Drivers and barriers of AI adoption and use in scientific research
基于OpenAlex数据集，研究影响科研中AI采纳的关键因素，发现早期采纳者多来自AI合作网络丰富的机构，而大型语言模型普及后，社会资本（如与AI经验者合作）仍是持续驱动力，对科学政策制定者有用。
We study the early adoption and use of artificial intelligence (AI) in scientific research. Using a large dataset of publications from OpenAlex (all fields, up to 2024) and building on theories of scientific and technical human capital, we identify key factors that influence AI adoption. We find that early adopters were domain scientists embedded in AI-rich collaboration networks and affiliated with institutions with strong AI credentials. Access to high-performance computing (HPC) mattered only in a few scientific disciplines, such as biology and medical sciences. More recently, as tools like Large Language Models (LLMs) have diffused, AI has become more accessible, and institutional advantages appear to matter less. However, social capital—especially ties to AI-experienced collaborators and early-career researchers—remains a persistent driver of adoption. We discuss the implications for science policy and the organization of research in the age of AI. • Factors driving and hindering AI adoption in science. • AI adoption shaped by social, institutional, and individual factors. • Collaboration networks and team composition strongly predict adoption. • Access to computing resources is generally not a major barrier. • Institutional and technical factors matter less after LLMs emerge.