The “Proven Proof”

Deepfakes are synthetic media generated by software that make use of machine learning. At the dawn of the Great Internet Plague, the practice of deepfake was becoming increasingly widespread. The term deepfakes was coined in 2017 on the Reddit wall, it was the nickname of an anonymous producer of fake videos.

To generate deepfakes you just needed the right software and a large archive of photos and videos of the victim. With that it was possible to produce absolutely realistic video material where the victim did, or more often said, things that otherwise he/she would never have said or done. Deepfakes circulate videos of various characters from a political scene saying things they had never actually said. It happened to Barak Obama, Theresa May, Donald Trump and Adolf Hitler.

At the same time the phenomenon of porn deepfakes exploded: Scarlett Johansson, Katy Perry and Natalie Portman. Nobody was safe, not even Nicolas Cage, who was one the the actors most popularly targeted and appeared in hundreds of scenes from different TV series and films (not different from reality, huh?).

Lawyers had become busy right away, but it was difficult to limit the damage of a deepfake once it was released on the net. The state of California declared deepfake videos illegal at the end of 2019, but it seemed impossible to enforce such a law. Software development companies scrambled to propose solutions to stem the phenomenon, starting with the identification (deepfake detection). It was mitigated some by (deepfake mitigation), but each technique was intricate, and not very suitable to solve the problem at its root.

The apps were installed on the device with which photos were taken and videos were made, this would have included information in the image (metadata) such that they could, in a second moment, be compared with deepfakes to demonstrate the authenticity of the former and the falsity of the latter. Another method, perhaps even more complicated, was to analyze the alleged fake videos by carefully observing some details such as the ears or the reflection in the pupils.

However, it was still a battle against windmills. Scarlett Johansson, a victim of countless pornographic deepfakes, in December 2018 in an interview with the Washington Post declared that she believed any attempt to remove deepfakes “a lost cause”. She did not consider herself worried about herself, protected by her fame, but for all ordinary women who in turn could find themselves unwitting victims.

The evolution of deepfake technology actually made the tool much more accessible. It was even possible to create a deepfake from a single image, without the need for a large archive of photos or videos of the subject. Have you ever wanted to see Mona Lisa speaking and breathing like a real human being? In the first years of the III millennium, there was the technology to make it happen.

The fact that the video was not “proven evidence” was already clear to some of us in unsuspecting analog times. In the pre-digital era we knew well that it was possible to manipulate a shot through editing to create different interpretations or versions of reality. Yet already at the end of the twentieth century a fact was considered true because “TV had said it.”

We video geeks, spoke of the truth as something that was defined through multiple points of view. In our documentaries we did not pretend to tell the truth, but rather tried to offer a partial vision of things, our vision, our irreducible point of view.

In my eyes, deepfake technology offers a more sophisticated method of manipulating reality. Again, the antibodies not to succumb to this vulnerability are not digital, rather, they are found in the expectations we place in technology and possibly in the levels of delegation (of truth, correctness, knowledge…). For example, in this story Cindy believes in Eva, despite the overwhelming video evidence, because the two are friends and trust has been built over time in the relationship.

“The old idea that pictures do not lie is going to have to undergo drastic revision.” Science fiction writer Jack Wodhams wrote in a 1986 story¹.

The fact that such a scenario was taken for granted at that time says it clearly: the problem, once again, was not the technology available and its accuracy, but the trust in their sources and the network of relationships in which the falsified image is manifested. How to deal with a phenomenon like deepfake without falling into “absolutist relativism”² typical of the era of post-truths?

A little healthy skepticism, a pinch of awareness and two grains of trust in relationships built with passion. If the idea of the one and only truth loses points, the partial truths remain. Partial truths you can read through (suspending judgment) which lead to a more complex vision of the world around us.