Reproduce “ProToM: Promoting Prosocial Behaviour via Theory of Mind-Informed Feedback” 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
Matteo Bortoletto, Yichao Zhou, Lance Ying, Tianmin Shu, Andreas Bulling
ProToM is an AI facilitator for settings where people pursue separate goals but can still help one another. By tracking agents’ beliefs, inferring their goals through inverse planning, and estimating the task-level value of possible interventions, it delivers sparse, explained recommendations; in simulated environments it achieved 100% success with speedups of 0.52 in mDKG and 1.17 in Overcooked, while outperforming general-purpose vision-language and reasoning models in efficiency and human-rated usefulness.
Key insights
- Prosocial coordination can be treated as a planning and inference problem even when agents do not share a common task objective; helping another agent may improve overall completion efficiency.
- Explicit belief tracking helps prevent the facilitator from assuming that agents know information they could not have observed, while inverse planning provides a structured alternative to asking a general-purpose model to infer everything at once.
- Selective communication is central: ProToM’s utility and behavioral-divergence checks allow it to intervene only when advice is both useful and likely to alter behavior.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.