This week, MIT’s literal Nightmare Machine did the impossible: by applying its self-teaching algorithms to pictures of both presidential nominees, it made this year’s election cycle even scarier than it already is. The Nightmare Machine is an AI experiment that uses algorithmic software to identify what particular components of images make them ‘scary’ and then applies tha information onto recognizable images, like The Sydney Opera House, or the Louvre. When applied to images of Trump and Clinton on the campaign trail, the Republican and Democrat contenders look a lot like the pot-growing cannibal hillbillies from American Horror Story: Roanoke, instead of the pot-hating cannibal plutocrats we’ve come to know and love them as.
Videos by VICE
The Nightmare Machine learns with help from its audience. Through the project’s website and social media platforms like Instagram, users can vote on what images scared them the most. According to Motherboard, researchers in Cambridge trained deep neural network algorithms, similar to Google’s Deep Dream program, to recognize certain features from conventionally scary images. The algorithms then create their own internal representations, applying these features to, quite literally, turn an image horrible. With Halloween right around the corner, the project churned out so many contenders for our Insta of the Week column that we had trouble choosing just one. Check out some of our other favorites, below:
To learn more about the Nightmare Machine project, head over to the project’s website.
Related:
This Is What Computers Dream of When They Look at Art
Deep Web Horror Game Knows Where You Live
Virtual Reality Horror Film ‘Catatonic’ Comes to Your Smartphone
