Computer Chess Research Articles

Watson: Pathfinder for Global Business?

When discussing artificial intelligence, Dr. Ray Kurzweil often projects a future where thinking machines have become so proficient at imitating the cognitive processes of humans, they do each of those functions better than humans, by every measure. Many believe, perhaps under the influence of recent screenplays and novels, that this rule-changing transformation could bring dire consequences for an increasingly vulnerable human race at the mercy of a superior but soulless machine intelligence. However, what this rather dramatic scenario ignores is that the highly dynamic force driving the development of this “machine intelligence” is the creative and adaptive spark of human intelligence, which has a much longer track record than that of machines and has shown the ability to reinvent itself whenever circumstances call for it. This observation in no way minimizes the significant gains that have been made in artificial intelligence (AI). It only raises the consideration that human intelligence is a moving target, as well, and that humans might be learning as much about how to think from teaching machines as they are teaching “our new masters.” As an illustration of this viewpoint, I offer Watson, IBM’s recent celebrity supercomputer, who bested humanity’s top contestants earlier this year in the rather narrow arena of Jeopardy, a decades-old U.S. television game show that tests broad-based knowledge of what has usually been called cultural trivia—aka “insignificant or inessential matters.” This game show format involves quick retrieval and expression of facts across a wide range of subjects. Accordingly, competitive advantage requires quick location of the information needed, assessment of that information, and presentation in a specified format (as a “What is...?” question). The underlying competence required is not subtle, and not particularly taxing to a suitably prepared human savant. In IBM’s case, success was achieved by using many thousands of algorithms running simultaneously to maximize the speed and breadth of the search process. Human Jeopardy competitors often find achieving success at the above task quite challenging, and long-term definitive winners have been few. As many already know, Watson’s victory in this 2011 contest was quite decisive—but getting the machine to that level of skill (95% accuracy, at response speeds of under four seconds) was no small task for IBM. Five years earlier, when the project was first launched, the computer’s accuracy rates hovered around 15%, and response times were close to two hours, versus four seconds for the top human competitors. While the victory of IBM’s Deep Blue computer over reigning chess grandmaster Garry Kasparov in 1997 was a significant accomplishment, Kasparov contended afterwards that he saw intuitive play in the machine’s game that suggested a guiding outside intelligence. That question was never definitively settled, but it is clear that the winning chess computer in 1997 had twice the computation speed of its predecessor machine, which lost to the same opponent just one year before. Watson: Pathfinder for Global Business?

Launching “The Evolution of Cooperation”

This article describes three aspects of the author's early work on the evolution of the cooperation. First, it explains how the idea for a computer tournament for the iterated Prisoner's Dilemma was inspired by the artificial intelligence research on computer checkers and computer chess. Second, it shows how the vulnerability of simple reciprocity of misunderstanding or misimplementation can be eliminated with the addition of some degree of generosity or contrition. Third, it recounts the unusual collaboration between the author, a political scientist, and William D. Hamilton, an evolutionary biologist.

The Case for Computer Chess Engines in the Classroom

The Case for Computer Chess Engines in the Classroom

Special Issue on General Game Playing

Computer game playing has always been of special interest to Artificial Intelligence from the early days on. The cultural history of game-playing machines reaches even farther back than the advent of the modern computer. In the 18th century, an (allegedly) purely mechanical automaton, The Turk, outplayed almost any human chess player. The main reason for The Turk’s popularity was that people considered human-level intelligence a basic prerequisite for being a good chess-player. This motivated the first AI researchers to regard chess-playing as an excellent benchmark for the creation of intelligent machines. But when in 1997 a computer finally defeated the then world chess champion in a six-game match, it was clear that progress in computer chess had little if any implications on how to build intelligent machines in general. This insight has led to the recent interest in general game playing, which is concerned with systems that can learn to play just any game by being given nothing but the bare rules of that game. Figure 1 provides a schematic illustration of this principle. Because neither the specifics nor even the type of games to be played are known at programming time, a general game player cannot be endowed with fixed heuristics, tailor-made for one specific game. Rather, the player must be able to design a suitable strategy all by itself whenever it is asked to play a new game. Successful general game-playing programs can thus be seen as a new generation of AI systems that can be told what we want from them and that learn to carry out possibly radically different tasks without human intervention.

