2017
AlphaZero
Four hours to rediscover chess.
DeepMind gave AlphaZero the rules and nothing else — no openings, no endgame tables, no games by anyone. It learned by playing itself.
In a hundred-game match against Stockfish 8 it did not lose once. The conditions of that match were argued over afterwards, and reasonably so.
The part that held up was the style. It gave away material for position in ways that read, to the people who studied it, as taste rather than calculation.