医学人工智能的算法黑箱问题:伦理挑战与化解进路  

Algorithmic black-box problem in medical artificial intelligence:Ethical challenges and solution approach

作　　者：杨军洁 周程[1,2] Junjie Yang;Cheng Zhou(Department of Philosophy,Peking University,Beijing 100871,China;School of Health Humanities,Peking University,Beijing 100191,China)

机构地区：[1]北京大学哲学系,北京100871 [2]北京大学医学人文学院,北京100191

出　　处：《科学通报》2023年第13期1604-1610,共7页Chinese Science Bulletin

基　　金：教育部人文社会科学重点研究基地重大项目“当代认知哲学基础理论问题研究”(22JJD720007)资助。

摘　　要：人工智能技术正逐渐成为推动生物医学发展的重要力量.与此同时,医学人工智能带来的伦理与治理问题也日益凸显.笔者以“medical artificial intelligence ethics”为主题在Web of Science数据库中进行检索,共获得相关文献554篇(1975~2022年).Artificial intelligence (AI) is gradually becoming an important force driving the development of biomedicine. However, asAI increasingly relies on complex and opaque machine learning algorithms, a critical ethical issue known as the“algorithmic black-box” problem has emerged. Despite the development of several explainability tools, they have not beenwidely adopted in the medical field due to their inability to provide satisfactory explanations in clinical practice.Furthermore, different stakeholders including algorithm experts, medical professionals, patients, and the general publichave varying requirements for explainability and transparency. As a result, this has created a series of internal, internalexternal interaction, and external-level ethical issues in data, algorithmic, and social dimensions.In the ethical challenges associated with the increasing complexity and opacity of medical artificial intelligence,constructing medical artificial moral agents has been proposed as a viable solution. To implement ethical frameworks inthis domain, three approaches have been identified: The top-down approach, the bottom-up approach, and the hybridapproach. The top-down approach prioritizes moral design based on specific ethical principles. However, this approachfaces difficulties in responding appropriately to complex ethical situations due to the lack of consensus among ethicalexperts, contradictions between ethical principles and practical goals, and the abstract nature of moral principles. Thebottom-up approach, on the other hand, requires medical artificial intelligence to develop a set of operating methods thatalign with human moral intuition in a series of case-based reinforcement learning scenarios. Nonetheless, this approach isonly effective in retrospective regulation, and converging moral reasoning to a certain pattern remains challenging.In light of the current state of artificial intelligence development, it is imperative to adopt a “hybrid approach” thatintegrates both top-down and bottom-up approa

关 键 词：人工智能技术 医学人工智能 伦理挑战 进行检索 数据库 生物医学 

分 类 号：R-052[医药卫生] TP18[自动化与计算机技术—控制理论与控制工程]