GAN: the technology behind fake porn and facial rejuvenation

 [ Art and AI , episode 8]

 Have you read the other episodes yet ? Learn more about neural networks and deep learning !

It was 2014 when Ian Goodfellow revolutionized AI. Legend has it that the American computer scientist went out drinking with some friends when they asked him for help with a thorny project they were working on: a computer capable of creating photos by itself.

The problem and the solution

At that time, AI researchers were already using neural networks as “generative” models to create new plausible images, but the results were often not very good: images of a computer-generated face tended to be blurry or, to example, ears were missing. The plan that Goodfellow’s friends were proposing was to use a complex statistical analysis of the elements that make up a photograph to help machines produce images on their own. This would have required a tremendous amount of calculations, and Goodfellow told them it simply wouldn’t work.

But as he pondered the problem, he had an idea: What would happen if he pitted two neural networks against each other? His friends were skeptical, so once he got home, he decided to give it a try. Goodfellow coded his idea in a few hours and tested the software. It worked the first time!

What he invented that night is now called GAN, or generative adversarial network. The technique has sparked great enthusiasm in the field of machine learning and has turned its creator into an AI celebrity.

Operation and progress of GANs

In recent years, AI researchers have made significant progress using deep learning, but while “trained” AIs have been extremely capable at learning to recognize images, they have not been at creating them. The goal of the GANs is to give machines something like an imagination.

A GAN mimics the confrontation between a forger and an art critic who repeatedly attempts to locate the fake painting. Both networks are trained on the same set of images. The first, known as the generator, is in charge of producing images that are as realistic as possible. The second, known as the discriminator, compares them to authentic images from the set of images she was trained with and tries to determine which are realistic and which are blatantly false. Based on these results, the generator adjusts its parameters to create new images that can be considered “true” by the “critic”. And so on, until the discriminatory network can no longer distinguish the artificial image from those with which it was trained.

To understand the potential of GAN, just look at, for example, the results obtained by Nvidia, the famous company operating in the field of graphics processors, in 2017. The researchers trained the network with images of famous people to be able to generate faces of celebrities imaginary, obtaining surprising results. Do you believe that every face you see in the image does not exist in the real world?

DeepFake, malicious and good uses: from fake porn to cinema

Watch the video above. Do you think the face in the foreground is really Obama’s? It might seem like it, but that’s not the case at all. One of the many possibilities offered by GANs is the creation of so-called DeepFake videos.

The very term ” DeepFake” comes from the name of a Reddit user who in December 2017 used deep learning technology to replace the faces of celebrities from music and cinema in those of some pornographic videos. Certainly not a very correct practice.

In recent years, many concerns have been expressed about the development and spread of deepake, including in areas other than porn. Imagine if, unlike Obama’s video which is openly a fake, counterfeit videos of politicians were circulated in which they seem to take sides against their own values ​​or against other states, creating diplomatic problems between world powers or manipulating public opinion.

Luckily, the GANs have not only had negative implications, but have also been used for a good purpose, especially in the art of cinema: from facial rejuvenation techniques ( similar to those seen in The Irishman or Ant-Man) to the inclusion of dead actors in new films (such as Paul Walker in Fast & Furios 7).
Even in the world of pictorial art they have done good: in fact, I will tell you about “GAN Art” next time!

Do you want to try using GANs? Read this article and try the software!

The GAN article : the technology behind fake porn and facial rejuvenation comes from TechCuE .