Reproduce “Modeling the Mistakes of Boundedly Rational Agents Within a Bayesian Theory of Mind” 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
Arwa Alanqary, Gloria Z. Lin, Joie Le, Tan Zhi-Xuan, Vikash K. Mansinghka, Joshua B. Tenenbaum
MIT CSAIL researchers introduced an extended Bayesian Theory of Mind model designed to infer human goals by explicitly accounting for agents' bounded rationality, including confused goals, misguided plans, and mistaken actions. This model accurately captured human judgments across experimental tasks, demonstrating that people distinguish between different levels of error when interpreting others' sub-optimal behaviors.
Key insights
- Human intuitive psychology implicitly accounts for the bounded rationality of others, distinguishing between different levels of cognitive and motor errors (goals, plans, actions) when interpreting intentions.
- A model explicitly incorporating probabilistic mechanisms for multi-level mistakes offers a more accurate and psychologically plausible explanation of human goal inferences than models assuming optimal agents or only basic action noise.
- The ability to interpret sub-optimal behavior by considering distinct error types is essential for robust social cognition and has implications for building more human-like and adaptable AI systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.