Human-AI Interaction and User Trust: The Mediating Role of Perceived Transparency

Abstract

The current study explored the effect of the quality of HAI interaction on user trust and whether the user's perception of transparency would serve as a mediation effect. Using purposive sampling and taking into account their familiarity with digital technologies and AI-enabled applications, 180 participants were selected for the study, which was design based quantitative experimental in which they interacted with AI-based decision-support scenarios of varying interaction and explanation levels. Participants then engaged with the assigned scenario and filled out a structured questionnaire on a 5-point likert scale regarding perceived quality of human-AI interaction, perceived transparency, and user trust. All data were analyzed using regression based mediation analysis. The reliability and the validity tests were conducted for internal consistency and convergent validity, respectively, and all constructs were found to be reliable and valid. The interrelationship between the human-AI interaction, perceived transparency and user trust was found to be significant and positive with the use of correlation analysis. The regression results showed that human-AI interaction significantly positively influenced user trust as well as perceived transparency, which in turn significantly positively influenced user trust. The 5,000 resample bootstrapped confidence intervals validated the mediation of the perceived transparency on the connection between human-AI interaction and user trust, which accounted for over 50% of the total effect. The results indicate that the quality of the human-AI interaction is not the only factor affecting the user's trust, and that the comprehensibility and explainability of an AI system plays an important role in the conversion of interaction quality into trust. Theoretical and design implications and limitations are discussed as well as directions for future research.