Может ли искусственный интеллект прогнозировать решения суда? Системати­ческий обзор международных исследований

Мониторинг общественного мнения: Экономические и социальные перемены. 2024. № 5. С. 100–122.

Kazun A.

Advancements in artificial intelligence technologies and the emergence of open databases containing judicial decisions have led to rapid improvements in algorithms capable of classifying legal documents and forecasting decisions made by judges. This article examines a body of international research dedicated to the question of how accurately AI can predict judges’ decisions, and consequently, whether it could potentially replace human judges in the future. The answer to this question is formed through the analysis of two key aspects: the capability and accuracy of predicting judicial decisions and the various constraints associated with the use of AI.

Analysis of international experience shows that the accuracy of predictions has increased in recent years; however, the quality of the models depends greatly on the specificity of the tasks and the available data. Most studies analyze decisions from higher courts worldwide, significantly reducing their practical potential for dealing with mass categories of cases. Moreover, concerns have arisen regarding the use of models that operate on a “black box” principle, as their decisions are difficult to interpret. Despite the rapid development of AI technologies, the complete replacement of judges is unlikely because of the range of methodological limitations, including insufficient quality and volume of data, issues with interpretability, challenges in understanding legal and cultural context, and limitations in transferring models to other legal systems. However, AI technologies can be used to reduce the costs associated with case material handling.

Keywords: искусственный интеллект, машинное обучение, machine learning, deep learning, глубокое обучение, artificial intelligence, legal document classification, algorithmic accuracy, prediction of judicial decisions, предсказание судебных решений, классификация юридических документов, точность алгоритмов