Google AI wants to dominate board games

Google AI wants to dominate board games

Building AI that’s good at playing board games isn’t the end game.

Blue Jean Images/Corbis

In 1997, IBM’s Deep Blue computer famously defeated world chess champion Garry Kasparov. Now, thanks to Google, a computer can kick your butt at the ancient board game Go.

Google DeepMind, the search giant’s London-based artificial intelligence arm, has developed a program called AlphaGo that can beat a human player at the two-person board game. In fact, the program managed to sweep European champion Fan Hui in a five-game match, the first time a computer has defeated a professional player in the full-size game of Go.

Go’s complex and abstract nature makes it a more challenging AI project than designing a similar program for chess. The strategy game, which originated in China thousands of years ago, is played on a 19×19 grid with black and white stones and challenges players to occupy more of the board than their opponent.

A Go board can be configured in a constellation of arrangements that outnumbers the atoms in the universe, according to Demis Hassabis, who runs Google DeepMind. The vast number of moves and outcomes makes it more challenging to artificial intelligence than chess, he said.

Related stories
  • Eric Schmidt’s AI dream? A ‘Not-Eric’ to do his bidding
  • Mark Zuckerberg aims to build own AI butler, a la ‘Iron Man’
  • Elon Musk helps fund AI to protect humanity
  • Artificial intelligence is the next major wave of computing, Android creator says

Beating a Go champion achieves “one of the long standing grand challenges of AI,” said Hassabis on a conference call Tuesday. His team published its work Wednesday in the international science journal Nature.

AlphaGo’s success at the ancient game comes as artificial intelligence moves from scientific curiosity to real-world applicability. AI’s potential has worried some technologists, including SpaceX CEO Elon Musk and Microsoft co-founder Bill Gates, who worry about its potential dangers. In August 2014, Musk expressed fears that AI could be more dangerous than nuclear weapons. Even famed physicist Stephen Hawking has voiced reservations about AI.

Google is “very cognizant about ethical issues,” said Hassabis, adding the company, which has re-organized under holding company Alphabet, agreed not to use DeepMind’s technology for military purposes. Google bought DeepMind in 2014.

AlphaGo uses a combination of machine learning technologies called neural networks and tree search. If you love complex algorithms, feel free to dig into the way AlphaGo plays and learns in Nature. Facebook’s AI team has also been developing a program that can play the game.

“Scientists have been trying to teach computers to win at Go for 20 years,” said Facebook CEO Mark Zuckerberg in a post Tuesday. “We’re getting close, and in the past six months we’ve built an AI that can make moves in as fast as 0.1 seconds and still be as good as previous systems that took years to build.”

Building AI that’s good at playing board games isn’t the, ummm, end game. DeepMind said the same methods used to master Go could one day be used for tasks like climate modeling and disease analysis. In the short term, aspects of the technology behind AlphaGo could show up in the digital assistant on your phone or in the recommendation app you use to pick a restaurant in the next year or two, said Hassabis.

AlphaGo’s next challenge will be to play Lee Sedol, one of the top Go players in the world, in Seoul in March.