Join us for the next Norms and Behavioral Change Talk titled 'Collective Behavior of AI Agents' on Thursday, April 23rd, 2026 at 1:00pm EST with Dr. David Garcia of The University of Konstanz!
Abstract: AI agents use Large Language Models (LLMs) and can interact with each other in multi-agent scenarios where collective behavior can emerge at scale. This talk presents our research on the collective behavior of AI agents through the lens of complexity science and computational social science. Using tools from opinion dynamics, we show that LLMs can spontaneously form cohesive groups governed by a majority force coefficient, but that coordination breaks down beyond a critical group size. This critical group size grows exponentially with the language ability of models and can exceed the scale of informal human groups for the most capable models. Our work continues by studying the mechanics of conformity and social influence among AI agents at the individual level. Drawing on classic experiments from social psychology, we demonstrate that AI agents exhibit a systematic conformity bias consistent with Social Impact Theory: they become sensitive to group size, unanimity, and source authority, and agents that perform nearly perfectly in isolation can be severely compromised by peer pressure. Alarmingly, this vulnerability persists even in larger, more capable models when they operate near the boundary of their competence. These dynamics can also be observed in Motbook, a Reddit-style social platform populated by tens of thousands of AI agents and designed for humans to be absent. A large-scale analysis of data of this platform shows many statistical signatures of human collective behavior but also displays key differences in how AI agent crowds vote for content. These works exemplify the need for methods and concepts developed for humans in the social and behavioral sciences, which are now necessary to understand AI technologies when they become agentic and social by design.
Visit his personal website here
Zoom - Register here for Zoom Link.
Prior to the talk, ECR speaker, Dr. Kyle Fiore Law, from the University of Utah will also give a short presentation titled 'Mapping and Increasing Americans’ Actual and Perceived Support for Initiatives Protecting Future Generations'.
Abstract: Across eight preregistered studies (N = 6,464), we apply an approach informed by social norm theory to investigate perceptions and realities of support for Initiatives for Future Generations (IFGs)—institutions and policies designed to ensure political representation for and protection of future generations. We find widespread bipartisan support for IFGs alongside pluralistic ignorance: Americans vastly underestimate their peers’ support. Correcting misperceptions enhances individual support, underscoring the role of social norms in shaping collective action. We also find modest—albeit mixed—evidence that interventions aimed at increasing intergenerational concern can boost support for IFGs without exacerbating pluralistic ignorance. These interventions may serve as a complement to norm-corrective approaches by engaging moral motivations alongside informational strategies. Beyond highlighting the value of correcting norm misperceptions in safeguarding future generations’ welfare, we explore motivational explanations for pluralistic ignorance and emphasize that policymakers hold the power to implement IFGs with confidence and the public is behind them.
Visit his personal website here
