文章摘要

刘严,杨建林,王忠军.从AI for Science到AI for Intelligence:情报研究范式的转型及其在事实核查中的实现[J].中国图书馆学报,2026,52(2):51~71
从AI for Science到AI for Intelligence:情报研究范式的转型及其在事实核查中的实现
From AI for Science to AI for Intelligence:The Paradigm Shift in Intelligence Studies and Its Implementation in Fact checking
投稿时间：2025-07-28  修订日期：2025-11-30
DOI：
中文关键词: AI for Science  AI for Intelligence  事实核查  情报研究范式
英文关键词: AI for Science  AI for Intelligence  Fact checking  Paradigm of intelligence studies
基金项目:
作者单位
刘严 南京大学信息管理学院 江苏 南京 210023 
杨建林 南京大学信息管理学院 江苏 南京 210023 
王忠军 《情报理论与实践》主编 北京 100089 
摘要点击次数: 244
全文下载次数: 212
中文摘要:
      本文以AI for Science（AI4S）在情报场景中的适配张力为研究起点,基于库恩范式理论,揭示其在本体预设、目标导向与方法路径上的局部不可通约,据此提出以情报核实为核心任务的AI for Intelligence（AI4I）范式,强调认知建模、因果核实与对抗适应三类关键能力。这些能力在AI4S任务逻辑中长期处于边缘地位,而在当前情报相关研究中逐步显现为结构性诉求。为厘清AI4I的学理意义,梳理该范式对情报学经典理论的守正与创新。进一步,以事实核查任务为典型场景,阐明三类能力在舆论监测、主张筛选、验证采证、结论裁定与结果叙述等环节中的机制映射与协同关系,探讨在图书馆情境下的技术落地可行性保障,以呈现AI4I对情报实践的指导价值。在此基础上,提出“生成—核实”双环协同框架,设计对象层、模型层与流程层三类接口与切换策略,探讨效率与可信性并重的智能情报生态构建路径,以期为我国构建AI赋能的科技安全情报体系提供理论支撑。图3。表1。参考文献45。
英文摘要:
Taking the adaptation tensions of AI for Science（AI4S） in information research as its starting point,this paper,drawing on Kuhn's theory of scientific paradigms,argues that AI4S and information research are locally incommensurable in terms of ontological presuppositions,goal orientations,and methodological pathways. AI4S presumes data completeness,environmental controllability,and ex post verifiability,which fit the logic of natural science laboratories but conflict with intelligence practice. Analyses of academic misconduct checking and technical novelty verification further reveal that current “smart” systems mainly emulate an AI4S style single loop,optimizing throughput and automation while leaving evidence chain construction,adjudicative verification,and responsibility tracing weakly supported.
On this basis,the paper proposes AI for Intelligence（AI4I） as a distinct,verification centered paradigm for information research. AI4I is defined as an intelligence framework that,in multi source and adversarial information environments,prioritizes intelligence verification and organizes three interlocking capabilities—cognitive modeling,causal verification,and adversarial adaptation—into a fast “cognition—verification—feedback” loop aimed at reducing cognitive risk and ensuring accountability. The paper clarifies the paradigm level connotations of AI4I,reviews emerging work on these three capability clusters,and situates AI4I in dialogue with classical theories of cognitive intelligence,evidence theory,and the intelligence cycle,thereby reframing long standing concerns about credible judgment in a processual and system oriented manner.
Fact checking is then selected as a representative intelligence task. This paper discusses how AI4I's three capability layers can be mapped onto the five stages of fact checking,namely media monitoring,claim selection,evidence gathering and verification,conclusion adjudication as well as narrative reporting,through concrete technical mechanisms. Using libraries as a testbed,this paper further examines engineering feasibility under constraints of data privacy,limited computational resources,and talent shortages,and proposes a three pronged support structure. Furthermore,the paper advances a “generation-verification” dual loop ecology,with interfaces and switching strategies specified across the object,model,and process layers,thereby outlining a construction path for an intelligent intelligence ecosystem that reconciles efficiency with trustworthinessThe ultimate aim is to furnish theoretical support for building an AI enabled science and technology security intelligence system in China. 3 figs. 1 tab. 45 refs.
查看全文   查看/发表评论  下载PDF阅读器