Truth And Technology: Deepfakes in Law Enforcement Interrogations

Abstract

There are theoretical and normative questions to consider about how generative AI might infect a critical aspect of a police investigation, the custodial interrogation. For instance, can police officers utilize generative AI tools during a custodial interrogation to attempt to compel a suspect's confession? Historically, courts have permitted police to use trickery and deception to secure confessions. Such police tactics have included falsified forensic reports, fabricated witness statements, and pretend polygraph results that when shown to suspects have significant impact on compelling them to confess. However, the increasing sophistication of generative AI enables police to fabricate-with increasing ease-incriminating images, audio and video recordings, as well as other evidence that may intensify the inherent compulsion of a custodial interrogation. For example, suppose the police show a suspect a deepfaked video of a witness who claims to have seen the suspect commit the crime, or a deepfaked video of "an accomplice" who confesses to the crime and simultaneously implicates the suspect. These types of ploys may well exceed the bounds of due process and exacerbate the potential for false confessions. To be sure, the power conferred to the police by generative AI is not simply limited to the ability to show deepfaked audio or video, or other fabricated evidence. Emerging AI models can receive input and provide contemporaneous feedback. Thus, AI could help policein real time-identify the most effective strategies for eliciting a confession based on a suspect's characteristics, such as body language, intonation, race, gender, ethnicity, etc. In sum, an interrogator's ability to compel a confession with the amalgamation of information AI can generate manifests a legal conundrum.

Keywords: generative AI, criminal procedure, interrogations, artificial intelligence, technology, due process