This person does not exist: an AI "invents" real faces of non-existent people

Artificial intelligence not only helps us to take better photos, drive automated cars or invest in the stock market. It is present in the algorithms of Netflix, Amazon or even in music applications like Pandora. Now if you thought you had seen it all, then you should take a look at this website.

The page is called This Person Does Not Exist, and if we enter it, the only content we will see will be a person's full screen photograph. Nothing more. The surprising thing comes when we find out that the face of that woman, boy or old man that we just saw it just doesn't exist. It is an image generated automatically by an artificial intelligence.

Whenever we refresh or reload the page, the web uses a special algorithm based on artificial intelligence called GAN (Generative Adversarial Network). Thanks to this technique, the system recreate from nothingan excessively hyper-realistic image of a person that we could hardly distinguish from an ordinary human being.

Below, we can see a small sample of some faces generated automatically by this algorithm.

The creator of ThisPersonDoesNotExist.com is Phillip Wang, an Uber software engineer who wanted to demonstrate what the GAN is capable of. Last Tuesday he made it known in the Facebook group “Artificial Intelligence & Deep Learning”, and since then it has become tremendously popular.

The code that makes this web page possible is called StyleGAN and was written by Nvidia. A neural network with the potential to revolutionize the world of video games and 3D modeling, but which, like everything in life, can also be used for more sinister purposes. The "deepfakes" or computer generated images interspersed with real photos or videos They allow you to create false news or alter people's perception of controversial facts or events.

And that is precisely the goal that Phillip Wang is seeking with the creation of this website, to make people aware of the possibilities offered by these types of advanced tools. "I have decided to scratch my own pockets and increase public awareness of this technology"Says Wang."Faces are what stand out the most in our cognitive values, so I have decided to work on that specific model. Every time you update the site, the network will generate a new facial image from scratch from a 512-dimensional vector.

How does an artificial intelligence GAN work?

The GAN is based on using 2 types of networks or programs with very specific tasks: the generator and the discriminator. Each of these programs compete with each other for millions and millions of attempts to refine their ability to generate images, until in the end they manage to create a photograph that is practically indistinguishable from the real world.

The concept of adverse generative networks, or GANs, was introduced in 2014 by computer scientist Ian Goodfellow. Since then, Nvidia has been the one who has worked the most on these concepts and is currently the undisputed leader of this type of technology.

You have Telegram installed? Receive the best post of each day on our channel. Or if you prefer, find out everything from our Facebook page.

Recent Posts