Food for Thought – Deepfake

An idea, a book, a concept, a “food for thought” that we share with you to open reflection, discover new subjects, and bring you to (re)think from another angle.

One could translate the concept as the deep false or the depth of the false. But that would suggest that the false can be deeper than the true. Moreover, there is not a false that would be deeper than another, no more than there exists a true that would be deeper than another true. There is what is true and what is false and between the two an intangible and irreducible dividing line.

But, since the extension of the fake, long limited to the domain of simulation, of simulacrum, of lying, of parodic imitation or simple material counterfeiting, to so-called alternative facts and therefore to information and the formation of public opinion, the boundary between true and false has become as precarious as porous.

With the deepfake, an additional step, undoubtedly the step too far or the fatal step, is crossed. One now produces falsehood in depth and in series thanks to the new resources of artificial intelligence. The deepfake is the AI-assisted production of massive falsehood, in the double quantitative and qualitative sense of really enormous false, of false beyond even the false in its simple opposition to the true, massive false with criminal intent.
Originally, Deepfakes is the nickname of a Reddit user who specialized in making porn videos where the faces of X actresses are replaced by those of celebrities. The find could have remained a campus prank. But very quickly, thanks to generative artificial intelligence programs like Midjourney or Dall-E, digital counterfeiting becomes a weapon of mass destruction. By teaching, through deep learning, programs to perfectly reproduce faces, movements, body attitudes, voices, lexicons, language tics, it becomes possible to superimpose audio or video files created onto others that preexist them. The identifiers of a person’s singularity, the markers of their uniqueness, are thus reproducible, hackable, and misusable infinitely. One can make anyone say anything, show anyone in fatal positions or situations.

While material counterfeiting only deceives its victim without harming their dignity, at most making them look like a sucker, digital counterfeiting is designed to harm them seriously, even definitively. On one side, there is a consumer to swindle, on the other, there is a target to bring down. It is enough to choose one’s target: Obama, Taylor Swift, Margot Robbie, Macron, Zelensky, your neighbor, your rival, your whipping boy, or your wife’s alleged lover and make them say things like a slaver, enemy of the people, Nazi, or sexual predator. The goal is not to deceive or abuse but to humiliate, harass, compromise, slander. Think big, Kick ass, Lie for true would say Donald Trump. When doing deepfake, one must not do it halfway, because it is the delirious exaggeration, the surpassing of the norm as well as the gap, the contradiction beyond the limits of reason, that makes the machine turn. That is why usual translations of deepfake as hyper-manipulation or videotox are so weak: they make one believe in a small playful simulacrum well made but relatively innocent, or a slightly borderline and manipulative image, more indigestible than dangerous. Thus, one minimizes the mass destructive power that is the deepfake: devastating for democracy which dies without the people’s trust in its political class, devastating for civil peace which assumes the sharing of common decency among citizens regardless of their social condition and culture, devastating finally for peace between nations as a vector of hate and provocation.

The 27-member Europe recently validated an AI Act that condemns deepfakes and requires their distributors and producers to identify and systematically signal them, failing to force them to delete them or prohibit them from producing them. In the United States, a federal bill aiming to institute penalties of up to 10 years in prison to punish digital counterfeiting of sexual character has still not been voted on by Congress. As usual, the law lags behind technology.

References
Servitudes virtuelles / Jean-Gabriel Ganascia
La post-vérité ou le dégoût du vrai / Claudine Tiercelin
Documentalité, pourquoi il est nécessaire de laisser des traces / Maurizio Ferraris
On Bullshit, de l’art de dire des conneries / Harry Frankfurt
Conjectures et réfutations / Karl Popper
Car l’illusion ne s’oppose pas à la réalité / Jean Baudrillard