Google wins last Go match, Facebook AI head casts bitter barb

Google’s artificial intelligence system AlphaGo won the final Go game of five against human world champion Lee Sedol on Tuesday. “I don’t necessarily think AlphaGo is superior to me,” Sedol told the Associated Press. “I believe that there is still more a human being could do to play against artificial intelligence.” Sedol won only one game in the series.

AlphaGo had already sealed its series victory by winning the first three games of the best-of-five. That triumph dealt a blow to Facebook, which has also been developing AI to play Go, a hugely complex ancient Chinese board game.

In January, Facebook CEO Mark Zuckerberg said the company had created an AI system that was close to being able to beat the best human Go players. But before Facebook could show off its achievement, Google’s DeepMind unit put its AlphaGo system up against 18-time world champion Sedol.

Reaction at Facebook to AlphaGo’s conquest over Sedol was mixed. Graciously, after the machine beat the man for the third-straight game and clinched victory in the series, Zuckerberg posted a message congratulating the DeepMind AI team. Boorishly, after AlphaGo’s third victory, Facebook’s head of AI research Yann LeCun cast a bitter barb in Google’s direction. “Congrats to the DeepMind AlphaGo team for this Grand Slam,” LeCun said on Facebook. “Now, can you do it purely through reinforcement learning, without pre-training the convolutional net on recorded games between humans?”

In AI geek talk, that was a zinger. Loosely translated, it means “You’re not that smart, and neither is your dumb computer.” More specifically, LeCun appeared to be suggesting that AlphaGo didn’t do enough of its own learning – that it mostly just sucked up data from millions of Go moves fed into its machine brain, then spit out its own moves in accordance with the information it had been stuffed with. Reinforcement learning, on the other hand, requires a machine to analyze what it did to bring a positive or negative outcome, so it can learn how to make positive, winning moves and avoid moves that would lead it toward defeat.

LeCun’s criticism hinges on his use of the word “purely.” AlphaGo did in fact undertake reinforcement learning, on top of the pre-training, according to Google. As DeepMind founder Demis Hassabis put it in a blog post, “Our goal is to beat the best human players, not just mimic them. To do this, AlphaGo learned to discover new strategies for itself, by playing thousands of games between its neural networks, and adjusting the connections using a trial-and-error process known as reinforcement learning.”

LeCun took several additional swipes at AlphaGo, posting a link to a blog post suggesting the significance of Google’s victory was being overblown; putting up a comment saying the next iteration of AlphaGo “should be called BettaGo (haha!) and include online adaptation to the opponent;” and linking to an article that called AlphaGo’s victory an “intermediate” accomplishment rather than a breakthrough.

 

Photo: A Go board, with stones (Wikimedia Commons photo/Donar Reiskoffer)

 

Tags: , , , , , , , , , ,

 

Share this Post



 
 
 
  • seahen

    When will Facebook be ready to put their Go stones where their mouth is?

 
 
css.php