Trust, Limits, Ethics of AI-Based Recidivism Prediction versus Traditional Assessment Tools: A Comparative Analysis (KOSCA)

• In Event: Roundtable: AI in Criminal Justice

Abstract

Advancements in AI technology are opening new possibilities in the field of recidivism risk assessment for offenders. Traditional static and dynamic risk assessment tools have primarily relied on fixed characteristics and certain behavioral factors of offenders to predict the likelihood of reoffending. Recently, AI-based predictive models that analyze daily life data and behavioral patterns in real time have attracted growing attention. This presentation compares the key differences between AI-driven models and conventional assessment tools, and proposes strategies to enhance the reliability and accuracy of AI models, including data quality management, improving algorithmic explainability, and establishing external verification systems. Furthermore, the presentation addresses the ethical and legal issues associated with the use of AI in risk assessments, such as privacy violations, decision-making errors, and stigmatization, and discusses necessary legal and institutional safeguards. Finally, by comparing cases from Korea and the United States, the presentation explores the feasibility and social acceptability of introducing AI-based risk assessment models in Korea.

Authors

• Yoori Seong, Korean Institute of Criminology and Justice

• Jisun Choi, Korean Institute of Criminology and Justice

• Kyung-Shick Choi, Boston University

• Hyesun Kim, Worcester State University

• Jae-Seung Lee, Miami University

• Jiseun Sohn, Southern University and A&M College