As artificial intelligence becomes more and more capable, the implications of those capabilities start to become equal parts scary and impressive. That’s particularly true when it comes to creating images of human faces, and a new website that’s making the rounds today sums up the capabilities of machine learning pretty darn well. If you thought that AI wasn’t to the point where it could create fake-yet-believable human faces, prepare for a rather rude awakening.
The website was created by Phillip Wang, who used NVIDIA’s generative adversarial network, StyleGAN, to make it. It’s a fairly simple website as far as design is concerned, as it only shows a single image of a human face when you visit it. Refresh again and it’ll show you a different image of a different face. The twist, as you might have figured out by now, is that none of these people actually exist.
Instead, the images hosted on the aptly-named thispersondoesnotexist.com were created from scratch using StyleGAN, which NVIDIA has published as open source software. “Each time you refresh the site, the network will generate a new facial image from scratch from a 512 dimensional vector,” Wang wrote on Facebook. In simple terms, StyleGAN employs machine learning to create fake images using a large dataset of pictures of real people. You can read more about how GANs work their magic in an in-depth summary over on Lyrn.ai.
The results are astonishing, to say the least. Though a few of the images you’ll encounter have the odd artifact that gives their deception away, most of them can certainly pass for a real image of an existing person. It’s somewhat alarming to consider the implications of technology this capable, and in fact, there’s already been plenty of debate about the topic within the tech community thanks to the rise of things like Deepfakes, which use this technology to accomplish more unsavory goals.
Wang points out that the capabilities of GANs like the one NVIDIA created don’t end at human faces, as others have used them to create fake images of cats, cars, and bedrooms. If nothing else, Wang’s website shows us that this branch of machine learning has advanced significantly in a very short amount of time, and it’s certainly very interesting to consider where it might be after just a few more years of development.