Understanding the principles behind collective behavior in animals such as fish, birds, and insects is a challenge that spans biology, physics, and engineering. Traditional models describe these phenomena using “social forces” like attraction, repulsion, and alignment. However, these approaches abstract away the cognition of individual agents into hard-coded interaction rules, which lack the behavioral flexibility and context-sensitivity seen in nature. In this talk I'll introduce an active inference framework for modelling collective behavior, which casts a self-propelled particle as a probabilistic decision-maker trying maximize evidence for an internal model of its world. This approach generates naturalistic patterns of cohesion, milling, and directed motion without predefining specific behavioral rules. Our findings highlight how individual beliefs about uncertainty shape collective dynamics, and show that adapting these beliefs in real-time enables groups to more robustly encode information and respond to perturbations. This model provides new insights into how cognition can shape emergent group behavior and offers a versatile framework for studying multi-scale intelligences in animal collectives and engineered systems.
Back to Workshop IV: Modeling Multi-Scale Collective Intelligences
Copyright. All Rights Reserved.
