Exposed: The Detection of Deep Fakes through Digital Forensics

• In Event: Seeing through New Eyes: Science, Truth and Technologies of Visualization - II

Abstract

“Deepfakes” are manipulated videos generated with artificial intelligence. Recent deepfakes have included celebrities’ heads swapped onto pornographic videos as well as altered political propaganda of Trump and Obama, prompting a multimillion-dollar investigative response from defense agencies and technology companies. In a virtual arms race against deepfakes, new investigative efforts generate a networked infrastructure of police, moderators, and algorithms designed to detect and delete faceswapped, lip-synced, or motion-transferred videos.
In this paper, I argue that the digital forensics deployed to expose fakes are a combination of human intuiting and algorithmic sorting. I call these dual processes “artifice intelligence”—the embodied, networked sensing of deception. What pleasures do digital forensic scientists feel in detecting a fake? How might the desire to inspect the “authenticity” of deepfake photos be an eroticizing gesture, one rooted in technoscientific desires to uncover truth and reality? In addition, I focus on the materiality of deepfake pornography detection, on the inescapable “fleshiness” of faces, bodies, and skins that are swapped, obscured, and revealed in deepfake production and detection. How do these digitally corrupted bodies come to matter, as legal evidence, as pixelated gender and race, and as traces of victimization?

Author

• Mitali Thakor, Wesleyan University