Large Physics Models and EuCAIF — AI as a New Scientific Tool

by Sascha Caron (Nikhef National institute for subatomic physics (NL))

Wednesday 14 May 2025, 11:00 → 12:00 Europe/Zurich

40/S2-A01 - Salle Anderson (CERN)

In this talk, we introduce the concept of Large Physics Models (LPMs) as specialized, general-purpose AI systems for physics research, and outline a practical roadmap for their development. LPMs would inherit concepts from Large Language Models and physics foundation models, functioning ultimately as ChatGPT-like tools owned and developed by the scientific community. We explore both the arguments for and against the pursuit of LPMs within the physics community, discuss how their scientific capabilities might be assessed, and reflect on the importance of considering philosophical perspectives on their role. Drawing inspiration from collaborative models in experimental physics, we propose concrete approaches for building large-scale LPMs. We also present the European Consortium for AI in Fundamental Physics (EuCAIF), which supports AI-enabled research across particle physics, astrophysics, nuclear physics, gravitational waves, cosmology, theory, and computational infrastructure. We discuss how EuCAIF's interdisciplinary structure can provide the coordination necessary to advance the development of AI in physics and potentially facilitate the realization of tools like LPMs.”

Bio:

Sascha Caron works at the interface of fundamental physics and data science/AI. He has been involved in several important experiments, such as D0 (at Fermilab) and H1 (DESY). As a researcher at the ATLAS experiment (CERN), he made leading contributions to the first LHC searches for dark matter and supersymmetry and is pioneering data-driven methods and machine learning to explore new physical phenomena.
In addition to his experimental work, Caron leads research in the fields of dark matter phenomenology, astroparticle physics and astrophysics. Caron has initiated and led a number of innovative projects dealing with data analysis and artificial intelligence. This includes founding the Darkmachines initiative (www.darkmachines.org running 2017-2022 to develop AI benchmarks for fundamental physics). Currently he is leading the AI infrastructure group for nuclear, particle and astroparticle physics (within the JENA initiative). He is a main initiator and a member of the Management Board of the European Collaboration for AI in Fundamental Physics (EuCAIF) (see eucaif.org).

 

CERNcolloquium_2025_AI.pdf