Digital assistants are being examined and developed to support various international nuclear safeguards inspection tasks. In prior work, this team conducted human performance experiments to assess trust in a voice user interface that supported in-field decision making for a seal examination task. The seal examination experiments presented much of the information needed to successfully complete the task visually, with simulated output from an automated tamper detection algorithm presented auditorily. For most trials, our participants in the seal examination task could ignore the voice user interface and still have reasonable performance levels on the task. In a new set of experiments presented in this paper, participants compare simulated nuclear material measurements taken from a remote detector (auditory-only) to a list of declared values (visual-only). The nuclear material measurement task differs importantly from the seal examination task in that a portion of the information required to successfully complete the task is presented solely via the voice user interface, forcing participants to rely on (and question) information that they cannot visually confirm. As in prior work, we experimentally manipulate various trust factors to determine if and how these changes impact user trust. In this paper we describe the motivation for our research, the details of our new experimental task focusing on nuclear material measurements, preliminary results, and a discussion of future research.
Year
2024
Abstract