In late July 2017, DeepMind announced that AlphaGo would retire from competitive play. The programme had beaten Lee Sedol, the reigning world champion, in a widely watched match in March 2016. It had then played a series of online matches against top professionals under the name Master, winning sixty consecutive games. Its final competitive appearance was a three-game match against Ke Jie, the world number one, which it won three to zero. Then it stopped.
The narrative around AlphaGo had always been about beating humans at a game considered too complex for computers. Go has more possible positions than atoms in the observable universe, which made it a standard argument for why AI could not master it. When AlphaGo won, the reaction was a mix of genuine awe and existential unease that AI had crossed a threshold.
But the more interesting thing about AlphaGo was not that it won. It was how it played when it was winning.
In the match against Lee Sedol, AlphaGo played move 37 in the second game. Human Go commentators described it as strange, as a mistake. Moves like that were not played by humans because human intuition said they were wrong. AlphaGo played it because its training on millions of games had found a pattern that human intuition had missed. Lee Sedol needed fifteen minutes to respond. He later said the move had fundamentally changed how he understood the game.
That is what made AlphaGo more than a demonstration of computational power. It was finding genuine strategic insights in a domain humans had studied for thousands of years, and finding them by approaching the problem differently. It was not a search algorithm brute-forcing positions. It was something that had developed a model of what good play looked like, and that model contained moves no human had thought to try.
The retirement made sense from DeepMind's perspective. AlphaGo had proved what it needed to prove. The team had already moved on to AlphaGo Zero, which learned from self-play alone, with no human game data at all, and surpassed everything that had come before it. The lesson was not that computers can beat humans at games. The lesson was that the process of learning to play well had produced knowledge that even the humans who had spent their lives on the game did not have.