UC Santa Barbara physicists test OpenAI models to accelerate particle physics research

Researchers at UC Santa Barbara and the Kavli Institute for Theoretical Physics are experimenting with OpenAI models to shorten the time needed to generate and test theories in particle physics, reducing tasks that once took weeks to minutes.

A team of physicists at the University of California, Santa Barbara (UCSB) and the Kavli Institute for Theoretical Physics (KITP) is using OpenAI models to explore whether artificial intelligence can accelerate parts of the scientific process in particle physics.

The researchers developed a system called FERMIACC, which uses OpenAI models alongside established collider physics tools to generate and test hypotheses about unexplained experimental results. The work focuses on analyzing anomalies detected in particle collider data and evaluating potential theoretical explanations.

AI agents tested as collaborators in physics research

Amalia Madden, a postdoctoral researcher at KITP, began experimenting with AI tools as a way to clarify research questions and bridge gaps between different areas of physics. She described initially using AI in a limited way before realizing that improved reasoning models could be applied to more complex research tasks.

Working with UCSB PhD candidate Inigo Valenzuela Lombera, Madden began using OpenAI models to help construct and test explanations for unusual results observed in particle collider experiments.

In particle physics, theorists often respond quickly when collider experiments show unexpected deviations from the Standard Model, which describes most known fundamental particles and forces. Researchers generate potential explanations, simulate particle interactions, and compare the results with observed data to see if the hypothesis holds.

Traditionally, that process can require weeks of work by graduate researchers.

FERMIACC pipeline reduces testing cycle from weeks to minutes

To speed up that workflow, Madden and Valenzuela Lombera collaborated with UCSB professors Nathanial Craig and Prateek Agrawal and KITP postdoctoral researcher Jessica Howard to build the FERMIACC system.

FERMIACC operates as a closed-loop agent pipeline built with the OpenAI Agents SDK and integrates existing collider physics tools including FeynRules, MadGraph, and Pythia.

The system generates hypotheses, runs simulations, compares predicted particle signatures with experimental data, and evaluates how well each model matches the observed results. According to the researchers, hypothesis generation can take seconds and a full simulation and analysis cycle can finish in under ten minutes.

Potential applications beyond particle colliders

The researchers point to past collider anomalies as an example of why faster analysis could be valuable. In 2015, data from the Large Hadron Collider suggested a possible new boson particle, prompting hundreds of theoretical papers before later analysis showed the signal was likely a statistical fluctuation.

FERMIACC is designed to test possible explanations more quickly by automating parts of the modeling and simulation process while maintaining checks on particle behavior and interactions.

The team also suggests that similar approaches could be applied to other areas of physics. Future work could include analyzing cosmological data, where faint signals related to dark matter, cosmic inflation, or early universe physics may require large-scale modeling and simulation.

Researchers say the system illustrates how AI models may move beyond conversational tools when integrated through APIs and connected to scientific software environments.

ETIH Innovation Awards 2026

The ETIH Innovation Awards 2026 are now open and recognize education technology organizations delivering measurable impact across K–12, higher education, and lifelong learning. The awards are open to entries from the UK, the Americas, and internationally, with submissions assessed on evidence of outcomes and real-world application.