What new versions of chess will AI entities or humans invent that may introduce new challenges for humans and machines? Shown here is Fortress Chess, played in Russia in the 18th and 19th centuries. Four players (AIs and/or humans) are represented here as black, white, dark grey, and light grey.

1997

DEEP BLUE DEFEATS CHESS CHAMPION

Vladimir Kramnik (b. 1975), the undisputed World Chess Champion from 2006 to 2007, once told reporters, “I am convinced, the way one plays chess always reflects the player’s personality. If something defines his character, then it will also define his way of playing.” Would this “personality” also be reflected in an AI’s style of play?

For decades, technologists had considered chess a kind of measuring stick for artificial intelligence, being a game that required strategy, detailed reasoning, logic, foresight, and—at least for human players—guile. After many years of debating as to when a machine might defeat a reigning world chess champion, it finally happened in 1997: IBM’s Deep Blue computer defeated Russian world chess champion Garry Kasparov (b. 1963) in a six-game match. After Game 5, Kasparov had become so discouraged that he explained: “I’m a human being. When I see something that is well beyond my understanding, I’m afraid.”

Using special-purpose hardware, the 1997 version of Deep Blue was capable of evaluating 200 million chess positions per second, and it typically searched to a depth of around six to eight future moves, if not far more. Deep Blue’s strategy could also take into account a large database of past grandmaster games, and it used endgame databases that included chess positions with five or fewer pieces.

The dream of chess machines stretches far back in time. The Mechanical Turk chess robot, created by Hungarian inventor Wolfgang von Kempelen in 1770, played a strong game of chess, but it also employed a human who hid inside the machine. In 1950, computer scientist Alan Turing and mathematician David Champernowne (1912–2000) designed a computer program, known as the “Turbochamp,” to play chess. However, because no computer was available to actually run the algorithm, Turing simulated a computer by manually consulting the algorithm during a testing phase.

In 2017, the program AlphaZero beat world-champion chess-playing computer programs, having taught itself how to play in less than a day! The program used machine learning, starting from random play, and was given no domain knowledge except the game rules.