Human vs AI: go champ confident in $1 million match — for now


The world champion of go said Monday he would beat a supercomputer by a “landslide” in a $1 million match next month, but acknowledged that Artificial Intelligence would soon overpower the Asian board game’s best human brains.

South Korean Lee Sedol is due to take on the Google-owned AlphaGo in one of the most hotly-anticipated showdowns the ancient game has ever seen.

The 32-year-old is one of the greatest players in modern history, having won 18 international titles.

AlphaGo — developed by the AI firm Google DeepMind — stunned the world last month when it was revealed that it had trounced three-time European Go champion Fan Hui 5-0 in a closed-door match last October.

“Based on its level seen in the match (against Fan), I think I will win the game by a near landslide — at least this time,” Lee told reporters.

Lee is ranked at the top of the nine-level scale for professional go players, far above Fan, who is at the second level.

“But if Artificial Intelligence keeps advancing at this pace, I’m not sure I’ll be able to win a year or two down the road,” Lee said.

In the centuries-old game players take turns placing stones on the intersecting points of a grid, vying with the opponent to surround and capture territory.

The person — or computer — who controls the most territory is the winner.

Go is something of a Holy Grail for AI developers, whose first global success in board games came in 1997 when IBM’s Deep Blue defeated then-world chess champion Gary Kasparov.

Although its pieces are undifferentiated — just black and white stones — go is more complex than chess because a player can place a piece anywhere on the 19 x 19 board.

That makes it tough for a regular computer to predict an opponent’s move, because it has to crunch through every possibility.

In fact, there are more possible ways to play a game than there are atoms in the universe, Google DeepMind head Demis Hassabis said in a joint video conference call with Lee.

“Even if you have the biggest supercomputer in the world, it will not be enough… to exhaust and research all the possibilities,” he said.

Hassabis says AlphaGo uses two sets of “deep neural networks” that allow the computer to crunch data more intelligently, by discarding moves a human player would instinctively know were silly.

The computer was programmed with 30 million moves from games played by human experts, and then left to do some self-coaching, he said.

AlphaGo has won almost all the matches it has played so far, but its developers sought out Lee for a real test “against somebody who is on top of the game,” Hassabis said.

The five-game match, to be held in Seoul from March 9-15, will have a prize purse of $1 million, and will be televised in countries including South Korea, China and Japan as well as streamed online.