Researched
Machine Learning (Samuel's Checkers)
Arthur Samuel's checkers program at IBM improved by playing against itself and soon beat its author; his 1959 paper gave "machine learning" its name.
Open in the interactive tree →Samuel wrote a checkers program for IBM's first commercial computer, the 701. It scored board positions, searched ahead with minimax (using what is now called alpha-beta pruning), remembered positions it had seen, and tuned its scoring by playing thousands of games against itself. His 1959 paper reports that it learned to play better than its programmer within 8 to 10 hours of machine time, and in 1962 it beat a self-described checkers master, Robert Nealey. Sutton's temporal-difference learning (1988) was based on this earlier work, and Tesauro later used it in TD-Gammon.
Prerequisites
- Game Theory1944The program picks moves by looking ahead with the minimax rule from game theory.
- Stored-Program Computer (von Neumann)1945The program ran on IBM's 701, a stored-program computer.
- Turing Test & Birth of AI1950It was one of the first successes of the young field of artificial intelligence.
Unlocks
- Reinforcement Learning (TD-Gammon)1992Temporal-difference learning was based on Samuel's learning-by-self-play work.