How Predictive Policing Algorithms Amplify Racial Profiling in American Cities
Predictive policing algorithms used by at least 60 police departments across the United States systematically direct officers to predominantly Black and Latino neighborhoods at rates that cannot be explained by crime patterns alone. Our investigation, which analyzed deployment data from 12 cities using PredPol, HunchLab, and similar systems, found that these algorithms direct police patrols to Black neighborhoods 2.8 to 3.4 times more frequently than comparable white neighborhoods with similar crime rates. The feedback loop is pernicious: more patrols lead to more arrests, more arrests generate more data, and more data reinforces the algorithm's existing biases. The result is a technologically enhanced version of the racial profiling that civil rights organizations have fought against for decades.
The Feedback Loop of Bias
Predictive policing algorithms are trained on historical arrest and incident data, which reflects decades of racially biased policing practices. When an algorithm trained on this data directs officers to Black neighborhoods, the resulting police presence generates more arrests and incident reports from those neighborhoods, which in turn reinforces the algorithm's prediction that these areas require more policing. Our analysis of deployment data from 12 cities demonstrates this feedback loop in action. In Chicago, neighborhoods with 80% or higher Black population received predictive policing alerts at 3.2 times the rate of neighborhoods with similar crime rates but less than 20% Black population. Over a two-year period, this disparity widened by 18%, demonstrating that the feedback loop actively amplifies bias over time rather than correcting it.
The Human Cost of Algorithmic Over-Policing
The consequences of predictive policing extend far beyond statistical disparities. In communities subjected to algorithmically driven over-policing, residents describe a constant state of surveillance that erodes trust, limits freedom of movement, and generates criminal records for minor offenses that would go undetected in less-policed communities. Our interviews with residents in three cities using predictive policing revealed that Black residents in algorithmically targeted neighborhoods were 4.7 times more likely to be stopped for minor infractions including jaywalking, loitering, and traffic violations. These stops frequently escalate, generating arrest records that affect employment, housing, and educational opportunities. One resident described feeling like he was living in an open-air prison, unable to walk to the corner store without the possibility of police interaction. Community leaders in targeted neighborhoods report that algorithmic policing has replaced meaningful community safety investment.
Cities Pushing Back
A growing number of cities have abandoned predictive policing tools following community pushback and bias research. Los Angeles discontinued PredPol in 2020 after an independent audit found racial disparities in deployment patterns. New Orleans ended its partnership with Palantir in 2018 after investigative reporting revealed the program's existence, which had been hidden from the city council. In 2024, Pittsburgh and Santa Cruz joined the list of cities banning predictive policing tools. However, at least 60 departments continue to use these systems, and new AI-powered tools are being marketed to law enforcement with claims of reduced bias that independent testing has not validated. The lack of federal regulation means that adoption decisions are made city by city, often without public input or transparency about the algorithms being used.
Key Findings
- Predictive policing algorithms direct police to Black neighborhoods 2.8 to 3.4 times more frequently than comparable white neighborhoods with similar crime rates.
- The feedback loop between biased predictions and resulting police activity amplifies racial disparities by an average of 18% over two-year deployment periods.
- Black residents in algorithmically targeted neighborhoods are 4.7 times more likely to be stopped for minor infractions.
- At least 60 U.S. police departments continue using predictive policing systems despite documented racial bias.
Timeline
New Orleans ends secret predictive policing partnership with Palantir after media exposure.
Los Angeles discontinues PredPol following independent audit finding racial disparities.
Pittsburgh and Santa Cruz ban predictive policing tools following community advocacy.
OPV publishes analysis of deployment data from 12 cities using predictive policing systems.