I have been thinking lately of doing a project using this technology, it's fascinating to see how a computer can "learn" to play video games from scratch. I came across some videos in which they teach how they teach a computer to play PacMan and It is funny to see how at first it stays still and little by little it recognizes the map, in the case in which I saw, it required 200,000 iterations for it to be able to pass a medium size map. The closest I have been to creating something like this was designing an algorithm in Lisp to solve mazes by using the search algorithm A * (a very fun practice taught in class B-20 by Phd. in CS Salvador Godoy Calderon) in the same way I made a games agent that solves A Tic-Tac-Toe 4x4. I think it is a very interesting way how it teaches and gives us all the resources to be able to learn in our own way and at the same time have fun doing these agents, because there are many ways to solve this problems and each one carries its due complexity.Soon I will do a project in Pac-Man or Mario bros I think it will be entertaining