You donâ(TM)t program AI, you design the system that solves your task. In this paper authors were evaluating performance of a particular learning method and showed itâ(TM)s weaknesses. I just hope it would generate more research in that area that would solve these problems.

It is not a âsoftwareâ(TM), it is a neural network that uses reinforcement learning ideas. It is a system that learns how to play a game from images. Authors did not pre-program their system to play a game as opposite to the software.