Deep Fakes: A Primer
There used to be a popular saying: ‘the camera never lies.’ However, with the advent of cutting edge technology and advances in Artificial Intelligence, this might be a thing of the past.
In May 2018, a video of POTUS Trump giving advice to the Belgians on the issue of global warming and climate change went viral on the internet.
“As you know, I had the balls to withdraw from the Paris climate agreement,” he said, looking directly into the camera, “and so should you.”
Many people from around the world disdained Trump for his remarks. The video provoked mass outrage and sparked comments from hundreds, many objecting to how an American president could weigh in Belgian’s climate policy. This anger, however, was misdirected. As it turned out, the video was nothing but a high-tech forgery. The video was part of propaganda by a Belgian political party namely the Socialistische Partij Anders (sp.a).
The sp.a admitted to having paid a production studio to use machine learning to create a video, technically known as a ‘deep fake’.
Deep Fake is a technology by which a computer-generated replica of a person can be made into videos, saying things or performing actions that the person might not have actually said or done.
The video by sp.a is one such deeply disturbing example of the potential such technological advances has. Such videos exploit our inclination of believing what our eyes see and have the potential of threatening our already vulnerable information ecosystem, not to forget the cycle of misinformation it can start undermining the possibility of a reliable, shared reality.
Such fake videos don’t require a powerful computer and can be created on a regular computer. Deep fakes run on ‘Generative Adversarial Network’ or GAN. The brainchild of Ian Goodfellow, GANs can algorithmically generate new data types from existing data sets.
A GAN, through deep learning, replicates the data sets it is needed. It can capture and copy variation within a data set into a seemingly natural video.
Using GANs not only the voice can be altered, but the person can be seemed to perform actions that he has never done in his life.
Another wave of devastating fake videos was the digitally altered pornographic video. Some trolls created deep fake videos by superimposing faces of actresses over women’s bodies, the videos so created proved to be convincing enough to fool people.
In this era of fake news and false propaganda, if doctored images were a challenge, then deep fakes have opened an entirely new dimension of potential cyber-crimes.
Moving a step forward in the field, an international research team from the Max Planck Institute for Informatics, Germany revealed a technology for producing what they termed as ‘deep video portraits’. They showed how through their technique, a person could control the facial movements of another person, ‘facial ventriloquism’ as they termed it.
To explain this, it would be to say that if the person demonstrating the technique would open his mouth, then so would the synthetic face (in the video), if he demonstrator would show any facial expression, the same would be replicated by the software and depicted in the video.
In a subsequent press release, the team did admit that their technology had the potential to be misused, however, they emphasized their motive behind developing the technology i.e. ‘to make a difference to the visual entertainment industry’. The technology, as the researchers explained was capable of synthesizing faces that look ‘nearly indistinguishable’ from the truth.