The first of five matches between Google DeepMind's AlphaGo and puny human Lee Sedol is starting right now.
It might not replace Kasparov vs. Deep Blue as the classic example of man vs. machine, but there's more going on here than meets the eye.
It's really happening. Starting right now, a computer program called AlphaGo is taking on Lee Sedol, a 9-dan South Korean Go player described as "the top Go player in the world over the past decade." There's no telling exactly what will happen, but the outcome will be historic regardless.
AlphaGo made headlines in late January when Google DeepMind, the group behind the program, announced that it had beaten European Go champion Fan Hui 5-0 back in October 2015--the first time a program had beaten a human in a fair match. "The problem is humans sometimes make very big mistakes," Fan Hui told Nature when asked about the loss, "because we are human." Essentially, there's a kind of mental pressure added to any match between human and machine. "The program is not like this," he said. "It's very strong and stable, it seems like a wall.
The program uses Monte Carlo tree search--a way of determining the most promising moves--and deep neural networks that learned how to play by studying expert Go games. It's more complicated than that, but those are the basics. A paper released by DeepMind in Nature goes into far more detail about how it all works, but the important thing to understand is that AlphaGo combines two different artificial intelligence techniques in a novel way.
There will be five total matches between AlphaGo and Sedol from now until March 15, each of which is expected to last four or five hours: one each on March 9 and 10, then March 12 and 13, and then the final match on March 15. That's two matches, then a break, two matches, a break, and then the final ones. Since this is happening at 1 PM local time in Seoul, South Korea, that actually means each match will be broadcast live the night before in North America at 11 PM ET.
"The matches will be played under Chinese rules with a komi of 7.5," Google DeepMind's page about the South Korea matches notes. Those are the points earned by the player that goes second at the end of a match. Additionally, both AlphaGo and Sedol will have two hours per match and three 60-second "countdown periods after they have finished their allotted time."
As far as predictions go, there's a feeling that Sedol will still emerge victorious. The rank gap between Hui and Sedol is significant, and Sedol's had the added benefit of being able to study Hui's games against the program. Sedol himself predicted in February that he'd win 5-0 or 4-1 when the matches finally started. Even so, AlphaGo's had several months to improve. The outcome is still very much in the air.
But even if Sedol does sweep the program, there's much to be gained. "I think the Kasparov vs. Deep Blue matches will still be remembered as the original human vs AI machine battle," says Drexel University assistant professor Santiago Ontanon. But from an AI perspective? "All the AI techniques used in Deep Blue were known for decades," says Ontanon, "so IBM just built a ridiculously fast machine that ran existing AI algorithms. However, AlphaGo is a different thing! The algorithms running in AlphaGo are actually state of the art… So, while Deep Blue was a show of computational power, AlphaGo is a real AI contribution!"