"It knows how to not understand us!" A study on what the concept robustness entails in design of conversational agents for preschool children

Abstract

This thesis explores preschool children’s interaction with a conversational agent (CA), and we position this work within the field of Human-Computer Interaction (HCI). Through a preliminary case study, we explored children as users of conversational technology and gained insights regarding how a CA could function as a pedagogic tool in a kindergarten setting. In a succeeding design process consisting of three distinct phases and four iterations, we developed and evaluated a prototype in order to gain a deeper understanding of what robustness entails in child-CA interaction. In a concluding field trial, we deployed the prototype in a kindergarten in order to study the interplay between all actants involved in use; the CA, the children, and the educators. We applied Steven Jackson’s (2014) perspective broken world thinking to guide our exploration throughout the design process. This had us focusing on how conversations are always breaking, as well as how the users conduct repair work when a communicative breakdown occurs. The findings indicate that various trouble sources complicate the communication between children and CAs. The trouble sources primarily derived from the subtle differences that make conversations between users and CAs different than ordinary conversations between people. This thesis shows how attempting to develop a prototype that addressed several of these trouble sources resulted in an increased understanding of how these should be mediated through design. At the same time, it illustrates how an exclusive focus on breakdown can devaluate inherent characteristics within the user group and thus become detrimental to the design result. We offer suggestions on how to support robust communication between children and CAs, and provide three overarching design recommendations that we believe practitioners within the field can benefit from: (1) support the user in understanding communicative constraints, (2) support the collaborative nature of conversation, and (3) understand what constitutes an effective conversation.