Demis Hassabis, one of the co-founders and current leader of Google DeepMind, said after the fifth and final match between AlphaGo (DeepMind’s computer player) and Lee Sedol, that it was the “most mind-blowing kind of game experience we’ve had so far.” This is an incredible statement coming from Hassabis—one of the top AI researchers in the world—as he has dedicated his career to “solving intelligence” by first solving games.
The fourth win in the five-game match came for AlphaGo after it made a big mistake early in the game, which some thought would be enough to give Sedol his second win in the tournament.
#AlphaGo made a bad mistake early in the game (it didnt know a known tesuji) but now it is trying hard to claw it back… nail-biting.
— Demis Hassabis (@demishassabis) March 15, 2016
But AlphaGo came back. And even if Sedol had taken this last game, AlphaGo had already locked up the $1 million prize by beating Sedol in the first three consecutive games. More importantly, with wins already under its belt, Google DeepMind had established that it is possible to have an AGI (artificial general intelligence) defeat human minds in what is considered to be the most complex professional game in the world.
This is an especially mind-boggling feat considering the fact that AlphaGo was not specifically programmed to play Go, but learned to play from “watching” and playing millions and millions of games.
But if Sedol had pulled out that last win, it still would have been significant. Even though Sedol had already lost, a second win might have meant that there was a chink in AlphaGo’s algorithmic armor; a way for a human player to consistently find and exploit a weakness.
But this was not the case, and AlphaGo was able to correct its error, and outplay Sedol to finish off the game.
As for the future of Google DeepMind and AlphaGo, according to one of Hassabis’ talks on YouTube, the plan is to essentially improve the AGI by having it play more and more algorithmically challenging games, such as StarCraft. And at some point, Hassabis says that the AGI could eventually transition over to machines that interact with the real world. Those machines will probably be like advanced versions of Boston Dynamics’ Atlas—Alphabet does own both Google DeepMind and Boston Dynamics, after all.
As for the game of Go, Chris Garlock, one of the commentators of the match, said that these games will be studied for years to come. He also said that, “This match this week, has done what the game of Go always does. It brings people together. We’ve all come together… [from] around the world.” But the game doesn’t just bring people together. As of now, robots are a part of that club too.
What do you think about AlphaGo’s final win? Let us know in the comments section below!
HT: Popular Science
Images: Google DeepMind