This website uses AI to generate faces of people who don't exist

With the help of artificial intelligence, you can manipulate video of public figures to say whatever you like — or now, create images of people's faces that don't even exist.

You can see this in action on a website called thispersondoesnotexist.com. It uses an algorithm to spit out a single image of a person's face, and for the most part, they look frighteningly real.

Hit refresh in your browser, and the algorithm will generate a new face. Again, these people do not exist.

Original image has been replaced. Credit: Mashable

Original image has been replaced. Credit: Mashable

The website is the creation of software engineer Phillip Wang, and uses a new AI algorithm called StyleGAN, which was developed by researchers at Nvidia.

GAN, or Generative Adversarial Networks, is a concept within machine learning which aims to generate images that are indistinguishable from real ones.

You can train GANs to remember human faces, as well bedrooms, cars, and cats, and of course, generate images of them.

Wang told Mashable he created the site to create awareness for the algorithm, and chose faces "because our brains are sensitive to that kind of image."

He added that it costs $150 a month to hire out the server, as he needs a good amount of graphical power to run the website.

"It also started off as a personal agenda mainly because none of my friends seem to believe this AI phenomenon, and I wanted to convince them," Wang said.

"This was the most shocking presentation I could send them. I then posted it on Facebook and it went viral from there."

This Tweet is currently unavailable. It might be loading or has been removed.

As for the societal impact of this technology, Wang said the more people are aware of it, the better they can be prepared for these images.

A powerful enough GAN could be used to create an image of a loved one, which could be used for manipulation, he said. Or a big enough dataset could be used to create all sorts of realistic images, from scratch.

"I think eventually, given enough data, a big enough neural [network] can be teased into dreaming up many different kinds of scenarios," Wang added.

"We just don't have very good control over the dreams, but I actually have a small side project where I'm going to see if I can control the StyleGAN dreams with a paragraph (I want to see an old man with sunglasses and a beard wearing a blue shirt)."

By the way, it's not just images and video that AI can fool you. Text too, can be generated by artificial intelligence, and it's also just as believable.