Website AI Creates Headshots of Nonexistent People On-Demand

By: Marie Donlon

A software engineer from Uber, hoping to highlight what artificial intelligence (AI) is capable of, created a website featuring headshots of people who do not exist.

Aptly named thispersondoesnotexist.com, the website was built by software engineer Philip Wang using a class of AI algorithm employed in machine learning called generative adversarial networks (GANs) to create the fake images. Visitors to the site refresh the page to be greeted with a brand-new fake headshot, created using code written by tech company Nvidia called StyleGAN.

Source: thispersondoesnotexist.com

To create the images, the GANs relied on two networks working against each other: the generator, which manufactures an entirely original sample based on images held in a database; and the discriminator, which collects samples from training data along with the generator’s sample and makes predictions about whether the samples are real or fake. Receiving feedback from the discriminator, the generator learns to manufacture even more realistic images with each attempt. Meanwhile, the discriminator continues to learn by measuring the manufactured samples against real samples, which eventually makes it more difficult for the generator to deceive the discriminator with fake images. This back and forth process continues until the generator successfully deceives the discriminator with a fake image.

Describing the inspiration behind the website, Wang explained, “Faces are most salient to our cognition, so I’ve decided to put that specific pre-trained model up. Each time you refresh the site, the network will generate a new facial image from scratch from a 512-dimensional vector.”

According to Wang, the technology has several applications, including gaming and virtual applications, but warns that it could also be used to create fake documents and fake news.