This Video Game Made by AI Looks Fun as Hell

A computer created 'Killer Bounce' and 'Death Walls' after absorbing hours of video of humans playing classic games.

|
Sep 18 2018, 12:00pm

Screengrab: YouTube

There’s an algorithm at the Georgia Institute of Technology that’s making video games.

Designed by PhD student Matthew Guzdial and associate professor Mark Riedl (both at GT’s School of Interactive Computing), the program absorbed hours of footage of people playing classic video games like Super Mario Bros., Kirby Adventure, and Mega Man to “learn” their key features. Then, the machine learning system used that footage as the basis for its own games.

Today we’ve got a better idea of what those games look like thanks to Guzdial. He’s uploaded video of himself playing two of the machine learning-generated games to his YouTube channel.

The graphics are simple, but you can tell what the AI was going for and the games actually look pretty fun. In the first game, called Death Walls, the player tries to outrun a fast approaching killer wall.

I’d love to play the second game, Killer Bounce, where the player’s avatar constantly bounces and destroys every block it touches. Put some nice sprite artwork on top of that game and I’d buy it for my Nintendo Switch right now.

The machine that can make games is the latest product of Riedl and Guzdial’s research into computer programs that can be creative. While AI won’t be generating AAA titles any time soon, algorithms may one day help humans design games.

Get six of our favorite Motherboard stories every day by signing up for our newsletter .