Artificial intelligence has learned to generate perfect photos of fake people

Although the word "neural network" has long been included in our lexicon, these systems themselves are inaccessible to most users. Firstly, they are too complicated to operate and of little use for widespread use. Secondly, they require a fair amount of computing power for their work. However, the site thispersondoesnotexist.com was recently launched, which is the simplest and most convenient tool for working with the StyleGAN neural network.

The author of the development is an enthusiastic programmer Philip Wang, who built a laconic interface for displaying the results of StyleGAN work. Neurity itself belongs to the category of generative systems and is based on last year's Nvidia technologies, which in turn grew out of the algorithms of the artificial intelligence developer Ian Goodfellow. Its main task is to generate images that ideally meet the basic criteria.

If you follow the link, you will see a pseudo photo of a human face, which will be replaced by similar images every time the page is refreshed. All of them are generated by StyleGAN, which consists of two neural networks - the first creates an image, the second checks it and sends it for revision. Having gained experience as a result of these actions, StyleGAN has learned to create almost perfect pictures of people who have never existed in nature.

In addition to human faces, this neural network also skillfully creates images of cats, anime characters, superheroes, graffiti - and, in general, anything, it would be the starting material for training. And this is alarming for specialists, because the risk of misuse is too great for such tools. If a conventional artist uses StyleGAN to create a hundred unique images for his interactive film, then what will prevent a conventional politician from drawing a support group of many thousands to him and presenting the media as the truth?