Up above, you can see the results of millions of hours of labor and billions of dollars. Yes, man has finally created a computer that can learn to play Breakout almost as well as you can!
Joking aside, this is actually fairly impressive. The scientists at DeepMind, Google’s AI arm, essentially put the “agent” in charge of the game and didn’t tell it the rules. It had to figure out everything by itself by playing the game over, and over, and over again, with a positive feedback loop when it did the right thing. As you can see, it takes a while, and the AI lacks even the most basic understanding of the game at first, but once it figures out, “Oh, the paddle is supposed to be under the ball!,” we’re off to the races.
Obviously, this brute force approach won’t work for real-world tasks. Can you imagine this thing trying to learn how to drive using this method? But Google doesn’t want AI for real-world tasks, for the most part. They want it to bolster certain parts of its search business. In other words, the next step for this AI is to find spam and porn links. We can’t wait to see the time lapse video of that one.