New AI Builds Video Game Levels After Watching YouTube Videos

Gaming channels on YouTube pull in billions of views each month, but human beings aren’t alone in their appreciation of Let’s Play videos. Researchers at Georgia Tech have designed an AI that builds video game levels after watching YouTube videos of those games.

The researchers tested their program with perhaps the most iconic video game of all time: Super Mario Bros. The AI “watched” videos of the classic platformer and then generated its own levels based on the information it gleaned. It learned to put platforms on top of supporting pillars and made sure to space its gaps so that Mario’s jump would be able to clear them.

An article published by Wired includes some examples of the terrain generated by the AI. There aren’t yet any videos of players completing these levels, though the researchers behind the program will present their work at the upcoming Foundations of Digital Games Conference

“Our system creates a model or template, and it’s able to produce level sections that have never been seen before, do not appear random, and can be traversed by the player,” said Mark Riedl, associate professor of Interactive Computing at Georgia Tech. “One could say that the system ‘studies’ the design of Mario levels until it is able to create new playable areas.”

Computer programs aren’t just using YouTube to build video games; they’re also watching videos in order to learn how to perform basic tasks like cooking and pouring drinks. And AIs can play video games, too; this thing is pretty good at Mario, even if it sucks at Tetris.