Arthur Samuel’s Checkers Program Learns from Experience

It's not hard to understand why DeepMind's first thought with AI was games. As several (El Ajedrecista) other (the chess-playing Turk) moments (the first game programs) have mentioned, games and machine learning have a long history. But in the early '50s it's remarkable these machines could even approach playing a game like checkers, several decades before Deep Blue started beating chess world champions.

Legendary ML computer scientist Arthur Samuel, who IBM credits with coining the term machine learning, wrote his first checkers program for the IBM 701 in 1952. His paper “Some Studies in Machine Learning Using the Game of Checkers,” in the July 1959 IBM Journal of Research and Development, described two methods: saving positions the program had scored and adjusting its scoring formula’s weights. Those weights changed as one copy, Alpha, played another, Beta. A computer, Samuel wrote, “can be programmed so that it will learn to play a better game of checkers than can be played by the person who wrote the program.”