Google’s DeepMind division has pulled off an unforeseen victory that goes far beyond the rules and strategies of any game. Its AI software AlphaGo, which made headlines for its pending match-up against the reigning world champion of the ancient Chinese strategy game Go, has now defeated its human opponent in the first of five scheduled games.
South Korean Lee Sedol has long been considered the world’s Go master, even from his prodigy status as a professional player at age twelve. His loss on Wednesday was a shock to both game fans and AI experts, who’d originally predicted that the complexities and required human emotional aspect of the game would mean artificial intelligence would require years of fine tuning in order to beat a human opponent at this level.
Interestingly, the factor that may have most helped AlphaGo defeat Sedol could be those same human emotions that critics argued are vital to winning the complex game. Reportedly, both the software and the human player made critical mistakes early in the game, but unlike a human opponent, AlphaGo was unhindered by its error and (oversimplifying things, here) simply readjusted its strategy. That ability to remain emotionally detached from a mistake–something that could easily cause a human player to overadjust or question every further move–could have played a role in the software’s strategies moving forward.
While Sedol has been very gracious about what many experts predicted would be a non-event, Google’s DeepMind has got to be ecstatic. Its “reinforcement learning” concept has meant that AlphaGo has a mechanism other AI attempts haven’t had, and that’s the ability to teach itself through intense trial and error. That innovation has certainly proven itself worthy for now, but there are still four more Go matches ahead of it.