Samuel's checkers program
Era 2 · AI is born (1943–1979) › Neural networks and learning
- Lineage
- Data and learning
- Kind
- Implementation
Who
IBM
What happened
At IBM, Arthur Samuel writes a checkers program for the IBM 701, the company's first commercial computer, and improves it for years. In July 1959 comes "Some Studies in Machine Learning Using the Game of Checkers."
Why it mattered
One of the first programs to learn on its own: it improves by playing thousands of games against itself. With this work Samuel popularizes the phrase "machine learning."
The full story
Samuel joined IBM in 1949 and chose checkers as a test bed: the game is simple enough to program and hard enough that calculating every move is not an option. With the IBM 701's small memory, the program could not explore every continuation, so it looked a few moves ahead, scored the resulting positions with an evaluation function, and discarded branches that could not improve the result, a technique later called alpha-beta pruning.
Learning came in two ways. Rote learning stored positions already seen, along with their value, so they did not have to be recalculated. The other, more original, adjusted the weights of the evaluation function by comparing a position's score with the one obtained by looking further ahead, so the program kept correcting its own judgment. To practice, it played against itself, thousands of games.
The public demonstrations made an impression: Thomas J. Watson, IBM's president, predicted they would lift the company's stock, and it rose. In 1962 the program won a game against Robert Nealey, who was billed as a checkers master, although its record against humans was mixed.
The 1959 paper is one of the founding texts of machine learning, and its idea of correcting one estimate with a later one anticipates temporal-difference learning.
Builds on
Programming a Computer for Playing Chess (March 1950)
Inherited, with no proven direct link
Samuel's program uses the scheme Shannon had proposed for chess: a search a few moves deep with an evaluation function
Leads to
Temporal differences (August 1988)
Documented direct influence
The temporal-difference method grows out of Samuel's work. Sutton and Barto write that Samuel was the first to propose and implement a learning method built on temporal-difference ideas, in Samuel's checkers program
Threads that pass through here
- From reward to alignment (step 2 of 12)
- Machines versus champions (step 1 of 7)
Sources
Original source
Scholarship and context
How to cite
“Samuel's checkers program.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/samuel-checkers
This entry was last reviewed on November 13, 2026.