Detecting Distrust Towards a Virtual Assistant from the User's Speech
Abstract
In the context of human-machine interaction, it has been shown that the level of trust the user has toward the machine greatly affects the quality of their interaction. In particular, a user that distrusts the machine will tend to underuse it. An automatic prediction of the level of trust that a user has towards a certain system could be employed to attempt to correct potential distrust by having the system take relevant actions like, for example, explaining its decisions more thoroughly or apologizing for their mistakes. In our recent work, we explored the feasibility of automatically detecting the level of trust that a user has on a virtual assistant (VA) based on their speech. For this purpose, we collected a dataset containing human-computer speech interactions using a novel protocol. Under this protocol, subjects were asked to answer various factual questions with the help of a virtual assistant, which they were led to believe was either very reliable or unreliable. We found that the subject’s speech can be used to detect which type of VA they were using, which could be considered a proxy for the user’s trust toward the VA’s abilities, with an accuracy up to 76%, compared to a random baseline of 50%. This work indicates that the task of detecting trust toward a virtual assistant from the user's speech is feasible though very challenging. These results should be considered preliminary since the dataset was purposely collected under a very controlled environment that may not be reflective of most use cases. Further, the dataset is small which complicates the use of the most sophisticated modeling methods involving deep neural networks (DNNs) that are state of the art for many speech processing tasks. We propose to continue this line of work in two directions: (1) collect a new dataset with less controlled and more realistic conditions and including a new dimension of analysis related to teaming, and (2) explore novel DNN-based modeling methods, including transfer learning techniques that could leverage the use of larger datasets for related tasks to obtain more robust models.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 21, 2022
- Source ID
- FA95502110445XX0
Entities
People
- Luciana Ferrer
Organizations
- Air Force Office of Scientific Research
- United States Air Force