How should the advancement of large language models affect the practice of science?
Large language models (LLMs) are being increasingly incorporated into scientific workflows. However, we have yet to fully grasp the implications of this integration. How should the advancement of large language models affect the practice of science? For this opinion piece, we have invited four diverse groups of scientists to reflect on this query, sharing their perspectives and engaging in debate. Schulz et al. make the argument that working with LLMs is not fundamentally different from working with human collaborators, while Bender et al. argue that LLMs are often misused and overhyped, and that their limitations warrant a focus on more specialized, easily interpretable tools. Marelli et al. emphasize the importance of transparent attribution and responsible use of LLMs. Finally, Botvinick and Gershman advocate that humans should retain responsibility for determining the scientific roadmap. To facilitate the discussion, the four perspectives are complemented with a response from each group. By putting these different perspectives in conversation, we aim to bring attention to important considerations within the academic community regarding the adoption of LLMs and their impact on both current and future scientific practices.
Authors
- Marcel Binz (ORCID: https://orcid.org/0000-0001-8872-8386)
- Daniel J. Schad (ORCID: https://orcid.org/0000-0003-2586-6823)
- Dirk U. Wulff (ORCID: https://orcid.org/0000-0002-4008-8022)
- Emily M. Bender (ORCID: https://orcid.org/0000-0001-5384-6227)
- Samuel J. Gershman (ORCID: https://orcid.org/0000-0002-6546-3298)
- Zeynep Akata (ORCID: https://orcid.org/0000-0002-1432-7747)
- Jevin D. West (ORCID: https://orcid.org/0000-0002-4118-0322)
- Balázs Aczél (ORCID: https://orcid.org/0000-0001-9364-4988)
- Richard M. Shiffrin (ORCID: https://orcid.org/0000-0002-3708-3212)
- Colin Allen (ORCID: https://orcid.org/0000-0003-4497-1725)
- Adina L. Roskies (ORCID: https://orcid.org/0000-0003-2738-916X)
- Stephan Alaniz (ORCID: https://orcid.org/0000-0003-3541-2163)
- Marco Marelli (ORCID: https://orcid.org/0000-0001-5831-5441)
- Carl T. Bergstrom (ORCID: https://orcid.org/0000-0002-2070-385X)
- Matthew Botvinick (ORCID: https://orcid.org/0000-0001-7758-6896)
- Vencislav Popov (ORCID: https://orcid.org/0000-0002-8073-4199)
- Qiong Zhang (ORCID: https://orcid.org/0000-0001-9062-9571)
- Eric Schulz (ORCID: https://orcid.org/0000-0003-3088-0371)
Institutions
- Rutgers, The State University of New Jersey (US)
- Eötvös Loránd University (HU)
- Google (United States) (US)
- Harvard University (US)
- University of California, Santa Barbara (US)
- University of Basel (CH)
- University of Washington (US)
- University of Zurich (CH)
- Harvard University Press (US)
- University of California System (US)
- Google DeepMind (United Kingdom) (GB)
- Rutgers Sexual and Reproductive Health and Rights (NL)
- Max Planck Institute for Biological Cybernetics (DE)
- Google (United Kingdom) (GB)
- Oxford Centre for Computational Neuroscience (GB)
- Indiana University Bloomington (US)
- Max Planck Institute for Human Development (DE)
- Czech Academy of Sciences, Institute of Psychology (CZ)
- Munich Center for Machine Learning
- University College London (GB)
- Indiana University (US)
- Technical University of Munich (DE)
- University of Milano-Bicocca (IT)
Publication Details
- Journal
- Proceedings of the National Academy of Sciences
- Published
- 2025-01-27
- DOI
- https://doi.org/10.1073/pnas.2401227121
- Citations
- 42
- Primary Topic
- Topic Modeling
- Type
- article
- Field-Weighted Citation Impact
- 104.20
Funders
- Deutsche Forschungsgemeinschaft
- Bundesministerium für Bildung und Forschung