Google DeepMind Learns How to Play Atari Games

Posted on February 25, 2015

A computer program named Google DeepMind has learned how to play and master some Atari games on its own. Some of the games the program was able to master include Breakout and Seaquest. It was able to learn how to play the games without much input about the games themselves.

Google acquired the DeepMind company last year. The mission of Google DeepMind is to "combine the best techniques from machine learning and systems neuroscience to build powerful general-purpose learning algorithms." BBC News reports that Google calls its machine an "agent."

Dr Demis Hassabis, DeepMind's vice president of engineering, tells BBC News, "The only information we gave the system was the raw pixels on the screen and the idea that it had to get a high score. And everything else it had to figure out by itself."

DeepMind was presented with 49 different Atari games. It was comparable or better than a human in 29 games. However, it failed to play better than at human some of the games, such as Pac-Man and Montezuma's Revenge.

A report on DeepMind's Atari gaming can be found here in the journal Nature. Here is a video showing DeepMind getting better and better at Breakout after more and more training sessions: