数字政府大模型的自主性能力与风险规避  

Autonomous Capability and Risk Aversion Measures of Digital Government Large Model

作　　者：范炜烽[1] 罗鹏[1] Fan Weifeng;Luo Peng(Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学,南京210094

出　　处：《国家现代化建设研究》2024年第4期105-118,共14页Journal of Modernization Studies

基　　金：国家社会科学基金重大项目“互联网发展与国家治理体系和治理能力现代化研究”(项目编号:22&ZD028);2024年江苏省决策咨询研究基地项目“数字化治理平台赋能基层治理现代化创新研究”(项目编号:24SSL114)。

摘　　要：生成式人工智能技术是构建数字政府大模型的核心,可以辅助政府实现数据与模型算法驱动的政府治理现代化。在自主性技术理论视角下,数字政府大模型具有完整的数据自主性、学习自主性、决策自主性与服务自主性四种能力指向,分别表现为治理路径映射关系的重塑、知识结构适应性逻辑的优化、机器理性科学性趋向的重造与供给模式精细化结构的重组。然而,生成式人工智能的自主性也会导致数字政府大模型面临技术监管与引导主动性缺失的模型梦游、依赖与欲望内生性增长的模型命令、扭曲与定式反向性塑造的模型适应、偏见与歧视自主性扩张的模型无意识等自主性风险。因此,迫切需要加强对数字政府大模型的全生命周期监管,提升政务用户的大模型识读能力,科学设置政府大模型与政务业务的边界,构建政务数字化协同互动的生态体系以规避实践中可能产生的自主性风险,提高数字政府大模型的综合效能。Generative AI technology is the core of constructing digital government large model,which can assist the government to realize the modernization of governance driven by data and model algorithms.Under the perspective of autonomy technology theory,the large model of digital government has the complete data autonomy,learning autonomy,decision-making autonomy and service autonomy,which are manifested in the reshaping of the mapping relationship of the governance path,the optimization of the adaptive logic of the knowledge structure,the remodeling of the scientific tendency of machine rationality,and the reorganization of the refined structure of the supply mode.However,generative AI autonomy will also prompt the digital government large model to face four specific autonomy risks:model sleepwalking of the lack of technical regulation and guidance initiative,model command of endogenous growth of dependence and desire,model adaptation of distortion and stereotypical reverse shaping,and model unconsciousness of autonomy expansion of bias and discrimination.Therefore,there is an urgent need to strengthen the supervision of the whole life cycle of digital government large models,improve the ability of government users to read and understand large models,scientifically set the boundaries between government large models and government services,and build a collaborative and interactive eco-system of government digitization in order to avoid the autonomy risks that may arise in practice,and to improve the comprehensive effectiveness of digital government large models.

关 键 词：生成式人工智能 数字政府大模型 技术自主性 风险规避 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] D035[自动化与计算机技术—控制科学与工程]