The MIT "Nightmare Machine" Needs Your Input for Its Horror Algorithm
MIT researchers have created a website devoted to scaring the bejeezus out of people. The so-called “Nightmare Machine” uses artificial intelligence to generate horrifying faces, make images depicting places like the Taj Mahal a little spookier, and learn based on user feedback.
Nightmare Machine’s creators explain that the creepy visages started with “state-of-the-art deep learning algorithms” used to generate new faces. They then added “a hint of scariness onto the generated faces, in the spirit of Halloween,” and decided to share their abominations with the world.
The generated faces are then voted on in a “hot or not” styled interface, asking people whether or not they were scared by what they saw. Some of the images are genuinely creepy; others look like bad Photoshops. Every time someone votes on a generated face, the algorithms learn more about which features actually scare people, and which don’t.
Rating ten faces creates a “special hall of faces” that can be shared to Twitter. Responses range from calling this the best use of deep learning to questioning whether “the images, or the fact that machines can generate such images,” is scarier.
The modified images of real places fare worse — they look like they’ve been run through an Instagram filter pack, rushed out before Halloween.
Nightmare Machine is the successor to improved facial recognition and other technologies centered on the human face, like a system from Stanford researchers that swapped people’s facial expressions in real-time.
Now A.I. can simply create new faces. Eventually it might even create something that looks as life-like as this stunning computer generated image of a Japanese girl. At that point, we’ll be living in a nightmare dreamt up by A.I. instead of by our own feeble imagination.
Inverse reached out to Nightmare Machine’s creators and will update this story when we hear back. In the meantime, head to nightmare.mit.edu to unlock a dark corner of your mind.