Effects of Anthropomorphic Characteristics of Chat Agents on Perceptions of Bad Advice: An Exploratory Study

Anthropomorphic form is an inherent characteristic of artificial intelligence chatbots driven by large language models and appears to have profound effects on user perceptions of chatbots. The proliferation of anthropomorphized chatbots has raised concerns about miscalibrated trust and misplaced attributions of responsibility for chatbot mistakes. An exploratory experiment examined the potential effects of anthropomorphic design in both the user interface and the linguistic output of a chatbot that provided bad advice about an everyday medical situation. In a simulated chatbot interaction, participants experienced either humanlike or machinelike language output. Additionally, the user interface either simulated typing (i.e., the typewriter effect) or presented a static text response all at once. Humanlike language increased perceived anthropomorphism and social presence but had no effect on perceived competence and trust. There were no significant effects of the typewriter effect user interface. More frequent AI usage was associated with higher chatbot trust and perceived social presence. Perceived anthropomorphism, social presence, trust, and competence were all highly intercorrelated. Interestingly, participants who were apparently unaware of the obviously bad advice tended to show considerably higher levels of chatbot perceived anthropomorphism, social presence, trust, and competence. Finally, we found some evidence that higher perceived anthropomorphism is associated with greater perceived responsibility of the chatbot itself for the advice it gave, whereas higher perceived chatbot trust and competence were associated with lesser perceived responsibility of the developing company for the chatbot’s output. Potential future research directions are discussed.