Mixed-Stakeholder Deliberation on AI Policing Risks
A recent study released on arXiv (2608.05418) details outcomes from a workshop involving various stakeholders discussing the implementation of AI in law enforcement, particularly concerning racial bias. This event convened 30 representatives from the community, law enforcement officials, and scholars to evaluate the potential risks associated with 13 AI applications in policing. Generally, participants showed a willingness to embrace AI, outright rejecting only three applications, particularly the recidivism risk assessment, with concerns centered on its foundational assumptions rather than its execution. The findings indicate that emphasizing racial equity expanded rather than limited the dialogue, leading to essential inquiries about the tool's effectiveness, its genuine advantages, and whether those advantages would be universally accessible. The research underscores the necessity of involving impacted communities in decisions regarding AI adoption, given the recognized risks of racial bias. The workshop aimed to collaboratively address the boundaries of risk in AI for policing through inclusive discussions among diverse participants.
Key facts
- Study published on arXiv with identifier 2608.05418
- Workshop involved 30 community representatives, police officers, and academics
- Assessed risks of 13 AI use cases in policing
- Explicit focus on racial bias
- Participants rejected only three use cases outright
- Recidivism risk assessment was notably rejected due to objections to its premise
- Foregrounding racial equity did not narrow the deliberation
- Discussions focused on effectiveness, genuine benefit, and equitable distribution of benefits
Entities
Institutions
- arXiv
Locations
- UK