Abstract:

In this study, we compare the users’ emotional responses induced by an AI chatbot assistant and a real person disguised as a chatbot assistant. A total of 40 participants were randomly split into the AI chatbot group and the disguised chatbot group of 20 each. After the experiments, the emotional responses of the users were measured by the PANAS scale. The results show that there is a significant difference (p<0.001) in the overall positive emotion and the negative emotion between the AI chatbot group and the real person chatbot group. The overall positive emotion induced by the real person chat group (M = 37.15, SD =7.34) is higher than the AI chatbot group (M = 24.8, SD = 5.20) while the overall negative emotion of the real person chatbot (M=18.95, SD=7.65) is lower than the AI chatbot group (M=28.2, SD=8.29). Specifically, among the 20 emotional responses, only 1 positive emotion of alert and 4 negative emotions of guilt, scared, ashamed, and afraid show no significance between the two groups. This study implies that a real people disguised chatbot better communicates emotionally than the experimental AI chatbot in a way that participants experience stronger positive emotions and weaker negative emotions. The experiment provides a benchmark to design AI chatbot assistant such that it can interact affectively and attain the same emotional level as a real assistant. Limitations of the study and future direction of the experimental research will be discussed.

Keywords:

Chatbots, Emotional responses, AI.