Nicholas Camp, an assistant professor of organizational studies at the University of Michigan, was tapped with other experts from the University of Michigan, University of California-Davis and Stanford who together analyzed body camera footage from New York Police Department pedestrian officers.
Around 2021, researchers like Camp were brought in to analyze footage from the New York Police Department as part of a federal oversight effort. It came after a judge ruled the department’s stop-and-frisk practices unconstitutional.
“This is not about replacing human judgment — it is about focusing it,” Camp told Spectrum News. “One of the biggest challenges in police accountability is just the sheer number of interactions that law enforcement have with the public, and the number of videos they generate."
[...]
“What AI does here is help us find needles in the haystack and figure out where auditors should be concentrating their efforts,” Camp said. Instead of spending hours watching routine encounters, reviewers can focus on the moments that matter most, making oversight faster and more effective.
A highlighted finding from the study concerns gaps in how officers request for consent to search — a key protection under the Fourth Amendment. The word “consent” appeared in less than 13% of consent-search interactions analyzed. Instead, officers often used indirect phrases such as “Do you mind if I check?” or “Can I check?” Those phrases can be ambiguous, Camp said, raising concerns about whether civilians clearly understood that they had the right to refuse.
The study also found indirect or command-based language appeared more often in encounters involving Black and Hispanic civilians than with white civilians. He said those patterns raise potential concerns under the Fourteenth Amendment, which prohibits discriminatory practices based on race.
[...]
While not perfect, the findings suggest artificial intelligence can reliably identify encounters that carry constitutional implications, including stops that may not have been properly documented, offering a new tool to surface patterns that traditional reviews might miss.
