A Novel Method for Trust-Sensitive Design: Applying Conjoint Analysis to Machine-Assisted Genome Annotation

Human trust is recognized as being central to integration of AI systems into organizational settings and workflows. However, trust is a highly context-dependent, state-based attitude that can be influenced by hundreds of user, technology, and environmental factors. In this paper we describe the novel application of conjoint analysis, an established behavioral science method, to quantify the relative contribution of four design features to trust in an AI application to support scientists engaged in genome annotation. We found that Transparency had the greatest relative importance to trust, followed by Confidence, Control, and Accuracy. Our results demonstrate the utility of a conjoint analysis approach for identifying and prioritizing trust-sensitive design features, offering a cost- and time-effective approach to guide development efforts.