《保险研究》20260109-《论人工智能责任保险中的因果关系判定——以模型“可解释性”为中心》（杨一凡）

［摘   要］以模型“可解释性”为中心，人工智能可类型化为“白箱”“灰箱”和“黑箱”三种形态，各形态对应的责任保险中也应差异化适用因果关系判定规则。“白箱”人工智能责任保险中，模型具有完全“可解释性”意味着损失结果的“近因”构成与效力占比明确，适用“比例分配”的因果关系判定规则更能尊重保险双方的合理期待。“灰箱”人工智能责任保险中，模型具有部分“可解释性”，若损失结果的“近因”来自未知部分且效力不明，开发者举证不能满足基于“合理人工智能”标准的客观推断要求，应将此类“近因”归于“除外风险”，适用“不包括占优”的因果关系判定规则更符合商业保险对可保风险应具备“可评估性”的本质要求。“黑箱”人工智能模型不具有“可解释性”，但其责任保险具有“政策保险+商业保险”的混合属性，可允许的风险范围扩大，需结合最终承担责任划分、人工智能变化形态和初始合同相关约定，适用“比例分配”或“不包括占优”的因果关系判定规则。

［关键词］人工智能；责任保险；因果关系；可解释性；保险理赔

［基金项目］2022年度教育部人文社会科学重点研究基地重大项目“法治化营商环境建设法律问题研究”（22JJD820018）。

［作者简介］杨一凡，中国人民大学法学院博士研究生，研究方向：保险法。

A Research on the Determination of Causation in AI Liability Insurance：Centered on Model Explainability

YANG Yi-fan

Abstract：Centered on model explainability，artificial intelligence （AI） can be categorized into three forms：white-box，gray-box，or black-box.Corresponding liability insurance for each form should apply different causation determination rules.In white-box AI liability insurance，complete explainability of the model means that the proximate cause of the loss and its proportional impact are clear.Applying the proportional-allocation-causation rule better respects the reasonable expectations of both the insurer and the insured.In gray-box AI liability insurance，the model has only partial explainability.If the proximate cause of the loss originates from the opaque part and its impact is unclear，and the developer’s evidence fails to meet the objective inference requirements based on the reasonable AI standard，such a proximate cause should be identified as an excluded risk.Applying the excluding-the-dominant-causation rule conforms to the essential requirement of commercial insurance，which is that insurable risks can be assessed.In black-box AI，the model lacks explainability.But its corresponding liability insurance combines the nature of policy-based insurance and commercial insurance，which will expand the scope of insurable risks.The application of the proportional-allocation-causation rule or excluding-the-dominant-causation rule should be chosen with reference to the final division of liability，the evolving form of AI，and the terms of the initial insurance policy.

Key words：AI;liability insurance;causation;model explainability;insurance claims