Algorithms seem to know everything about us; they read and summarize our emails, they feed us videos aligned with our interests, and they’ve even taken over as in-person therapists. For many, AI is a friend (a pal and a confidante), but for others, it’s an all-knowing program that seems to hack our very thoughts—even our worst nightmares.
In 2016, a research team led by MIT and CSIRO put that theory to the test. They found that with the right guidance, machines can absolutely learn the very images that terrify us, and they can spit them out with alarming accuracy.
A Nightmare is Born
“For centuries, across geographies, religions, and cultures, people try to innovate ways of scaring each other. [...] This challenge is especially important in a time when we wonder what the limits of Artificial Intelligence are: Can machines learn to scare us?” Thus, Nightmare Machine was born, a study put together by a small team of professors and research scientists to see just how well AI could learn our deepest fears. It didn’t take long for them to discover that machines do, in fact, learn quickly.
At the time, Nightmare Machine categorized its horrors into two pages: Haunted Faces and Haunted Places. The team opened polls for site visitors, allowing them to vote for which newly generated face was scariest. The algorithm took beloved cartoon characters and famous buildings and warped them into something terrible. Kermit the Frog had bloodied, gnarly teeth in a crooked smile. The Golden Gate Bridge rippled with dark waves amid brooding storm clouds. Not even Bart Simpson was safe. Every face went from an innocent smile to a crumpled ball of flesh that only vaguely resembled human features. It wasn’t just uncanny valley—it was much worse.
But how did it get such accurate results? We have something called “Deep Neural Network” to blame, an algorithm drummed up at Cornell University in 2015. In a nutshell, the system “creates artistic images of high perceptual quality” by reconfiguring random human-made images into its own artistic ones. The study itself aimed to provide an “algorithmic understanding of how humans create and perceive artistic imagery,” and Nightmare Machine just carried on the vision.
Putting the Nightmares to Bed
If you’re wondering where the whole thing went, don’t. It’s not around anymore, which is probably for the best. Upon training AI to scare everyone in the site’s path, the researchers accomplished what they wanted and put the whole thing to bed. You can still find the webpage, but most images associated with it are only available on the odd Reddit post or lost in ten-year-old articles. Nightmare Machine’s own Instagram isn’t even up anymore, so you’ll need to get your fix somewhere else—and with AI what it is today, you likely won’t need to go very far to find the next terrifying story about what these machines can do.
The whole thing was an interesting thought experiment about what humans find horrifying, and just how quickly AI can be trained to recreate the terror.
