Human intergroup coordination in a hierarchical multi-agent sensorimotor task arises from concurrent co-optimization

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Social Psychology

Background:

• Division of labor and specialization are fundamental to biological and social systems.
• Quantitative understanding of group coordination and specialization remains a challenge.

Purpose of the Study:

• To investigate how specialized groups coordinate in a shared sensorimotor task without communication.
• To quantitatively model emergent group dynamics, learning speed, coherence, and coordination.

Main Methods:

• Experimental paradigm with two specialized human groups (sensors and actors) performing a cursor-steering task.
• Simulation of computational models (Bayesian learning, bounded rationality, hierarchical reinforcement learning).
• Analysis using mutual information and comparison with perceptual control theory.

Main Results:

• Both human participants and model-based simulations (Bayesian, bounded rationality) successfully completed the task.
• Increasing mutual information and emergent cooperation observed within groups over time.
• Model-free reinforcement learning failed to capture human behavior; model-based approaches generalized better.

Conclusions:

• Internal models and concurrent co-optimization are crucial for adaptive coordination in specialized groups.
• Model-based computational approaches provide better insights into distributed information processing and coordination than model-free methods.
• Findings offer a quantitative framework for understanding specialization and emergent cooperation.