AlphaGo

AlphaGo was a computer program developed by DeepMind that defeated Lee Sedol, one of the world's strongest Go players, in a five-game match in 2016, winning four games to one. The victory was a watershed moment in AI: Go had resisted the brute-force methods that succeeded in chess because its branching factor made exhaustive search infeasible, and the game had been considered a benchmark that would require genuine machine intelligence to master.

AlphaGo combined deep neural networks with Monte Carlo tree search. A policy network, trained on human expert games, narrowed the search to promising moves. A value network, trained on self-play data, evaluated board positions without searching to the end of the game. The system learned its evaluation function from data and self-play rather than from handcrafted rules. This was the bitter lesson in action: human Go knowledge, accumulated over millennia, was outperformed by a system that learned its own representations through computation. The subsequent AlphaZero system dispensed even with the human game data, learning entirely from self-play — a pure instance of the general method winning over the knowledge-based approach.