A website has been launched which creates remarkably real-looking artificial intelligence (AI) generated faces.
The faces are produced using NVIDIA’s StyleGAN, a generative adversarial network (GAN) face generator. The StyleGAN was open-sourced in February 2019 allowing Uber software engineer, Philip Wang, to use the tool to create www.thispersondoesnotexist.com.
Philip Wang posted on Facebook: “I have decided to dig into my own pockets and raise some public awareness for this technology. Faces are most salient to our cognition, so I’ve decided to put that specific pretrained model up”
Wang continued, ” Each time you refresh the site, the network will generate a new facial image from scratch from a 512 dimensional vector.”
The GAN works by splitting the AI‘s workload into separate parts. One set of algorithms, a generator, tries to create something which in this case is a human face while another set, a judge, tries to determine if it’s a real image or a fake one. If the judge determines the image is fake, the generator starts over again. If the judge is fooled into thinking the image is real, then an AI developer will check the results and evaluate if the algorithms need changing.
The technology can even decipher different styles, with a method called style transfer, whereby characteristics of one image are blended with another. This can be traits such as skin and hair color, which are blended together creating what looks to be an entirely new person in the process.
Another StyleGAN-based generator website has gone live, https://thiscatdoesnotexist.com, although it has been commented that the cat generator does not have such a high success rate compared to the face generator.
Whilst many have praised the advances in AI, concerns have also been raised about the possibility that these tools could be used for malicious purposes such as fake news and propaganda.