DeepMind researchers at Google's Alphabet subsidiary announced on Wednesday their AlphaGo program beat European Go champion Fan Hui in a series of five matches and achieved a 99.8-percent winning rate against other Go programs.
Go is seen as a formidable test for artificial intelligence researchers because it is much more complex than chess, with a larger range of possible positions that require more sophisticated strategy and reasoning.
The reason games are used as a testing ground is that they're kind of like a microcosm of the real world," says DeepMind Technologies founder Demis Hassabis.
AlphaGo mixes a deep-learning algorithm with a Monte Carlo algorithm designed to rigorously explore large numbers of possible combinations of moves. The DeepMind researchers say they also trained the program using input from expert human players. "The machine has continued to get better," Hassabis notes. "We haven't hit any kind of ceiling yet on performance."
Alphabet has announced a March tournament between AlphaGo and current Go champion Lee Sedol, in which they will compete for a $1-million prize.
From The New York Times
View Full Article - May Require Free Registration
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found