From Chinese Room to Human Window

The debate in philosophy and cognitive science about the Chinese Room Argument has focused on whether it shows that machines can have minds. We present a quantitative argument which shows that Searle’s thought experiment is not relevant to Turing’s Test for intelligence. Instead, we consider a narrower form of Turing’s Test, one that is restricted to the playing of a chess endgame, in which the equivalent of Searle’s argument does apply. An analysis of time/space trade-offs in the playing of chess endgames shows that Michie’s concept of Human Window offers a hint of what a machine’s mental representations might need to be like to be considered equivalent to human cognition. keywords: Turing Test, Chinese Room, Human Window, computer chess, time-space trade-offs, Pareto optimality

Expert-driven genetic algorithms for simulating evaluation functions

In this paper we demonstrate how genetic algorithms can be used to reverse engineer an evaluation function’s parameters for computer chess. Our results show that using an appropriate expert (or mentor), we can evolve a program that is on par with top tournament-playing chess programs, outperforming a two-time World Computer Chess Champion. This performance gain is achieved by evolving a program that mimics the behavior of a superior expert. The resulting evaluation function of the evolved program consists of a much smaller number of parameters than the expert’s. The extended experimental results provided in this paper include a report on our successful participation in the 2008 World Computer Chess Championship. In principle, our expert-driven approach could be used in a wide range of problems for which appropriate experts are available.

Realizing a Kid's Dream of Computerized Baseball

Growing in the 1950s, we did not have PlayStations, Nintendos, or Xboxes--no video games at all. We did not have 150 channels on cable or satellite TV or VHS movies or DVDs. We had no surround sound or CDs, no cell phones, no personal computers, and no Internet. But we did have our imaginations, and we made games. One game I played was dice baseball. In college in 1970, I programmed a baseball computer game based on my dice baseball game that has been accepted in the Baseball Hall of Fame as the oldest baseball game computer code. How did this journey happen? When I was about ten years old, my cousin George introduced me to a game that involved rolling two dice; their combinations when tossed equated to hits, doubles, homers, or outs. I will always remember the dice combinations: two is a home run, three a double, four an out, and so on through twelve. The probabilities of the dice combinations with hits were comparable to a typical player's season batting average. In my junior high school days, I played this game hundreds of times and began keeping cumulative statistics for batter and team records. I played a full thirty-two-game season with all the National League teams, and I even played an All-Star game after each team had played sixteen games by selecting the batters and pitchers with the best individual records. I am convinced that the skills I learned from adding batter's records and calculating batting averages from my simple game prepared me for a business career that involves measures and metrics. It led to my authoring books and presenting seminars in over thirty international cities annually on how organizations can improve their performance. As a manager of performance management solutions at SAS, a world leader in business intelligence and analytics software, I observe incredible computer applications, ranging from biotech industry drug testing analysis to retailers forecasting line-item inventory for tens of thousands of shelf-items in hundreds, sometimes thousands, of stores. FROM CHILD'S PLAY TO GAME THEORY During my junior year at Cornell University, I took a game theory course taught by Henry David Block, a famous professor who wrote computer chess games that played against computerized chess games written by Russian professors. I majored in industrial engineering and operations research, which was a demanding curriculum on top of being a varsity football letterman. Professor Block, with his classic white Santa Claus beard, was an inspiration to me. His course, titled Bionics and Robots, was fun and less stressful than other engineering courses. For my term project in this course, I wrote a computer program using a programming language that was a blend of COBOL and FORTRAN--two ancient programming languages. The computer code applied a random number generator (in place of the two dice) for one hundred possible outcomes of batter hits that simulated every batter's at-bat for the 1969 National League season. I still have the original hard-copy print-out on that big paper with the paper-feed holes--a true relic of the past. It was programmed on IBM punch cards that you had to walk to the data center to submit your card deck. This program's deck of punch cards was a foot thick. My classmate, Pete Watzka (who, coincidently, led the 1971 Ivy League baseball conference in batting average), calibrated each player to his 1969 batting average. Our computer program also differentiated batting power between single and home-run hitters, and for each game we adjusted the players down or up based on the quality of the opposing pitcher's 1969 record. Our computer program simulated all of the games played for each National League team. Pete and I enjoyed changing the input variables of players and pitchers and seeing how the outcomes changed each time we processed a computer run. On May 13, 1970, Pete and I submitted the final computer run for the course paper. …

The Computational Intelligence of MoGo Revealed in Taiwan's Computer Go Tournaments

