How good are you at Super Mario World? Chances are, you’re not as good as a neural network and genetic algorithms.
MarI/O may be just that, and it’s really good at Super Mario World. In the video above, you’ll see MarI/O beating a level of the game with ease, and a quick explainer on what made it so efficient (I won’t say ‘good’ because it left a lot of points on the board).
F**k it, we'll do it live!
Our biggest ever edition of TNW Conference is fast approaching! Join 10,000 tech leaders this May in Amsterdam.
Creator Seth Bling has published the source code for his efforts so you can try it yourself. Seth tells us he’s got a degree in Computer Science, and has always been interested in machine learning. MarI/O was his first project involving genetic programming, and only took him “a couple weeks” to get up and running.
MarI/O is a neural network that appears to be learning how to play Super Mario World by trial and error — just like you or I would. After playing the game for a bit, MarI/O learns which enemies do what (and when), then seems to decide on the best method for bypassing that enemy — just like you or I would.
MarI/O may not be as well known, but it seems to work similarly to Google’s DeepMind, which taught itself to play Atari games better than actual humans. I guess machines aren’t as snobby about graphics as we are.
But don’t take my word for it; watch the video, and marvel at MarI/O knowing it may only be a matter of time before the machines learn our routines and turn destroying humans into their own little game.
Read next: Minecraft meets Jurassic World