Scientific exploration, collaboration and labor division in the large language model era

This study investigates how the rapid diffusion of large language models (LLMs) has influenced the structure of scientific work. By analyzing a massive dataset of over 775,000 scientists and 137,000 multi-author papers, the authors track shifts in research exploration, collaboration patterns, and the internal division of labor within teams from 2022 onwards.

Since 2022, there has been a marked increase in the breadth of research conducted by individual scientists. Researchers are publishing in a wider variety of fields, entering new domains at higher rates, and pivoting their research agendas more frequently. This trend is particularly pronounced among established scientists and those from non-English-speaking, low- and middle-income countries. The data suggests that LLMs may be lowering the barriers to entry for unfamiliar fields by assisting with literature synthesis, terminology, and methodological navigation.

Beyond individual exploration, the social structure of science is evolving. Collaboration networks have become more interdisciplinary, with scientists working with peers from a broader range of disciplinary backgrounds. Simultaneously, the internal organization of research teams has shifted. Analysis of CRediT contribution statements reveals that team members are adopting more specialized, narrower roles. While technical roles such as software development and validation have increased, conceptual and management roles have seen a relative decline. This suggests a transition toward more fluid, differentiated team structures where individuals may rely less on one another for broad, overlapping tasks.

Authors who exhibit stronger signals of LLM-assisted writing were already more interdisciplinary and exploratory before 2022. However, the gap between these "high-AI-writing" authors and their peers has widened significantly since the public release of advanced LLMs. This indicates a reinforcement effect where early adopters of AI tools are leveraging them to further accelerate their exploration of new research frontiers.