Designing a Gender-Inclusive Conversational Agent For Pair Programming: An Empirical Investigation.

Recently, research has shown that replacing a human with an agent in a pair programming context can bring similar benefits such as increased code quality, productivity, self-efficacy, and knowledge transfer as it does with a human. However, to create a gender-inclusive agent, we need to understand the communication styles between human-human and human-agent pairs. To investigate the communication styles, we conducted gender-balanced studies with human-human pairs in a remote lab setting with 18 programmers and human-agent pairs using Wizard-of-Oz methodology with 14 programmers. Our quantitative and qualitative analysis of the communication styles between the two studies showed that humans were more comfortable asking questions to an agent and interacting with it than other humans. We also found men participants showed less uncertainty and trusted agent solutions more, while women participants used more instructions and apologized less to an agent. Our research results confirm the feasibility of creating gender-inclusive conversational agents for programming.