数据治理视角下AI大模型价值对齐的困境与应对  

The Dilemma and Breakthrough of AI LLM Value Alignment from the Data Governance Perspective

作　　者：朱树婷[1] 叶美兰[1] Zhu Shuting;Ye Meilan(Nanjing University of Posts and Telecommunication,Nanjing 210003,China)

机构地区：[1]南京邮电大学,南京210003

出　　处：《信息通信技术》2024年第3期36-41,49,共7页Information and communications Technologies

摘　　要：AI大模型正广泛应用于人类社会生活,存在违反人类价值标准的风险。如何实现AI大模型应用与人类价值判断相匹配,AI大模型价值对齐成为当前人工智能领域重要议题之一。文章依据先进信息技术融入人类社会的结构化过程,指出AI大模型价值对齐的内在困境,从数据治理视角分析形成原因和突围方法,为实现AI大模型价值对齐提供新思路。While AI large language model(LLM)is being widely used in human social life,there exists the risk of AI violating human value standards.AI value alignment has become one of the important issues in AI research field,aiming at how to match the AI application with human value judgments.According to the adaptive structuration of advanced information technology use,this paper analyzes the inherent dilemma of AI value alignment,explores the reasons and breakthrough methods from the perspective of data governance,and provides new ideas for achieving AI LLM value alignment.

关 键 词：AI大模型 价值对齐 数据治理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP311.13[自动化与计算机技术—控制科学与工程]