Article: Deepfakes, Photographs, and Trust in Evidence Virginia Law Review Online

Abstract

Recent voices in the evidence literature have expressed alarm over deepfake technology–AI-driven algorithms that can manufacture fictitious images and audio that are difficult to distinguish from reality. These voices for reform claim that deepfakes represent a sea-change in technology that threatens to upend the legal proof process. The problem lies not only in deepfakes’ ability to mislead and confuse, but also in the technology’s pervasiveness and ease of use. Are these fears over deepfakes justified? Does the legal system need to adopt new authentication rules and approaches to address their threat? In this Essay, I argue that the answer is no, but more significantly, I use deepfakes as a vehicle to explore the nature of authentication and more generally, legal proof. My argument proceeds in three related steps: First, I propose a tentative model of evidentiary authentication based on base rates of trust. Such a base rate model suggests that existing evidentiary principles of authentication are more than sufficient to address the problem of deepfakes, particularly as the public becomes aware of and exposed to their existence and dangers. Second, the authentication problem of deepfakes parallels the legal system’s angst over photographs a century ago. Indeed, the compromises that the legal system made to handle photographs and other images provide a ready framework for thinking about and addressing the deepfake problem. Finally, I raise some broader implications of taking a trust model of evidence seriously. On the one hand, a theory of legal proof based on trust may illuminate evidentiary problems like authentication. On the other hand, a trust theory of proof lays bare some troubling realities. If legal proof is ultimately about trust, what happens to information sources that certain factfinders are inclined to distrust, and what will this mean in an increasingly polarized society?