Two scientists built a tool to help people navigate the era of deepfakes.

A/B Testing

Deepfakes and other AI-generated images have become commonplace the algorithms that churn them out have become widespread.

On one sugar-coated hand, this means cooler movie and video game visual effects. On the other hand, it means that bad actors can produce photorealistic propaganda, fake porn of real people, or other convincing but fake media.

That's why two University of Washington scientists created a website, "WhichFaceIsReal.com," which is meant to train people to spot the telltale signs that an alleged photo was actually built by an algorithm — by asking them to guess which of two side-by-side photos a real person and which is an AI-created dupe.

Facial Recognition

The website has been up and running since February, shortly after a website called "ThisPersonDoesNotExist.com" went live.

That site highlights an AI algorithm built by Nvidia called StyleGAN that can create almost-perfect portraits of human faces out of thin air.

WhichFaceIsReal focuses on the specific errors and glitches in StyleGAN portraits, so the tips and tricks — look for asymmetrical glasses, ears, and jewelry; check whether the background is full of mysterious splotches of color — don't always generalize to other face-generating AI algorithms.

But if nothing else, WhichFaceIsReal can serve as a reminder that AI-generated media is coming soon, and we're all going to have to be more critical about the images and videos we see every day.

Try It Out: WhichFaceIsReal.com

More on StyleGAN: A New AI Draws Cats, and They’re Utterly Grotesque


Share This Article