We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Artificial Intelligence

Title: Solving the Baby Intuitions Benchmark with a Hierarchically Bayesian Theory of Mind

Abstract: To facilitate the development of new models to bridge the gap between machine and human social intelligence, the recently proposed Baby Intuitions Benchmark (arXiv:2102.11938) provides a suite of tasks designed to evaluate commonsense reasoning about agents' goals and actions that even young infants exhibit. Here we present a principled Bayesian solution to this benchmark, based on a hierarchically Bayesian Theory of Mind (HBToM). By including hierarchical priors on agent goals and dispositions, inference over our HBToM model enables few-shot learning of the efficiency and preferences of an agent, which can then be used in commonsense plausibility judgements about subsequent agent behavior. This approach achieves near-perfect accuracy on most benchmark tasks, outperforming deep learning and imitation learning baselines while producing interpretable human-like inferences, demonstrating the advantages of structured Bayesian models of human social cognition.
Comments: 6 pages, 2 figures. Presented at the Robotics: Science and Systems 2022 Workshop on Social Intelligence in Humans and Robots
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2208.02914 [cs.AI]
  (or arXiv:2208.02914v1 [cs.AI] for this version)

Submission history

From: Tan Zhi-Xuan [view email]
[v1] Thu, 4 Aug 2022 22:27:11 GMT (135kb,D)

Link back to: arXiv, form interface, contact.