Google has created artificial intelligence software that can play video games better than you.
The project originated in DeepMind Technologies, a machine-learning startup that Google acquired in 2014 for $500 million. Of course, Google didn’t blow a cool half-billion on a company whose sole goal is building software capable of playing retro Atari games with the speed and skill of an over-caffeinated 15-year-old; in theory, a system capable of mastering a game like Pong on its own can also learn more complex tasks, such as driving an autonomous car through busy city streets.
“It’s mastering and understanding the structure of these games, but we wouldn’t say yet it’s building conceptual knowledge or abstract knowledge,” Demis Hassabis, DeepMind’s co-founder, told Bloomberg. “The ultimate goal here is to build smart, general-purpose machines, but we’re many decades off from doing that.”
This isn’t the first time that researchers have used video games as a training tool for an artificial-intelligence construct. In January, a team of researchers at the University of Tubingen, Germany, modified a copy of Super Mario Advance to give Super Mario a rudimentary intelligence. The altered Mario could plan his own route through game levels, and respond to researchers’ spoken questions about his (pre-programmed) “mood.”
Google isn’t the only company hard at work on artificial intelligence. IBM, for example, is devoting an increasing amount of money and resources to Watson, its cognitive supercomputing platform. Over the past year, Big Blue has focused on transforming Watson into a “digital assistant” for hospitals, particularly oncology centers, which is all well and good from a practicality standpoint—but can Watson play a winning game of Pong?