Trust and Distrust in Conversational AI Agents: The Effects of Agent Interaction Style and User Information Need
Date
2022-08-15Type of Degree
PhD DissertationDepartment
Consumer and Design Sciences
Restriction Status
EMBARGOEDRestriction Type
Auburn University UsersDate Available
08-15-2027Metadata
Show full item recordAbstract
Conversational agents (CAs) with artificial intelligence (AI) are computer programs with the ability to interpret and respond to users’ natural language and communicate with the users as humans would do with another human. CAs have been increasingly adopted to serve different roles, such as customer service agents, sales and marketing agents, enterprise assistants, and a data collection engine. With the growth of the CA market and the expanded roles that CAs play in consumption decisions comes the need to understand what impacts consumer trust or distrust in a CA. Addressing this need, this research empirically examined how CA interaction style and user information need impact consumer trust and distrust in a CA, respectively, and the mechanism by which these impacts occur, based on trust-distrust theory, social response theory, construal level theory, and cognitive load theory. The research employed a 2 (CA Interaction Style: social-oriented vs task-oriented) × 2 (User Information Need: specific vs general) between-subjects experimental design. A U.S. national sample of 206 consumers participated in an online experiment. Results indicate that trust and distrust are related but separate constructs that must be studied together. Also, consumers perceive more social support from CAs that use a social-oriented interaction style compared to the task-oriented interaction style, and the perceived social support in turn positively influences consumers' trust in the CA and mediates the effect of CA interaction style on trust in the CA such that the higher the perceived social support, the higher the trust in the CA. On the other hand, the findings of this study show that user information need does not influence consumers’ cognitive vigilance toward the CA. Further, user information need has neither direct nor indirect effects on distrust in CA. It was also found that trust positively influences perceived quality of the interaction with the CA while distrust negatively influences perceived interaction quality. The findings of this research offer guidelines on the effective use of CA interaction style in CA design and serve as s foundation for refining the social response theory, construal level theory, and cognitive load theory. The managerial and theoretical implications are discussed.