It took four matches, but a human has finally done what many experts first thought was a certainty, then fairly impossible, and now mind-boggling: world champion Lee Sedol has finally beaten Google’s AI software AlphaGo in the ancient Chinese strategy game Go. In the five-match scheduled event, the first three matches went to Google DeepMind division’s software; in the fourth match, the human world champion finally won.
The game itself has a nearly-infinite number of possible moves, and therefore relies on intuition and emotional state to help the Go master defeat his or her opponent. This lack of “feel” for the game is what had so many game players predicting an easy win for the human player. But the very first matchup showed something that gave the software a leg-up, and that was an inability to second-guess himself due to an early mistake. Both players in the first game made crucial early errors, but unlike the human player, AlphaGo couldn’t experience self-doubt or embarrassment at his own mistake, something that commentators attributed to the machine’s win.
In the fourth match, held yesterday in Seoul, it was three hours into the game before it looked as though Sedol might win this one. Part of AlphaGo’s arsenal of game-winning strategies is an ability to correct itself for future moves through trial and error; once the software makes a move and sees the resulting outcomes, that information is stored in order to be called on when needed later.
This entire project has been an interesting look at how far AI software has come, as well as how far the technology still has to go. While others speculate on what possible abilities DeepMind can come up with next for its software, the bigger picture is this: AI and human interaction stands to be an ongoing partnership. The days of having computers take over all aspects of mission-critical thought processes are obviously not coming soon, but humans paired with AI software may be unstoppable.
Interestingly, the five-match series–which was broadcast live on YouTube and aired on local channels–included a $1 million prize for the winner. As AlphaGo has already won three matches, the software is officially the prize winner. Software tends not to spend its money wisely, and therefore AlphaGo’s winnings will be donated to charity.