Collective behavior from surprise minimization.

Area of Science:

• Collective behavior
• Computational neuroscience
• Theoretical ecology

Background:

• Collective motion is common in nature, with models often using self-propelled particles and social forces.
• Organisms are probabilistic decision-makers, not simple particles.

Purpose of the Study:

• To introduce active inference as a novel framework for modeling collective behavior.
• To demonstrate how minimizing surprise can explain emergent group dynamics.
• To explore the relationship between individual beliefs and collective properties.

Main Methods:

• Developed a computational model based on active inference and active Bayesian inference.
• Simulated agents driven by the imperative to minimize surprise.
• Analyzed emergent group phenomena such as cohesion, milling, and directed motion.
• Explored parameter spaces to link individual beliefs to group characteristics.

Main Results:

• Collective phenomena like cohesion and directed motion emerge naturally from the active inference framework.
• Active inference unifies and generalizes classical 'social force' models.
• Individual beliefs about uncertainty influence collective decision-making accuracy.
• Agents can update their models, leading to more robust information encoding and sensitivity to fluctuations.

Conclusions:

• Active inference provides a unified cognitive framework for understanding collective behavior.
• This approach offers a powerful alternative to traditional particle-based models.
• The model highlights the role of individual belief updating in group dynamics and information processing.