Home Current Issue

25 August 2026, Volume 46 Issue 04
    

  • Select all
    |
  • Agent-Driven Transformation of Intelligence Work: Challenges, Gaps, and Future Directions
    Wu Xuan Li Guangjian
    Library & Information. 2026, 46(04): 1-11. DOI:10.11968/tsyqb.1003-6938.2026039
    Abstract ( ) Download PDF ( ) Knowledge map Save
    The application of agent technology in intelligence work has drawn widespread academic attention, yet the profound impacts it exerts on intelligence work have not been systematically examined. This paper analyzes the multiple challenges that agents pose to intelligence work across three dimensions: technical methods, cognitive thinking, and work systems. On this basis, it identifies the structural gaps in current responses to these challenges, which center on three issues: incomplete cognitive loops, limited depth of cognitive interaction, and insufficient initiative of cognitive agency. To bridge these gaps, the paper proposes four directions for future research, namely deepening foundational research, drawing lessons from international experiences, advancing localized practice, and establishing governance frameworks, focusing respectively on the theoretical construction of agent-empowered intelligence work, the systematic review of pioneering international practices, the contextual adaptation of open-source frameworks, and the institutional design for cognitive risk governance. 
  • AI Agents Empowering Intelligence Work: Technical Logic and Implementation Pathways
    Pan Jiali Li Guangjian
    Library & Information. 2026, 46(04): 12-20. DOI:10.11968/tsyqb.1003-6938.2026040
    Abstract ( ) Download PDF ( ) Knowledge map Save
    To address the cognitive limitations of traditional artificial intelligence, which prioritizes factual reconstruction over value judgment, and to overcome the bottlenecks of factual hallucinations and logical flaws faced by native large language models in complex intelligence environments, this paper explores the technical logic and implementation pathways of general-purpose intelligent agents empowering intelligence work. First, it elucidates the core characteristics of agents within the intelligence context and constructs an AI agent taxonomy. Second, it dissects the technical logic by which agents, relying on the “LLM + Memory + Tools + Action” architecture, achieve the cognitive leap from surface-level information analysis to in-deep intelligence processing. Finally, grounded in the intelligence lifecycle, the study elaborates on the implementation pathways across key phases, including situation awareness, logical reasoning, value judgment, autonomous execution, and human-machine collaboration. The research concludes that intelligent agent technology breaks through existing cognitive bottlenecks, propelling intelligence work toward a new paradigm of smart services.
  • AI Agent Application Practices and Characteristics in Foreign Intelligence Communities
    Luo Liqun Li Guangjian
    Library & Information. 2026, 46(04): 21-33. DOI:10.11968/tsyqb.1003-6938.2026041
    Abstract ( ) Download PDF ( ) Knowledge map Save
    Global intelligence work is undergoing a profound transformation, during which developing AI agent with capabilities of autonomous perception, cross-domain reasoning, and closed-loop action has emerged as the central direction of both intelligence operations and intelligence research. Whether national intelligence services can be the first to achieve the "agentic turn" will directly determine their future intelligence competitive status. Major intelligence powers have entered an accelerated phase of agent-driven transformation, intensively conducting research and practice on agents across diverse intelligence tasks. Adopting a global perspective, this paper conducts a survey of agent application projects across foreign intelligence communities and provides an in-depth analysis of the application scenarios, technical principles, and collaborative mechanisms of intelligence agents within four core task domains: situational awareness, intelligence judgment and early warning, decision-making and action, and defense and counter-operations. On this basis, the paper summarizes the practical characteristics of foreign intelligence communities along four dimensions: application modes, technical architectures,risk security, and organizational ecosystems.
