Computers have recently gotten much, much better at a somewhat unsettling skill: generating fake human faces. As in, creating an image of a human who has never existed before.
We saw the concept go a bit viral this week with ThisPersonDoesNotExist, a website hooked to a machine that generates a new face every few seconds. Or the feline version that dreams up (sometimes nightmare inducing) cats, ThisCatDoesNotExist.
Now its been turned into a game. Think you can tell which human is… well, human? (Spoiler: all of the images used as examples above are, according to the game, computer generated.)
Aptly called WhichFaceIsReal, the site throws two images side-by-side: one real, and one generated by a computer. It was put together by two professors at the University of Washington, building upon the same work as the sites mentioned above: StyleGAN, an algorithm recently open-sourced by a team at Nvidia. This algorithm pits two neural networks against each other — one attempting to generate fake face images, while another network attempts to flag the fake.
It’s not impossible to tell which is real, at this point. After you’ve been playing for a while, you might start to notice things the computer tends to get wrong. The authors of the site even outline some of the common issues here — things like weird water-splotch-like graphical artifacts, or smiling mouths featuring too many front teeth.
But even if you’re getting 90% of them right off the bat, you’ve gotta wonder: would you notice the fake if it weren’t so deliberately contrasted with a real one? If any of the fakes were just a random photo on a profile on the Internet, would you even take a second look?
This post was originally posted at http://feedproxy.google.com/~r/Techcrunch/~3/cVG_Yj_P6Zg/.