DISSECTING THE NEUROCOGNITIVE ARCHITECTURE OF JOINT ACTION

Background and aims: Social behavior depends on coordinated actions to achieve shared goals, but the cognitive and neural bases of social decision-making are still unclear. While individual decisions rely on value-based processes in cortico-striatal circuits, it is unknown whether joint decisions require specialized social mechanisms or similar value-based computations.
Methods: Pairs of mice were trained in a cooperative decision-making task in which spatially coordinated choices produced joint rewards. Social influence was systematically manipulated by varying partner identity, available sensory cues, and spatial configurations. Behavioral choices were analyzed using logistic regression to quantify the contribution of social and non-social factors. In parallel, multi-agent reinforcement learning models were implemented to capture learning dynamics and task structure underlying social coordination. Now, we are combining this behavioral and computational framework with neural recordings and chemogenetic inhibition of the medial prefrontal cortex (mPFC).
Results: Mice successfully learned to coordinate with mutualistic rewards. Swapping spatial positions while maintaining pairs' identity revealed coordination in an egocentric reference frame. Preliminary computational modeling suggests mice use separate value representations for self and other actions. Logistic regression showed that current decisions depend on recent self and other choices, with social information weighted more heavily for recent choice history.
Conclusions: This work establishes an integrated experimental and computational framework for dissecting social decision-making in mice. Findings suggest egocentric, value-based coordination mechanisms. Future work will determine whether coordination emerges from parallel versus joint state representations and will use mPFC recordings to link neural activity with computationally inferred variables during social decisions.