UserHarness: A Framework for Agent Theory-of-Mind via User-Mind Reconstruction
A new framework called UserHarness reframes Theory-of-Mind (ToM) reasoning for AI agents by explicitly reconstructing the user's mental state. Proposed in a paper on arXiv (2605.27721), UserHarness decomposes the user's beliefs, intentions, and their relation to the environment and actions. Unlike complex pipelines that model behavior indirectly, this approach directly models how users act based on beliefs updated through observations, and how beliefs and intentions jointly determine actions that change the environment. The framework also handles nested social reasoning about others' beliefs or intentions. UserHarness aims to build more effective agent assistants by addressing the core structure of ToM problems.
Key facts
- UserHarness reframes Theory-of-Mind reasoning as explicit user-mind reconstruction.
- The framework decomposes the user's mental state, its relation to the environment, and resulting actions.
- It addresses the core structure: users act based on beliefs updated through observations.
- Beliefs and intentions jointly determine actions, which change the environment.
- Social reasoning often requires nested beliefs about others' beliefs or intentions.
- Existing approaches use complex pipelines that model behavior indirectly.
- UserHarness is proposed in arXiv paper 2605.27721.
- The goal is to build more effective agent assistants.
Entities
Institutions
- arXiv