作者：衣俊霖
作者介绍：上海交通大学凯原法学院助理教授，法学博士
出版日期：2025年4月9日
内容提要 随着人工智能在我国智慧司法建设中的普及推广，司法审判逐步呈现出“法官为主、机器为辅”的人机协同审判格局。目前，司法人工智能暂时消隐于法官背后，审判信任被置换为对法官人格、法院制度或人工智能算法的信任。人格信任奠基于法官作为“完人”的应然预设，与实然的法官“常人”之间存在现实差距。法院组织以司法责任制作为约束人机互动的主要制度安排，但在预测审判逻辑的影响下，存在盲目采纳机器建议的“人机合谋”风险。算法信任机制尽管能发挥局部的信任补救功能，但囿于“主体—工具”的二分思维，难以对法官与算法的协同过程进行约束。应当以信任者与信任对象的沟通互动为出发点，对人格信任、制度信任和算法信任进行反思性整合，将司法人工智能从内部办案工具拓展为外部沟通工具，推动人机协同法律议论的充分展开，为人机协同审判奠定更为坚实的信任基础。
关键词 人工智能 人机协同审判 审判信任 法律议论
Abstract: With the rise of Artificial Intelligence (AI) in smart courts, judicial trials have been gradually shifting toward a human-machine collaborative model, where ‘the judge leads and the machine assists’. Currently, judicial AI remains temporarily hidden behind judges, with trial trust being replaced by personal trust in judges, systemic trust in courts, or algorithmic trust in AI. Personal trust, however, assumes judges are ‘perfect individuals’, ignoring their reality as ‘ordinary people’. While courts use judicial responsibility mechanisms to regulate the interaction between judges and algorithms, the predictive nature of algorithms risks the uncritical adoption of machine suggestions, leading to the potential risk of ‘human-machine collusion’. Although algorithmic trust serves as a partial remedy for trust deficiencies, it struggles to regulate the process of judge-algorithm collaboration due to the ‘subject-tool’ dichotomy. It is essential to conduct a reflective integration of personality, institutional, and algorithmic trust based on communicative interactions, redefine judicial AI as both an internal case-handling tool and an external communication tool, and fully promote the legal argumentation of human-machine collaboration, thereby laying a stronger foundation of trust for human-machine collaborative adjudication.