The concept of deep learning, which is a subset of the Artificial Intelligence discipline, involves machines being able to extract contextual or inherent meaning from inputs. It’s much, MUCH more complex than it sounds; humans are great at it, machines not so much. But now the process has finally been put to some good use. Forget facial recognition or pattern recognition or interpreting blocks of text to find out what’s going on, now the machines can learn to play old Atari games.

DeepMind Technologies, a company that develops machines that try to learn as humans do, has published a paper showing the machine has learned to play games on the old Atari 2600. It may not sound like a big deal, except the machine learned only by evaluating the pixel on the screen and had no other inputs. No rules, no game-specific programming, nothing but what the screen outputs. The amazing thing is that it adapted to the visual cues to learn the games just as a human being would do.

The games weren’t terribly difficult: BeamRider, Pong Sports, Breakout, Space Invaders, and Seaquest. More complex games on the console, like Q*Bert, still cause it to stumble. But it’s ability to deduce rules solely from screen output is remarkable and unnerving at the same time, and shows promise.

Google recently acquired the company to help it parse data from images, and all large tech companies including IBM, Microsoft, Apple, and others use deep learning to help their systems perform better whether it’s financial analysis or speech recognition.