Computer edges top board gamer in victory for artificial intelligence

In what's being billed as a further step in the emergence of artificial intelligence, a computer has managed to beat the European champion of the ancient board game of Go.

Its developers are now ready to pit their programme against Go’s world champion in a match during March.

AlphaGo was developed by David Silver, Aja Huang, Demis Hassabis and colleagues in Google DeepMind, London, and thinks like a human when it comes to Go.

“It is based on something more akin to imagination,” said Mr Silver when describing how it works.

It uses two neural networks – a computer system modelled on the human brain and nervous system – that assess what moves to make. These can play solo games and learn from experience when "taught" by human supervised learning and reinforcement.

It also uses a “tree search” approach which asks branching questions to make decisions.

Designing a programme that could defeat humans at Go represented one of the "grand challenges" for those involved in artificial intelligence, the authors write in journal Nature.

Cyber challengers

Having created the programme the researchers wanted to test its mettle. It first encountered other Go-playing computers and defeated 99.8 per cent of its cyber challengers.

AlphaGo next faced more formidable opposition with a match against professional player Fan Hui, the current European champion. He was vanquished by five games to nil.

“This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away,” the authors write.

Having claimed one scalp the programme’s next encounter is with world champion Go player Lee Sedol, with a match planned for March in Seoul.

Go originated in China and has a long heritage. It involves two players alternately placing black and white pieces onto a grid and attempting to occupy more space than their opponent.

“It is an ancient game and the most complex game played by humans ,” said Mr Huang, also of Google DeepMind.

There are more configurations on the board than there are atoms in the universe, he said.

The researchers believe that their approach will help computers to achieve human-level performance in other artificial intelligence areas.