Finding the best option among many. It is the other great tradition in artificial intelligence besides learning, and it underpins route planning, scheduling, game playing and the training of every model. When a problem has a clear measure of good and too many possibilities to check, this is the family of methods you reach for.
Technical view
Covers exhaustive and heuristic search, gradient-based continuous optimisation, and combinatorial methods such as branch and bound, simulated annealing and genetic algorithms. Training a neural network is continuous optimisation over parameters; a planner in an agent is discrete search over actions. The practical distinction is whether the space is smooth enough to follow a gradient through, which determines everything about the method.
Related
Part of AI Terminology Index, section: The Foundations.