Evidence across applications
What improves, and how do we know?
Place-based prediction
Modest reductions with visible patrols
Evaluate the tool and the operational response together.
Person-based prediction
Fairness and effectiveness concerns
Examine errors, consequences, and evidence quality.
Identification & detection
Some process benefits; crime effects unproven
Separate faster operations from public outcomes.
Administrative AI
Limited evaluation evidence
Test usefulness in a bounded workflow before scaling.
Guler, A., Boke, K., & Kula, S. (2026). AI in Policing: A Systematic Review of AI Use in the US Law Enforcement. In L. Huey, L. Ferguson, & I. T. Adams (Eds.), Future Policing: Technology and Transformation. Routledge. Publication (opens in a new tab)
The research
The review finds mixed evidence. Place-based predictive policing shows modest crime reductions when paired with visible patrols, while person-based models raise concerns about fairness and effectiveness. Facial recognition and gunshot detection can improve some operational processes, but consistent crime reductions have not been demonstrated.
Why it matters
Operational benefits and broader public outcomes need separate evaluation. Decisions about AI should consider the specific application, transparency, procedural fairness, oversight, and independent assessment.
What to keep in mind
Many applications remain insufficiently evaluated, particularly tools intended to improve administrative efficiency. Evidence varies by technology and setting, and longer-term research is needed to assess outcomes and impacts.
Publication
Guler, A., Boke, K., & Kula, S. (2026). AI in Policing: A Systematic Review of AI Use in the US Law Enforcement. In L. Huey, L. Ferguson, & I. T. Adams (Eds.), Future Policing: Technology and Transformation. Routledge.
Read the original publication (opens in a new tab)