Computers are winning ancient game of ‘Go,’ giant leap for machine learning

You’ve probably noticed that Mark Zuckerburg has made mention of his growing interest in artificial intelligence (AI) a few times over the past year. However, today in a Facebook post, Zuckerberg announced that not only has Facebook AI made significant progress in playing the ancient Chinese game, “Go,” but it’s also very close to actually winning the game.

If you’re wondering why beating a board game makes any kind of difference at all in the world of AI, Zuckerberg briefly mentions part of the reason in one of two current AI-themed posts on Facebook. For decades, scientists have been attempting to teach computers how to win at the 19 x 19 square strategy game. Although the rules of Go seem simple, requiring each player to place a disc on one of the intersections, creating territories, and surrounding pieces and capturing other pieces, some people can make a game of Go last a lifetime.

One of the issues with using AI to accomplish tasks is that it has to learn almost every scenario possible, as well as all the possible outcomes. It “learns” through exposure to the actual scenarios and outcomes rather than learning from mistakes it made in previous attempts. In fact, in many cases, if the AI begins to lose the game, it simply starts another path that has no strategic meaning to it adding to the complexities of teaching it how to win in a game that has trillions of possible moves. It’s why it was such a big deal when IBM’s Deep Blue computer beat chess champion Garry Kasparov.

But while Facebook is seemingly leading the way, perhaps coincidentally, Google’s Research Team announced that its AlphaGo system has succeeded in winning games against the top Go players, pushing AI research even further further. AlphaGo was pitted against three-time Go champion, Fan Hui, and won five games to zero, making this the first time a computer has ever beaten a professional Go player. In fact, out of 500 games, AlphaGo was able to win all but one game, even after giving the other players a head start of four free moves at the beginning of each game. There are plans for AlphaGo to also compete against the top player in world, Lee Sedol, in South Korea in March.

According to Google, AlphaGo uses more “human-like” approaches than those in previous attempts, which involve suggesting not just intelligent moves to play while evaluating the position that is reached. In this manner, AlphaGo is able to play according to the most successful move.

Whether it’s research from Facebook, Google, or another company, mastering this ancient game would push AI to the next level. In his Facebook post Zuckerberg says, “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.” He attributes much of the progress to his office-mate, Yuandong Tian, who he says is seated about 20 feet away and faced with the task of training the intelligence to be a master player. The team at Google, however, are satisfied with having mastered Go and achieving, as they say, one of the grand challenges of AI. The ability for a computer to beat a Go professional wasn’t expected to be possible for at least another 10 years.

Also watch: Awesome tech you can’t buy yet: Smartphone microscopes, CNC mills, $15 computers

Please enable Javascript to watch this video

from Planet GS via John Jason Fallows on Inoreader http://ift.tt/1PFp9wA
Christina Majaski