Publics’ Perceptions of Machine Learning Based Risk Assessments

In the legal system, the prediction of a person’s risk of committing a crime has mostly been based on expert judgment. However, newer techniques that employ machine learning (ML)—a type of artificial intelligence—are being implemented throughout the justice system. Yet, there is a lack of research on how the public perceives and uses machine learning risk assessments in legal settings. In two mock-trial vignette studies, the perception of ML-based risk assessments versus more traditional methods was assessed. Study 1 was a 2 (severity of crime: low, high) x 2 (risk assessment type: expert, machine learning) x 2 (risk outcome: low, high) between-subjects design. Participants expressed ethical concerns and discouraged the use of machine learning risk assessments in sentencing decisions, but punishment recommendations were not affected. Study 2 was a within-subjects design where participants were randomly assigned read through one of three crime scenarios (violent, white-collar, sex offense) and one of three risk assessment techniques (expert, checklist, machine learning). Consistent with Study 1, participants had ethical concerns and disagreed with the use of machine learning risk assessments in bail decisions, yet their own decisions and recommendations did not reflect these concerns. Overall, laypeople express skepticism toward these new methods, but do not appear to differentially rely on ML-based versus traditional risk assessments in their own judgments.