The rate at which Artificial Intelligence (AI) is mimicking real life is becoming more and more scary or exciting, depending who you ask. One such example is a website known as The site, created by Philip Wang, who is a software engineer at Uber, uses AI to generate an endless supply of human faces that look so real yet do not exist in real life.

It uses what is known as a GAN (Generative Adversarial Networks) to produce a new face every time you refresh the page. It is specifically trained on a dataset of real human face photos.

"Recently a talented group of researchers at Nvidia released the current state of the art generative adversarial network, StyleGAN, over at 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. Their research group have also included pretrained models for cats, cars, and bedrooms in their repository that you can immediately use. Each time you refresh the site, the network will generate a new facial image from scratch from a 512 dimensional vector," wrote Wang.

These people are not real – they were produced by a generator that allows control over different aspects of the image. (by NVIDIA Research)

What is a GAN?

GANs are a deep leaning technique that, to put it quite simply, learns a skill, gets feedback on what is still missing, relearns the missing skill based on feedback, unlearns what is not necessary to mastering the core skill, get feedback. This process is repeated until the AI has mastered the skill.

In this case, the skill is to reproduce human faces.

In other cases, GANs have been used to produce fake cats and even draw art.

Diversity concern

One thing you'll notice with the site though is that it takes quite a lot of refreshes to come up with an image with African features.

Most of the images are of people of European and Asian descent. This could probably be because of the training images used contained fewer African faces as compared to European and Asian ones.

AI in the near future

Already we have witnessed how AI can be used to make fake videos of well known people and mimic their voices as so well that to an untrained eye and ear it is difficult to detect. This has become an issue especially around fake news where it is feared that adversaries could spread false messages using their political opponents faces and voices. These also use GANs to implant people's faces in videos and in some cases, in very unsavoury ways including implanting celebrities faces on pornography videos.

This will make it quite difficult in the near future, as the technology gets better, for humans to detect what is real and what is not, especially online.

Share this via: