How is AI used in predictive policing and why is it controversial

Forecasting Patterns, Not Specific Crimes

Predictive policing uses AI to analyze historical crime data in order to forecast broader patterns — areas or times where crime is statistically more likely to occur, or in some more controversial implementations, individuals considered statistically higher-risk of future involvement in crime — rather than predicting specific individual crimes with certainty before they happen.

How Location-Based Predictive Policing Generally Works

The more common form of predictive policing analyzes historical crime data to identify patterns in when and where certain types of crime have tended to occur, generating forecasts used to help allocate police patrol resources toward areas identified as higher-risk based on these historical patterns.

How Individual-Based Predictive Policing Has Been Used

Some more controversial implementations have attempted to identify specific individuals considered statistically higher-risk of future involvement in crime, based on factors like prior criminal history or social network associations — an approach that has drawn particularly significant criticism and, in some documented cases, has been discontinued due to serious concerns about fairness and effectiveness.

The Central, Well-Documented Controversy

A core concern driving controversy around predictive policing is that historical crime data doesn’t purely reflect actual underlying crime rates — it also reflects where and how police have historically chosen to patrol and enforce, meaning areas that were more heavily policed in the past, sometimes due to documented historical bias, will show more recorded crime in the data, regardless of whether actual crime rates in those areas were genuinely higher.

Why This Creates a Feedback Loop Risk

Because predictive policing systems are trained on this historical data, they risk directing more future policing attention toward these same historically over-policed areas, generating even more recorded police activity and enforcement there, which can further reinforce and perpetuate the original biased pattern rather than reflecting an objective, unbiased assessment of where crime actually occurs.

Why Some Jurisdictions Have Scaled Back or Ended These Programs

Given these documented concerns, along with mixed evidence about whether these systems genuinely improve public safety outcomes relative to their costs and risks, several police departments and cities have discontinued or significantly scaled back predictive policing programs, reflecting a broader, ongoing reassessment of this application of AI within law enforcement.

Bottom Line

Predictive policing uses AI to analyze historical crime data to forecast likely crime locations, times, or, more controversially, higher-risk individuals, and it’s controversial primarily because historical crime data can reflect prior biased or disproportionate policing rather than purely objective crime patterns, risking a feedback loop that reinforces existing bias — a concern serious enough that several jurisdictions have scaled back or discontinued these programs.

Go deeper

• How Do Retailers Use AI to Spot Organized Retail Crime Rings?
• What happens when an ai underwriting model is trained on biased historical claims data?