Events
MaD Seminar: Combining Fast and Slow Thinking for AI in the Physical Sciences
Speaker: Haewon Jeong (UCSB)
Location: 60 Fifth Avenue, Room 150
Date: Thursday, November 19, 2026
Abstract: Should the next advance in AI for science come from larger models, more inference-time compute, or more agents? These represent three distinct scaling axes. In this talk, I will argue that the structure of a scientific problem should guide how we navigate this three-dimensional scaling space.
First, I will discuss our work on building AI agents for NMR spectroscopy, an inverse problem that involves inferring molecular structures from spectroscopic data. By combining specialized chemistry tools, small predictive models, and the reasoning capabilities of LLMs, our agent achieves unprecedented accuracy in molecular structure elucidation for small organic molecules.
Next, I will turn to scaling multi-agent systems for scientific discovery. We formalize scientific discovery as a search over a hypothesis graph and use statistical mechanics to study the collective dynamics of interacting agents. We demonstrate phase-transition behavior as we vary the degree of agent-to-agent communication across different graph topologies, revealing how the benefits of collective intelligence depend on task structure.
This perspective invites the broader scientific community to systematically explore how to bridge the gaps in the jagged intelligence of today's frontier models for their own scientific problems by navigating this three-dimensional scaling space.
Bio: Jeong received her Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University (’20), where her research pioneered information-theoretic approaches to reliable large-scale computing. She then joined Harvard University as a postdoctoral fellow (’20–’22), shifting her focus to responsible machine learning, particularly fairness in algorithmic decision-making. Now an assistant professor at UC Santa Barbara and a visiting scholar at the Flatiron Institute, Jeong leads research at the intersection of AI and scientific discovery. She founded and directs UCSB’s REAL AI Initiative and leads the AI effort for UCSB’s NSF-funded Programmable Cloud Laboratory, developing AI agents that autonomously plan and execute scientific experiments. Her broader research seeks to make AI a competent and trustworthy scientific collaborator, combining generative and agentic AI methods with fundamental questions of reliability, safety, and responsible decision-making. Her contributions have been recognized with the NSF CAREER Award, Google Research Grant, JPMorgan Chase Faculty Research Grant, Amazon Research Award, and Hellman Fellowship.