MarI/O AI Program Learns To Play 'Super Mario World'
No matter how good your Super Mario World skills are, we bet MarI/O can beat you.
Created by developer Seth Bling, MarI/O – software made up of a neural network and genetic algorithms – can complete Super Mario like that's its job.
Bling uploaded a video showing MarI/O playing a level of the popular game with impressive skill. Possessing a degree in computer science and an interest in computer learning, Bling was able to create the neural network that learns how to play a game through trial and error.
Learning how to play just like we do, MarI/O can figure out the best ways for escaping enemies and completing the level.
Bling says the program started out with no knowledge of Super Mario or Nintendo. It learned how to play the game through a process he calls "neuro-evolution." The AI program uses a neural network, or mathematical network of how our brain works, to see which blocks it can stand on, where the enemies are and which buttons to press to navigate.
In the video, we get a look at the AI program's brain — seeing how it makes decisions while moving through the level.
Bling also explains how the program learned over a 24-hour evolution period. The first version of MarI/O starts off dumb, just standing there until it learns how to walk. Over the program's training time, MarI/O learned to not only play Super Mario World, but it also developed the ability to master the level with ease after 34 generations.
It's important to add that the program isn't the best player, since it does leave some coins behind. Still, it's amazing to know that a program was able to learn and master a game using the same principals of our biological evolution.
Watch Bling's MarI/O learn to play Super Mario World in the video below.
Photo: Seth Bling | YouTube