The search space for most games is very large -- more than 1040 legal chess board positions and more than 2×10170 Go board positions. This means that an exhaustive deterministic search is not possible and some fom of heuristic is needed. For example, AlphaGo uses two heuristic evaluation functions using deep learning, one, the policy network, guesses the opponents next moves and the other, the value network, assess how good the board poistion is.
Used on Chap. 11: pages 224, 225