  • The Adaptability and Application Reflections of Open-Source Agents in the Intelligence Field
    Yang Xiao Pan Jiali Li Guangjian
    Library & Information. 2026, 46(04): 34-46. DOI:10.11968/tsyqb.1003-6938.2026042
    Abstract ( ) Download PDF ( ) Knowledge map Save
    With the rapid development of large language models and the open-source ecosystem, AI agents are gradually evolving from tool-like forms centered on model capabilities into intelligent systems oriented towards complex tasks. This paper, based on AI Agent projects in the GitHub open-source community, reviews the current development of open-source AI Agents from multiple dimensions, including release year, programming language, update frequency, development technology, open-source license, and development entities. The study finds that the open-source intelligent agent ecosystem presents a trend of rapid expansion in development scale, highly diversified technical architecture, and increasingly complete functional system. Next, this paper constructs a 1+3 multidimensional adaptation analysis framework based on macro-task flow control and the classic DIKW (Data-Information-Knowledge-Wisdom) value transformation chain to analyze the adaptability of representative open-source AI Agent in information science from four dimensions: intelligence tasks, data processing, information analysis, and intelligence cognition. Overall, existing AI Agents have strong supporting capabilities in task orchestration and data processing and demonstrate certain data modeling and semantic integration capabilities at the information analysis layer. However, at the intelligence cognition layer, they are still mainly based on process simulation and weak reasoning, and have not yet formed true cognitive reasoning and decision support capabilities. Addressing the inherent limitations in intelligence cognition, such as mechanical collaborative interaction, lack of value judgment, and superficial cognitive alignment,, it is further proposed that open-source intelligent agents should be transformed from passive capability tool systems into proactive cognitive agent systems through pathways such as cognitive template embedding, cognitive  right confirmation for value assessment, and cognitive collaborative construction. This provides theoretical reference and practical inspiration for its in-depth application in the field of intelligence science.
  • Cognitive Risks of Agent Applications in the Intelligence Domain and Their Governance
    Cui Zhiwen Luo Liqun Li Guangjian
    Library & Information. 2026, 46(04): 47-58. DOI:10.11968/tsyqb.1003-6938.2026043
    Abstract ( ) Download PDF ( ) Knowledge map Save
    As agents are increasingly embedded in the full workflow of intelligence perception, retrieval, analysis and judgment, they have profoundly reshaping the cognitive foundation of intelligence work. From the perspective of cognitive power, this paper systematically examines the impacts of agent application on the intelligence cognitive process and identifies three core latent risks in intelligence practice, namely cognitive entry risk, cognitive pathway risk, and cognitive outcome risk. To address these risks, this paper proposes governance solutions from two dimensions: core principles and institutional mechanism design. The core principles highlight cognitive diversity, transparency of cognitive power, the inalienability of cognitive sovereignty and non-dilutability of cognitive responsibility. Mechanism design focuses on institutionally safeguarding cognitive diversity, establishing a disclosure regime for cognitive impacts, securing substantive human adjudication over pivotal cognitive links, and clarifying the accountability chain for intelligence judgement.
  • Consolidating the Foundation of Intelligence Security: An Interpretation of Data Security  Governance  from  the  Perspective  of  Intelligence  Studies
    Yang Ruixian Guo Shiyu Sun Zhuo Mo Fuchuan
    Library & Information. 2026, 46(04): 59-69. DOI:10.11968/tsyqb.1003-6938.2026044
    Abstract ( ) Download PDF ( ) Knowledge map Save
    In the digital-intelligence era, data has become a core element of intelligence activities, and data-related risks have become deeply embedded throughout the processes of intelligence collection, processing, analysis, and circulation, posing severe challenges to traditional approaches to intelligence security paradigms. From the perspective of the Holistic Approach to National Security, this paper re-examines the connotation and boundaries of intelligence security and systematically elucidates the underlying logic through which data security supports intelligence security from three dimensions: practical motivations, intrinsic foundations, and specific manifestations. It argues that data security governance is a prerequisite for consolidating the defense of intelligence security, with its core lying in ensuring trustworthy data sources, standardized processing procedures, and the blocking of risk transmission. On this basis, the paper constructs a data security governance framework oriented toward intelligence security. Taking the coordination of development and security as the main thread, the framework rests on three pillars—technology, institutions, and collaboration—and incorporates four key evaluation dimensions: trustworthiness, controllability, usability, and traceability. It further establishes a closed-loop operational mechanism linking monitoring, assessment, and response, thereby promoting a shift in intelligence security assurance from static and passive control toward dynamic and proactive governance.
