A new website, ThisPersonDoesNotExist.com, has used AI to generate an endless supply of fake faces, fueling concerns that AI-generated fake visuals could soon be used in the creation of both "fake news" and "deepfakes".
The site was created by Uber software engineer Philip Wang who used Nvidia research to create the algorithm which is able to continually create eerily-realistic looking portraits of individuals. The AI has been trained on a large dataset of real images of people and then uses a generative adversarial network (GAN) to recreate its own examples.
"Each time you refresh the site, the network will generate a new facial image from scratch," explained Wang.
The framework of the AI was initially invented by researcher Ian Goodfellow and Nvidia's version of the algorithm, named "StyleGAN", was recently made open source. The system has proven to be enormously flexible and can mimic any source, leading to researchers using it to create things like anime characters, fonts and graffiti.
The system uncomfortably demonstrates AI's potential to uncannily replicate faces since, as Wang told Motherboard, "most people do not understand how good AIs will be at synthesizing images in the future".
While Wang's version of the technology is, if a bit creepy, harmless, algorithms such as StyleGAN could have serious implications. For creatives the endless virtual worlds that could be generated create infinitely exciting possibilities. However, it has potential to be used to create "deepfakes" whereby the GAN is used to paste people's faces onto videos, which have been used to create non-consensual pornography. Its potential to be used for "fake news" is also a cause for concern since, as The Verge puts it, "the ability to manipulate and generate realistic imagery at scale is going to have a huge effect on how modern societies think about evidence and trust".