At HAI ’26 in Osaka
Nao Miura of Nervroid Lab will present a poster at HAI ’26, the 14th International Conference on Human-Agent Interaction, held in Osaka from November 16 to 19, 2026.
The title is “Conversation as Committed Action: Separating Deliberation and Expression in Persistent Human-Agent Interaction.” The paper appears in the conference proceedings.
The other problem that fluency creates
In a long-running relationship between a person and an agent, it matters that the person can tell whether the agent is still considering something, has decided it, or is only choosing how to phrase it.
Suppose an agent decides internally to “answer what is known,” but at the expression stage asks, “Could you clarify?” Both are fluent sentences. The second, though, changes the act itself, not just the wording.
The proposal: commit first, then phrase
The paper proposes placing a boundary, called commitment, between deliberation and expression. What is still under consideration cannot act on the outside world until it is fixed as an act. Once it is fixed, the expression stage cannot swap it for a different one.
- Interpretation alone cannot authorize external effects.
- Expression realizes one recorded act; choosing another requires going back to selection.
- If information used in the decision is corrected, the authorization becomes invalid before the utterance is released.
What was confirmed, and what remains
In the prototype under development, this idea is implemented on selected dialogue routes. Three hand-authored deterministic traces confirmed that parts of the boundary work.
What has not been tested: guarantees across every route, invalidation after a correction (the third rule), and the benefit to the people who interact with the agent. The paper also outlines matched evaluations as future work.
If you are at the conference, please stop by the poster.
This is a conference announcement. For details, please see the published paper.