Google DeepMind’s AI teaches itself to beat human players of Chinese game in three days
AlphaGo Zero discovered novel moves that human Go masters had never even conceived. Photo: ABC
An artificial intelligence program developed by Google has for the first time taught itself from scratch how to defeat human players in the complex Chinese game of Go.
AlphaGo Zero, the latest version of the ground-breaking program that first defeated champion players last year, used machine learning to master Go within just three days.
After 40 days of operation, AlphaGo Zero’s algorithms had independently developed principals that took human players thousands of years to discover.
Even more impressive, the program also discovered novel combinations of moves that human Go masters had never even conceived.
Built by Google’s DeepMind lab in London, the program signals a big step for computers being able to learn without the need for any human input, according to lead researcher David Silver.
“Now we have the final version of AlphaGo, AlphaGo Zero, which has learned completely from scratch, from first principles,” Professor Silver said in a company video.
“[It has learned] without using any human data and has achieved the highest level of performance overall.
“The most important idea in AlphaGo Zero is that learns completely tabula rasa, that means it starts completely from a blank slate, and figures out for itself only from self-play and without any human knowledge.”
AI program could have myriad applications
Computers long ago began outplaying human chess players, but there were doubts about whether a program could master Go’s mix of strategy, intuition and mind-boggling complexity.
But it is AlphaGo Zero’s capabilities that the DeepMind team believe liberates artificial intelligence from the limits of human knowledge in areas beyond board games.
They believe it could potentially be used to great effect in tackling other major problems, such as curing diseases like Alzheimer’s, Parkinson’s and cystic fibrosis.
Engineer and futurist Mark Pesce from the University of Sydney said AlphaGo Zero is a big technological advance.
“This is actually undiscovered territory, and in a lot of ways the computer is the blackboard for us to be able to test some of our theory,” Dr Pesce said.
“It’s not just the computer doing this, it is us designing the computer that’s doing this. So this is definitely a partnership, it’s not just the rise of the terminator.”
-ABC