Spokane County Sheriff’s Office to use AI to review body camera footage in effort to improve training practices

The Spokane County Sheriff’s Office has recorded thousands of hours of footage captured by its deputies’ body cameras, making it nearly impossible for the agency to review every traffic stop, welfare check or response to a crime scene.

That’s partially why the agency has tapped artificial intelligence to do the heavy lifting of reviewing the footage to evaluate how effective training is for county deputies.

The Spokane County Commission voted unanimously Tuesday to accept a nearly $1 million grant from the U.S. Department of Justice that will be used to analyze footage from body cameras worn by county deputies.

For the next three years, an AI and data management system called “TrustStat” will comb through recorded interactions between the deputies and the public, with a special focus on identifying “key behaviors and language related to de-escalation, use of force, and other critical areas of deputy performance,” according to a news release from the Justice Department.

Spokane County Sheriff John Nowels said the agency is likely the first in the country to apply the software specifically to a review of training practices. He added that the program is part of the agency’s continual efforts to improve practices, build community trust and develop better deputies.

“My interest has always been human performance,” Nowels said. “How do we make the most professional law enforcement officers we can, in every sense of the word? How do we get people who are high performers to perform even better?”

Nowels said he’s hopeful the program will provide empirical data on what strategies and training are effective, and that the results could impact practices at law enforcement agencies across the country.

TrustStat, a software developed by Dallas-based Polis Solutions, is an analysis tool already used by a handful of law enforcement agencies across the country, including agencies in the company’s hometown, in St. Petersburg, Florida; Kinston, North Carolina; and Alliance, Nebraska, as reported by ProPublica.

The software was born out of a former research project of the U.S. Defense Department called the Strategic Social Interaction Modules program, nicknamed “Good Stranger,” which sought to better train soldiers for modern warfare in which they need to rely on social skills to de-escalate terse situations or navigate uncertain circumstances.

The $40 million research project was workshopped at Fairchild Air Force Base with some Washington law enforcement officers and former military members more than a decade ago, as reported by NBC News. Polis Solutions, whose founder is a former Defense Department employee, trained the TrustStat software on the program’s database.

Nowels said part of the reason TrustStat was selected as the vendor is an existing relationship with the Sheriff’s Office’s Training Director Tony Anderman, who participated in the Defense Department’s project.

TrustStat is entirely automated: Large language models analyze speech, and image -processing algorithms identify physical movements and facial expressions captured on video, ProPublica reports.

The artificial intelligence will identify deputies’ emotional responses in their interactions, beginning with a baseline reading during training for the agency to cross-reference with data pulled from real interactions once that deputy is on the streets, Nowels said.

TrustStat will analyze things like voice intonation, inflection or body position to determine what emotions might be present in the deputy and the individuals they are interacting with, Nowels said.

What the software will not do is review whether a deputy’s use of force was appropriate, if a person contacted by deputies or the deputies themselves acted within their rights, or if there were policy or conduct violations. The grant’s language limits its use to a review of training practices, Nowels said.

“I know there’s a lot of people who want to say, ‘This artificial machine is going to catch deputies behaving badly,’ or it’s going to catch themselves behaving badly,” Nowels said. “Let me be very clear – that is not what it is designed to do, and that’s not what we’re going to use it for.”

Nowels said that doesn’t mean he won’t take appropriate action if violations are discovered through the analysis.

The tasks of determining what all the agency hopes to answer about its training practices, reviewing the findings and taking actionable steps will fall on a board of Sheriff’s Office deputies ranging from the rank and file to top command officers, which is yet to be finalized, Nowels said.

Some of those questions will likely key in on use-of-force incidents, like what common factors the events share, what emotions were present and how the agency’s response could have been improved. The county may have empirical data to that effect in the years to come, Nowels said.

“But for right now, it’s about, how do we assess the training models that we’re using to make sure that the information that we are giving to our deputies is received, ingrained into their behavior and retained once they’re out on the street,” Nowels said.