AlphaGo: Move 37 and the End of Human Certainty
AlphaGo made a move that no human had played in 3,000 years of Go history.
Go is 3,000 years old. Its board has 19×19 intersections. The number of possible Go positions exceeds the number of atoms in the observable universe. For decades, it was considered the game that AI could never master, too complex for brute-force search, too intuitive for any algorithm to grasp. Chess fell to Deep Blue in 1997. Go, researchers said, would take another century.
DeepMind's AlphaGo beat Fan Hui, the European Go Champion, in October 2015, five games to zero. The Go community was not particularly alarmed. Fan Hui was strong, but not the pinnacle. In March 2016, AlphaGo faced Lee Sedol, one of the greatest Go players alive, 18-time world champion. Four games were scheduled in Seoul, broadcast live globally. Lee Sedol confidently predicted he would win in a walkover.
Game 2, Move 37: AlphaGo played on the fifth line, upper right. Go convention holds that early moves on the fifth line are too aggressive, too greedy for territory that should be contested later. Every human professional watching immediately said it was a mistake. Lee Sedol left the board to compose himself. The commentators puzzled over it for minutes. And then, as the game unfolded, it became clear it was not a mistake. It was a move so unusual that it had never appeared in recorded Go history, and it was correct.
AlphaGo won the series 4-1. After Game 4, which Lee Sedol won, the only human to ever beat AlphaGo in a formal match, Sedol described the win as "invaluable," a thing he would treasure precisely because it was so rare. AlphaGo retired from competitive play in 2017 after going 60-0. Its successor, AlphaGo Zero, trained entirely without human game data and surpassed it in three days.
Move 37 has since been adopted by human Go players. The AI showed something it had learned with no concept of tradition: that the conventional wisdom of 3,000 years had blind spots. This is the deeper lesson of AlphaGo, not that machines beat humans, but that machines trained without human priors can find what human priors hide.
What it actually meant
Move 37 was only possible because AlphaGo had no concept of convention. It was not brave. It had no inherited sense of what looked risky. The move came from the absence of human priors, not from superior human intuition scaled up. That is a genuinely different kind of capability.
What enterprises get wrong today
Enterprises train AI to follow human best practices and flag deviations as errors. They optimise for reproducing what their best people do, which means the AI inherits the same blind spots. A system trained only to mimic your team will never show you what your team cannot see.
Related capabilities
More from Moments
Ask about this story
Move 37 is one of the most discussed moments in AI history. A move that violated 3,000 years of Go convention, played by a machine with no concept of tradition, and it was right. What would you like to explore? The technology, the match, or what it means?