Angèle Christin

Since 2022, Generative AI and Large Language Models (LLMs) have taken contemporary societies by storm. Venture capital and financial markets have been pouring stupendous amounts of money into AI efforts and startups, while governments around the globe encouraged their researchers and companies to win the geopolitical “AI race.” The amount of funding and media coverage surrounding LLMs has been overwhelming, often silencing critical perspectives on the labor practices of data annotation companies or the environmental impact of data centers.

In this context, science has emerged as a poster child to illustrate the benefits of AI technologies. Initiatives such as AlphaFold’s protein folding, led by Google’s DeepMind, inspired many disciplines to argue that GenAI would lead to revolutionary advances. Most scientists now use GenAI as a daily part of their work, especially when their research involves coding. At the same time, scientists have also warned of problems associated with using GenAI for science, especially in terms of academic rigor, honesty, and critical skills. Researchers highlighted the epistemic risks associated with GenAI and how these technologies might erode trust in science. Early deployment of GenAI in the physical and astrophysical sciences show both promise and pitfalls. For instance, in an experiment conducted at the Space Telescope Science Institute, professional astronomers valued an LLM-powered literature bot but also flagged shortcomings such as shallow synthesis and misplaced confidence.

The widespread adoption and vibrant debates surrounding the use of GenAI in science call for systematic inquiry. What do scientists want from GenAI? What are their hopes and fears? And what is the opportunity cost of engaging with GenAI versus with human collaborators for scientific training and discovery? These questions about values, trade-offs, and trust are central to Christin's investigation of how astrophysicists navigate this new collaborative landscape.

Astrophysics provides a unique lens on the challenges of using LLMs and AI agents in science. The field is data-intensive, open, and highly computational. Progress depends on large public datasets (such as the Vera C. Rubin Observatory), reproducible pipelines, and references to a large literature (NASA ADS, which covers the literature corpus in astrophysics, indexes over 15 million resources). These features make astrophysics both an early adopter of AI tools and an ideal “stress test” for responsible LLM integration.

Angèle Christin is currently conducting intensive ethnographic fieldwork with astrophysicists, cosmologists, and astronomers at Stanford University’s Kavli Institute for Particle Physics and Astrophysics and the Stanford Linear Accelerator (SLAC), examining how AI is transforming their daily scientific practices. In Paris, Christin plans to add a comparative dimension to this project by conducting fieldwork at a French astrophysics lab.