In order to promote computer Go and stimulate further development and research in the field, the event activities, Computational Intelligence Forum and World 9times9 Computer Go Championship, were held in Taiwan. This study focuses on the invited games played in the tournament Taiwanese Go players versus the computer program MoGo held at the National University of Tainan (NUTN), Tainan, Taiwan. Several Taiwanese Go players, including one 9-Dan (9D) professional Go player and eight amateur Go players, were invited by NUTN to play against MoGo from August 26 to October 4, 2008. The MoGo program combines all-moves-as-first (AMAF)/rapid action value estimation (RAVE) values, online upper confidence tree (UCT)-like values, offline values extracted from databases, and expert rules. Additionally, four properties of MoGo are analyzed including: (1) the weakness in corners, (2) the scaling over time, (3) the behavior in handicap games, and (4) the main strength of MoGo in contact fights. The results reveal that MoGo can reach the level of 3 Dan (3D) with: (1) good skills for fights, (2) weaknesses in corners, in particular, for semeai situations, and (3) weaknesses in favorable situations such as handicap games. It is hoped that the advances in AI and computational power will enable considerable progress in the field of computer Go, with the aim of achieving the same levels as computer chess or Chinese chess in the future.

Detecting clones in game-playing software

The number of competitions using software agents in the AI or entertainment computing fields has grown over the years from a few to many. For instance, tournaments exist to let programs to play chess, checker, go, bridge, poker, simulated soccer, etc. While most competitions are organized for research purposes, some have a commercial side, thus raising the stakes for the software authors. Hence, plagiarism is becoming a problematic issue in the field of game-playing software agents used in competitions and tournaments. As new ideas and technologies are successfully implemented in open source programs, they will be reused and revisited by later programs until they become standard practices, but on the other hand the same phenomenon can lead to accusations and claims of plagiarism, especially in competitive scenarios, such as computer chess tournaments. Establishing whether a program is a “clone” or derivative of another can be a difficult and subjective task, left to the judgment of the individual expert and often resulting in a shade of gray rather than black and white verdicts. Tournaments judges and directors have to decide how similar is too similar on a case-by-case basis. This paper presents an objective framework under which similarities between game programs can be judged, using chess as a test case.

Reinforcement learning for training a computer program of Chinese chess

Computer chess has always been an interesting subject in artificial intelligence (AI). We propose a method to design a Chinese chess program that improves its performance through training. In this study, we utilise temporal-difference learning, which is a method of reinforcement learning, where each position receives reward value from the next position and the value of the position is modified by a heuristic evaluation function. The experimental results show that the program indeed improved its performance by reinforcement learning.

Simple versus realistic modeling

Actually, the fact that the question has been raised may be confusing to an engineer. We try to instill the notion to engineering students that they should always use the simplest model that addresses the question. The key is not whether the model is simple or realistic (more on this distinction later). The key is the question(s) that the model is expected to help answer. Consequently, there is a family of possible models that can be used to address different questions about the same set of phenomena. Take chess as an example. A simple state transition diagram can be used to define some of the basic rules: White always starts, black and white alternate, and the final states are white wins, black wins, or draw (OK, we can include quitting by either side before the game is essentially over but that is not good form). Figure 1 shows this model. It is simple and addresses a limited set of questions. And it addresses them correctly and realistically. Now, if we wanted to address questions of strategy such as course of action selection, a different model will be used. We will define the chessboard, the pieces, the allowed moves by each type of piece, the special moves and so forth. The resulting model is complex and its state space is exceedingly large. But such models exist (computer chess games) and are indeed very sophisticated and are realistic. Now, we also know that models are built for different purposes. Engineers build them to address design questions—whether developing system requirements or designing systems to meet given requirements. There is another class of models, the so-called exploratory ones. These models are built in order to help us understand reality. An example from physics is the model of the atom. The Bohr model is very

Deep Blue x Kasparov: a revanche

The fact that IBM chess computer Deep Blue won the world champion, Garry Kasparov, in May 1997 (in a series of 5 matches, Kasparov won one, drew two and lost two more) led to the appearance of Several comments on the theme man vs. machine. A year earlier Kasparov had played with an older version of Deep Blue and won the five-game series (despite losing his first match - his first loss to a computer). The tone of the comments was then of general relief: we were still superior to the machine - at least in chess.

