The journey of man versus computer in chess began in 1950, ignited by mathematician Claude Shannon, who proposed the idea of using computers to play chess. This concept took a significant step forward in 1952 when Professor Alexander K. Dewdney built the first chess-playing computer. However, the technology of that era was still in its infancy, and these early computers lacked the speed and computational power necessary to play chess at a competitive level.
Kaissa: A Pioneering Chess Computer
A notable milestone in the development of competitive chess computers was Kaissa, created in the Soviet Union in 1970. Developed by the Research Group of Computer Science at the Moscow Institute of Applied Mathematics and Mechanics (VNIIA), Kaissa was one of the first computers specifically designed to play chess at a high level. The name “Kaissa” derives from the Greek word for “queen,” reflecting the power and importance of the queen in the game of chess.
Kaissa’s design incorporated a parallel computing architecture, enabling it to examine multiple moves simultaneously, which significantly increased its processing speed compared to other computers of the time. It utilized a combination of depth-search techniques and position evaluation to make decisions during games. Additionally, Kaissa employed a transposition table to avoid redundant position evaluations during its depth searches. Kaissa’s capabilities were demonstrated on the world stage when it participated in the first World Chess Championship for computers in 1974, securing the title. Despite its achievements, Kaissa and other early computers still could not rival the best human chess players and were soon surpassed by more advanced systems.
The Rise of Chess Video Games
The 1980s saw the emergence of chess video games, marking one of the earliest applications of artificial intelligence in video gaming. These games allowed players to challenge computer opponents on personal computers like the IBM PC. One of the pioneering chess video games was “Sargon,” developed in 1978. Sargon used a depth-search algorithm to decide its moves and could play at an intermediate level.
In 1981, “HiTech” emerged, utilizing a position evaluation algorithm that enabled it to play at an advanced level. Chess video games also made their way to home video game consoles during this decade, with notable titles such as “Chessmaster,” released for the Nintendo Entertainment System (NES) in 1988. Chessmaster employed a depth-search algorithm, allowing it to perform at an advanced level, bringing sophisticated chess play to a broader audience.
Deep Blue: A Milestone in Chess Computing
A significant leap in chess computing came with IBM’s development of Deep Blue, a supercomputer designed specifically to challenge and defeat the world’s best chess players. Deep Blue’s moment of triumph arrived in 1997 when it defeated reigning world chess champion Garry Kasparov. This victory marked the first time a computer had beaten a world champion in a full match.
Deep Blue’s success was attributed to its parallel computing architecture, which allowed it to examine up to 200 million moves per second and evaluate up to 30 million positions per second. Like Kaissa, Deep Blue used a combination of depth-search techniques, position evaluation, and a transposition table to avoid redundant position evaluations. Developed by a team of IBM researchers, Deep Blue combined custom hardware and sophisticated software to achieve its groundbreaking performance.
The Advent of Neural Network-Based Chess AI
While early systems like Deep Blue did not use neural networks, the 1980s saw the introduction of neural networks in chess AI. One of the first was “CHESS-1,” developed in 1985 by David B. Fogel and Steven M. Drucker. CHESS-1 was a single-layer neural network trained on a dataset of known chess positions, employing an evolutionary learning algorithm to reach an intermediate playing level.
In 1989, Murray Campbell, Albert Zobrist, and Thomas Anantharaman developed “GK-1,” a two-layer neural network trained using an evolutionary learning algorithm on a dataset of chess positions. GK-1 achieved an advanced level of play and won the World Chess Championship for computers in 1990.
AlphaZero: The Pinnacle of AI Chess
A groundbreaking advancement in neural network-based chess AI came with DeepMind’s development of AlphaZero in 2017. AlphaZero used deep neural networks and machine learning to play chess, Go, and poker, achieving unparalleled performance in all three games. Trained through reinforcement learning, AlphaZero learned from countless game matches, continually improving through practice.
AlphaZero’s neural network analyzed game positions and determined optimal moves, surpassing all previous AI systems and computers. In chess, AlphaZero defeated the world champion computer Stockfish in a series of matches, showcasing a level of play that outstripped any other AI or computer to date. AlphaZero’s achievements underscore a significant advancement in neural network-based artificial intelligence, highlighting its potential not only in chess but in various other applications as well.
All images and all text in this blog were created by artificial intelligences