Can AI-Assistance Improve Evaluations of Eyewitness Lineup Identifications?

Eyewitness misidentifications are an important source of error in the legal system; they contribute to the majority of cases that have been later overturned by DNA evidence (Innocence Project, 2023). Yet, no tool exists that can assist law enforcement in distinguishing between reliable and unreliable eyewitnesses. In this predissertation, I present two experiments that show that AI-assistance can improve people’s evaluations of eyewitness lineup identifications. Part I shows that AI-assistance can eliminate a known cognitive bias—the Featural Justification bias. Whether or not this occurs, however, depends on the participant’s perception of how useful they found the AI to be. Part II shows that in the absence of AI-assistance participants do show the ability to discriminate between correct and incorrect eyewitness lineup identifications. But, for feature-based and recognition-based lineup identifications, AI-assistance improves this ability. For familiarity-based identifications, however, discriminability was comparable for participants who received AI-assistance and those who did not. Altogether, these two studies suggest that AI-assistance can help people more accurately judge an eyewitness’s lineup identification.