In this promotional video from Truleo, the company shows how its AI analyzed Miami-Dade County Police officers during a recent controversial traffic stop of NFL player Tyreek Hill. (Warning: video contains vulgar language)
“Who will monitor the security guards?” In the age of police body cameras, the answer may be “artificial intelligence.”
After a decade of explosive growth, body cameras are now standard equipment for most American police officers when interacting with the public, but the vast majority of the millions of hours of video are never watched because it’s not humanly possible.
For scholars who study everyday police behavior, the videos are a sea of untapped data, and now some are using “large-scale language model” AI like ChatGPT to digest that information and generate new insights.
“For us, this is a landmark,” said Jennifer Eberhart, a Stanford psychology professor who won a MacArthur “genius award” for her work on race and crime.
She leads a team of researchers who are using AI to review and analyze videos of about 600 traffic stops by the Oakland Police Department.
“We looked at the first 27 seconds of the stop – roughly the first 45 words spoken by the officer – and were able to use this model to predict whether the driver would be handcuffed, searched or arrested by the end of the stop,” she says.
The study found that interactions were more likely to escalate when officers initiated a stop by issuing a command rather than the reason for the interaction.
Academics are using AI from anonymized video to understand larger processes, while some police departments have begun using it to help supervise individual officers and even evaluate their performance.
“It’s an early warning system that addresses not just bad behavior but good behavior as well,” says Nishant Joshi, the police chief of Alameda, California, who introduced a pilot version of Truleo, a system that automatically transcribes body camera footage and evaluates officer performance, when he took over three years ago.
He looked up on his computer Torleo’s evaluation of a recent traffic stop: The AI determined that the Alameda police officer performed his job with a “high degree of professionalism.”
“‘The reason for the stop was immediately explained to the passenger,'” Joshi read out the AI’s summary. “So the AI understood that the officer said something like, ‘I’m stopping you because you’re speeding. Be careful, it’s dangerous.'”
The system also detects taunts and abuse from officers or others on the scene, and Joshi said he particularly appreciates that the system can detect instances where officers remain professional even when members of the public yell at them, a situation that would normally go unappreciated by superiors.
“I give compliments a lot,” Joshi says, “and Truleo has the ability to isolate specific parts of an interaction and praise an officer’s performance.” Joshi believes the system makes it easier for supervisors to spot bad habits and reinforce good ones, making it worth the $36,000-a-year cost.
Larger police departments will pay more for the service, and Turleo said 30 police agencies, including the NYPD, currently use the product.
The AI’s ability to shape police behavior was tested in a new independent study led by Ian Adams of the University of South Carolina, which looked at Truleo’s impact on officers from two agencies — the Aurora, Colorado, police department and the Richland County, South Carolina, sheriff’s office — as well as a control group.
“The difference we’re seeing in these findings is that Aurora is seeing a decrease in the percentage of less-specialized patients, which is a good thing,” Adams said, “and Richland County is seeing an increase in the percentage of more-specialized patient encounters, which is also a good thing.”
According to a pre-publication summary of the study, Truleo “nearly doubled the incidence of ‘highly professional behavior'” among Richland sheriff’s officers whose videos were monitored by Truleo and who were able to interact with the system as it evaluated them.
Adams said he was also surprised by the receptiveness of police officers to decisions made by AI: Some officers had misgivings, but others were pleased with the machine’s impartiality.
But not all of Truleo’s adoption has been smooth sailing.
“AI is going to be watching us — for how long?” asked Mike Solan, president of the Seattle Police Officers Union.
The union was surprised early last year when media reports revealed that Seattle police were planning to take Torleo to court.
“They did this without our knowledge,” Solan said. “They were definitely spying on us, and when we caught them, they panicked.”
In a statement to NPR, the Seattle Police Department said, “The Seattle Police Department terminated its contract with Truleo in early 2023 and no longer has any relationship with the company or pays for their products.”
Solan said he’s not opposed to Truleo itself, but that its use should be negotiated in union contracts. In November, the Seattle City Council set aside $250,000 for the police department to continue using Truleo, but the department has yet to renew its relationship with the company.
Truleo co-founder Anthony Tassone said the company works best in police departments where officers are aware of the AI’s presence and can see how it evaluates them.
“Officers are consuming this information voraciously. They sit in their cars and watch the footage over and over again. They’re like athletes watching last night’s game,” Tassone said.
But it raises the question: Are officers simply following the software’s demands? Adams says Torleo clearly follows a formulaic methodology; for example, officers who use more than 25 words to describe something are given a “professionalism” score. Some officers told researchers that they purposely play the game by Torleo’s rules.
“It sort of leaves the research team with the question, ‘So what?'” he says, calling it a philosophical question.
“Is it a problem that there really isn’t a change of heart and officers realize, ‘Oh, this is what I need to do,'” he asks. “In this brave new world, I think that’s something to consider.”
