Google’s AlphaGo AI may have secured the five-game match with its third win yesterday, but that doesn’t mean the competition is over. As announced today (via Demis Hassabis, CEO of DeepMind), the second-best-in-the-world South Korean Go player Lee Sedol actually managed to score a victory against the Google AI in the fourth game…
According to Hassabis, the mistake came at move 79, but the AlphaGo AI didn’t realize its mistake until around move 87.
“Lee Sedol is playing brilliantly! AlphaGo thought it was doing well, but got confused on move 87. We are in trouble now,” Hassabis said on Twitter. He later clarified, saying the “Mistake was on move 79, but AlphaGo only came to that realization on around move 87”.
Some experts say that today’s AI loss may actually be just as exciting as its previous three wins. Mile Gu, an Asst. Professor at Nanyang Technological University took to Facebook to explain that “we now can finally see some of its weaknesses and gain insight on the whole Monter Carlo deep learning algorithm itself.”
AlphaGo is dreadfully impatient. It needs to optimized win probability. Thus, will all reasonable moves have low win probability (as she is losing). AlphaGo will be pushed to play moves that are more ‘likely’ to win – that is moved where it can reverse the game unless the opponent plays at the exact right point of the board.
AlphaGo now has a record of 3 wins and 1 loss against Lee Sedol, but has an overall record of 9 wins and 1 loss considering the 5 straight wins against the European champion Fan Hui last year. The fifth and final match is set to be livestreamed on YouTube on March 15th, so we’ll see in a just a couple days whether or not Lee Sedol can repeat his strategies and pull a second win.
Watch the full fourth game below.