"We are the Champions" is probably still ringing through the halls of Google's DeepMind offices, days after AlphaGo defeated Go world champ Lee Sedol one last time.
The artificial intelligence program won the first three of five matches against Sedol, but after Sedol's comeback in the fourth match, it was anyone's guess who would take the final victory. Turns out it was the software.
"After a loss in Game 4, and a move early on that looked like a mistake, but could have been a creative and effective new move, AlphaGo won Game 5 against the legendary Lee Sedol," DeepMind CEO Demis Hassabis wrote in Google's latest blog post.
#AlphaGo wins game 5! One of the most incredible games ever. To comeback from the initial big mistake against Lee Sedol was mind-blowing!!!
— Demis Hassabis (@demishassabis) March 15, 2016
Lasting more than two hours, Sunday's match remained close until the very end; after 280 total moves, Sedol resigned, having gone through two extended overtime periods.
"It was difficult to say at what point AlphaGo was ahead or behind, a close game throughout," Michael Redmond, American commentator, said of match five.
Following what experts say "looked like a mistake" with move 48—similar to the AI's misstep in game four—the competition "developed into a long, very difficult end game," Redmond said.
"This Challenge Match has brought us Go players to new areas we've never explored. We are now seeing a lot more interesting in playing Go," Korean commentator Kim Seongryong added. "And even in one week, I feel like my Go playing has improved."
Invented in China more than 2,500 years ago and enjoyed by more than 40 million people worldwide, Go requires players to place black or white stones on a board and capture the opponent's pieces or surround empty spaces to build territories.
But despite its apparent simplicity, the game features more possible moves than there are atoms in the universe, making it a tough competitor for traditional "brute force" artificial intelligence methods.
Google's DeepMind AI division, however, took on that challenge with AlphaGo. The program uses a state-of-the-art tree search with two deep neural networks—one to suggest intelligent moves, the other to evaluate each option. AlphaGo then chooses what it deems the most successful maneuver.
When pitted against top artificial intelligence Go programs, Google's software won all but one of its 500 games. Even reigning three-time European Go champion Fan Hui was no match for the machine, which triumphed five games to zero, marking the first time a computer program beat a pro Go player.
But DeepMind built AlphaGo not just to kick human butt at some games: The platform's AI can be applied to real-world problems—just like IBM turned its Jeopardy-conquering machine into an actual superhero.
"AlphaGo has the potential to be a huge study tool for us professionals, when it's available for us to play at home," Redmond said.
The computer program's victory against Sedol means Google DeepMind will donate the $1 million prize to UNICEF and STEM charities, and Go organizations.