A solid 25 years after its release, people are still messing around with classic Mario games. Japanese fans re-scored Queen's "Don't Stop Me Now" using the game's sound effects, a level auto-player, and four screens playing simultaneously. We've shown before that computers can learn how to game. For some reason, Mario games just end up being really easy sandboxes for people to spend hours in.Gaming YouTuber SethBling created a program called "MarI/O" that taught itself how to play Super Mario World. As you'll see in the video above, the program decides which buttons to press by referring to what SethBling calls "neural networks," which are adapted by a "fitness" score. The program measures fitness by how far Mario makes it toward the end of the level and how fast, and when it runs into an obstacle, it can learn to evade by jumping, modifying jump duration, and so on. These decisions make nodes, which are referred to on following runs.Only the networks with the highest fitness are selected to create the next session. SethBling's program started out knowing nothing about the game, and failed many times before eventually learning the best route through the first level. The program also recognizes when fitness tapers off (when Mario dies), and adds a mutation by making Mario jump or do something different on the next level generation. Essentially, it's a machine version of evolution, making micro- or macro-adjustments as it needs to get through.If you're a biology student or enthusiast, watch this. You might get a kick out of seeing a video game version of gene editing.