Published May 1, 2018 | The Psychiatric Quarterly
Artificial intelligence can be as useful and as accurate in determining a teen’s risk of committing violence at school as one-on-one interviews by a team of child, adolescent and forensic psychiatrists, according to a pilot study led by Drew Barzman, MD, Director of Child and Adolescent Forensic Psychiatry Service.
Barzman and colleagues studied data from interviews with 103 “at risk” students ages 12-18 from 74 schools in Ohio, Indiana, Kentucky and Tennessee.
The researchers used two standard scales, the Brief Rating of Aggression by Children and Adolescents (BRACHA) and the School Safety Scale (SSS), to manually annotate interview transcripts to detect content, words and phrases that indicated a risk of violence.
Then they used machine-learning software to analyze the interviews. The software was 91 percent as accurate as the clinical judgments.
These results suggest that machine-learning algorithms eventually could play valuable roles in identifying violence-related risk levels, risk factors and protective factors among teens.
“Eventually, machine learning will be able to complete the annotation process in real time during the interview to provide useful output,” Barzman says. “We’re going to be able to develop artificial intelligence that can be used in schools around the country to identify kids at risk of violence so that we can intervene or provide resources that lower the risk.”
Barzman’s team included 19- and 20-year-olds who provided generational insights into references about social media, video games and popular culture.
They noticed that high-risk teens who were referred for evaluation “had a pattern of misinterpreting negatively the neutral actions of their peers, and that’s something that can be addressed,” Barzman says.
The immediate and recurring challenge, he adds, is finding funding to continue studying school violence.