Human group coordination in a sensorimotor task with neuron-like decision-making.

Area of Science:

• Collective intelligence
• Computational neuroscience
• Human-computer interaction

Background:

• Cooperative group formation is a fundamental biological principle across scales.
• Complex tasks often require group cooperation beyond individual capabilities.

Purpose of the Study:

• To investigate how groups of humans learn to solve a shared sensorimotor task without explicit communication.
• To compare computational models for predicting emergent group behavior and coordination strategies.

Main Methods:

• An experimental paradigm where human agents act as binary decision-makers in a cursor control task.
• Assigning preferred movement directions to agents, unknown to them, influencing population vector output.
• Analyzing performance based on learning speed, accuracy, action synchronization, and group coherence.

Main Results:

• Bayesian inference and Thompson sampling models significantly outperformed perceptron and reinforcement learning models in predicting human group behavior.
• Thompson sampling, approximating Bayes-optimal behavior, showed particular predictive power.
• Internal models appear crucial for adaptive coordination in cooperative tasks.

Conclusions:

• Human groups can learn complex, emergent control strategies through shared feedback in sensorimotor tasks.
• Computational models incorporating internal models, like Bayesian and Thompson sampling, are vital for understanding distributed information processing and adaptive coordination.
• The experimental paradigm offers insights into collective intelligence and decentralized problem-solving.