  • Construction of a Process Evaluation Model for Human-AI Collaborative Academic Information Production          ——Reverse Cognitive Audit and Empirical Research Based on Metacognition Score
    Lin Xin Ma Haiqun Ma Xiaoqi Zhang Guoqing
    Library & Information. 2026, 46(04): 70-81. DOI:10.11968/tsyqb.1003-6938.2026045
    Abstract ( ) Download PDF ( ) Knowledge map Save
    After generative artificial intelligence became involved in information production in academic information production, outcome-oriented evaluations represented by plagiarism rates and AIGC detection rates fail to fully reflect authors' actual mastery of textual content, argumentative logic, and citation evidence chains. This study constructs an MS-based process evaluation framework grounded in reverse cognitive audit, incorporating overall success rate, confidence-correctness consistency, confidence stability, and performance fluctuation into metacognitive scoring. Based on 120 course papers, the study conducts normal observation and curriculum intervention. The results show that the samples have generally low plagiarism rates but relatively high AIGC detection rates, with a weak correlation between the two. The process evaluation prompt group outperforms the outcome-compliance intervention group in higher mean MS score and higher proportion of MS≥80, as well as citation extraction failure rate and reference error rate. The findings indicate that process evaluation oriented toward textual understanding, evidence tracing, and confidence calibration can compensate for the deficiencies of outcome-oriented detection in process interpretation and provide a complementary evaluation approach for academic information quality governance in the AIGC environment.
  • Scale Optimization and Mediation Test of User Engagement in Public Digital Cultural Services
    Liu Jiajing Zhu Shenping Yao Shanji Zheng Jianming
    Library & Information. 2026, 46(04): 82-93. DOI:10.11968/tsyqb.1003-6938.2026046
    Abstract ( ) Download PDF ( ) Knowledge map Save
    To elucidate the measurement dimensions and mediating value of user engagement in public digital cultural services (PDCS), this study carried out scale optimization and an empirical test on its mediating effect.Based on two rounds of questionnaire surveys for scale revision, this paper constructs a moderated mediation model to explore the mediating mechanism of PDCS user engagement in the relationship between immersive experience and cultural tourism intention, as well as the moderating effect of destination interest. The results indicate that the optimized scale is well-adapted to PDCS immersive experience scenarios. PDCS user engagement plays a partial mediating role in the positive impact of immersive experience on cultural tourism intention, and this mediating effect is positively moderated by destination interest.
  • National Strategy, Public Expectations, and Professional Consciousness: The Triple Logic of China's Public Library Value Construction
    Zhang Xindi Li Zexin
    Library & Information. 2026, 46(04): 94-102. DOI:10.11968/tsyqb.1003-6938.2026047
    Abstract ( ) Download PDF ( ) Knowledge map Save
    The mission, vision and values of public libraries serve as the core elements that define functions of libraries, standardize industrial development, and link national cultural governance with public cultural services. This paper, based on the evolution of trajectory of public libraries in China and abroad, combining historical review and multi-cases comparison, this paper constructs a three-layer logical analysis framework of national strategy, public expectations, and professional consciousness and analyzes the internal mechanism through which the interaction and game among these three forces shape the mission, vision and values of public libraries. The study finds that the mission, vision and values of Chinese public libraries constitute a dynamic structure shaped by the continuous interplay of national, social and professional logics. Their evolution stems from the dynamic  transformation of dominant disclosure of the triple logic in the course of history. There exists a spiral relationship between value concepts and practices, which is characterized by “practical dilemma-value feedback-value correction-new value concepts generation”.
  • From Document Collections to Data Collections: The Connotation, Transformation, and Practice of the Data Library
    Li Jinyong Li Peng
    Library & Information. 2026, 46(04): 103-109. DOI:10.11968/tsyqb.1003-6938.2026048
    Abstract ( ) Download PDF ( ) Knowledge map Save
    Focusing on the emerging form of the “data library,” this study integrates collection restructuring, value release, and service innovation into a unified framework, and carries out research from three perspectives: conceptual connotation, transformation pathways, and practical cases. At the conceptual level, it elaborates the morphological characteristics of the data library from four dimensions: collection, space, service, and subject participants. At the transformation pathway level, it proposes four coordinated transformation strategies: institutional restructuring, technological transformation, organizational reengineering, and scenario-driven development. At the practical case level, it conducts an empirical analysis of practical data practices in six representative libraries, identifying several typical characteristics such as the collection dimension takes the lead in development breakthroughs, spatial construction develops faster than service transformation, and significant path divergence between academic libraries and public libraries. 
