When it comes to the pictures on This Person Does Not Exist, you won’t believe your eyes—and you shouldn’t.
Though the people pictured in the photos look authentic, none of them actually exist. Each time a user refreshes the page, a new computer-generated face appears.
The images are created through algorithms known as Generative Adversarial Networks (GANs), which feature two computer programs, the generator and the discriminator, that compete against each other using commonalities in data sets from real faces to generate artificial ones that grow in refinement over time.
The results are shocking.
Click that link: the person you see is generated by a neural network ~from scratch~. Refreshing the page makes a brand new one.
This person does not exist ⤵ pic.twitter.com/aIpjpeJLLj
— 𝔻𝕠𝕦𝕘 𝔹𝕣𝕒𝕕𝕖𝕟 (@DugBraden) February 14, 2019
— Jo (JNH) 🎮🖌🖌 (@Jo_Ngoc_Hang) February 15, 2019
— atomapolaclipse (@atomapolaclipse) February 15, 2019
— Florent Bourgeois (@florent_b) February 14, 2019
While most of the images look authentic, there are sometimes glitches in the matrix.
so – this person does not exist ( its a generated image from https://t.co/W2DtLIZdWO ) .. and I'm glad, because if he did, he should see a doctor about that thing on his face #yikes pic.twitter.com/l2dP77nlvC
— andyidsinga (@andyidsinga) February 13, 2019
All in all, the technology—which was introduced in 2014—has gotten more sophisticated with time. This GAN (called StyleGAN) for faces was developed in part by a Graphic Processing Unit Company called Nvidia. That sophistication could lead to some sinister developments down the line.