Today, March 13, is the anniversary of Game 4 in the 2016 Go tournament between AlphaGo and Lee Sedol. AlphaGo had already won the tournament 3-0, but it was a five-game contest and by agreement all five would be played. At this point, the folks rooting for the human just wanted one victory, one hint, one bit of validation that maybe humanity still had a trick or two up its sleeve, that there was some sort of spark in us that had not been reproduced in AlphaGo’s electronic circuits. And they got their wish. AlphaGo resigned Game 4.
The game began much as Game 2 had. AlphaGo had focused on lots of small point gains and was winning games by small margins. In other words, it was playing a conservative game that didn’t value winning by a lot over winning by a little. It just wanted to win. In Game 4, Lee countered this by using an all-or-nothing kind of strategy, where we would win big or lose big. The algorithm that AlphaGo uses has to take certain shortcuts when evaluating moves. If you know exactly how it works there are ways you can exploit this, by pursuing a strategy that the shortcuts overlook.
The gameplay peaked on Move 78. Lee had stared and stared at the board and placed his piece in a spot near the middle of the board. This was an astonishing play and another top professional Go player described this as “a divine move” and said he had not seen it coming. Later, when asked, Lee said he made the move simply because he couldn’t see any other move to make. But here’s the interesting thing: Recall back in Game Two. Move 37 was a turning point for AlphaGo, and there was only a one in 10,000 chance that a human player would’ve made that move. After Game Four, when the programmers looked into AlphaGo at what the chances were that human player would have made Move 78, the one that Lee had just made that would win the game, AlphaGo saw the probability of a human making that play were one of 10,000 as well. Once again it was a stroke of genius, a creative insight, a flash of brilliance that enabled the winner to defeat its opponent.
It is interesting to note that these two games turned on a move that most players would not have made. That is a course to be expected, because most players aren’t as good as AlphaGo or Lee Sedol. So it may be a mistake to read too much into it, But if there’s any lesson to glean from this, it is a quote by Elbert Hubbard: “One machine can do the work of 50 ordinary men. No machine can do the work of one extraordinary man.” This may be the case, but it might also be the case that no person can do the work of one extraordinary AI either.
When Lee walked out of the room after his victory, it didn’t really matter that he had lost the first three games. He was greeted with universal applause and adoration, and he had, in the minds of many, represented humanity with great dignity and ability. Later he described his feeling at winning by saying, “I am so happy. I will never exchange this win for anything in the world.”