Beyond Words: Evaluating and Bridging Epistemic Divergence in User-Agent Interaction via Theory of Mind

Minyuan Ruan, Ziyue Wang, Kaiming Liu, Yunghwei Lai, Peng Li, Yang Liu · 2026

Read Beyond Words: Evaluating and Bridging Epistemic Divergence in User-Agent Interaction via Theory of Mind by Minyuan Ruan, Ziyue Wang, Kaiming Liu, Yunghwei Lai, Peng Li, Yang Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-13832.txt.

Description

Language-model agents increasingly handle complex, long tasks, yet effective help requires more than faithfully executing user instructions. Requests reflect users' current understanding, which may be partial, outdated, or incorrect. When that understanding diverges from the actual task state, literal instruction following can fail the user's goal. BASIS, a benchmark for Belief-Aware Solutions in Interactive Settings, tests whether agents can infer the mental states behind user requests, recognize belief-reality mismatches, and use that understanding to build solutions.

Rather than testing mental-state inference through explicit questions alone, BASIS assesses both the utility of such inferences and the quality of downstream solutions. Models often fail to account for users' mental states, and solution quality is higher when such information is elicited and available during problem solving. Studies cover prompting, interventions, training, and multi-agent collaboration under partial observability. Minyuan Ruan leads the work with five co-authors.

Catalog record

CREATORMinyuan Ruan, Ziyue Wang, Kaiming Liu, Yunghwei Lai, Peng Li, Yang Liu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-13832

Browse the full library · About the Tower of Babel