Conformity and Majority Following in AI Agents Societies.

As Large Language Models are increasingly deployed in multi-agent settings, understanding their collective behavior becomes essential. In this talk I show how tools from complexity science, statistical physics, social psychology, and sociology can be used to understand the emergent phenomena observed in societies of AI agents. A central finding is that AI agents exhibit a strong tendency toward majority following, a mechanism that operates both when forming social ties and when updating opinions, naturally giving rise to scale-free networks and collective norm convergence. While this tendency can enable spontaneous coordination without external intervention, it also carries risks: it can amplify errors, reinforce biases, and produce collective misalignment even when individual agents are well-aligned. These signatures of conformity and social influence are not only a characteristic of small scale experiments, but can also be observed in large scale multi-agents social networks like Moltbook. Together, these results suggest that AI agent societies present similar collective dynamics as observed in human collectives, with important implications for both AI safety and our understanding of collective AI behavior.

Giordano De Marzo is a postdoctoral researcher and lecturer at the University of Konstanz, where he is part of the Social Data Science Lab. His work focuses on the modelling and the study of collective machine behavior, with a current emphasis on large language models and AI agents. Prior to his position in Konstanz, he was a doctoral student at the Enrico Fermi Research Center, the Department of Physics at Sapienza University of Rome, and the Sapienza School for Advanced Studies. During his PhD, he focused on modelling complex digital systems using numerical simulations, analytical methods, and machine learning. He obtained both his MSc in Theoretical Physics and his BSc in Physics from Sapienza University. In addition to his background in physics, he has experience in machine learning, natural language processing, and data analysis. He is a Junior Research Fellow at the Complexity Science Hub Vienna and a fellow of the Imminent Research Center and the Enrico Fermi Research Center. He has also serves as a consultant for the International Labour Office, the United Nations University, and the private sector.