Battle.net to become Skynet within a few years
Blizzard, in association with research company DeepMind, have developed software based on the real-time strategy game StarCraft II that teaches computers how to think for themselves.
Judgement Day is upon us… Maybe.
The two companies have co-developed a software package, known as SC2LE, to advance the computer’s ability to read a situation and then develop a strategy based on the given information. Blizzard has also given the software over 60,000 game replays (with plans to expand it to 500,000) to peruse for reference and experience.
The hope is that through the study of actions and counter-actions within the millions of moment-to-moment encounters in these replays, the computer will learn, through imitation, how to best react to any given setup or problem posed by its StarCraft opposition. Though I’d like to see it handle a full-on Zerg rush. Sorry “Johnny 5”, that’s not happening.
StarCraft II was specifically chosen because of its multi-layered task management, requiring the player to build and maintain an efficient base, whilst also managing resources, building defences and mounting an army for its own series of attacks. DeepMind’s A.I “agents” were able to perform these functions on a small scale, but struggled when presented with a full game scenario, which requires endless adaptation and multitasking.
Researchers are working to develop their agent’s ability to put all the pieces together, ultimately being able to react to any given situation in a full-game environment and therefore being able to beat every single StarCraft player on the globe. Yes, I’m including Korea.
Whilst it seems odd to spend so much money, time and effort to teach a computer to play a perfect game of StarCraft II, it’s obvious that, on a larger scale, the idea of computer A.I having a self-taught understanding of predictability, adaptation, counter-intelligence and decision-making is absolutely fascinating. Fascinating and terrifying.
Still, I for one welcome Sarah Kerrigan as our new overlord.
SC2LE A.I. research [DeepMind Blog]