IBM's Chess Players: On AI and Its Supplements

This article investigates the ways in which the reporting of technological developments in artificial intelligence (AI) can serve as occasions in which Occidental modernity's cultural antinomies are played out. It takes as its reference point the two chess tournaments (in 1996 and 1997) between the then world champion Gary Kasparov and the IBM dedicated chess computers Deep Blue and Deeper Blue and shows how these games of chess came to be seen as an arena where fundamental issues pertaining to human identity were contested. The article considers the dominant framing of these encounters in terms of a conflict between two opposed categories—“human” and “machine”—and argues the essential role of human agency, the human supplement, in the performances of machine intelligence.

Statistical analysis of the relative strength of chess positions

The statistical value of a piece on its occupied square of origin is extended to those squares controlled by the piece. A cumulative strength metric applies to each square controlled by a piece. The computer model, Relative Strength (RS), predicts the next move in an opening example from the world computer chess championship. A predictive rate is significant. RS is a teaching aid for recognition of tactical chess patterns based on positional patterns. From implementation of RS, recent advances in external computer storage of chess positions include invention of the radix hash sort.

GENTLEMEN, STOP YOUR ENGINES!

For fifty years, computer chess has pursued an original goal of Artificial Intelligence, to produce a chess-engine to compete at the highest level. The goal has arguably been achieved, but that success has made it harder to answer questions about the

BayesChess: A computer chess program based on Bayesian networks

In this paper, we introduce a chess program able to adapt its game strategy to its opponent, as well as to adapt the evaluation function that guides the search process according to its playing experience. The adaptive and learning abilities have been implemented through Bayesian networks. We show how the program learns through an experiment consisting on a series of games that point out that the results improve after the learning stage.

Comparing Typical Opening Move Choices Made by Humans and Chess Engines

The opening book is an important component of a chess engine, and thus computer chess programmers have been developing automated methods to improve the quality of their books. For chess, which has a very rich opening theory, large databases of high-quality games can be used as the basis of an opening book, from which statistics relating to move choices from given positions can be collected. In order to find out whether the opening books used by modern chess engines in machine versus machine competitions are ‘comparable’ to those used by chess players in human versus human competitions, we carried out analysis on 26 test positions using statistics from two opening books: one compiled from humans' games and the other from machines' games. Our analysis using several nonparametric measures shows that, overall, there is a strong association between humans' and machines' choices of opening moves when using a book to guide their choices.

An algorithm for the detection of move repetition without the use of hash-keys

This paper addresses the theoretical and practical aspects of an important problem in computer chess programming - the problem of draw detection in cases of position repetition. The standard approach used in the majority of computer chess programs is hash-oriented. This method is sufficient in most cases, as the Zobrist keys are already present due to the systemic positional hashing, so that they need not be computed anew for the purpose of draw detection. The new type of the algorithm that we have developed solves the problem of draw detection in cases when Zobrist keys are not used in the program, i.e. in cases when the memory is not hashed.

Chess, Chance and Conspiracy

Chess and chance are seemingly strange bedfellows. Luck and/or randomness have no apparent role in move selection when the game is played at the highest levels. However, when competition is at the ultimate level, that of the World Chess Championship (WCC), chess and conspiracy are not strange bedfellows, there being a long and colorful history of accusations levied between participants. One such accusation, frequently repeated, was that all the games in the 1985 WCC (Karpov vs Kasparov) were fixed and prearranged move by move. That this claim was advanced by a former World Champion, Bobby Fischer, argues that it ought be investigated. That the only published, concrete basis for this claim consists of an observed run of particular moves, allows this investigation to be performed using probabilistic and statistical methods. In particular, we employ imbedded finite Markov chains to evaluate run statistic distributions. Further, we demonstrate how both chess computers and game data bases can be brought to bear on the problem.

“Technologies of the Self:” Foucault and Internet Discourse(1)

While some have argued that computing via the Internet offers a vision of freedom and a shared humanity, others have claimed with equal vehemence that it may become the instrument of global surveillance and personal alienation. Foucault's notion of self-fashioning (souci de soi) exemplifies both sides of this debate, since fashions may both be imposed and freely chosen. To present a Foucauldian perspective on fashioning of self online I use instances of recent postings to the Usenet news group rec.games.chess. Key aspects of self-fashioning that I identify include romantic and modernist images of interior experience, the importance of keeping your “cool,” the discussion of techniques designed to improve skill or strength, and the purchase and use of chess computers as icons of mastery. Finally, I consider some implications of this Foucauldian approach for future research on Internet self-constructions.