AlphaGo

AlphaGo was the DeepMind system that beat world-class Go players in 2016, years before most researchers expected it. Go had resisted computers far longer than chess because the board is too large to brute force, so a machine winning was treated as a landmark.

The reason people still bring it up is one move in the second game against Lee Sedol, now known as Move 37. Commentators initially called it a mistake, because no human player would consider it. It turned out to be the move that won the game. Move 37 became shorthand for the thing AI has rarely done since: producing an idea outside the range of what humans were doing, rather than a very good imitation of it. When researchers ask whether today’s models have had a Move 37 moment, that is the bar they mean.