ISTI-TALK: Modeling complex social phenomena with LLM agents

Day - Time: 08 April 2026, h.12:00

Place: Area della Ricerca CNR di Pisa - Room: C-29

Speakers

• Erica Cau

Referent

Giulio Del Corso

Abstract

Opinion dynamics algorithms can simulate complex social phenomena, ranging from polarization to consensus formation, through agents with simple rules of interaction and opinion update. Yet, these approaches carry limitations inherent to their nature: mechanistic rules do not allow for modeling the intricacies of human language and its persuasiveness, while treating opinions as numbers flattens the richness of natural language.In this talk, we discuss how LLMs can be integrated into agent-based models to simulate complex phenomena and show that LLM agents converge via an asymmetric, selective persuasion process. We make a further step by introducing YSocial, an LLM-powered digital twin in which autonomous agents interact, generate content, and disseminate information in a controlled virtual environment, enabling  in vitro simulations of complex social interactions.