Drivers and Barriers of AI Adoption and Use in Scientific Research

TL;DR

The paper investigates drivers and barriers to adopting AI in scientific research by embedding adoption decisions within the scientific and technical human capital (STHC) framework, using a large OpenAlex-based dataset (1980–2020) and a conditional logit with matched pairs. It formalizes hypotheses across external (social ties, mentorship, compute access) and internal (scientific background, taste for exploration) resources and estimates first-use and reuse decisions, revealing that social capital and about-knowledge, particularly ties to AI-experienced peers and AI-focused institutions, are robust predictors of adoption. The results show that institutional AI specialization, prior collaborations with CS/AI experts, and the involvement of newbies strongly drive initial adoption, while HPC access is only relevant in select domains; reuse is shaped by realized STHC, with effects generally smaller but still present. These findings suggest policy and organizational levers—such as fostering boundary-spanning, incentivizing knowledge sharing, and targeting AI specialization within institutions—to accelerate diffusion of AI tools across science, while recognizing potential interdisciplinary collaboration costs and compute-divide concerns.

Abstract

New technologies have the power to revolutionize science. It has happened in the past and is happening again with the emergence of new computational tools, such as artificial intelligence and machine learning. Despite the documented impact of these technologies, there remains a significant gap in understanding the process of their adoption within the scientific community. In this paper, we draw on theories of scientific and technical human capital to study the integration of AI in scientific research, focusing on the human capital of scientists and the external resources available within their network of collaborators and institutions. We validate our hypotheses on a large sample of publications from OpenAlex, covering all sciences from 1980 to 2020, and identify a set key drivers and inhibitors of AI adoption and use in science. Our results suggest that AI is pioneered by domain scientists with a `taste for exploration' and who are embedded in a network rich of computer scientists, experienced AI scientists and early-career researchers; they come from institutions with high citation impact and a relatively strong publication history on AI. The access to computing resources only matters for a few scientific disciplines, such as chemistry and medical sciences. Once AI is integrated into research, most adoption factors continue to influence its subsequent reuse. Implications for the organization and management of science in the evolving era of AI-driven discovery are discussed.

Figures (3)

• Figure 1: The focal scientist. A focal scientist is active in a domain other than computer science ('domain scientist'), has a first paper in year $t>1980$, a first AI-related paper in the period $t+n=(2012,2020)$, and a subsequent last paper in the period $t+n+i=(2013,2020)$, with $n, i \geq 1$.
• Figure 2: STHC framework. Left figure: The institutional environment potentially provides information, directs attention, and offers resources (computing facilities, human capital) related to AI; also, institutions possess a certain level of reputation and scientific excellence. Middle figure: The prior co-author network provides human capital that is relevant to the focal scientists' domain, computational analysis, and/or AI. Right figure: The focal scientist's human capital is described by her past research output in terms of scientific content, quality, and internationality. The variables are described in detail in Section 3.2.
• Figure 3: Matching procedures to investigate first-use of AI (top) and re-use of AI (bottom)