PHILADELPHIA (KYW Newsradio) — Artificial intelligence is intersecting with criminal justice in a new way: Researchers at the University of Pennsylvania have created a tool to help prosecutors and legal professionals determine the accuracy of eyewitness suspect identifications.
When investigators use eyewitnesses to identify a suspect or defendant, they usually take a statement.
“They make an identification, and then they use normal words to describe their confidence,” says Paul Heaton, academic director of the Quattrone Center for the Fair Administration of Justice at Penn’s Carey Law School.
Heaton and his team have developed an A.I.-based tool that he says can parse those statements and help determine how confident witnesses are, rather than relying on the subjective judgment of investigators. Heaton says it could help to reduce incidents of misidentification.
“One of the challenges that arises with eyewitness statements is in the courtroom,” he says.
“Everyone is super confident, right at trial — ‘For sure, I know that I got the right person.’ But what we know is that people make mistakes, and sometimes they identify the wrong person.”
Now prosecutors — or defense attorneys — can run witness statements through the program to help distinguish between accurate identifications and faulty ones.
“Having the AI provides a more neutral, objective way of making that judgment,” Heaton said.
Heaton’s team used more than 4,500 statements to create the A.I. model, which they say gets it right about 70% of the time. But with every statement processed through the program, they believe, the A.I. will grow “smarter.”
“One of the advantages of having one of these A.I.-based large language models is, as you get additional data, you can refine the training as the models themselves improve their ability to assess semantic content. The models are going to naturally improve.”