A new website lets users click through an endless library of fake human faces created by artificial intelligence.
Called ThisPersonDoesNotExist.com, the results are startlingly lifelike and may make you question what’s real and what isn’t.
Every time a user refreshes their browser, the site spits out a randomly generated face, spanning from older men and women to even children.
Scroll down for video
At first glance, these photos appear to be of random people. However, a website called ThisPersonDoesNotExist.com uses AI to generate startlingly realistic copies of human faces
HOW DOES IT WORK?
ThisPersonDoesNotExist.com was created by a software engineer at Uber.
It randomly generates fake human faces using a neural network first created by Nvidia researchers.
The neural network, called General Adversarial Networks, blends characteristics from one image with another to create a totally new face.
Nvidia researchers supplied the algorithm with 70,000 faces from pictures uploaded to Flickr.
In a matter of seconds, the algorithm, called StyleGAN, can generate a new face.
StyleGAN was recently made open source and has been used to generate fake animals and anime characters.
A few images may appear glitchy or blurred in some areas, but most would make you think the subject is a real human being.
Philip Wang, a software engineer at Uber, said he built the website using research first released by chipmaker Nvidia last year.
Nvidia researchers built an algorithm using a neural network called General Adversarial Networks (GAN) that’s able to create customized and realistic looking faces.
GANs are able to learn from large sets of data to look for patterns and produce new data.
They supplied it with 7,000 images of human faces from Flickr. It learned by starting with low-resolution images and recognizing more and more details as the resolution of each image increases, according to Lyrn.ai.
In the case of Nvidia’s algorithm, called StyleGAN, it’s trained to generate random human faces, but the software is able to randomly generate any number of things, ranging from fake animals, anime characters and fonts, to even fake documents.
‘Faces are most salient to our cognition so I’ve decided to put that specific pretrained model up,’ Wang explained in a Facebook post.
Every time a user refreshes their browser, the site spits out a random face, spanning from older men and women to even children. Some may appear glitchy, but most seem startlingly realistic
Philip Wang, a software engineer at Uber, said he built the site using research released by Nvidia last year. Nvidia crafted its AI by supplying it with 7,000 images of human faces
‘…Each time you refresh the site, the network will generate a new facial image from scratch from a 512 dimensional vector.’
In all, the process takes only a matter of seconds before a seemingly real, yet fake face appears before the viewer.
‘Most people do not understand how good AIs will be at synthesizing images in the future,’ Wang told Motherboard.
A video explains in further detail how Nvidia’s StyleGAN is able to generate fake human faces on demand.
‘Our generator thinks of an image as a collection of “styles,” where each style controls the effects at a particular scale,’ Nvidia researchers explained.
These three categories include coarse styles, or the subject’s pose, hair and face shape; middle styles, or their facial features and eyes; as well as fine styles, which includes the color scheme.
‘We can choose the strength to which each style is applied, with respect to an “average face,”‘ the researchers said.
‘By selecting the strength appropriately, we can get good images out every time (with slightly reduced variation.)’
While ThisPersonDoesNotExist.com and Nvidia’s research showcase the impressive abilities of AI, they also undoubtedly raise questions about the tech’s possible implications.
For example, GANs have been used to create ‘deep fakes,’ or artificially manipulated videos that use AI to swap celebrity faces onto porn star bodies.
In the image, the top row are the only legitimate photographs of real people, the rest have been computer generated. The AI uses various traits to create randomly generated people
The same technology has also been used to create digitally-altered videos of world leaders, including former President Barack Obama and Russian president Vladimir Putin.
‘The idea that someone could put another person’s face on an individual’s body, that would be like a homerun for anyone who wants to interfere in a political process,’ Virginia senator Mark Warner, who has led a crackdown on political ads on social media platforms, told CBS.
‘This is now going to be the new reality, surely by 2020, but potentially as early as this year,’ he added.
WHY IS FAKE CELEBRITY PORN MADE BY AN AI SO CONCERNING?
Back in December, it was discovered that Reddit users were creating fake pornography using celebrity faces pasted on to adult film actresses’ bodies.
The disturbing videos, created by Reddit user deepfakes, look strikingly real as a result of a sophisticated machine learning algorithm, which uses photographs to create human masks that are then overlaid on top of adult film footage.
Now, AI-assisted porn is spreading all over Reddit, thanks to an easy-to-use app that can be downloaded directly to your desktop computer, according to Motherboard.
To create the likeness of Gal Gadot, for example, the algorithm was trained on real porn videos and images of actor, allowing it to create an approximation of the actor’s face that can be applied to the moving figure in the video.
As all of this is freely available information, it could be done without that person’s consent.
And, as Motherboard notes, people today are constantly uploading photos of themselves to various social media platforms, meaning someone could use such a technique to harass someone they know.