The Richland County Sheriff's Department has invested in AI technology to monitor body camera footage. COLUMBIA — The assistant in charge of reviewing hundreds of hours of body camera footage for the
Richland County Sheriff's Department each week doesn't have a desk and never takes time off. You could say the assistant's wide-ranging capabilities aren't even human, and that's because this assistant is no normal law enforcement employee. Since the beginning of 2024, Richland County's largest law enforcement agency has been using artificial intelligence software to monitor the department's slog of body camera footage. The technology is expensive but those familiar with it say it's extremely effective at monitoring deputies' behavior while on the job. The service is provided by
AI company Truleo , which claims to be the most “ethical, unbiased, and evidence-based AI assistant in policing.” The sheriff's department contracted with the company in 2023 and was one of the first departments in the state to adopt AI software to review body camera videos. Before discounts, the price to monitor all 180 patrol and “patrol adjacent” officers for one year came out to $228,000. But taxpayers aren’t footing the bill. The sheriff's department
partnered with researchers at the University of South Carolina to test the effectiveness of the software over a six-month trial basis starting in January 2024. “Body worn cameras represent the largest single data source we have on officer behavior,” said
Ian Adams, assistant professor of criminology at USC , and author of the study. “Most of it remains hidden from us, because we will never have enough human labor to actually review it in detail.” Truleo claims to be able to deliver audio playback “more accurate than human review” with the ability to detect and categorize different policing events with more than 90 percent accuracy. “This type of software is trying to solve a real problem in American policing,” Adams said. The sheriff's department said the study has been extended, and they're still participating in it today. The cost of the software is taken care of by a third-party grant.
Detecting professionalism on the job
In 2015, South Carolina passed a
law mandating the use of body-worn cameras in police departments across the state. But the law didn’t explain how the oversight for that footage would work. “If you’re going to assign a human being to audit an hour of body worn camera footage, you need at least an hour to do it,” Adams said. “If you’re generating tens of thousands of hours, that ends up becoming a burden that just can’t really be met.” Truleo doesn’t monitor the actual footage — rather, it creates a transcript of the footage from the audio captures, and then uses its algorithms to detect different events, like how often arrests occur. According to Jennifer Spurrier, who manages accreditation for the sheriff's department, the audio is analyzed by the software within 10-15 minutes of it being uploaded. “If you detect in the transcript something like ‘turn around and place your hands behind your back,’ that’s a pretty good clue, algorithmically, that an arrest is about to take place, right?” In Adams’ study, they used the software to detect the standard of “professionalism” the officers were employing. They measured this in twofold: the presence of explanatory language before taking action, and the absence of demeaning language.
Richland County Sheriff's Department Master Deputy Clayton Graham speaks with Deputy Sheriff Alex Fulmer during a traffic stop on St. Andrews Road on Feb. 3. At the Richland County Sheriff's Department, the study found that the number of highly professional encounters doubled when deputies had access to the AI generated feedback. The department's policy states that uniformed deputies must have their body cameras on any time they are interacting with members of the public, and that footage is never deleted. “Across our work, there’s pretty good evidence to suggest that providing this information directly to the officers themselves is valuable and effective,” Adams said. “That doesn’t mean that, in the real world, maybe you want both officers and supervisors to have access to that information.” Deputy Collins Harper said when he first started policing, he wasn’t mandated to wear a body camera. “I think that these hold people a lot more accountable,” said Harper at an April 4 Media Day panel. After the study, the department continued to use Truleo. Other than professionalism, the system can also measure escalation, de-escalation, and conflict resolution based on the language officers are using. “Language offers us a lot of ways to try and think about what’s going on behaviorally,” Adams said.
'Human in the loop'
Adams notes the transcription isn’t always perfect, nor should it be expected to be. The program employs what they refer to as a “human in the loop.” “The algorithm might raise a concern, but ultimately it’s a human who validates whether that actually happened,” Adams said. Spurrier said that supervisors are required to look at the body camera footage at the end of their shift. “There is an audit trail, so you can see who views any of that footage,” Spurrier said at the April 4 panel. However, those AI generated transcripts are only reviewed by supervisors "at their discretion." Deputies are able to self-audit with the system. If a “defensive action” is taken, like a pursuit, then that body camera footage gets monitored by four different people — the sergeant in charge, the lieutenant captain, and two different people in the professional standards departments.
Richland County Sheriff Leon Lott. Spurrier said the department also has the ability to tap into any camera that’s actively recording at any time. As for what kind of events the AI could be trained to flag, Adams said “the sky's the limit.” “I think most agencies with body cameras are going to have to start considering this (software), because all the motivations for everybody are aligned in the same direction,” Adams said. “Let’s identify where officers are doing well, let’s identify where we can improve.”
