Home » Artificial Intelligence » Google Claims Artificial Intelligence Breakthrough – Builds a Computer that Can Beat Humans at ‘World’s Most Difficult Board Game’

Google Claims Artificial Intelligence Breakthrough – Builds a Computer that Can Beat Humans at ‘World’s Most Difficult Board Game’

Go has been described as the world's most difficult board game

Go has been described as the world’s most difficult board game

Two tech giants have been locked in a secret race to create an artificially intelligent computer program that can beat professional Go players at their own game. Google has beaten Facebook in the race to create an artificial intelligence (AI) system that can beat the best human players at the ancient board game Go.  First played in China more than 3,000 years ago, Go involves two opponents placing black or white stones on a square grid, with the aim of surrounding their opponent’s pieces.  It is notoriously difficult for an AI to learn, because there are trillions of possible moves – and more configurations of the board than there are atoms in the universe.  As far back as 1965, British mathematician and cryptologist I. J. Good proclaimed that it would be even more difficult to programme a computer to play a reasonable game of Go than of chess. [1]

“In order to programme a computer to play a reasonable game of Go, rather than merely a legal game – it is necessary to formalise the principles of good strategy, or to design a learning programme,” he wrote.  Since then, machines have been created that are capable of beating humans at Go, but a computer program that can consistently beat champions in the game was thought to be at least a decade away.  It was therefore one of the last games where the best human players could still beat the best artificial intelligence players. Now researchers at DeepMind – the British company that Google bought in 2014 for £242 million – have developed a program called AlphaGo that they claim has a 99.8 per cent winning rate against other Go programmes. [1]

ia-dude

Demis Hassabis, chief executive of Google’s DeepMind Technologies

AlphaGo has already defeated the three-times European Go champion and Chinese professional Fan Hui in a tournament by a clean sweep of five games to nil – the first time a computer programme has defeated a professional player with no handicap.  It is now set to take on the world’s best Go player with a cool one million dollar prize pot up for grabs.  “On reviewing the games against Fan Hui I was very impressed by AlphaGo’s strength and actually found it difficult to decide which side was the computer, when I had no prior knowledge,” said Jon Diamond, president of the British Go Association.  The breakthrough, reported in the journal Nature, provides hope that human-level performance could potentially be achieved by robots in other seemingly impossible areas – such as complex disease analysis. [2]

sucker-berg

Facebook Founder Mark Zuckerberg announced his company was working on the same challenge today

Meanwhile, Facebook founder Mark Zuckerberg announced today that that his AI team, led by Yuandong Tian, is “getting close” to creating a machine capable of beating the best human Go players.  “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,” said Zuckerberg in a Facebook post.  “Our AI combines a search-based approach that models every possible move as the game progresses along with a pattern matching system built by our computer vision team.”

In a paper submitted to the International Conference on Learning Representations, Tian wrote that his machine had achieved a “stable” five dan level in the game, representing an “advanced amateur” level.  However, this is still below the “professional level” achieved by Google’s machine, and Tian admitted that the software still has flaws.  “Sometimes the bot plays tenuki (“move elsewhere”) pointlessly when a tight local battle is needed,” he wrote in the paper.  “When the bot is losing, it shows the typical behavior of MCTS [a machine learning technique known as Monte Carlo Tree Search] that plays bad moves and loses more. We will improve these in the future.” [3]

———————————
[2] Elizabeth Gibney, Google AI algorithm masters ancient game of Go, Nature, 27 January 2016

 

billwallace

Leave a Reply

Your email address will not be published. Required fields are marked *