生成式大语言模型的社会认知风险与应对  

Social Cognitive Risks andMitigation Strategies for Generative Large LanguageModels

作　　者：马忠 高怡英 Ma Zhong;Gao Yiying(School of Marxism,Xian Jiaotong University,Xian 710049)

机构地区：[1]西安交通大学马克思主义学院,西安710049

出　　处：《浙江社会科学》2025年第2期95-105,158,共12页Zhejiang Social Sciences

基　　金：教育部高校思政课教师研究专项重大项目“研究生思政课建设存在的问题及对策研究”(22JDSZKZ11);国家社科基金专项“全媒体时代马克思主义意识形态话语传播力研究”(2022LJRC01)的阶段性成果。

摘　　要：生成式大语言模型是当前人工智能技术的一种主要新形态,作为一种新的语言形式,不可避免地引发了知识生产的变革、认知结构的嵌入及社会认知的重塑。综观人类科技史,任何新技术的诞生在带来革命性进步的同时,由于技术本身对传统生产方式的颠覆性及实践层面的制度滞后性,必然会带来一定的技术风险。同样,生成式大语言模型在新技术深度赋能新发展的同时,不可避免地产生诸如数据平等、数字孤立、数字鸿沟、算法控制等新问题,在社会认知领域则有可能带来认知倾向的技术操控、社会认知的虚假状态、数字帝国主义的加深等风险。这些风险的产生原因在于技术差异、黑箱逻辑和数据选择所导致的算力不平等、算法不透明和数据局限性。为此要从技术、制度和应用等多个层面入手,通过多元共享打破话语垄断,以制度框架防范潜在风险,提升全民素养以弥合数字鸿沟,确保技术、社会和责任之间的平衡,使人工智能更好服务人类发展。Generative large language models represent a significant new form in current artificial intelligence technology.As a novel linguistic form,they inevitably trigger transformations in knowledge production,reorganization of cognitive structures,and reshaping of social cognition.Throughout the history of human technology,the birth of any new technology has led to revolutionary progress.However,due to the disruptive nature of technology on traditional production methods and the lag in institutional practices,certain technical risks are bound to arise.In the realm of social cognition,similarly,while generative large language models deeply empower new developments through advanced technology,they inevitably generate new issues such as data inequality,digital isolation,digital gap and algorithmic control.In the social cognition domain,these technologies might lead to risks like the technical manipulation of cognitive biases,the creation of false social cognitions,and the intensification of digital imperialism.The causes of these risks lie in computational power inequality,algorithm opacity,and data limitations resulting from technological disparities,black-box logic,and data selection.Therefore,it is essential to address these issues from multiple dimensions,including technology,institutions,and applications.This involves breaking the monopoly of discourse through diverse sharing,preventing potential risks with legal frameworks,and enhancing public literacy to bridge the digital divide.By ensuring a balance between technology,society,and responsibility,artificial intelligence can better serve human development.

关 键 词：生成式大语言模型 社会认知 风险 人工智能 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]