Some of the best minds in Australia and the U.S. have put their combined brain power to scientific (and evil) use.
Researchers at CSIRO’s Data61 and the Massachusetts Institute of Technology (MIT) Media Lab have built what they called the Nightmare Machine, an artificial intelligence designed to transform beloved childhood characters into downright monsters.
Not only can their AI now create its own images, but it’s specifically trying to freak us out.
According to the CSIRO, the “spooky formula” is generated by two deep learning algorithms: One that takes a visual style from one image and adds it to another, and one that can create faces.
More than 200,000 images have been assessed for their fear factor, it said in a blog post, to help the AI learn what is truly frightening to mere mortals.
The Nightmare Machine site comes with a warning message: “Images on this website are generated deep learning algorithms and may not be suitable for all users. They contain scary content.”
It’s not wrong. For starters, it’s created a diabolical Angelina Jolie, two truly freaky presidential candidates and a very intense Ellen DeGeneres.
Ultimately, the team is asking “Can machines learn to scare us?”
To that end, the AI is quickly coming to grips with horror film-style imagery.
“We use deep learning algorithms to learn first how haunted houses, then ghost towns, and more recently toxic cities look,” Data61 principal research scientist Manuel Cebrian said in the website’s post.
“Then, we apply the learned style to famous landmarks. It’s surprising how well the algorithm can extract the element from the ‘scary’ templates and plant it into the landmarks.”
Anyone is able to vote online for which AI-generated images are scariest, helping the algorithm learn and develop.
So go ahead. Help something called the Nightmare Machine get smarter.