Humor intelligence for virtual agents

Abstract

Humor is pervasive in human social relationships and one of the most common ways to induce positive affect in others. Research studies have shown that innocent humor increases likeability, boosts trust, reduces tension, encourages creativity and improves teamwork. In this paper, we present a study focusing on deploying humor in interaction with a virtual agent. 25 participants evaluated the logs of conversations exchanged between a human user and two virtual agents acting as tour guides. Even though answers were equal in terms of content delivered, one agent used humorous statements to respond the queries while the other agent presented the content in a neutral way. To create answers with a humorous effect we combined information extracted from various websites focusing on tourist fun facts, pun and jokes collections. Results showed that our manipulation was successful, i.e. the humorous agent was indeed perceived as being significantly funnier. Additionally, the agent was perceived as delivering more interesting answers as compared with its counterpart. Further, participants showed statistically significant preferences towards the humorous agent when asked to choose between the agents. As such, we believe that using humor in interaction with virtual agents increases the agent likeability and possibly contributes towards a better user experience.