Google’s AlphaGo AI program beats Go champion in world first

28 Jan 2016

Oh, that’s gotta smart, Mark Zuckerberg. One day after the Facebook founder said he wanted to develop AI capable of beating a human player of Go, Google publishes a paper confirming it has done it.

Throughout the decades of development in AI, where the bar is set on a particular program has always been on how it fares against a human in one of the many common board games out there, the logic being that if its choices can match that of a human mind, it’s reached a new level in intelligence.

One of the most famous examples being, of course, chess champion Garry Kasparov taking on IBM’s Deep Blue in the game and losing, the first time a chess player lost to an AI.

Now, however, Google’s DeepMind labs has published a paper in Nature revealing that a computer program, called AlphaGo, has defeated a human professional Go player for the first time ever.

Go?

For those unfamiliar with Go, it is an abstract strategy game played on a grid of black lines, usually 19 by 19.

Game pieces in black and white, called stones, are played on the intersections of these lines with the object of the game being to gain the most territory on the board with your stones and there is no predetermined end point.

The game only finishes when players mutually decide that no more moves of value can be made and those familiar with Google Doodles will remember back in 2014 one featured the legendary Go player Honinbo ‘Invincible’ Shusaku.

The greatest AI challenge

Since Deep Blue’s win nearly 20 years ago, this board game has been seen as the next evolutionary test for AI and one of the greatest challenges for programmers, with many of the current programs only capable of playing at a standard similar to a human amateur.

The difficulty lies in the fact, in a game like chess, there are 20 possible moves for the average position, which is complex in itself, but in Go, the number of possibilities increases to ten times this number.

To underline how mind-boggling this is, one of the AlphaGo team members said that the number of possible configurations on the board is more than the number of atoms in the universe.

Oh, so close, Zuckerberg

When competing with other available Go programs, AlphaGo won 99pc of its games, and then faced its first real challenge when it competed with European Go champion, Fan Hui: he lost five-to-zero.

The big showdown, however, will take place when AlphaGo has its Garry Kasparov moment when it’s put up against the world champion, Lee Sodol, in March of this year in Seoul.

All of this is likely to leave Facebook founder Mark Zuckerberg feeling rather miffed given that just yesterday (27 January), he said Facebook’s own developers were ‘getting close’ to reaching the same human-beating standard.

This journalist can’t be the only one hoping for a future AI showdown between AlphaGo and Facebook’s own program in two months’ time.

In fact, let’s just do away with humans and get AIs to duke out our board game challenges from now on.

Go board game image via Shutterstock

Colm Gorey was a senior journalist with Silicon Republic

editorial@siliconrepublic.com