"English is the new coding language"
In this mini-workshop, we demonstrate practical agentic workflows for scientific research and discuss how these tools can be deployed responsibly and effectively, taking examples from high energy physics. We discuss human-in-the-loop approaches, validation strategies, reproducibility, and the evolving role of scientific expertise in the age of increasingly capable AI systems.
Talks and hands-on demonstrations will be followed by an extended discussion and Q&A session amongst all participants (hopefully ranging from undergrad students to full professors), where we discuss:
- How to use agentic AI to reduce growing coding and workflow burdens in science?
- What technical skills remain essential for researchers supervising and validating AI-assisted work?
- How should research groups, training programs, and scientific collaborations adapt to new capabilities?
- Best practices to ensure robustness, reproducibility, and scientific accountability?
- +++
We hope to use this as an opportunity to get everyone at the same table and discuss how our communities can most effectively benefit from these technologies while maintaining the high standards of rigor scientific research requires.
Speakers:
Daniel Murnane (Niels Bohr Institute)
Eric Moreno (MIT)
Fazl Barez (University of Oxford/Martian)
Sofia P. Schweizer (Rutgers University)
Papers:
"AI Agents Can Already Autonomously Perform Experimental High Energy Physics" https://arxiv.org/pdf/2603.20179
"Automated Interpretability-Driven Model Auditing and Control: A Research Agenda" F. Barez
"COLLIDER-BENCH: Benchmarking AI Agents with Particle Physics Analysis Reproduction" arxiv 2605.13950