If you found my previous post about AI agents and bikes interesting then I recommend you to read this article Nature - Reinforcement learning agents playing Atari games. It describes the importance of non-engineering methods in solving the high dimensional and noisy problems that we - human agents - encounter and solve in every day lives. You can also easily follow the references bellow to widen the reading about this topic.
And here you can see the algorithm learning to play the games. It's just an amazing thing if you realize that the computer is learning it from the raw input, without any further information about what is that world about, or even what do the different pixels actually mean.
