Artificial Intelligence

For most of its history, artificial intelligence meant search, logic, and reasoning rather than learning from data — writing down what an agent knows and letting it deduce, plan, and decide what to do.

The organizing idea is the rational agent: something that perceives its environment and acts to maximize its expected performance. Everything else is machinery for choosing the action — searching a space of possibilities, proving a logical consequence, or computing an expected utility under uncertainty.

Search comes first: cast a problem as states and moves, and algorithms like A* find a path to the goal, guided by a heuristic that estimates the distance still to go. Add an adversary and the same idea becomes the minimax search that plays chess.

When the world is uncertain, logic gives way to probability. A Bayesian network compresses a joint distribution over many variables into a graph of local dependencies, and the same expected-utility calculus, run over time, becomes the Markov decision process that underlies planning and reinforcement learning.