Maybe the robots are poised to take over after all.
In the first match of their planned five-game series, Google’s Go-playing computer program defeated Korean champion — and human being — Lee Sedol. Though just one victory — and a close one at that — the outcome shocked not only the Go community, but even artificial intelligence researchers, who not-too-long ago were forecasting that it would take another decade for AI to defeat professional Go players.
Google’s AlphaGo program “did not play like a human at all,” stunned Go expert Kim Sung-ryong told the Associated Press.
The victory adds another triumph to the artificial minds over their supposed human masters and comes almost two decades after IBM’s Deep Blue computer took down chess champion Gary Kasparov.
An ancient Chinese board game, Go is considered to be considerably more complex than chess. Intuition factors heavily in the strategies of top players, which is why many experts thought it would be difficult for computers to master.
But Google’s DeepMind artificial intelligence researchers used machine learning techniques to improve AlphaGo’s game play. The program would play against itself and adjust algorithms based on the outcomes. That strategy worked well enough to allow AlphaGo to beat a European champion last year.
Lee, though, was thought to be a much tougher challenge. The 33-year-old has won 18 of the last 21 world Go championships since turning pro at age 12. But AlphaGo, despite making some mistakes, won in the end. And it did so by making a move that Lee said a human never would have made.
We knew they could calculate much faster than we can. But now they’re starting to be even more creative? Scary stuff indeed!
Photo: Lee Sedol, South Korea’s world champion Go player, at a press conference in February promoting his match against Google’s AlphaGo computer program. (AP Photo/Ahn Young-joon)