Google's Deep Q-Network Can Kick Your Butt at Atari Games

Google is such a fun company. The company establishes an extremely fun workplace, and it's incredibly innovative, even if innovative products have failed (i.e. Google Glass). Even on the search engine, on certain days, the Google logo may have a quirky little addition to it, like playing a small video game in the logo.

Either way, Google has taken the next step to blow our minds. How?

Well, Google has recently attained the ability to beat Atari games.

And the difficulty of Atari games are no laughing matter.

With an algorithm that Google calls "deep Q-network," a computer was able to achieve human-level proficiency at more than 24 Atari games. To describe the process, Google provided the computer with basic knowledge of how to play the particular game, but the cool bit was the computer's ability to "see" the pixels on the screen. The algorithm basically told what actions the virtual buttons performed and the score.

In other words, Google was a strategy guide. And the computer was able to LEARN.

Now, should we be frightened about this fact?

Maybe. We should at least be afraid of DQN entering eGaming tournaments.

Overall, DQN played either at or higher than human levels (at more than 75% of the level of a professional human player) of performance in 29 of the 49 games played. In certain games, DQN conjured far-sighted strategies that allowed it to achieve the maximum score, such as in Breakout, where "it learned to first dig a tunnel at one end of the brick wall so the ball could bounch around the back and knock out bricks from behind." However, DQN surprassed existing machine learning algorithms in 43 of the 49 games.



Quoting Dharshan Kumaran and Demis Hassabis of Google DeepMind:

We also hope this kind of domain general learning algorithm will give researchers new ways to make sense of complex large-scale data creating the potential for exciting discoveries in fields such as climate science, physics, medicine and genomics. And it may even help scientists better understand the process by which humans learn. After all, as the great physicist Richard Feynman famously said: "What I cannot create, I do not understand."

What is most surprising is that even though DQN rocks at games such as Q*bert, Pong, and Breakout, it's laughably bad at Ms. Pac-Man and Asteroids.