Beyond the Unbalanced Scale: Re-Designing AI Collaboration in the Judiciary with Human-Centered Principles

Abstract Human-AI collaboration in the judiciary is widely regarded as a promising approach to reconcile efficiency with fairness, thus gradually gaining mainstream recognition. However, the current human-AI collaborative mode in the judicial domain remains an AI-oriented and unbalanced one, thereby limiting its capacity to achieve truly human-centric collaborative outcomes. This study constructs a specialized classification system for human-machine collaboration in the judicial domain, identifying three distinct collaborative modes: the Pipeline Mode, the Hybrid Mode, and the Augmented Mode. Grounded in this classification, the study reveals three forms of structural imbalance: information imbalance caused by system closure, communication imbalance caused by differences in decision-making methods, and dependency imbalance resulting from the disparity in computing capabilities. These imbalances help explain why existing collaborative arrangements frequently yield suboptimal results. Based on these findings, this paper proposes a set of human-centered collaborative principles and explores the inherent trade-offs among them. Furthermore, it also explores future research directions from three dimensions: human, technology, and norms, aiming to further enhance the application effectiveness of legal artificial intelligence and ensure that technology always serves judicial justice.