Media Effects Research Lab - Research Archive
Tailoring your AI friend: Effects of customization on users’ interaction with companion chatbots
Student Researcher(s)
Jin Chen (Ph.D Candidate);
Yuan Sun (Ph.D Candidate);
Yanting Wu (Ph.D Candidate);
Faculty Supervisor
INTRODUCTION
Although there is still a long way to go before text-based conversational chatbots become real healthcare helpers, studies have showed the promising effects of chatbots as psychological companions. In the human-chatbot interaction, giving users the choice of tailoring the chatbot more similar to them is expected to reduce uncertainties in the early stages of building a relationship with companion chatbots. While some users appreciate customization, others do not. This study is thereby dedicated to investigating the effects of customization and the underlying mechanisms and contributing to the current customization literature in understanding human-chatbot interaction.
RESEARCH QUESTION / HYPOTHESES
H1: Users who customized their own companion chatbot will perceive a higher level of control over the interaction.
H2: Perceived control will mediate the effects of customization and lead to (a) better evaluations towards the chatbot, (b) enjoyment of interaction and (c) greater behavioral intentions towards the chatbot.
H3: Users who customized their own companion chatbot will perceive a higher level of sense of identity.
H4: Higher sense of identity will mediate the effects of customization and lead to (a) better evaluations towards the chatbot, (b) enjoyment of interaction and (c) greater behavioral intentions towards the chatbot.
H5: Users who customized their own companion chatbot will have a higher level of perceived homophily.
H6: Higher level of perceived homophily will mediate the effects of customization and lead to (a) better evaluations towards the chatbot, (b) enjoyment of interaction and (c) greater behavioral intentions towards the chatbot.
H7: Users who customized their own companion chatbot will have lower level of expectancy violation.
H8: Higher level of expectancy violation will mediate the effects of customization and lead to (a) better evaluations towards the chatbot, (b) enjoyment of interaction and (c) greater behavioral intentions towards the chatbot.
RQ: Does 1) power usage, 2) machine heuristic moderate the mediation of perceived customization in the relationship between self-as-source in a) chatbot evaluation, b) enjoyment of interaction and c) behavioral intention?
METHOD
A two-condition (customization vs. non-customization) between-subject online experiment was conducted via Qualtrics. A total of 167 participants were recruited and equally distributed in the two conditions: customization (N = 85) and control (N = 82, 49.1%). Participants in the experimental condition were offered three choices of customizing a companion chatbot (name, gender, personality) and a yoking strategy was employed to assign participants to the control condition. In both conditions, participants interacted with a companion chatbot before answering questions relating to their attitudes and behavioral intentions towards the chatbot.
RESULTS
The serial mediation of perceived customization and the separate mediation of perceived control, sense of identity, perceived homophily, and expectation violation were found statistically significant in the relationship of customization with the chatbot evaluation and user intention. The results lent support to all the hypotheses. When users perceived the chatbot as more customized, they perceived more control, greater sense of identity, greater similarity with the chatbot, and less expectation violation, and subsequently they gave more positive evaluation and expressed greater user intention. For the two research questions, we found significant moderated mediation effects of formative machine heuristic. When users perceived a higher level of customization, those who have greater formative machine heuristic perceived greater similarities with the chatbot and greater sense of identity compared to those with lower formative machine heuristic. In addition, power users experienced a greater sense of identity when they perceived the chatbot as highly customized compared to non-power users. And people's expertise in technology did influence their perceptions of chatbot customization. Lastly, there was a significant main effect of gender match on social attraction. When users interacted with a chatbot of the same gender, they perceived the chatbot to be more socially attractive.
CONCLUSIONS/DISCUSSION
Theoretically, we applied the agency model of customization in the chatbot context and confirmed the two proposed psychological mechanisms, the perceived control and sense of identity. Additionally, perceived homophily and expectation violation were found as two other psychological mechanism involved in customizing a companion chatbot. Practically, our study addressed how to improve chatbot experiences by considering users’ perspectives. People’s perception of a companion chatbot could be affected by simply interacting with customizations and being the source of actions, which has valuable practical implications for chatbot designers. Future studies could examine the effects of more chatbot customization options over long-term interactions, and explore other psychological mechanisms such as social presence.
For more details regarding the study contact
Dr. S. Shyam Sundar by e-mail at sss12@psu.edu or by telephone at (814) 865-2173