Thanks so much! Naive animals are untrained animals that don’t know how to cooperate

But an important part :)

Thanks Jen! Would appreciate any feedback!

Thanks Brian!!

Thanks Catherine! ❤️

Thank you!

Thank you!

Thank you Arka!

This was a truly collaborative effort and will open new avenues to probe social cognition and collective behavior. Huge thanks to the amazing team who made this cooperation😁 possible! @esschaffer.bsky.social @nuttidanuttida.bsky.social @talmo.bsky.social @nan-mssm.bsky.social (5/5)

To uncover latent strategies, we built a multi-agent inverse RL (MAIRL) model with joint value decomposition. MAIRL infers individual goal and interaction maps that mirror asymmetric neural codes and are decodable from mPFC population activity, linking behavior, circuit, and computation (4/5)

The medial prefrontal cortex (mPFC) represents both trial-by-trial leadership dynamics and enduring social roles. It also encodes a role-based, egocentric “social value map” of partner's position that adapts to task demand. Disrupting mPFC impairs cooperation and shifts self–partner weighting (3/5)

Both leaders and followers integrate partner position into their decision-making, exerting asymmetric yet reciprocal influence. Forward multi-agent RL (MARL) shows simulated agents can learn the task and also develop stable leader–follower dynamics, suggesting social roles can self-organize (2/5)