The Match: AlphaGo Beats Lee Sedol

On Thursday, the greatest human player of the ancient game of Go lost a match to a machine, and the machine did not just win; it played moves that the human experts called beautiful. AlphaGo, a program built by the London company DeepMind, defeated Lee Sedol four games to one in a five game match in Seoul, and the result ended a decade of predictions about when computers would conquer the last board game. The match is the March 2016 story, and the story is the lesson: the machines learned to play a game of intuition, and the humans who watched were changed by it.

The Match is the subject of this article: what AlphaGo is, why Go was considered impossible for computers, how the match unfolded, and what the result means for artificial intelligence. The match ran from March 9 to March 15, and it ended on Tuesday with a final score of four to one. This is the story of the match, and the story is about the moment the last human stronghold of the board games fell.

1. The Game

Go is the oldest board game still played in its original form, and it is the hardest game for computers. The board is a grid of nineteen by nineteen lines, and the number of possible positions is astronomically larger than the number in chess. A chess game has around ten to the fortieth possible positions, and a game of Go has around ten to the one hundred and seventieth. The branching factor, the number of legal moves at each turn, is about two hundred and fifty, against about thirty-five for chess.

The scale of the game made it the last bastion of human mastery. The best chess computers had beaten the best humans by the late 1990s, and the experts confidently predicted that computers would need another decade to reach the top of Go. The game rewards intuition, pattern recognition, and a kind of positional judgment that was thought to be beyond calculation. The game was the wall, and the wall was about to be climbed.

2. The Program

AlphaGo was built by DeepMind, the artificial intelligence company acquired by Google in 2014, and the program was built on a new set of ideas. The system used deep neural networks, the layered learning systems that had transformed image recognition and speech, to evaluate positions and choose moves. One network, the policy network, learned to predict the moves a strong player would make, and another, the value network, learned to judge which positions were winning.

The program also learned from itself. After being trained on millions of positions from human games, AlphaGo played millions of games against its own copies, and the self-play sharpened its judgment beyond anything in the human record. The program combined the neural networks with a search algorithm that explored the most promising lines. The result was a player that did not think like a human, and did not think like an old chess computer, and thought like nothing that had ever played the game.

3. The Confidence

The match was arranged after AlphaGo had already made history, and the history was the reason the experts took it seriously. In October 2015, the program had beaten Fan Hui, the European champion, five games to nothing, and the result was published in the journal Nature in January 2016. Fan Hui was a strong professional, and the result was the first time a computer had beaten a professional player at all, and the margin was a shock.

The match against Lee Sedol was the real test, and the test was the highest possible. Lee Sedol had won eighteen world titles, and he was widely considered the greatest player of his generation, a player whose name was synonymous with the top of the game. The experts gave AlphaGo only a small chance, and Lee Sedol himself was confident. The confidence was the setup, and the setup made the fall more dramatic.

4. The Games

The match unfolded in Seoul, at the Four Seasons Hotel, and the first three games belonged to the machine. AlphaGo won game one, and the experts said it was luck; it won game two, and the experts said it was style; it won game three, and the experts said it was a revolution. In game two, the program played a move on the shoulder of a stone, a move that commentators initially called a mistake, and the move turned out to be the key to the win. The move, move thirty-seven, became known as the move from another dimension.

The fourth game belonged to the human. Lee Sedol found a move that the machine did not expect, a wedge that commentators called the divine move, and he won, and the room exploded. The win was the first against AlphaGo, and it proved that the machine could be beaten, and it gave the match its human drama. The fifth game was close, and the machine won it, and the match ended four to one. The games were the story, and the story had everything.

5. The Reaction

The reaction to the match was global, and the reaction was about more than a game. The match was watched by around two hundred million people, by some estimates, and the audience stretched far beyond the world of Go. The result forced a public reckoning with the power of artificial intelligence: if a machine could master the game of intuition, what could it not do? The question was asked in living rooms, newsrooms, and boardrooms around the world.

The reaction inside the Go world was different and deeper. The professionals who had spent their lives mastering the game watched a machine play moves they had never seen, and the game itself was changed. Lee Sedol said he would have to learn to adapt, and the younger players who grew up after the match would train with machines the way earlier generations trained with books. The reaction was the beginning of a new era for the game, and the new era was the point.

6. The Science

The science of the match was the real victory, and the science was about learning. AlphaGo was not programmed with the rules of strategy; it was given the rules of the board and the ability to learn, and it learned from human games and from its own play. The method, deep reinforcement learning, was the same family of methods that would go on to power the next generation of artificial intelligence, and the match was the public demonstration that the methods worked.

The match was also a demonstration of the power of scale. The program ran on powerful computers, and the computing power was part of the story: the machine could play out millions of positions that no human could see. The science was not only about the algorithm; it was about the combination of the algorithm with the hardware, and the combination was the recipe for the future. The science was the substance behind the spectacle, and the substance was real.

7. The Future

The future after the match was the question on everyone's mind, and the question had two parts. The first part was about the game: would any human ever beat the machines again, and would the game survive its own mastery? The second part was about everything else: if a machine could master Go, how soon before machines mastered driving, medicine, law, and the rest of human work? The match was a milestone, and milestones are for looking forward.

The forward look was optimistic and anxious at once. The optimists saw a tool that could help humans solve the hardest problems, and the anxious saw a world where the machines were always one step ahead. The match did not answer the question, and the match made the question impossible to ignore. The future was the subject of the conversation, and the conversation had just begun.

8. The Lesson

The lesson of the match is about the limits of human intuition, and the lesson is humbling and liberating at once. The greatest human player of Go believed the machines were a decade away, and the machines were already there, and the belief was not foolish; it was simply out of date. The lesson for experts is that the most confident predictions about the pace of technology are the ones that age the worst, and the lesson for everyone is that the machines are better at some things than anyone thought possible.

The lesson is also about the game itself, and the game survived the machine. The match did not make Go boring; it made Go more interesting, because the machine revealed new ideas that the humans had never seen. The same could be true of the other fields the machines are entering. The match is the March 2016 story, and the story is the lesson: the machine won the game, and the humans won a new way of thinking about the game, and the thinking is just beginning. The divine move was played by a human, and the move from another dimension was played by a machine, and the game will never be the same.

Tags

#technology #ai