Artificial intelligence in police research: a preliminary examination of feasibility and replication

Abstract Objectives This manuscript tests whether artificial intelligence (AI) review of body-worn camera footage replicates the findings of a randomized-controlled trial of police de-escalation training using systematic social observation (SSO). Methods Body-worn camera video that was previously analyzed using SSO was subjected to review by an AI program. The study then replicates the analysis of the randomized-controlled trial using outcomes generated by AI. The results are compared to those of the original study. Results All five measures produced by the AI correlated with at least one SSO measure, though several SSO measures appeared completely unrelated to any measure produced by AI. The analysis of treatment effects by AI reaches the same conclusion as the original study. Conclusions SSO provides greater nuance to understanding police interactions but is weakened by its time-consuming and expensive nature. AI provides a promising avenue for conducting similar analyses more quickly and efficiently.