Today is March 8. On this date in 2016, the world geared up to watch Lee Sedol play AlphaGo in the game of Go. For the next seven days, I will be going back through exactly what transpired during that tournament, day by day.
To lay the scene, it’s important to remember a few things. At the time, the idea that an artificial intelligence could defeat a professional Go player seemed preposterous. Earlier that year, even the optimistic Elon Musk said it would be a decade before an AI could defeat a grandmaster at Go.
Why was Go so difficult for an artificial intelligence? A couple of reasons: First, the size of the board, and the fact you can place your pieces anywhere upon it, means that there is a near-infinite number of ways to play the game. The old “brute force” computational method of just looking at every conceivable move, then every move beyond that, and every move beyond that breaks down very quickly. Second, the strategy involved in the game is much more ethereal and nuanced than in a game like chess or poker – so much so that Go is regarded by many top players as more a philosophy of life than a mere game.
AlphaGo, the AI developed by DeepMind, certainly would have its work cut out for it. Lee Sedol, its opponent, wasn’t just one of the best Go players in the world; he was one of the best ever. Although a reserved and humble person by nature, Lee confidently predicted he would defeat AlphaGo in a landslide of 5-0. This sentiment was shared by the hundreds of millions of people around the world who were following the tournament. As you may remember, that is not exactly how things worked out.