价值嵌入与价值对齐:人类控制论的幻觉  

Value Embedding and Value Alignment:The Illusion of Human Cybernetics

作　　者：吴静 WU Jing(Nanjing Normal University)

机构地区：[1]南京师范大学公共管理学院哲学系 [2]南京师范大学数字与人文研究中心

出　　处：《华中科技大学学报（社会科学版）》2024年第5期11-19,共9页Journal of Huazhong University of Science and Technology(Social Science Edition)

摘　　要：价值对齐反映了对人类智能与人工智能关系进行简化的理论惰性,其本质潜藏着一种价值维度上的人类中心主义投射,同时也体现了人类在技术自治性面前的不安感。价值对齐是一种企图将所有的技术-社会的复杂性问题都置于“价值偏差”的系统性盈余中予以整体性解决的技术治理方法。然而,将人类价值观嵌入人工智能系统,一方面会使得技术的发展成为最小化人类责任的借口;另一方面,这种话语以一种抽象性的人机价值差异掩盖了人工智能技术发展全过程中的价值性因素,从而以价值排他性的方式将人工智能的“问题”建构为一种绝对意义上的外在,使得关于价值的讨论只逡巡在人机的边界之上,而不向人类内部返回。从这个意义上来说,人工智能已经建构起来的社会历史中的价值无意识和不平衡的数据分布现实,是一个关于“具有自主性的智能体应该和不应该被如何设计”的问题。因此,人工智能的价值判断基础应当由外在的价值嵌入转向具身认知和知觉能力的形成。只有通过有效的人机交互与环境感知,打造开放式人机生态体系,鼓励开发与人类处于更广泛的目的分享和责任分担的人工智能,才能防范由技术所导致的任何形式的垄断。Value alignment reflects a theoretical inertia that simplifies the relationship between human intelligence and artificial intelligence,and its nature harbors an anthropocentric projection of the value dimension,as well as a sense of unease in the face of technological autonomy.Value alignment is an approach to technological governance that attempts to holistically address all techno-social complexities within the systemic surplus of“value bias”.However,embedding human values into AI systems will,on the one hand,make technological development an excuse for minimizing human responsibility;on the other hand,this discourse masks the value factors in the whole process of AI technological development with an abstract human-computer value discrepancy,thus constructing the“problem”of AI as an absolute one in the form of value exclusivity.Thus,the“problem”of AI is constructed as an absolute externalization in a value-exclusive way,so that the discussion of value only balks at the boundary between humans and machines,without returning to the human interior.In this sense,the reality of value unconsciousness and unbalanced data distribution in social history,which AI has constructed,is a question of“how autonomous intelligences should and should not be designed”.Therefore,the basis of value judgment of AI should be shifted from external value embedding to the formation of embodied cognitive and perceptual abilities.Only by creating an open human-computer ecosystem through effective human-computer interaction and environment perception,and by encouraging the development of AI that shares broader purposes and responsibilities with human beings,can we prevent any form of monopolization caused by technology.

关 键 词：价值对齐 控制论 具身认知 人机协作智能 

分 类 号：D018[政治法律—政治学] F014.31[经济管理—政治经济学]