Chatbot personality preferences in Global South urban English speakers

Abstract

Recent years have seen a major shift in focus when it comes to chatbot literature, moving from only considering their technical capacities to emphasizing their social and conversational abilities, as well as personality. This shift towards the more social end of the AI research spectrum possibly occurred as a response to social interface theory, which suggests that humans respond to technologically mediated communication in the same way as they respond to interactions with natural partners (i.e. other human beings), since both interactions are social in nature. Given this, many prominent personalities in the sphere of chatbots and artificial intelligence have wondered if personality may play as important of a role in HCI as it does in social relations. To that effect, two studies were carried out in the Global South, on a largely South Asian demographic to see if personality does influence user experiences with reference to enterprise chatbots across three domains: likelihood of re-engagement, comfort levels and estimation of productivity and efficiency. Three key personality types were developed using the Five Factor Model and IBM’s Primary and Secondary Characteristics Table, labelled Transactional, Prosocial and Friendly. The first study involved having participants interact with the three chatbot personalities and then rate themacross the aforementioned domains, while the second study had participants view interactions with the different chatbot personality and provide reviews. The results strongly suggest that inserting an evident, fun personality into the chatbot can influence user experience across these domains. • Friendly chatbot personalities have the highest median ratings for likelihood of re-interaction. • Positive perception of friendly chatbot personality attributed to how anthropomorphic the interaction appears. • Preference for friendly personality constant across use-cases and demographics. • Significant difference between user ratings for transactional, prosocial and friendly personalities across domains.