摘要
生成式人工智能的迅猛发展,使价值对齐从技术议题演变为人工智能治理中引领“智能向善”的核心实践。然而,当前以技术为中心的价值对齐范式,在实践中隐匿着“伪对齐”“泛对齐”与“窄对齐”三重伦理风险。这些风险的根源在于技术至上的算法逻辑遮蔽了伦理审思的优先地位、对价值的静态模拟压制了其在真实互动中的动态生成,以及工具理性的持续盛行销蚀了价值理性在具体情境中的多元诉求。为有效纾解风险困境并回应“智能向善”的时代诉求,需要从主体、内容和范式三个维度进行系统性重塑。在对齐主体上,推动迈向整体性与共生性的双向建构;在对齐内容上,倡导从规则内嵌到伦理智识的有效生成;在对齐范式上,实现从定向式训练到演化式布局的进路转型。通过重塑对齐主体、重构对齐内容、革新对齐范式等不同路径多措并举、多点发力,从而为探索出更具反思性、适应性与规范性的“人机协同、共生共进”的价值对齐新范式,为构建人工智能“善治”格局提供兼具理论深度与实践效能的治理方略。
The rapid development of generative Artificial Intelligence(AI)has transformed value alignment from a technical issue into a core practice of guiding“intelligence for the good”in AI governance.However,the current technology-centered value alignment paradigm conceals three ethical risks of“pseudo-alignment”“broad-alignment”and“narrow-alignment”in practice.The root cause of these risks lies in the fact that the algorithmic logic that prioritizes technology overshadows ethical reflection,the static simulation of value suppresses its dynamic generation in real interactions,and the continuous prevalence of instrumental rationality erodes the diverse demands of value rationality in specific contexts.To effectively alleviate this risk predicament and respond to the contemporary demand of“guiding intelligence for the good”,a systematic reshaping is needed from three dimensions:subject,content and paradigm.In terms of aligning the subject,a two-way construetion towards integrity and symbiosis could be promoted;in terms of aligning content,the effective generation could be advocated from the embedding of rules to ethical knowledge.On the alignment paradigm,a path transformation will be achieved from directional training to evolutionary layout.By taking multiple measures through different paths such as reshaping the alignment subject,reconstructing the alignment content,and innovating the alignment paradigm,and making efforts in multiple points,we aim to explore a new value alignment paradigm of“human-machine collaboration and symbiotic progress”that is more reflective,adaptable and standardized,thus providing a governance strategy that combines theoretical depth and practical effectiveness for building a“good governance”pattern of artificial intelligence.
作者
杨灿
林伯海
YANG Can;LIN Bohai(School of Marxism,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)
出处
《昆明理工大学学报(社会科学版)》
2025年第6期59-67,共9页
Journal of Kunming University of Science and Technology(Social Sciences)
基金
国家高层次人才特殊支持计划项目“弘扬全人类共同价值的理论与实践研究”(23VRC058)。
