Now That Google's Artificial Intelligence Can Beat You At Atari, What's Next?
(CBS SF) -- Google's newly acquired artificial intelligence company DeepMind is reveling in software that in a couple of hours mastered several Atari games. But it's not just some programmers geeking out over the novel uses of AI. Instead, it's an introduction to what this rapidly evolving technology is capable of in the real world.
The experiment began with DeepMind wanting to know if an AI could learn to play computer games all on its own. So they hooked up the software, named Deep Q, to several Atari 2600 games, but provided it with no specific instructions on what it should do.
With the game Breakout, Deep Q's algorithm struggles to return the ball at first, but after a few hundred plays, it eventually learns to break a tunnel into the side of the brick wall and then aim the ball behind the wall to beat the game.
VIDEO: Deep Q Learning To Play Breakout
The DeepMind software program uses AI techniques known as deep and reinforcement learning to take control of a real-world environment, according to a paper published in Nature.
The reinforcement part gives AI points, the equivalent of dog treats, whenever it does something good. After a a couple hours of practice, Deep Q becomes better than most human players.
Google bought London-based DeepMind for $628 million last year. But don't expect DeepMind software to drive one of Google's autonomous cars any time soon. Instead, it's likely the researcher's work will be used to develop Google's core products like search and smartphone assistant functions.
"Imagine if you could ask the Google app for something as complex as, 'Okay, Google, plan me a great backpacking trip through Europe,'" DeepMind founder Demis Hassabis told MIT Technology Review.