AI, body‐worn cameras, and a potential civilizing effect for officers: Evidence from the Arizona Truleo study

Abstract Research Summary Over time, the evidence on the impact of police body‐worn cameras (BWCs) has become increasingly mixed. Though there are numerous contributors to the inconsistent findings (e.g., context, policy, training, culture), one pressing issue involves the failure of law enforcement agencies to review any more than a fraction of the BWC footage generated by their officers. This footage review problem violates a central precondition for the hypothesized civilizing effect of BWCs on officers: their belief that footage will be reviewed. AI‐driven BWC analytics has emerged as a solution that may overcome the footage review problem, but there has been virtually no research on such AI‐based platforms. The current study fills this gap through an evaluation of Truleo in the Apache Junction and Casa Grande Police Departments (AJPD, CGPD). Officers were randomized to Treatment (Truleo) and Control (non‐Truleo) conditions for a 6‐month period, and we compared study groups across a range of measures including Truleo‐generated professionalism, use of force, complaints, stops, arrests, and citations. Treatment officers in both departments were more likely to generate High Professionalism ratings, but group differences fell short of statistical significance. Use of force decreased significantly in AJPD and in some of the pooled models, though the pooled models were sensitive to multiple‐comparisons corrections. The Bayesian probabilities paradigm offers an additional lens for considering the mixed effects with High Professionalism and use of force. Policy Implications Taken together, results suggest Truleo may hold promise for promoting positive behavior change among officers consistent with a civilizing effect. Law enforcement agencies should explore whether AI‐driven BWC analytics can enhance their BWC program, expand their review of footage, and positively shape their officers’ behavior. Additional research is needed in other settings to test the durability of these findings and the value of AI‐generated metrics of police performance (e.g., professionalism). Researchers should consider the significance of research findings from a broader lens that accounts for practitioner perspectives on the meaningful impact of an innovation. Lastly, the results presented here inform the ongoing dialogue over the role of AI in policing.