How Do People Ascribe Humanness to Chatbots? An Analysis of Real-World Human-Agent Interactions and a Theoretical Model of Humanness

A central problem in text-based chatbot research revolves around humanness: not only chatbots may reproduce fundamental aspects of being human, but also users may attribute humanlike features to chatbots. Humanness is paramount in human-chatbot conversations because it may profoundly impact the quality of the interaction: for example, when users expect that interactions with chatbots are similar to human conversations and these expectations are not met by the agent, they may easily become frustrated. This is particularly relevant in customer care, where efficient support is essential: here, unmet expectations about the chatbot's humanlike capabilities can undermine customer satisfaction. In this article, we qualitatively analyzed 12,477 real-world exchanges with a task-based chatbot in the customer care domain, involving a total of 1,060 conversations. We conducted a single case study to define a preliminary theoretical model of humanness for chatbot technology. The study findings point to a novel conceptualization of humanness in customer-chatbot interaction, highlighting that it is multiple, contextual, modular, and dynamic. Moreover, the theoretical model that we propose explains that the kind of humanness attributed to a chatbot depends on the context in which people expect to interact, the objectives and needs that they aim to fulfill, and the cues that the chatbot exhibits: all these factors may change as the interaction evolves over time, and such changes may further affect the users' ascriptions of humanness. Finally, we propose design implications of the model, like the need to create "context" and account for the plurality of humanness.