a move, their eyes perpetually examine and reexamine the same constellation of pieces. Rather than locate each chess piece and then add them up toward a meaningful goal, each repeated glance seemed to transform the system of meaningful relations from within.73 Utilizing language that would resonate with Murray or von Humboldt, Tikhomirov and Poznyanskaya concluded that "the formation of a system is not the passive unification of a number of elements, but a process of active actions with these elements by which their attributes are revealed."74 Simon, they argued, had not yet simulated this generative, perceptual act. NSS could obey heuristics, but it could not induce them.
Simon responded by designing a new program called PERCEIVER, which "fixated" on chess pieces in sequences that resembled human eye movements. He studied the capacity of skilled players to recreate chess positions after viewing them for only a few seconds, and he came closer and closer to apprehending the elementary units of human meaning-making: perception and memory.75 In spite of all this, Simon achieved limited success with so-called cognitive simulations. In 1995, a minimax algorithm was on the brink of defeating the top-ranked player in the world, Garry Kasparov. Meanwhile, programs designed to simulate thought processes were still performing, in Simon's words, at the level of "modest amateurs." Almost forty years after developing the NSS program, Simon was still addressing the entire field of artificial intelligence, tempting his colleagues with the unforeseen promises of a chess program that could beat a grandmaster while searching only a handful of moves—a program, in other words, that could think like a human.76
In the 1970s, when the uncertain consequences of computation began to portend both utopian and dystopian futures, Simon—who had spent his entire career studying decision-making in the face of uncertainty—remained optimistic. To those colleagues who feared a future of worker alienation, social dehumanization, or breaches of privacy and security, he advised that they "look back over human history and try to assess whether, on balance, man's gradual emergence from a state of ignorance about the world and about himself has been something that we should celebrate or regret. To believe that knowledge is to be preferred to ignorance is to believe that the human species is capable of progress and, on balance, has progressed over the centuries."77 [End Page 43]
Simon was in a good position to understand progress. In the 1950s, when the Ford Foundation began channeling hundreds of millions of dollars into South Asia, Simon travelled to India as an advisor for the foundation's management education program, witnessing economic development firsthand.78 He travelled to China in the midst of the Cultural Revolution to foster intellectual exchange across Cold War boundaries, and after Mao's death he returned to disseminate "the new ideas of cognitive psychology and computer simulation—a new kind of magic to replace Marxism and acupuncture."79 In 1983 he became chairman of the Committee on Scholarly Communication with The People's Republic of China. In 1987 he travelled to Moscow to find sympathetic counterparts in the Russian Academy of Sciences, and to deescalate nuclear tensions.80 He attended World Bank workshops in Beijing, spoke at the International Congress of Scientific Management in Australia, he served on President Lyndon Johnson's Science Advisory Committee, and his own academic department—the Graduate School of Industrial Administration at Carnegie Mellon—trained international students who brought management science back to their home countries. Simon was an agent of progress all over the world, and even a critic as harsh as Joseph Weizenbaum could concede that "Professor Simon is one of the most influential statesmen of science in America today. What he says really counts."81
Yet in spite of his many experiences abroad, Simon often insisted that travel was superfluous. He encapsulated this point in a "travel theorem," which occurred to him during his early expeditions for the Ford Foundation. Discovering himself thrown into a vast and unfamiliar subcontinent for the first time, Simon concluded: "Anything that can be learned by a normal American adult on a trip to a foreign country . . . can be learned more quickly, cheaply, and easily by visiting the San Diego Public Library."82 When one is expected to conjure expert insights after just two or three weeks abroad, one must generalize from whatever models are close at hand. William Jones, to adapt Said once again, could only know the Orient by going there, whereas Murray learned shatranj from his books:83
The proof rests on the observation that, for all the diversity of cultures to be found around the world, there are only a few basically different scenarios, all of which are readily sampled in the United States. I will not go quite so far as to say that when you've known one peasant you know them all, but I will assert that having spent a couple of days with a subsistence farmer in any land . . . you are well prepared to watch and listen to a Chinese rice planter or an Indian ryot. More important, you are well prepared to understand and learn from books that describe their lives and society.84
The travel theorem, in other words, is yet another iteration of bounded rationality. It is one more model for decision-making under frightening conditions of uncertainty. [End Page 44] Simon's multi-disciplinary career testifies to the fecundity of these well-defined, peripatetic models, especially when combined with computational savoir-faire. I have already shown that one of these models is chess, a game that Simon continuously revisited long after designing NSS and PERCEIVER; whether he was speaking for an audience in Singapore, using heuristics-based chess programs to illustrate human thought-processing;85 or whether he was at home in the US, extolling chess research as a pathway toward managing real-life situations, for "dealing with, say, large organizations, with the economy, or with relations among nations."86 Because Simon commanded the authority, influence, and funding necessary to bridge two domains as improbable as computer chess and international relations, he was capable of making chess theory travel, well beyond chess itself.
