Deciding When to Align: Computational and Neural Mechanisms of Goal-Directed Social Alignment

Area of Science:

• Cognitive Neuroscience
• Social Psychology
• Computational Neuroscience

Background:

• Human behavior exhibits a pervasive drive for social alignment across various domains, including motor synchrony, emotional contagion, and belief convergence.
• Current explanations focus on predictive coding and observation-execution mechanisms but lack insight into the regulation of social alignment, particularly its adaptive contexts and targets.

Purpose of the Study:

• To propose a goal-directed model of social alignment integrating computational and neural analyses.
• To reframe social alignment as a context-sensitive decision-making process rather than a reflexive tendency.
• To elucidate the regulatory mechanisms governing adaptive social alignment.

Main Methods:

• Computational modeling formalizing alignment as prediction-error minimization, incorporating a meta-learning layer for adaptive learning rates based on alignment value.
• Neural modeling identifying the mentalizing network (dmPFC, TPJ) and executive control network (dlPFC, rIFG) in regulating alignment via top-down control over mirror circuitry.
• Integration of self and other trait/mental state assessments as key inputs for regulatory functions.

Main Results:

• The model specifies computations and neural circuits for context-sensitive social alignment decisions.
• It highlights the role of higher-order representations within the mentalizing network exerting top-down control.
• The framework offers a unified understanding of adaptive social alignment (e.g., social learning) and maladaptive outcomes (e.g., groupthink).

Conclusions:

• Social alignment is a regulated, goal-directed decision process, not merely a reflexive tendency.
• The proposed model provides a computational and neural basis for understanding when and with whom individuals align.
• This framework yields testable predictions and unifies diverse aspects of social alignment across multiple domains.