Mixed-Stakeholder Deliberation Shapes AI Policing Risk Boundaries.

The adoption of AI tools in policing is growing globally, yet decisions about their implementation often exclude representatives from affected communities, despite known risks like racial bias. This research details a mixed-stakeholder deliberation workshop that brought together 30 participants, including community members, police officers, and academics, to evaluate the risks associated with 13 different AI use cases in policing.The findings indicate a general openness to AI adoption among participants, with only three use cases being outright rejected. Notably, recidivism risk assessment faced strong opposition, primarily due to fundamental objections to its premise rather than just its implementation details. A key insight from the study is that foregrounding racial equity did not narrow the scope of discussion.Instead, conversations expanded to address core questions about whether a tool genuinely works, delivers tangible benefits, and ensures those benefits are accessible to everyone. This integrated approach, likened to the "curb-cut effect" in inclusive design, underscores the value of incorporating a racial bias lens from the initial stages of AI risk-benefit analysis.