Sociomeeting: LLM-enhanced Agent-Based Models, a Methodological Discussion Based on Three Applications
Post-doctoral researcher at Università Carlo Cattaneo - LIUC, where he is the holder of the Dynamical System Design course. He studies how to apply and extend simulation computational methodologies to different kinds of complex social systems.
He is also a research fellow at HD-LAB - Healthcare Datascience Lab, where he is involved in the use of computational methods to study and support the Italian healthcare system
This open discussion will address critical challenges in agent-based modeling (ABM), particularly the issue of replicability caused by fragmented methodologies, inconsistent programming skills, and outdated tools. It proposes a collaborative solution to standardize ABM practices as AI-based approaches, especially those using large language models (LLMs), gain prominence. The presentation reviews existing ABM-LLM approaches, emphasizing explainability and the risk of misleading emergent properties. Three ongoing projects will be showcased: a sustainability game exploring LLM-driven decision-making, a market scenario with LLM agents optimizing pricing and product quality, and an LLM-based policy maker in an epidemiological context. Each project underscores the challenges of integrating LLMs, such as model intelligence, API call parallelization, and the potential of open-source models and retrieval-augmented generation systems.
Please note that this discussion will take place in the Seminar Room at 14:00h
We encourage all interested attendees to participate actively
Detalls de contacte:
Pablo Rosillo-Rodes Contact form