Neurocomputational mechanism of controllability inference under a multi-agent setting

Area of Science:

• Cognitive Neuroscience
• Social Psychology
• Computational Psychiatry

Background:

• Controllability perception is crucial for motivated behavior and emotional regulation.
• Estimating personal influence is complex in multi-agent environments, requiring exclusion of others' impact.
• Previous models focused on single-agent controllability, lacking mechanisms for social contexts.

Purpose of the Study:

• To investigate the computational and neural mechanisms of controllability inference in multi-agent settings.
• To model how individuals integrate self and other action-outcome contingencies.
• To explore the neural correlates and biases in social controllability estimation.

Main Methods:

• Developed a novel multi-agent Bayesian controllability inference model.
• Utilized model-based functional Magnetic Resonance Imaging (fMRI) analyses.
• Examined behavioral adaptation and self-bias in controllability inference.

Main Results:

• Controllability inference integrates self and other action-outcome contingencies via Bayesian inference.
• Multi-agent controllability recruits the temporoparietal junction (TPJ) and striatum, extending single-agent roles.
• Inferred controllability information modulates motivated behavior in the ventromedial prefrontal cortex (vmPFC).
• An innate self-positive bias in controllability inference aids adaptation; deviations correlate with guilt and may relate to depression.

Conclusions:

• The TPJ is essential for integrating self- and other-related information in social controllability.
• The findings generalize the roles of the striatum and vmPFC to multi-agent controllability.
• Social dynamics significantly shape our sense of control, with implications for social learned helplessness and mood disorders.