Photo by: WorldGo.org
After AlphaGo's success last week, Google are looking for another challenge in Starcraft.
On Saturday, AlphaGo reached a milestone for AI that was supposed to be 10 years away. AlphaGo defeated a 9-dan champion in the ancient Chinese game of Go, a game much more complex than Chess. AlphaGo, an artificial intelligence system developed by British based Deepmind, who were aquired by Google in 2014, is currently 3-1 up against Lee Sedol, with the final game being played on Tuesday 15th March.
The series is not yet finished, but Google, who acquired British based DeepMind in 2014, are already looking for the next hurdle to surmount, and they're considering Blizzard's Starcraft. The 1998 RTS game will provide many far more complicated problems for an AI to master. In Chess and Go, everything is presented on the board at all times, and so a program can make decisions knowing that it has almost all information, however in a game like Starcraft, limited information makes it very difficult for, human and non human, players to guess what their opponent is doing and how to best respond to their actions. For Google to beat a world class Starcraft player it would need to be able to take this limited information and make an educated guess, which is completely uncharted territory for an AI.
While extremely challenging, there are also areas where an AI would have some significant advantages over a human player in starcraft, so it'll be interesting to see how they handle these if Google do pursue this game as their next AI challenge. Human players are limited by their APM, how many actions they can do, or how quickly they can process information on the screen, limitations that a computer program would not necessarily have.
While it sounds like a big challenge even for Google, we might see the first AI defeat a GSL champion sooner than you imagine. The current standings between AlphaGo and Lee Sedol have AlphaGo winning 3-1 with the fifth and final match being played out on Tuesday 15th March at 1pm KST (4am GMT/12am EST). The match can be watched live on Deepmind's youtube channel.
For more information about Deepmind and AlphaGo, you can check out these articles:
The Verge interview DeepMind founder about the future of AI.
Google's paper in Nature magazine goes into detail about how they taught it to play one of the worlds most complex games.
