Programming a Computer for Playing Chess
Era 2 · AI is born (1943–1979) › Symbolic AI
- Lineage
- Symbols and logic
- Kind
- Publication
Who
Bell Labs
What happened
In March 1950, Claude Shannon publishes a paper in the Philosophical Magazine on how to program a computer to play chess, first presented at a convention of radio engineers in New York in March 1949.
Why it mattered
It frames the game as a search through the tree of moves, with a function that evaluates positions and the minimax procedure. It is the scheme behind nearly every board-game program until the 2010s.
The full story
Chess was the classic example of an activity that requires intelligence. Shannon chose it because the problem is well defined, has a clear goal, and is neither so simple as to be trivial nor so hard as to be out of reach.
The paper begins by showing that brute force will not work. Shannon estimates the number of possible games at around 10 to the 120th, a figure now called the Shannon number, so no computer could examine them all. Instead, the paper proposes looking a few moves ahead, scoring the resulting positions with an evaluation function that weighs material, mobility, and pawn structure, and choosing the move on the assumption that each player picks what is best for that player: the minimax procedure.
It also distinguishes two strategies: examine every move to a certain depth, or examine only the promising ones and follow forcing lines further. The second is the one that would try to imitate a human player.
Shannon did not write a complete program, but these ideas organized the work of the following decades. Arthur Samuel applied a similar scheme to checkers, and search with an evaluation function led all the way to Deep Blue in 1997.
Under the hood
In the minimax procedure, the positions at the end of the search are scored with the evaluation function. Moving up the tree, on your own turns you take the highest score among the next positions, and on your opponent's turns the lowest, on the assumption that each side picks what is best for itself.
The function Shannon proposed adds up material with weights, nine for the queen, five for the rook, three for the bishop and the knight, and one for the pawn, subtracts half a point for each doubled, backward, or isolated pawn, and adds a tenth of a point for each legal move available. Everything is counted for each side and subtracted.
Shannon estimated about a thousand combinations for one move by each side. Looking three moves ahead for each player means about a billion positions, too many for a 1950 computer; hence the second strategy, which examines only the promising moves.
Leads to
Samuel's checkers program (1952)
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
Sources
Original source
Further reading
How to cite
“Programming a Computer for Playing Chess.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/shannon-chess
This entry was last reviewed on November 13, 2026.