Programming a computer for playing chess

Before any computer had played a game of chess, Claude Shannon described how one would. His March 1950 paper in the Philosophical Magazine lays out the whole apparatus: represent the board as numbers, generate the legal moves, look ahead through the tree of replies, score the leaf positions with an evaluation function that weighs material and mobility, and choose the move that survives the opponent's best answers. That procedure is minimax, and the paper is its first application to a real game.

Shannon also counted. He estimated about 10¹²⁰ possible games, a number now called the Shannon number, and concluded that brute force alone was hopeless. He proposed two strategies: search every line to a fixed depth, or search selectively along the plausible ones the way a human does. Programmes would spend the next forty years arguing about which was right. Deep Blue, in 1997, settled it with enormous quantities of the first.

The paper was not really about chess. Shannon said it plainly: a machine that could play a good game might also be able to design circuits, translate languages or make strategic decisions. Chess was the test case for machine thinking, and it stayed that way until Go replaced it.