Norfolk's Commonwealth's Attorney's Office is teaming up with Harvard researchers to test a new AI-based tool.
Norfolk City Council voted Tuesday to accept a $38,251 grant from Harvard University to fund a part-time data analyst position in the Commonwealth's Attorney's Office.
The money comes from Harvard's Access to Justice Lab, and is tied to a 24-month study called "Triage Scores,”, with Norfolk and Philadelphia Legal Assistance as the two initial partner sites.
The Harvard Crimson reported that Harvard Law School professor D. James Greiner and statistics and government professor Kosuke Imai are developing AI-based "triage scores" as an alternative to commonly used risk scores. traditional risk scores estimate the likelihood of negative outcomes like reoffending, but triage scores are designed to weigh the potential benefits and drawbacks of different decisions, the Crimson reported.
"It's us and the Philadelphia Public Defender Service, coming at it from different directions, but we were the two national partners that got paid to work with these researchers,” Fatehi said.
The study focuses on plea agreements, which resolve more than 90% of criminal cases in the United States, according to Fatehi. Researchers want to know whether offering a plea deal actually reduces someone's risk of reoffending, has no effect, or makes things worse.
"A triage tool will say, okay, fine, low, medium, high risk for release, for example," Fatehi said. "But what happens if you do release? Does that increase or decrease their risk of reoffending, at any given risk level?"
The grant funds a part-time data analyst who will work inside the Commonwealth's Attorney's Office, compiling case data for Harvard researchers to build the triage model, then tracking outcomes once it's tested. Fatehi said the city's role is limited to accepting and disbursing the grant funds, with Harvard and his office designing the study itself.
Fatehi said there is no outside body independently overseeing how the tool's recommendations get used once testing begins. He said his office will treat itself as the safeguard, evaluating cases individually rather than following the model's output automatically.
"My office and I are the safeguards ourselves, in the same way that we are in cases where we're not using assessment tools or triage tools, where we are looking at a human being as a human being, both victim and accused, and making sure that we are making an individualized assessment and not just mechanically following some sort of flow trap,” he said.
Fatehi said Harvard researchers plan to publish findings from the study once it concludes.