Up to now I have presented a study of movement: chess's movement from India to Europe, the bishop's movement across the chess board, Simon's movement between East and West, and bounded rationality's movement between computer science, economics, psychology, and administrative policy. My intention has been to expand the historical context of algorithms in contemporary culture and to show how a specific cultural object—the game of chess—has long been a switching-post between technical systems and the lived movement of history. Murray could read progress and regress into the algorithmic script of medieval European chess puzzles, whereas Simon could read them into heuristic move-generators and tree-search algorithms. Both found progress inscribed in an effective code, and Simon in particular moved these inscriptions around the globe.
In closing, however, I want to focus on an occasion in Simon's life when this movement was suspended. In preparation for a 1970 visit to Chile and Argentina, Simon wrote a letter to the president of the Sociedad Argentina de Organización Industrial: "While I am in Buenos Aires, I would welcome the opportunity to meet and talk with some of the local people who are interested in computer science, and with local social scientists. I would be delighted, also, at the opportunity to meet Jorge Borges, whose books I enjoy and admire very much; but I expect he is constantly pestered by people, and I would not like to intrude on him."87
In his earliest writings on bounded rationality, Simon reached the starkly poetic conclusion that life is not a continuous surface. Life, rather, is a "branching system of paths, like a maze," and each successive moment splits the system into a new manifold of possible connections.88 He first published these observations in 1956 in a technical paper about rational choice, and then, as he explained in a letter to Borges: "a couple of years later, I stumbled upon Ficciones—and in particular, on La Biblioteca de Babel—to discover that you too viewed life as a search through a maze. The difference was that, while I had to express my vision in equations, you could transform yours into fascinating stories." Simon even suspected that a "transmigration of the soul" had passed between them, transplanting bounded rationality "from the mummy of a mathematical model to the live flesh of literature," and he requested a meeting to celebrate their peculiar intellectual sympathy.89 Borges agreed to the meeting, and they spoke for several hours in his office at the National Library in Buenos Aires. Simon discussed his dream of computer-simulated thought, and Borges talked about his childhood. The conversation [End Page 45] was amicable, but Simon came away with the sober understanding that Borges "wrote stories; he did not instantiate models." Borges revealed that "The Library of Babel" was inspired not by a mathematical abstraction, but by a real library he had worked at when he was younger, where he was both isolated and underpaid. The story conveys the nightmarish feeling that his entire life was being buried there. "If I write these stories," Borges explained, "it is because I have to, or because I need them."90
Think of this as a parable. During the nearly five decades since Borges and Simon met in Buenos Aires we have only become more, not less obliged to view the world through computational models. We have only become better, not worse equipped to imagine that language, literature, games, media, and culture can be studied like moves and counter-moves on a chessboard. It is tempting to welcome this as a mark of historical progress, bound up in those same progressive currents that transformed chaturanga into chess, minimaxing into satisficing, and which, given time, will rewrite all the confused fits and starts of history into the smooth algorithmic logic of optimization.
Notice, then, how this progression halts. After traveling from economics to psychology to computer science, from the chessboard to the Ford Foundation to the National Academy of Sciences, from Pittsburgh to Mumbai to Buenos Aires, bounded rationality falters against the "live flesh of literature." It simply does not apply there, in the Library of Babel, where it fails to grasp Borges's unsung clerical melancholy. [End Page 46]
Max Larson is a doctoral candidate in English at Penn State University. Through readings of Clarence Major, Christine Brooke-Rose, J. M. Coetzee, Evangelina Vigil-Piñon, and Henry Louis Gates, Jr., his dissertation historicizes the relationship between computational and critical methods of textual analysis after World War II. Email: email@example.com
. I thank Brian Lennon and Nergis Ertürk for their comments on early draft of this essay, and I thank Lubna Safi, the Diacritics editors, and two anonymous reviewers for their comments on a later draft. I also thank Alek Erickson for teaching me to play chess.
4. For examples of the various attitudes toward the AI-computer chess split represented in the ICCA Journal, see McCarthy, "The Fruitfly on the Fly"; Frey, "Memory-Based Expertise"; Donskoy and Schaeffer, "Perspectives on Falling from Grace"; Levinson et al., "The Role of Chess in Artificial Intelligence Research."
7. I take my definition of computationalism from David Golumbia, who relegates the term's strictly cognitive connotations to a subset of a much broader phenomenon: "a commitment to the view that a great deal, perhaps all, of human social experience can be explained via computational processes" (The Cultural Logic of Computation, 8).
9. My emphasis on the socio-technical history of algorithms is indebted to the concept of "algorithmic culture" developed among scholars like Kate Crawford, Tarleton Gillespie, and Ted Striphas. See, e.g., Striphas, "Algorithmic Culture."
14. Today, attempts to penetrate this darkness employ still more sophisticated methods, using computers, for example, and running the game's "morphological" and "physiological" traits through iterations of phylogenetic trees like an evolutionary biologist. Chess's disappearance into India's "misty fables of time" continues, therefore, to pose a unique test case for the reconstructive powers of Western science. See Kraaijeveld, "Origin of Chess." For a recent survey of debates regarding the history of chess, see Mark, "The Beginnings of Chess."
