형사절차에서 인공지능 위험평가 프로그램을 이용한 양형 판단

This article explores the feasibility of employing a risk assessment program, which has manifested as one of the most tangible applications of AI in the criminal justice system. The Wisconsin Supreme Court’s decision validating COMPAS, an American risk assessment program, on the grounds of not violating due process, marks a significant precedent. In parallel, the EU seeks to regulate algorithms and automated decision-making through the proposed EU Artificial Intelligence Act, drawing comparison with the United States Algorithm Accountability Act of 2022 for Automated Decision-making. The Algorithmic Accountability Act of 2022 mandates a framework where the FTC can compel companies to ensure transparency and accountability in the automated systems through impact assessments. Artificial intelligence programs integrated into sentencing have the potential to bolster public trust in trials by minimizing sentencing variations and enhancing fairness. However, the use of ‘risk’ of recidivism in sentencing decisions, a critical aspect of criminal law, necessitates a careful evaluation of legitimacy. Therefore, the application of risk assessment programs in sentencing should be distinct from their use in other criminal procedures such as predictive policing, parole, and security dispositions, which primarily focus on predicting the ‘risk of recidivism’. The factors of the risk assessment program should align with the sentencing factors determined by the Sentencing Commission. Any expansion or restriction of evaluation factors requires normative judgments from experts based on criminal law principles, notably the principle of responsibility. Additionally, this program must adhere to the standards of explainable artificial intelligence, aligning with the public perception, ethical considerations, and transparency imperative required for artificial intelligence systems.