Source record

Published by arXiv cs.HC (T1 source). The original is at https://arxiv.org/abs/2608.06381.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

A between-subjects user study (n=38) in the Overcooked-AI benchmark evaluates text versus audio explanations generated from a hierarchical reinforcement learning policy. No significant performance gains were found, but audio explanations significantly reduced participants' working-alliance bond with the agent, suggesting spoken explanations activate partnership expectations a reactive policy cannot meet.

First real-time modality comparison in human-agent collaboration flags a counterintuitive design risk: audio explanations can erode trust when the agent's behavior cannot sustain the partnership its voice implies.

Cited by

No citations on record.