15. According to Murray, chess evolved through three cultural stages: an Indian game, a Muslim game, and then, finally, a European game. Geographically, this represents a movement from India to Persia to Europe. Linguistically, it represents a movement from chaturanga (Sanskrit) to chatrang (Persian) and shatranj (Arabic), and then to scaci (Latin), followed by numerous European translations. For the purposes of examining Murray's argument and, in particular, his desire to distinguish Europeans from players in the so-called Muslim world, I use Murray's three cultural categories—as opposed to linguistic or geographic categories—throughout this essay.
20. Murray notes two salient exceptions: the tenth-century masters al-Suli and al-Lajlaj apprehended the value of tempo and the importance of studying opening moves, but according to Murray their work "passed almost unnoticed" through Muslim society (ibid., 234, 240, 352).
21. Ibid., 352.
22. Ibid., 452.
23. Ibid., 777.
24. Ibid., 469.
25. Ibid., 417.
26. Ibid., 395–98.
27. Although chess appears nowhere in European manuscripts before the eleventh century, Murray appeals to the laws of sound development to argue that shah could have only taken the Latin form scaci-in the tenth century: "Contemporary documents establish a knowledge of chess in Southern Europe at the beginning of the 11th century, but philological evidence requires that that knowledge must have commenced at least a century earlier" (ibid., 403). This is one of his principal emendations of Antonius van der Linde's work.
28. Murray does not pose this dilemma in these terms, but he does note the general perception, circa 1500, of a rift within the term "chess," when it had to be distinguished from, simply, "the old game" (ibid., 776).
29. This is only one hypothetical solution. The key move is Rh1, after which black is in zugzwang, and there are as many possible solutions as black's number of responses.
30. Ibid., 570.
31. Ibid., 565.
32. I am taking my definition of algorithm from Donald Knuth, who stipulates that an algorithm must be finite, definite, effective, and it must have an input and output. In this way, Murray's analyses of chess problems and their solutions resemble Knuth's analysis of ancient Babylonian clay tablets. See Knuth, Selected Papers on Computer Science, 187–91. Furthermore, Nathan Ensmenger argues that chess was amenable to artificial intelligence research, at least in part, because of the vast collection of recorded games that were already inscribed in machine-readable format ("Is Chess the Drosophila of AI?," 12).
34. Ibid., 277.
35. Ibid., 782–84.
36. Ibid., 394.
39. For the linguistic—in the institutional or scientific sense—history of imperialism, see Harpham, "Roots, Races, and the Return to Philology"; Said, Orientalism; Mufti, "Orientalism and the Institution of World Literatures"; Saussy, Great Walls of Discourse; Liu, The Clash of Empires.
40. I am taking the phrase "postcolonial philology" and my general method here from Aamir Mufti, "Orientalism and the Institution of World Literatures," and from S. Shankar, "Literatures of the World."
41. Counter-examples here would be, at one extreme, William K. Wimsatt and Vladimir Nabokov reading chess problems as if they were poems, and, at the other extreme, the namesake of Wolfgang von Kempelen's chess-playing automaton: Amazon Mechanical Turk. Wimsatt, "How to Compose Chess Problems, and Why," 78; Nabokov, Poems and Problems, 15.
44. For the history of the minimax theorem and its importance for game theory, see Kjeldsen, "John von Neumann's Conception of the Minimax Theorem." See also Paul Erickson, The World the Game Theorists Made.
62. Herbert Simon and Jonathan Schaeffer, "The Game of Chess," 7. Simon was not the only chess researcher concerned with moving away from a "brute force" minimax model. For an example, see Hans J. Berliner's articulation of the "Horizon Effect." However, as Berliner himself notes, his focus on quiescence search derived from his desire to program a computer to play Master-level chess, whereas NSS "apparently derived from a concern with human behavior" (Berliner, "Some Necessary Conditions for a Master Chess Program").
64. Ibid., 691, 694.
65. Ibid., 704–5.
67. Ibid., 8.
69. Ibid., 159.
70. "Chess history can be written largely in terms of the gradual emergence of a few central concepts: piece values, mobility, development, center-control, and so on" (Newell and Simon, Human Problem Solving, 705).
71. Newell and Simon, Human Problem Solving, 691. The irony is that the NSS program pioneered a method of tree search called "alpha-beta pruning," which drastically improved the quality of minimax-based programs. Despite its cognitive verisimilitude, its real legacy rests in making brute-force programs more efficient. See Ensmenger, "Is Chess the Drosophila of AI?," 16. Also see Donskoy and Schaeffer, "Perspectives on Falling from Grace," 156–57.
74. Ibid., 10.
79. Ibid., 342.
80. Ibid., 357.
87. Simon to Jorge A. Rizzi, June 25, 1970, Herbert A. Simon Collection, series 7, Carnegie Mellon University Library.
89. Simon to Jorge Borges, September 23, 1970, Herbert A. Simon Collection, series 7, Carnegie Mellon University Library.