In a stunning feat that shocked its creators, a computer program has masterfully claimed a victory at the complex game Go against a human. DeepMind's chief executive said the program's strong victory was not expected.
AlphaGo - a computer program developed by London-based Google DeepMind - beat three-time European Go champion and Chinese professional Fan Hui, said scientists on Wednesday, describing the event as a milestone in the development ofartificial intelligence
"AlphaGo won five-nil, and it was stronger than perhaps we were expecting," said Google DeepMind CEO Demis Hassabis.
"It's been the grand challenge, or Holy Grail if you like, of AI since Deep Blue beat Kasparov at chess," Hassabis noted, referring to the 1997 game in which IBM supercomputer Deep Blue claimed a victory over worldchess
champion Garry Kasparov.
The game Go, which originated in ancient China, is considered one of the most difficult games for a computer program to master due to the sheer amount of possibilities. The mathematical permutations are complex, and depend on the size of a board and the length of a game - typical Go games end with consent from the two players, not when all possible moves are exhausted - but the total number of theoretically possible scenarios can exceed the number of observable atoms in the universe.
"It's a very beautiful game with extremely simple rules that lead to profound complexity. In fact, Go is probably the most complex game ever devised by humans," said Hassabis, a former child chess prodigy.
David Silver, who, with Hassabis, co-authored the paper revealing the details of the program's victory in the science journal Nature, said the game leans on "intuition" over "brute force," a feat that was believed to be at least a decade away for AI.
"In the game of Go, we need this amazingly complex, intuitive machinery which people previously thought was only possible within the human brain, to even have an idea of who's ahead and what the right move is," said Silver.
This intuitive "human" thinking is not required for AI chess; powerful chess programs can cope with computing all possible variants before deciding upon a move.
AlphaGo is set to take on South Korea's Lee Sedol, considered the world's best Go player, in March.
"I heard Google Deepmind's AI is surprisingly strong and getting stronger, but I am confident that I can win, at least this time," said Sedol.
ls/msh (Reuters, AFP)