RESEARCHERS ARE TEACHING ARTIFICIAL INTELLIGENCE HOW TO TERRIFY HUMANS

File: 1476202265125.gif (494KB, 475x355px) Image search: [Google]

494KB, 475x355px

Machines do creepy things all the time. The more they seem to be able to think and act for themselves, the freakier it gets.

Consider, for example, this gif I just made of Boston Dynamic’s SpotMini robot. Legitimately impressive? Of course. Terrifying? Also yes.

But SpotMini’s just trying to do its thing. It can fetch you a beer! And load the dishwasher! Who cares if it also sometimes does that weird thing where it walks in a circle around its own head?

Especially when there are machines out there actively trying to terrify you. That’s the idea behind the Nightmare Machine, a deep-learning project by researchers at MIT who wanted to see if they could train a computer to generate horrifying images.

To do so, they trained an algorithm on a corpus of the kinds of images they wanted it to produce. They were going for a few different aesthetics: the hazy appearance of the sky above a haunted house, the unsettling palette of a cityscape drenched in toxic waste—you know, the sorts of scenes that might evoke witchy-fingered tree branches scraping on your window after midnight and ghostly disturbances in your peripheral vision.

So far, the algorithm seems to be doing a pretty good job.

The images being produced aren’t just otherworldly or psychedelic, like so many computer-generated images in the realm of machine learning, but straight-up spooky. For example, check out how the Nightmare Machine transformed an ordinary image of the Statue of Liberty into something out of Ghostbusters.

http://nightmare.mit.edu/faces

http://www.nextgov.com/emerging-tech/2016/10/researchers-are-teaching-artificial-intelligence-how-terrify-humans/132604/