Theory of Mind (ToM) presents a significant generalization challenge in computational modeling. This paper explores how neural networks with varying architectures and training regimes learn and represent ToM-related features. We introduce a novel method for quantifying feature representation within neural networks and apply it to a set of theoretically-grounded features designed to differentiate between hypothesized ToM strategies. We examine the relationship between feature representation and task accuracy across different model architectures and training datasets. This work provides insights into the mechanisms underlying ToM capabilities in neural networks and offers a framework for future research in computational ToM.
Keywords: theory of mind, computational modeling, social cognition
TL;DR: We examine the features represented within neural networks as they learn to solve challenging theory of mind tasks
Abstract:
Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11566
Loading