SPARQ

Featured

What Can AI Tell Us About NYPD Street Stops?

A new report by SPARQ Faculty Co-Director Jennifer Eberhardt, Faculty Affiliates Rob Voigt and Nick Camp, and Dan Sutton (Stanford Center for Racial Justice) investigates whether computational analysis of police body-camera footage can strengthen efforts to monitor police departments’ constitutional compliance.

In 2013, the New York Police Department (NYPD) was placed under federal monitoring due to  unconstitutional stop-and-frisk practices. Over a decade later, the agency was found to be out of compliance because of three persistent issues: underreporting of police-civilian encounters, unlawful stops initiated by officers, and a lack of accountability for supervisors. While the NYPD records millions of investigative encounters each year, ongoing monitoring efforts only review a small fraction of the footage, making it difficult to evaluate compliance at scale.

To address this issue, the team leveraged AI-powered computational approaches to analyze large amounts of police body-camera footage. In the report, the team demonstrates that machine learning models can reliably distinguish between types of police encounters—for instance, between a request for information during which the person is free to leave and a stop that includes probable cause to make an arrest—at rates substantially better than chance. The report also finds that natural language processing tools can help examine how a stop unfolds, including differences in what officers say. In addition, both machine learning models and linguistic analysis can be used to detect when and how racial disparities occur in police interactions. This work shows that key indicators of constitutional compliance can be analyzed at scale and that computational tools can significantly aid monitoring efforts.

Updates