Monte Carlo tree search MCTS
A search algorithm that grows a game tree through random playouts and spends its time where the payoff looks best.
Four repeating steps: select a path down the tree, expand a new node, play the game out randomly to the end, and propagate the result back up. Selection is typically governed by UCB1, balancing exploring unknown moves against deepening promising ones.
It needs no hand-written evaluation function, which was decisive for Go where nobody could write a good one. AlphaGo combined MCTS with neural networks guiding both selection and evaluation.