道德机器何以实现:生成式AI智能向善的逻辑基础与提示词框架  

How Moral Machines Can Be Tempered:The Logical Foundation of Generative AI's Alignment with Goodness and the Framework of Prompt Engineering

作　　者：刘琴[1] 常立瑛 LIU Qin;CHANG Li-ying(School of Literature,Journalism and Communication,South-Central Minzu University Wuhan 430074,China)

机构地区：[1]中南民族大学文学与新闻传播学院,湖北武汉430074

出　　处：《编辑之友》2025年第4期57-64,共8页Editorial Friend

基　　金：国家广播电视总局项目“虚拟主播对青年群体文化价值观的影响及风险控制研究”(BSY22037);中南民族大学教研项目“数据新闻人才培养的框架设置与实践体系探究”(JYX18004)。

摘　　要：生成式AI作为道德机器需要承担智能向善的使命,而提示词工程是生成式AI特别是对话式AI话语把关的重要环节,能够引导用户“说什么”,构建符合人类实际的价值观。文章从技术生命、意义本体和政治身份三个层面探讨生成式AI智能向善的逻辑基础,揭示其智能向善的核心是进行主流价值观的引领,而要实现生成式AI的社会道德责任,需要重新设计主流价值的语料库,通过开源的大模型和私域小模型等多种渠道开展语义训练,在不断演绎与修正中,提升用户的价值认知,实现价值纠偏与对齐。作为生命机器,生成式AI不是单纯地迎合价值主体,努力使自己的语言内容对齐输入指令,而是能够再创生产力,反作用于人类。因此,既需要从功能层面探讨其价值对齐和道德行善,更要从平等的视角重释人机关系:从被动到主动、从辅助到主导、从服从到超越。As a moral machine,generative AI carries the mission of guiding intelligence toward goodness.Prompt engineering plays a pivotal role in this process,especially in conversational Al,as it shapes the framework for guiding users on"what to say"and helps construct a value system aligned with human ethics.The paper explores the logical foundation of AIGC alignment with goodness from three perspectives:technological life,meaning ontology,and political identity.It reveals that the core of its ethical intelligence lies in promoting mainstream values.To fulfill generative AI's moral responsibility as a social utility,it is essential to design corpora that reflect mainstream values and conduct semantic training through various channels,such as open-source large models and private small models.Through continuous iteration and refinement,generative AI can nurture users'value cognition,achieving value correction and alignment.As a living machine,generative AI does not merely cater to human values or align its outputs with user inputs.Instead,it has the potential to redefine productivity and actively influence humanity.We must not only examine its value alignment and ethical behavior from a functional perspective but also reconsider human-machine relationships from an egalitarian viewpoint:shifting from passivity to activity,from assistance to leadership,and from compliance to transcendence.

关 键 词：道德机器 智能向善 价值对齐 提示词工程 

分 类 号：G2066[文化科学—传播学]