关键词: AI4S, AI指南, AI素养, 研究型图书馆, 生成式人工智能

Abstract:

[Purpose/Significance] AI for Science (AI4S) refers to the thorough integration of artificial intelligence (AI) into scientific, technological, and engineering research. It is driving a fundamental transformation in the way science is conducted by automating knowledge generation and enabling full-spectrum intelligence across the entire research lifecycle, and fostering interdisciplinary convergence Widely regarded as the fifth paradigm of scientific inquiry, following experimental, theoretical, simulation-based, and data-intensive research, AI4S heralds a new era of knowledge discovery. AI4S is an emerging paradigm that presents new challenges and higher demands for libraries' AI literacy education and AI services. Leading North American research universities, such as Harvard University, Dartmouth College, and the University of Toronto, have already leveraged their libraries to provide AI guides services supporting scientific research. Adopting the novel lens of AI4S-oriented services, this paper offers a systematic analysis of how North American research libraries have developed their AI4S guides, with the aim of providing transferable insights and practical references for research university libraries in China as they design research-support services under the AI4S paradigm. [Method/Process] This study employs Web survey and content analysis methods, using member libraries of the Association of Research Libraries (ARL) in North America as samples, to analyze the characteristics of AI4S guides in research libraries. It examines the key components of AI guides from three perspectives: AI cognition, AI tools, and academic usage norm. [Results/Conclusions] The survey was conducted from 1 December 2024 to31 March 2025. Among all ARL libraries, 97 were found to offer research-oriented AI guides: 85 from the United States and 12 from Canada. The analysis revealed the three hallmarks of the AI4S guides developed by the ARL members: 1) LibGuides serve as the primary delivery mechanism for AI4S guidance, 2) the guides are deeply integrated with AI-literacy instruction, 3) generic research guides coexist with and complement discipline-specific guides. These AI4S guides provide comprehensive support for researchers' capacity-building, covering the role of AI in research, informed selection of AI tools, and responsible use of AI technologies. Recommendations for AI tools occupy a central place in AI4S services; across the libraries, the endorsed tools fall into three categories: institution-developed tools, institution-approved tools, and third-party tools. Based on the findings, the paper proposes recommendations for research university libraries in China to develop and enhance AI guides under the AI4S paradigm, including strengthening policy interpretation, prioritizing data security, and promoting cross-departmental collaboration. This study still has the following limitations: 1) it is based on an analysis of the overall situation, and the analysis of typical cases of AI4S services is insufficient, 2) it did not conduct a comparative analysis of AI4S guides and services between domestic and international libraries.

Key words: AI for Science, AI guide, AI literacy, research libraries, generative artificial intelligence