Проблема «Чорної скриньки» в юридичному ШІ: чи достатньо існуючих методів XAI для задоволення вимог належного правового процесу?

The integration of artificial intelligence (AI) in legal practice creates the «black box» problem, where opaque models challenge the principles of due process. This opacity conflicts with the right to a reasoned and contestable decision. This article investigates the fundamental gap between the technical explanations provided by modern XAI methods (like LIME, SHAP) and the normative justifications required by the doctrine of due process. The goal is to critically evaluate XAI’s limitations and, based on these findings, to formulate a new standard for transparency in legal AI. The study is based on a comprehensive application of legal and scientific methods, including comparative-legal and formal-logical analysis. This approach was used to examine the interplay between AI technology and legal doctrines, analyzing legal acts and scientific works to identify inconsistencies between technical explanations and legal requirements. The article demonstrates that popular post-hoc XAI methods are fundamentally insufficient for legal needs. They generate unstable technical approximations of a model’s behavior, not legally meaningful justifications. These explanations can be unreliable and create a “Rashomon effect” (multiple contradictory explanations for one outcome), making the right to an effective appeal illusory. The law requires justification, not just explanation. The article proposes a new standard: «Justifiable AI» (JAI). This concept shifts focus from explaining opaque models to designing inherently interpretable hybrid systems that combine rule-based components with data-driven models. Future research should focus on developing JAI architectures, creating legal certification standards, and studying user trust in such systems.