Help! Is my chatbot falling into the uncanny valley? An empirical study of user experience in human-chatbot interaction

Abstract

Advances in artificial intelligence strengthen chatbots’ ability to resemble human conversational agents. For some application areas, it may be tempting not to be transparent regarding a conversational agent’s nature as chatbot or human. However, the uncanny valley theory suggests that such lack in transparency may cause uneasy feelings in the user. In this study, we combined quantitative and qualitative methods to investigate this issue. First, we used a 2 x 2 experimental research design (n = 28) to investigate effects of lack in transparency on the perceived pleasantness of the conversation in addition to perceived human likeness and affinity for the conversational agent. Second, we conducted an exploratory analysis of qualitative participant reports on these conversations. We did not find that a lack in transparency negatively affected user experience, but we identified three factors important to participants’ assessments. The findings are of theoretical and practical significance and motivate future research.