人工智能伦理治理:基本理路、现行策略与实践转向  

Ethical Governance of AI:Basic Rationale,Current Strategies and a Practical Turn

作　　者：陈凡[1] 程海东 胡孝聪 CHEN Fan;CHENG Haidong;HU Xiaocong

机构地区：[1]东北大学科学技术哲学研究中心 [2]东北大学马克思主义学院

出　　处：《延边大学学报(社会科学版)》2024年第5期97-107,142,共12页Journal of Yanbian University:Social Sciences Edition

基　　金：国家社会科学基金项目(22BZX026);辽宁省社会科学规划基金重大委托项目(L22ZD042)

摘　　要：人工智能伦理治理的基本理路,在理论资源上源于伦理学的基本理论:目的论、道义论、美德伦理。或者关注人工智能的行动结果,或者关注人工智能行动的原则,或者关注人工智能的道德培养。从人工智能发挥伦理功能的角度,可以区分为四个层次:有伦理效用、隐性伦理效用、显性伦理效用和完备伦理效用。在这样的基本理路下,从理论资源出发,现行的基本策略是采用自上而下、自下而上或者二者混合的方式使人工智能习得道德决策的基本原则,而习得的方式则是采用道义逻辑、认知逻辑、行动逻辑、道义—认知—行动逻辑,将伦理原则转译为人工智能能够识别和接受的算法。然而,现行策略中始终面临着伦理是否可计算这一根本问题,因此,应实现实践转向,使人类和人工智能在具体的道德实践中形成优势互补的道德共生关系。The basic rationale for the ethical governance of AI is derived from foundational theories in ethics:teleology,deontology,and virtue ethics.These frameworks either focus on the outcomes of AI actions,the principles guiding AI actions,or the moral cultivation of AI.From the perspective of AI exerting ethical functions,this can be categorized into four levels:ethical agents,implicit ethical agents,explicit ethical agents,and complete ethical agents.Based on this fundamental rationale,the current strategies for ethical governance involve top-down,bottom-up,or hybrid approaches to enable AI to learn the basic principles of moral decision-making.The methods for this acquisition involve using deontological logic,cognitive logic,action logic,and deontological-cognitive-action logic to translate ethical principles into algorithms that AI can recognize and accept.However,a persistent issue within these current strategies is whether ethics can truly be computationalized.Therefore,a practical turn is required,whereby humans and AI form a morally symbiotic relationship,leveraging their complementary strengths in specific moral practices.

关 键 词：人工智能 伦理治理 智能时代 人工智能伦理 价值观 算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] B82-057[自动化与计算机技术—控制科学与工程]