  • Evaluation of AI Application in Government: Indicator System Construction and Performance  Analysis  of  Major  Chinese  Cities#br#
    Wang Fang Zhang Xinyue Zhu Xuekun Wang Meiquan Yue Zhinan Yang Tiande Liu Jia
    Library & Information. 2026, 46(04): 110-123. DOI:10.11968/tsyqb.1003-6938.2026049
    Abstract ( ) Download PDF ( ) Knowledge map Save
    With the rapid advancement of digital intelligent technologies, artificial intelligence (AI) has increasingly become an important driver of public governance modernization. To systematically evaluate the application level of AI in urban governments and examine its development patterns, this study constructs an evaluation framework comprising four dimensions: supporting infrastructure, government operations, social governance, and public services, including 11 indicators, such as computing infrastructure, smart emergency response, and automatic policy entitlement services. Based on publicly available data of 76 major cities in China and objective weighting using the Entropy Weight Method, the study finds that AI application in Chinese municipal governments exhibits a tiered development pattern characterized by leading cities followed by most others. Across the four dimensions, public services and social governance show relatively high development levels, while government operations and supporting infrastructure remain relatively weak, with significant inter-city disparities within the infrastructure dimension. At the indicator level, smart tourism services, AI-powered government chatbots, smart emergency response, and intelligent transportation demonstrate relatively high levels, while computing infrastructure, AI policy frameworks, and AI-assisted decision-making remain underdeveloped and exhibit regional disparities.The study suggests that future efforts should focus on reducing regional disparities in infrastructure, strengthening intelligent decision-making capabilities, and enhancing regional coordination among urban agglomerations, thereby facilitating the transition of urban AI applications from experimentation to mature deployment and providing solid support for the high-quality development of digital government development.
  • Research on Credibility Assessment of Open Source Intelligence for National Defe- nse Science and Technology Security Driven by Large Language Models
    Zhang Haitao Zhang Chuanyang Fei Ruiteng Li Jiaxue
    Library & Information. 2026, 46(04): 124-133. DOI:10.11968/tsyqb.1003-6938.2026050
    Abstract ( ) Download PDF ( ) Knowledge map Save
    Guided by the Holistic Approach to National Security, this paper proposes an LLM-based credibility assessment method for open source intelligence on national defense science and technology security, aiming to denoise massive data and support scientific decision-making for national defense science and technology security governance. First, a a credibility assessment indicator system and framework adapted to the field of national defense science and technology security are constructed. Second, LoRA fine-tuning is employed to enhance the LLM's domain knowledge and its comprehension and execution of assessment tasks, and Dempster-Shafer (D-S) evidence theory is introduced to convert the model's indicator scores into evidential representations. Finally, the model is fine-tuned and validated based on highly adversarial national defense science and technology intelligence scenarios, and typical case analyses are carried out to implement credible assessment of open-source intelligence.
  • The Policy System and Optimization Path of Artificial Intelligence Bias Governance in China
    Yu Beili Pu Qingyun Feng Mei Su Xin
    Library & Information. 2026, 46(04): 134-144. DOI:10.11968/tsyqb.1003-6938.2026051
    Abstract ( ) Download PDF ( ) Knowledge map Save
    As artificial intelligence technology becomes increasingly integrated into fields such as healthcare, education, and employment, fairness issues arising from algorithmic bias have evolved into systemic concerns. This study selects 33 representative policy on AI algorithmic bias governance in high risk fields, using grounded theory for open coding, axial coding, and selective coding. Based on the logical structure of these policies, this study distills a model for governing algorithmic bias centered on “information resource allocation—service rights and interests protection.” This study finds that existing policy texts have already proposed corresponding regulatory measures in areas such as data and model management, feature and label management, output and service management, user rights protection, supervision and governance, as well as liability confirmation and rights protection. However, these policies still face structural issues, including regulatory fragmentation, i：imbalanced adaptation of governance tools, lack of and insufficient overall coordination and dynamic adjustment capability. It is necessary to improve the institutional system by enhancing cross-element coordination, implementing tiered dynamic supervision, and fostering collaboration among relevant departments, thereby promoting the transition of algorithmic bias governance from normative provisions to practical effectiveness.
Copyright © Library & Information
Tel: 0931-8270072　　E-mail: tsyqb@126.com
Powered by Beijing Magtech Co. Ltd.