2027 UF Guide to AI in Science

Authors

Nicholas A. Coles ^1,2, Abhay Alaukik ¹, Majed A. Arishi ³, Shanting Chen ¹, Annabel V. Dang ¹, Jungjae Im ⁴, Nabiha Iqbal ¹, Minyi Jiang ¹, Rui Jin ¹, Zhongchi Li ¹, Sooyeon Park ¹, Mariana G. Rebello ⁵, Jihyun Soh ⁴, Paola J. Sullivan ⁶, Lyra G. Toohey ¹, Sofia C. Villanueva ¹

Affiliations

¹ Department of Psychology

² Bureau of Economic and Business Research

³ Department of Civil and Coastal Engineering

⁴ Department of Marketing

⁵ School of Special Education, School Psychology, and Early Childhood Studies

⁶ Department of Family, Youth and Community Sciences

NoteThis was written by humans.

✨ But we provide links to helpful AI-generated content along the way.

AI is radically redefining the nature of scientific inquiry — for better and for worse.

In Fall 2026, faculty, graduate students, and postdocs across multiple departments at the University of Florida came together to understand this urgent development. In a seminar titled Collective [Artificial] Intelligence in Science, we documented and discussed how AI is changing each component of our research workflow. This includes project management, idea development, data analysis, writing, and dissemination. To push our understanding, we alternated between the perspectives of an AI zealot and an AI critic — identifying key resources, consensus statements, and areas of disagreement along the way. The result is a resource adopted in courses and labs the following semester: the 2027 UF Guide to AI in Science.

The seminar began with an introduction to big-team science — its rise¹, competitive advantages², and relevance to the current moment. For instance, the paper describing OpenAI’s GPT-1 had 4 authors³; the paper describing GPT-3 had 31 authors⁴; and the paper describing GPT-5 had 483 authors⁵ (Figure 1). In the artificial realm, a prominent AI-led cybersecurity attack on the online community, Hugging Face, involved coordinated action among not one but hundreds of AI agents⁶. Similarly, over 10,000 AI agents recently exchanged approximately 5 million messages over 88 hours in order to solve the Navier–Stokes equations⁷ — one of the seven Clay Millennium Prize Problems. Whether it be human or artificial, some of our most impressive accomplishments are requiring something rarely formally taught in science: large-scale collaboration⁸.

Figure 1: Credited co-authors on flagship AI model reports from OpenAI, Google, and Meta, 2018–2025. One square is one credited person; each block’s area is its author list.

After reviewing this backdrop, the course itself became a large-scale collaboration, dedicated to understanding how AI is changing the nature of science. Each week followed a consistent structure (Figure 2). First, we would select a component of the research pipeline to examine (e.g., data analysis). Second, we would spend one week compiling resources for understanding the emerging role of AI in that part of the research pipeline. Third, we would engage in peer-to-peer teaching. Last, we would use a live online voting platform to solicit, debate, and vote upon key resources and takeaways for the week^9,10. Through these exercises, seminar participants received hands-on experience with large-scale collaboration¹¹. Along the way, they not only developed and debated their personal philosophy statements — but also the general direction of AI at UF. The collective effort culminated in this 2027 UF Guide to AI in Science: a summary of our top resources, takeaways, and disagreements.

Figure 2: Structure of the UF Fall 2026 seminar, titled Collective [Artificial] Intelligence in Science.

1.

Coles, N. A., Hamlin, J. K., Sullivan, L. L., Parker, T. H. & Altschul, D. Build up big-team science. Nature 601, 505–507 (2022).

2.

Coles, N. A., Goes Braga Takayanagi, J. F., Grant, G. L. & Basnight-Brown, D. M. The rise, impact, and imbalances of big-team psychology. Psychological Science 37, 287–297 (2026).

3.

Radford, A., Narasimhan, K., Salimans, T. & Sutskever, I. Improving Language Understanding by Generative Pre-Training. https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf (2018).

4.

Brown, T. B., Mann, B., Ryder, N., Subbiah, M., et al. Language models are few-shot learners. arXiv preprint https://doi.org/10.48550/arXiv.2005.14165 (2020) doi:10.48550/arXiv.2005.14165.

5.

OpenAI. OpenAI GPT-5 system card. arXiv preprint https://doi.org/10.48550/arXiv.2601.03267 (2025) doi:10.48550/arXiv.2601.03267.

6.

OpenAI. The hugging face incident and the road ahead. https://openai.com/index/hugging-face-incident-and-the-road-ahead/ (2026).

7.

OpenAI. On the Navier–Stokes Millennium Prize Problem. https://openai.com/index/navier-stokes-solution/ (2026).

8.

National Research Council. Enhancing the Effectiveness of Team Science. (National Academies Press, Washington, DC, 2015). doi:10.17226/19007.

9.

Coles, N. A. et al. Team scientists should normalize disagreement. Science 384, 1076–1077 (2024).

10.

Dalkey, N. & Helmer, O. An experimental application of the Delphi method to the use of experts. Management Science 9, 458–467 (1963).

11.

Baumgartner, H. A. et al. How to build up big team science: A practical guide for large-scale collaborations. Royal Society Open Science 10, 230235 (2023).