Artificial Intelligence-Assisted Sentencing: Rationale, Procedural Justice Risks, and Mitigation Pathways

Abstract

The development of smart justice and the continuing reform of sentencing standardization have accelerated the practical application of artificial intelligence in sentencing. AI-assisted sentencing operates by transform ing legal rules into computable data elements and by learning adjudicative patterns from large volumes of judicial data. The explainability of such systems rests, in significant part, on the formal-rational transforma tion between legal rules and algorithmic data. Yet the introduction of artificial intelligence into sentencing also reshapes traditional judicial procedures and gives rise to a range of procedural justice concerns, partic ularly in relation to judicial agency, the legitimacy of decision-making authority, algorithmic opacity, and the reviewability of machine-generated recommendations. These risks require a clear institutional response. Artificial intelligence should remain confined to an auxiliary role in adjudication, while human judges must retain final decision-making authority and responsibility. At the same time, the quality of judicial data and the explainability of algorithmic processes should be improved. Procedural safeguards must also be strengthened by ensuring defendants’ rights to information, participation, explanation, and meaningful challenge, together with effective systems of judicial review, independent oversight, and institutional feedback. These measures can help reconcile the efficiency gains of AI-assisted sentencing with the procedural requirements of legiti mate criminal adjudication.