EXD.4:11 - SoTA-Echoing

The current Social Explainable AI chapters on Practices, §5.3, Incremental Communication, §§12.3.1–12.3.3 and Adaptation, §§13.2–13.3 connect explaining to situated, jointly adjustable contributions. Adapt that line into the contribution–diagnosis–repair decisions above, keeping learning claims at their studied scope.

Fisher et al., 2022 analyzed eleven natural consultations in which phases alternated and both sides contributed to organization. Use that as a bounded counterexample to a necessary fixed phase order. It does not select one universal conversational schedule.

Axelsson, Buschmeier and Skantze, 2022, §2 distinguish feedback concerning contact, perception, understanding and attitude. Adopt that distinction to avoid choosing a repair from an acknowledgement alone; a signal’s interpretation remains contextual.

Fichtel et al., 2025, §§5–6 found that enhanced prompting elicited more co-constructive behavior in a bounded Llama 3.1 70B study without a significant mean objective-understanding advantage. Keep that limit when designing human–AI explanatory interaction. More interaction is not an effectiveness measure.

For repairing a misrepresented concern, adapt the explicit acknowledgement and engagement examined by Yeomans et al., 2020, Study 4 and §6.2. Their written-message study concerns perceived receptiveness, persuasiveness and intentions to collaborate; it does not establish subject correctness, understanding or actual later cooperation. The contribution used here is making the addressed concern inspectable before adding a rebuttal. Against applying a linguistic recipe, keep actual agreement and uncertainty truthful and retain the subject criterion. A polite response can still answer the wrong question. Reopen this repair when the response shows that the concern remains misrepresented or the exchange no longer serves an explanatory purpose.

At the same local opportunity to respond, repeating the pooled-mean derivation answers the old question. Using the recipient’s explicit change of weighting rule supplies the requested branch-mean answer and preserves the earlier result under its original condition. This comparison explains the proposed move in the worked case. Reopen it when actual recurring interactions expose a different diagnostic or repair need.