A concept used in artificial intelligence/game theory for two-player games. The idea is to minimize the opponent's gain and maximize yours. Questions using this tag cover problems understanding/implementing the algorithm.
Minimax involves generating a game-tree to consider all the possible outcomes of a game, given its present configuration. The minimax agent then chooses the branch which leads to the agent's maximum gain and, consequentially, to the minimum gain of the opponent (a zero-sum game).
While generating a whole game tree may seem a computationally expensive task, several techniques have been developed to reduce the computation time. One of the most famous techniques is Alpha-Beta Pruning: given a branch in the game tree, stop evaluating that branch ("prune") as soon as you find an outcome in that branch that is worse than a previously-examined branch. Some variations (such as those used in Chess programs) only consider up to a certain number of turns and not always up to end-game.
Another variation of minimax, termed as expectiminimax, introduces "chance" elements in the game such as dice throws.
Finally, note that while minimax is originally intended for two-player games, the technique has been extended to games with more complex set-ups.
Further reading:
