If you’re old enough to still rely on some brain cells left over from the analog age, you likely also haul around a healthy degree of old-fashioned skepticism toward the world we encounter online. Hang onto that. You’re gonna need it.
Because the newest most untrustworthy thing on the Internet also happens to be the very feature we once depended on to verify actual humanity on the other end of the tubes: your face.
Not yours, in particular — although the features of your face may well be among those countless others hoovered up by StyleGAN, a (publicly available) algorithm formulated by AI firm Nvidia to generate a ceaseless stream of human faces, none of them real, and all of them indistinguishable from the strangers you ignore every day.
StyleGAN is the algorithm behind ThisPersonDoesNotExist, a site launched by software engineer Phillip Wang that generates a perpetual procession of virtual faces that spread like viral wildfire. (You can see similar applications of the technology at work on sites like This CatDoesNotExist and ThisRentalDoesNotExist.)
Refreshing the page and meeting a new entirely non-existent stranger is an uncanny experience in several senses; but the creep factor doesn’t really creep in until you try WhichFaceIsReal.com.
Created by Jevin West and Carl Bergstrom (and an extension of the University of Washington’s “Calling [B.S.]” initiative), the site challenges visitors to identify which face in a randomly paired duo belongs to a real person, and which is the GAN (generative adversarial network) at work. Or, in old-school Internet parlance, “Is he bot or not?”
It’s all great fun until you keep getting them wrong over and over. Then it becomes a little terrifying. What kind of damage could one of these nobodies do? What about an easily assembled army of them?
“Computers are good,” reads the site’s “About” page, “but your visual processing systems are even better. If you know what to look for, you can spot these fakes at a single glance — at least for the time being. The hardware and software used to generate them will continue to improve, and it may be only a few years until humans fall behind in the arms race between forgery and detection.”
The digital world is, after all, a realm of obfuscation — you can’t very well turn something digital without rendering it into code.
And in the early text-heavy days of anonymous chat rooms, Nigerian prince scams, and JPEGs that took all night long to download, the crudeness of the Web invited the same kind of reflexive caution that wandering into any seedy neighborhood might.
But fast forward two decades, and we are immersed in an Internet far more faithful to the contours of real life (or is it the other way around?), and far more determined to be mistaken for it. Which is why rebooting that old-school skepticism toward everything (and everyone) you encounter online will be an essential part of dealing with an Internet that can’t be taken at face value.
So how can you tell?
After a few minutes of clicking through WhichFaceIsReal, you start to realize that imperfections account for a lot in how we identify each other: A slightly smaller eye, a wonky tooth, a feature freckle.
StyleGAN faces aren’t betrayed by unreal symmetry or attempts at “perfection” – if anything, the algorithm’s talent for producing slightly off features proves to be one of its big strengths. So spotting a fake isn’t as easy as spotting, say, a FaceTuned profile pic.
Instead, when in doubt, look for imperfections in the imperfections: A sideburn that seems imported from another head (or decade); an expression of resolute vagueness; a wrinkle pattern that looks more mapped by chance than etched by time. Be on the lookout also for strange digital artifacts around the edges of images, textures that don’t look authentic (you’ve shopped for bags before), or oddly distributed light.
The good part is that every day, from now on, you get to pretend you’re some exciting cybersecurity forensics sleuth. The bad part is that you actually already have a job, and that this is really just another example of objective reality crumbling and sliding into the ocean.
One one hand it’s good that new advances in fraud are forcing us to look more closely and cautiously; but the last thing we need seeping into real life is another reason to doubt each other.