SNES Classic 'Super Mario World' Beaten With Artificial Intelligence

[{{909}}]

SNES Classic 'Super Mario World' Beaten With Artificial Intelligence

20 JUNE 2015 CAUTERIZE NEWS - RETRO





While we all know how to successfully clear the first few levels of any Super Mario title, how would a computer fare on its own with no prior knowledge of the game? Using self-learning artificial intelligence, one computer program has managed to successfully battle its way through Super Mario World - albeit with a lot of lost lives.
This new bit of software known as MarI/O was put together by a Seth Bling, a coded interested in the NeuroEvolution of Augmenting Topologies. By incentivising the computer to explore to the right of the playing field, should anything prevent progress (such as a Goomba or Koopa Troopa), the code would safely return to this spot and try something else to advance. After thirty or so lives, the computer code had managed to find its way to the exit of Donut Plains 1.
In Bling's video below, he first shows the computer absolutely storming through this early level of Super Mario World. Looking almost too good to be true, there really is no player behind the action on-screen other than the software MarI/O. Following this completion, Bling goes on to discuss the ins and outs of how this software works, providing a fascinating look and breakdown of one of our all time favourite video games.

Artificial Intelligence Beats Super Mario World

https://www.youtube.com/watch?v=qv6UVOQ0F44