Why it matters: Artificial intelligence's ability to generate believable human facial images from scratch is far more advanced than you might think. With a bit more refinement, it won't be long before AI is able to digitally create human photos that are indistinguishable from the real thing.
Deepfakes, or the act of using artificial intelligence and machine learning to plaster celebrity faces onto the bodies of other people, is a hot-button issue in the tech community. Perhaps even more alarming is a new website that demonstrates an algorithm’s ability to create portraits on command of people that don’t actually exist.
ThisPersonDoesNotExist.com, the brainchild of Uber software engineer Phillip Wang, uses generative adversarial network (GAN) technology from Nvidia called StyleGAN to create a new facial image from scratch based on a 512 dimensional vector.
The results, more often than not, are indistinguishable from authentic images. Occasionally, you’ll come across a generated image with some obvious artifacts and if you take extra time to analyze the offerings as a whole, you may notice that teeth sometimes look a bit unusual with random discoloration or spotting.
Still, for technology that’s only in its infancy, it’s darn impressive… or outright scary, depending on your take. If it’s this “easy” to generate a fake face, it won’t be long before we are able to create other fake things... evidence, environments, societies. Maybe Elon Musk’s theory about us being in a simulation isn’t all that far-fetched?