人工智能代理对道德决策的影响
The Moral Impact of Delegating to Artificial Intelligence
唐伟 1钟文瑞 2雷震 2张丹丹3
作者信息
- 1. 西南财经大学习近平经济思想研究院
- 2. 西南财经大学中国行为经济与行为金融研究中心
- 3. 四川师范大学脑与心理科学研究院
- 折叠
摘要
各类人工智能系统正作为代理广泛嵌入企业、政府与个人的决策流程，对人类决策的道德性和道德判断产生了深刻影响。尽管相关实证和理论研究快速增长，现有文献尚缺乏对人工智能代理区别于其他代理特殊性的系统性分析，也缺乏统一的分析框架来系统刻画人工智能代理对道德决策的影响路径。因此本文构建了“决策者—代理—反馈者”的决策与归责框架，对现有研究进行梳理和重组。本文认为：代理介入决策后，拉长了决策者的决策链条和反馈者(包含受影响者和第三方观察者)的反馈链条，由此削弱了决策者的道德感知和反馈者的归责，进而促进了决策者的不道德行为。而人工智能代理的黑箱性、高遵从、规模化、工具性等特点则在决策链上加剧了不道德指令的执行、增强了决策者的可否认机会、扩大了不道德行为的影响范围；同时，这些特性在反馈链上增加了反馈者对不道德行为的道德容忍、模糊了反馈者对决策者意图的判断与归责，进而促进决策者做出不道德行为。本文指出未来研究有必要继续完善该决策框架内各机制相对作用关系，考察在组织与社会层面道德行为的扩散与放大机制，并探索人机协作情境下的治理工具与制度安排。
Abstract
Various artificial intelligence (AI) agents are increasingly embedded in the decision processes of firms, governments, and individuals, taking on delegated decision execution and exerting profound influence on the morality of human decisions and moral judgment. Despite the rapid growth of related empirical and theoretical research, the existing literature still lacks a systematic analysis of what makes AI agents distinctive from other types of agents, and it also lacks a unified analytical framework to systematically characterize the pathways through which AI agents influence moral decision-making. Therefore, this paper develops a decision-makeragentevaluator framework of decision-making and accountability to synthesize and reorganize the existing literature. We argue that when an agent intervenes in the decision process, it lengthens both the decision-makers decision chain and the evaluators feedback chain (where evaluators include affected parties and third-party observers). This, in turn, weakens decision-makers moral salience and evaluators attribution of responsibility, thereby facilitating unethical behavior. Moreover, distinctive features of AI agentssuch as opacity (black-boxness), high compliance, scalability, and instrumentalityfurther intensify the execution of unethical instructions, increase decision-makers opportunities for deniability, and expand the scope of unethical impacts along the decision chain. At the same time, these features can increase evaluators moral tolerance for unethical outcomes and blur their inferences about the decision-makers intent and responsibility along the feedback chain, further encouraging unethical behavior. Finally, we suggest that future research should refine the relative roles of the mechanisms within this framework, examine how unethical behavior diffuses and scales up at organizational and societal levels, and explore governance tools and institutional arrangements for humanAI collaboration.关键词
人工智能代理/道德决策/人机协同治理Key words
artificial intelligence delegation/moral decision-making/human–AI collaborative governance引用本文复制引用
唐伟,钟文瑞,雷震,张丹丹.人工智能代理对道德决策的影响[EB/OL].(2026-02-14)[2026-02-17].https://chinaxiv.org/abs/202602.00173.学科分类
计算技术、计算机技术
评论