Predictive policing systems analyze historical arrest data, geographic patterns, and sometimes individual criminal records to generate risk scores or "hot spot" predictions. Place-based systems identify geographic areas with predicted high crime and direct more police patrols there. Person-based systems score individuals on their predicted likelihood of committing or becoming a victim of violence.
The technology's fundamental flaw is that it learns from historical policing data — and historical policing has been racially and geographically uneven. Areas with heavier historical police presence generate more arrests, which generate more data points predicting more crime, which justify more police presence. The system amplifies existing enforcement patterns rather than reflecting actual crime rates. Jurisdictions including Los Angeles, Santa Cruz, and New Orleans have suspended predictive policing programs after civil rights audits.
Federal oversight is limited. No federal law specifically governs predictive policing, and vendors often treat their algorithms as proprietary trade secrets, blocking independent audits. Courts have generally allowed predictive policing evidence as a basis for investigative stops, though the legal standards for algorithmic evidence remain unsettled.
Predictive policing puts people under surveillance and police contact because of where they live or who they know — not because of anything they've done. When these systems drive investigative stops, they launder historical discrimination into official police decisions. Communities with less power to contest algorithmic profiling bear the heaviest cost.
People assume predictive policing simply "follows the data" — and is therefore objective. The data is historical arrests, not historical crime. Those aren't the same thing: many crimes are never reported or detected, and enforcement has historically concentrated in minority neighborhoods. An algorithm trained on biased data produces biased predictions, precisely and at scale.
Predictive policing puts people under surveillance and police contact because of where they live or who they know — not because of anything they've done. When these systems drive investigative stops, they launder historical discrimination into official police decisions. Communities with less power to contest algorithmic profiling bear the heaviest cost.
People assume predictive policing simply "follows the data" — and is therefore objective. The data is historical arrests, not historical crime. Those aren't the same thing: many crimes are never reported or detected, and enforcement has historically concentrated in minority neighborhoods. An algorithm trained on biased data produces biased predictions, precisely and at scale.