Improving Usability Perception of Error-Prone AI Speakers: Elaborated Feedback Mitigates Negative Consequences of Errors

Abstract

ABSTRACT Users of voice user interface (VUI) often encounter errors, such as when a VUI attempts to recognize a user’s voice inputs or execute tasks. Conversation is prone to errors, and in the collaborative perspective, communicators manage common ground together to handle erroneous situations. Adopting a collaborative view of conversation, we propose that a VUI can address different types of errors by providing users with feedback to aid them in developing common ground to communicate more effectively. To test this proposal, we conducted a 2 (error type: recognition vs. execution error) × 2 (feedback elaboration: present vs. absent) mixed-design experiment in which users interacted with a VUI speaker and evaluated its usability in these four modes. Participants reported greater acceptance of feedback and higher usability perception for a speaker returning execution errors than for one returning recognition errors, particularly when the speaker presented feedback articulating reasons for the errors. This finding indicates that a VUI can employ feedback explaining the causes of errors to facilitate the development of common ground and to minimize the negative consequences of errors.