Rise of Generative Artificial Intelligence in Science

Abstract Generative Artificial Intelligence (GenAI) has rapidly emerged as a tool in scientific research. To examine its diffusion and impact relative to other AI technologies, we conduct an empirical analysis using the OpenAlex bibliometric database to retrieve GenAI and other AI relevant publications. To distinguish between publications that adopt AI technologies and those that mention them without AI use, we develop a two-stage classifier based on GPT-4 and SciBERT to categorize AI publications by their focus on either application or discussion. For publications from 2017 to 2023, we analyze growth trends and the geographical distribution of GenAI adoption across scientific fields. We also examine changes in team size and international collaboration to assess whether GenAI, as an emerging research area, exhibits distinct collaboration patterns compared to other AI technologies. Our findings show a rapid growth not only in GenAI-related publications but also in their applications, which are expanding beyond computer science into a wide range of disciplines. U.S. researchers produced nearly two-fifths of global GenAI publications and applications, while Chinese researchers contributed almost one-third. Several small and medium-sized advanced economies also demonstrate relatively high levels of GenAI applications in research. GenAI application is positively associated with larger team sizes ( p < 0.001) and higher rates of international collaboration ( p < 0.05). In GenAI discussion publications, the positive correlation with international collaboration is also significant ( p < 0.05). This suggests that applying GenAI in scientific research may require greater collaborative efforts, especially international partnerships, potentially reinforcing global collaboration networks embedded in the geography of knowledge production.