Last October, Google DeepMind's AlphaGo beat a Go master ten years before experts predicted. This week, AlphaGo plays world-champ Lee Sedol in a 5-day, live-streamed tournament in Seoul, South Korea.
Who is smarter, man or machine?
The question will be put to the test at the Four Seasons Hotel in Seoul, South Korea, this week when Google's AI system challenges the world champion of the game of Go in a tournament. The five-game match against AlphaGo (Google's AI) and Lee Sedol, the world champion, will begin on Wednesday March 9 at 1:00 pm local time (or 8:00 pm Eastern Standard Time on Tuesday, March 8). The games will be livestreamed on the DeepMind YouTube Channel, with commentary in English. After each match Google's blog will be updated with the results.
DeepMind, an AI startup based in London, was acquired by Google in early 2014—for an estimated $400 million. It began tackling arcade games, using convolutional neural networks (based on the brain's neural networks) to teach itself through reinforcement learning.
AI has been making progress in the game world for decades. In 1992, IBM's AI mastered backgammon. In 1997, IBM's Deep Blue beat chess master Gary Kasparov. In 2011, IBM's Watson beat Ken Jennings at Jeopardy.
But one of the most significant wins in the AI game world came in October 2015 when AlphaGo made headlines by beating European Go champion Fan Hui—ten years before most experts predicted AI could master the game.
The game of Go, however, is far more complicated. Instead of the 20 potential moves on a chessboard, Go has 200 options. In a video, Demis Hassabis, the head of DeepMind, called Go the "holy grail" of games. Because of all of the available moves on the board, Hassabis says that Go players frequently attribute their moves to intuition over simple logic. Another big distinction between programs like IBM's Deep Blue and Google's DeepMind is that Deep Blue was programmed to win in chess, whereas DeepMind "taught itself" using machine learning.
So who will win? There is no clear consensus. At the Association for the Advancement of Artificial Intelligence conference in Phoenix, Arizona, in February, Hassabis told TechRepublic that he couldn't give an answer. Lee Sedol—it may come as little surprise—said in a press conference that he will remain world champion.
Several experts in the AI community predict that the computer will prevail.
"AlphaGo will win," said Roman Yampolskiy, director of the Cybersecurity lab at the University of Louisville. "It had five extra months to improve its game. AlphaGo can also be trained specifically against Lee Sedol's playing style."
"Demis Hassabis is a genius in the domain of games/AI," said Yampolskiy, "and Google is not in the business of losing million dollar prizes while promoting an embarrassing loss worldwide."
Marie desJardins, AI professor at the University of Maryland, Baltimore County, told TechRepublic that "given how solidly AlphaGo beat Fan Hui last fall, it will at least be a close match."
"If I was going to place a bet, though," desJardins said, "I'd bet on the computer."
Toby Walsh, Professor of AI at The University of New South Wales, does not expect it to be a close call. "Either AlphaGo will be soundly beaten or it will soundly beat the Go master," said Walsh. "It's unlikely that AlphaGo will have a similar rating to the best humans, and this will feed into a very one sided match."
Walsh said, "I would put my money on the machines."
But not everyone sees the computer winning this week. Andrew Moore, Carnegie Mellon's dean at the School of Computer Science, said that historically, "the human has won in the first matchup, (chess, poker), though on Jeopardy the computer won straight out of the gate." Based on that, he sees the human winning the first match.
But win or lose this week, it seems difficult to argue that AI will not be the winner, eventually, especially since the information gleaned from Sedol's playing style will certainly inform AlphaGo's system.
"I regard the human's eventual demise in Go as inevitable!" said Moore.
- How Google's DeepMind beat the game of Go, which is even more complex than chess (TechRepublic)
- IBM Watson: The inside story of how the Jeopardy-winning supercomputer was born, and what it wants to do next (TechRepublic)
- Google AlphaGo AI clean sweeps European Go champion (ZDNet)
- Google AI gets better at 'seeing' the world by learning what to focus on (TechRepublic)
- How Google's AI breakthroughs are putting us on a path to narrow AI (TechRepublic)
- Smart machines are about to run the world: Here's how to prepare (TechRepublic)
- Google AI beats humans at more classic arcade games than ever before (TechRepublic)