医疗AI风险分级研究：从“工具论”到“能动者论”的视角

Medical AI Risk Classification: From the Perspective of “Tool Theory” to “Agency Theory”

摘要: 医疗人工智能(Medical AI)在诊断、预后、治疗等核心医疗环节的深度融合应用，使其伦理风险的关注焦点已从“是否发生”转变为“如何分级与治理”。本文突破事后归责的传统范式，融合卢西亚诺·弗洛里迪(Luciano Floridi)的“信息伦理学”(Information Ethics)与彼得–保罗·维贝克(Peter-Paul Verbeek)的“技术调解”(Technological mediation)理论，将医疗AI界定为塑造医患实践的“道德相关能动者”(Morally Relevant Agent)。立足这一核心界定，本文构建起涵盖算法可解释性、决策自主性、患者脆弱性、社会–制度耦合度的四维风险评估模型，据此划定医疗AI的低、中、高三个风险层级。针对不同风险层级，本文进一步构建从“软法”伦理指南到“硬法”强制责任保险的递进式治理框架，核心治理机制依托海伦·尼森鲍姆(Helen Nissenbaum)的情境完整性原则(Contextual Integrity)，将开发者、医生、医疗机构、监管者与患者共同纳入动态的责任配置，旨在促进伦理价值从技术设计到临床应用的全流程融入，从而为医疗AI的伦理治理探索可能的理论参考与制度路径。

Abstract: Medical AI’s deep integration into core clinical processes—including diagnosis, prognosis, and treatment—has shifted the focus of ethical risk from “whether it occurs” to “how to classify and govern it”. This paper moves beyond the traditional paradigm of ex post facto accountability, integrating Luciano Floridi’s Information Ethics with Peter-Paul Verbeek’s theory of Technological Mediation to conceptualize medical AI as a “morally relevant agent” that shapes doctor-patient practices. Grounded in this core conceptualization, the paper constructs a four-dimensional risk assessment model encompassing algorithmic explainability, decisional autonomy, patient vulnerability, and socio-institutional coupling, on the basis of which three risk tiers for medical AI—low, medium, and high—are delineated. For each tier, the paper further develops a progressive governance framework ranging from “soft law” ethical guidelines to “hard law” mandatory liability insurance. The core governance mechanism draws on Helen Nissenbaum’s principle of Contextual Integrity, dynamically distributing responsibilities among developers, physicians, medical institutions, regulators, and patients. The aim is to facilitate the full-process integration of ethical values from technological design to clinical application, thereby exploring possible theoretical references and institutional pathways for the ethical governance of medical AI.