人工智能价值对齐现有方法的批判与生成性认知路径分析  

A Critique of Existing Approaches and an Analysis of Enactive Cognitive Path to Artificial Intelligence Value Alignment

作　　者：赵伟 ZHAO Wei(School of Marxism,Central China Normal University,Wuhan 430079,China)

机构地区：[1]华中师范大学马克思主义学院,武汉430079

出　　处：《科学技术哲学研究》2025年第2期49-56,共8页Studies in Philosophy of Science and Technology

基　　金：国家社科基金重大项目“马克思共同体思想及当代价值研究”(23&ZD200)。

摘　　要：人工智能很难对齐,错位人工智能的部署可能不可逆转地破坏人类对世界的控制,甚至威胁人类文明。人工智能安全的主要方法是AI对齐,即创造与“人类价值观”对齐的人工智能。“价值对齐问题”是如何将人工智能的价值观与人类的价值观相一致的挑战,无论它们是什么,这样人工智能就不会对人类的存在构成风险。现有的几种主要价值对齐方法,例如技术路径和规范路径都没有解决价值对齐问题,建议在哲学上植根于当代认知科学的生成范式。为了使人工智能的价值观与人类保持一致,我们必须使和人类相关的东西与人工智能相关。这种方法解决对齐问题的结论是使人工智能在本体论上与人类相似。The artificial intelligence(AI)is difficult to align,the deployment of misplaced AI could irreversibly un⁃dermine human control over the world and even endanger human civilization.The primary approach to AI safety is AI alignment,i.e.creating AI that is aligned with“human values”.The“value alignment problem”is the chal⁃lenge of how to align AI values with human values,whatever they may be,so that AI does not pose a risk to human existence.Several existing approaches to value alignment,such as the technical approach and the normative ap⁃proach,do not solve the value alignment problem,and the proposal is philosophically rooted in the Enactive para⁃digm of contemporary cognitive science.In order to align the values of artificial intelligence with those of humans,we must make things related to humans relevant to artificial intelligence.The conclusion of this method in solving a⁃lignment problems is to make artificial intelligence ontologically similar to humans.

关 键 词：人工智能 存在主义风险 价值对齐 

分 类 号：N02[自然科学总论—科学技术哲学]