A philosophy-of-AI essay examining whether the moral and ethical harms of deepfake technology are serious enough to justify restricting it. Framed around the central question: do the moral and ethical reasons against deepfakes outweigh the benefits such that restriction is necessary?

Abstract

Four weeks after Russia invaded Ukraine, a deepfake video surfaced of president Volodymyr Zelensky supposedly telling civilians to lay down arms, the first appearance of deepfakes in an armed conflict, seen by hundreds of thousands before it was removed. Unlike spam bots and fake news, deepfakes directly attack the visual and auditory senses humans naively trust, letting anyone produce realistic videos of people doing things they never did. The essay first weighs the advantages and disadvantages of deepfakes, then turns to their ethical impact on identity, and finally their effect on the individual and one's view of reality.

Central question

Do the moral and ethical reasons against deepfakes outweigh the benefits such that restriction is necessary?

Position

The essay argues that deepfake creation should be monitored, that responsible synthetic-data creation be encouraged, and that access to the technology be limited. The main drivers for restriction are its use in information warfare, blackmail, sabotage, and ideological influencing.

The technology

Deepfakes have genuine positive uses: de-aging and CGI in film, crossing language barriers (the David Beckham malaria campaign spoke in nine languages), and giving a voice back to people who lost the ability to speak (Lyrebird). Against this stand the harms: research by Deeptrace (2019) found 96% of deepfake videos online were non-consensual pornography, exposure to political deepfakes worsens attitudes toward the targeted figure (Dobber et al., 2021), and real-time impersonation enables CEO voice-spoofing fraud. This even undermines the highest form of legal affirmation, visual evidence in court, prompting reliance on electronic seals, time stamps, and digital signatures (Mason, 2016).

Report