• 中国科学学与科技政策研究会
  • 中国科学院科技战略咨询研究院
  • 清华大学科学技术与社会研究中心
ISSN 1003-2053 CN 11-1805/G3

科学学研究 ›› 2026, Vol. 44 ›› Issue (9): 1971-1984.

• 前沿与观点 • 上一篇    下一篇

全球视野与中美方阵:AI跨学科主题及演进

姜春1(), 贺星2   

  1. 1 暨南大学公共管理学院/应急管理学院, 广东广州 510632
    2 河海大学公共管理学院, 江苏南京 211100
  • 收稿日期:2025-06-18 修回日期:2025-08-14 出版日期:2026-09-15 发布日期:2026-09-15
  • 作者简介:姜春(1993-),男,副教授,博士,E-mail: cjiang19@fudan.edu.cn。
    贺星(2004-),女,硕士研究生。
  • 基金资助:
    国家社会科学基金青年项目(25CKX029)

Global perspectives and the Sino-US AI landscape: Interdisciplinary topic identification and evolution

JIANG Chun1(), HE Xing2   

  1. 1 School of Public Administration and Emergency Management, Jinan University, Guangzhou 510632, China
    2 School of Public Administration, Hohai University, Nanjing 211100, China
  • Received:2025-06-18 Revised:2025-08-14 Online:2026-09-15 Published:2026-09-15

摘要:

面对人工智能(AI)技术快速演进的新形势,跨学科融合日益成为推动技术突破与知识创新的重要路径。然而已有研究鲜有探讨AI的跨学科知识及其演进体系,制约了对AI发展内在规律的理解,进而影响掌握发展主动权的能力。该研究聚焦AI领域的跨学科知识融合与结构演化问题,引入BERTopic模型,运用学科多样性与凝聚力指数和主题强度与内容演化分析方法,构建“主题识别-结构解析-机制比较”研究框架,识别全球AI领域的典型跨学科主题,系统分析其知识结构特征与演进模式,并聚焦中美两国的演化路径差异与驱动机制对比。研究发现,AI领域形成23个结构类型各异、融合程度不一的跨学科主题,呈现多范式共生、异质知识集成与结构耦合演进的特征,表明AI本身已成为复杂技术-社会系统中典型的跨学科集成平台;中美分别形成“范式驱动-系统扩散型”(美国)与“任务导向-技术整合型”(中国)两类典型路径,在制度逻辑、组织机制与资源配置方面展现出深层差异,美国侧重自由探索与多元协同,中国则强调战略导向与集中攻关。研究提出构建异质耦合平台、优化立项机制、完善接口学科生态等政策建议,以增强中国AI领域的跨学科整合能力与系统创新韧性。研究有助于深化对AI跨学科知识生产和演进规律的理解,并为评估中美科技竞争态势提供学理支撑与启示。

关键词: 人工智能, 跨学科, 主题识别, 主题演进, 中美竞争

Abstract:

Against the backdrop of rapid advances in artificial intelligence (AI) and the accelerating reshaping of the global technological competition landscape, interdisciplinary integration has increasingly become a critical driver of technological breakthroughs and knowledge innovation. Particularly amid intensifying Sino-American technological rivalry, comparing the differences in AI interdisciplinary knowledge integration paths across countries is of significant practical relevance for optimizing innovation strategies and enhancing competitive resilience. However, existing studies have predominantly focused on AI theory development, technological innovation, and overall trends, with relatively limited systematic exploration of AI’s interdisciplinary knowledge system and its evolutionary mechanisms. This gap partially constrains understanding of the intrinsic patterns of AI development and, consequently, limits the ability to seize strategic initiative.

This study draws on the Web of Science (WoS) database, selecting 89,448 valid publications up to December 31, 2024, and constructs an integrated research framework encompassing “topic identification - structural analysis - mechanism comparison”. First, employing the BERTopic modeling method, the study identifies representative interdisciplinary topics in AI and characterizes their features in terms of knowledge sources, integration depth, and structural types by introducing disciplinary diversity and cohesion indicators. Second, by analyzing changes in topic intensity and content evolution, it reveals the evolutionary paths and coupling patterns of interdisciplinary knowledge across different stages. Finally, using the United States and China as comparative cases, the study examines differences in topic distribution, structural patterns, and driving mechanisms, exploring how institutional logic, organizational forms, and resource allocation shape interdisciplinary integration.

The findings indicate that the AI field has formed a complex interdisciplinary knowledge system composed of 23 heterogeneous topics, with significant variation across topics in disciplinary sources, integration depth, and structural complexity, displaying a combination of parallel paradigms, heterogeneous integration, and structural coupling. The evolution of interdisciplinary knowledge typically follows a cyclical process, progressing from the introduction of heterogeneous elements to the integration of methodological paradigms and the differentiation of new subfields, with coupling realized through multi-layered nested coordination between core and interface disciplines. International comparison shows that the U.S. emphasizes paradigm-driven, system-level diffusion, relying on open collaboration and multi-actor participation to promote parallel exploration across multiple directions, whereas China exhibits an engineering- and application-driven pattern, emphasizing concentrated resource mobilization and rapid integration, resulting in goal-focused and outcome-oriented pathways. These mechanistic differences not only influence the pace and direction of knowledge evolution but also shape each country’s relative advantages in frontier leadership and applied breakthroughs.

The study proposes high-level strategic directions for enhancing China’s interdisciplinary integration capacity and systemic innovation resilience in AI, including the development of heterogeneous coupling platforms, optimization of project and collaboration mechanisms, and the cultivation of interface discipline ecosystems. The research systematically extends the characterization of interdisciplinary knowledge structures, the identification of evolutionary paths, and comparative analysis of U.S. and Chinese mechanisms, offering a generalizable analytical framework and methodological approach. It deepens understanding of the patterns and mechanisms of interdisciplinary integration in AI and provides a novel theoretical perspective for studying the complexity of technological convergence. By elucidating cross-national differences in interdisciplinary innovation paths and driving logics, the study provides scholarly support and practical guidance for formulating forward-looking AI development strategies and strenthening competitiveness global technological competition.

Key words: artificial intelligence, interdisciplinarity, topic modeling, topic evolution, China-US competition

中图分类号: