Study: LLMs are reshaping scientific collaboration and labor division

A new preprint from Zheng, Hong, Liu, and Ni (arXiv:2607.20923) provides the first large-scale evidence of how large language models (LLMs) are reshaping scientific workflows. The researchers linked PubMed Central full texts with OpenAlex data for 775,323 scientists and analyzed CRediT contribution statements from 137,120 multi-author papers. Their key finding: after 2022, scientists increasingly published across more intellectually distant fields and entered entirely new domains. This rise in interdisciplinarity was especially pronounced among established scientists and those from non-English-speaking low- and middle-income countries. Authors with stronger AI-writing signals were already more interdisciplinary before widespread LLM adoption, but the gap widened further after 2022.

Beyond exploration, the study reveals dramatic changes in collaboration and labor division. Scientists' collaboration networks became more interdisciplinary after 2022, yet for AI-heavy authors, research interdisciplinarity was less tied to their collaborators' disciplinary diversity—suggesting LLMs let individuals work across fields without relying on cross-disciplinary teams. Within teams, roles became more distinct and fluid. Contributors on post-2022 papers reported narrower role sets, shared fewer common roles, and showed less rigid role profiles. Software and validation roles increased, while conceptual and management roles decreased. This points to a shift where team members take on more specialized, separate responsibilities and may rely less on each other to perform tasks—a fundamental reorganization of scientific labor in the LLM era.