Reproduce “Theory of Minds: Understanding Behavior in Groups Through Inverse Planning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Michael Shum, Max Kleiman-Weiner, Michael L. Littman, Joshua B. Tenenbaum

This research introduces Composable Team Hierarchies (CTH), a generative model designed to understand multi-agent behavior by inferring latent group planning structures. The model demonstrates strong alignment with human judgments for inferring social relationships and predicting actions in complex three-player scenarios, notably outperforming simpler Level-K models in action prediction.

Key insights

  • Human social intelligence infers abstract, compositional social relationships within groups, rather than just predicting low-level actions, to understand and predict multi-agent behavior.
  • A unified framework capable of representing both cooperative (joint planning) and competitive (best response) behaviors hierarchically is essential for computational models of human social reasoning in groups.
  • Purely individualistic or iteratively competitive strategic models (like Level-K planning) are insufficient to fully explain human social understanding in scenarios involving genuine cooperation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.