Tech by VICE

Using FaceApp With Dogs Is Truly Unsettling

FaceApp doesn't play kindly with canines.

by Meaghan Beatley
Apr 28 2017, 3:00pm

FaceApp, the controversial face-morphing mobile app that drew criticism earlier this week for lightening the complexion of users with dark skin tones—reacts in some truly bizarre ways when you attempt to feed it non-human faces.  

Questions abound. What prompted the genetic mutation resulting in eye-mouth? Does closing one's eye-mouth qualify as blinking or chewing? And who the hell is that guy?

FaceApp allows users to alter an image of their face to make them smile, appear older or younger, or swap gender. To do so the app uses the neural network artificial intelligence technique, which repeatedly sifts through data looking for patterns and relationships. It's the same technique Facebook uses to help people with vision impairment "see" photos on their timeline.

But the technique isn't perfect, as that pair of teeth and human face in Winston's eyes so deftly illustrates. 

As AI expert Alex Champandard explained, what's likely happening to poor Winston is that FaceApp's algorithm is trained to search for human faces. When it encounters something that's not a human face, the algorithm effectively short circuits. Sad.

"For humans, [FaceApp] matches well because the algorithms are built and tested for that. For anything else, it's random luck how it'll work!" Champandard, who's the co-founder of investigative project Creative AI, told Motherboard. "In this case it turned out to be the eye of the dog that looks a bit like a human mouth, then the rest of the algorithm proceeds from there."

This pattern recognition isn't too dissimilar from how the human brain works.

"It is actually vaguely related to what brains do in that perception isn't just a passive process," Simon Makin, who frequently writes about brain science and artificial intelligence for Scientific American, told Motherboard. "Our brains have expectations about the world and are constantly looking to impose patterns on incoming data. Hence faces in toast and elephants in clouds, and ghosts and all that malarkey," he added.

FaceApp, which did not respond to our request for comment on what it did to poor Winston, occasionally does work successfully on dogs, further highlighting the randomness of outcomes when fed non-human faces.

But these results are arguably more distressing than the eye-mouth mutation. We'll stick to human faces.