This week we take a look at the new round of “Deep Fakes” making their way across the internet. Deep Fakes are realistic looking and sounding videos – often of well known people – synthesised using deep learning techniques. Join Jon, Nina and Dilpreet as they discuss the implications of a world where you can’t be certain what is real and what is fake anymore.
Fake internet videos are on the rise recently, but forgery in art is thousands of years old. In this episode of our creative AI podcast, we examine the use and implications of using deep learning technologies to generate realistic videos of people, often without their knowledge or consent.
Many high profile fakes, like the Nancy Pelosi “drunk” video were not made using AI techniques, just good old-fashioned digital editing and manipulation.
Many special effects artists with a sense of humour have been releasing “realistic” videos of people or situations that never happened. Some of our favourites include the baby being snatched by a golden eagle, the Dragon Air flight landing in a typhoon, and the “Bosstown Dynamics” robot attack.
However, real deep fake techniques allow you to train neural networks using videos or even still images of a real person, and then synthesise new videos of that person without having a special effects team helping you out. Training a network on a person’s voice allows you to generate realistic audio of that person saying anything you want them to say. There are even techniques that allow real-time face reenactment, allowing one person to become a digital puppet master for another.
GAN synthesised faces made the news recently when it was revealed the Linkedin account of “Katie Jones” was used to build political networks in the US. The profile picture was believed to be generated using a GAN (something that anyone on the internet can do).
We look at recent deep fake videos, including a very convincing deep fake of podcaster Jo Rogan, a relaxed Kim Kardashian on her monetisation strategy, and Facebook’s own Mark Zuckerberg extolling the virtues of his James Bond-esque conspiracy. While the technology is really in its infancy, it points to the future possibility of almost anyone being able to create their own fake videos easily. And then, who can you believe when it comes to news and information?
“We used technology and data, like Rembrandt used his paints and his brushes” - Ron Augustus, Director SMB Markets, Microsoft.
Another aspect of AI generated “fakes” has been in the production of art fakes. A recent project to create a new version of a Rembrandt painting received widespread criticism, including this review from the Guardian’s Jonathan Jones, who describes it as “a new way to mock art, made by fools.”
“The next Rembrandt makes you think about where innovation can take us…what’s next?”
Listen to the podcast to hear what our team thinks is next.
Please let us know your opinions about Deep Fakes via twitter and we also welcome suggestions for topics you’d like to hear about in future episodes.