Designing a Smart Speaker for Emergent Users: Human Plus AI Response

Abstract

This paper reports on the development of a smart speaker for the home setting of 'emergent' users - those whose technology experience and resource availability are low. Earlier research has shown that AI (Artificial Intelligence) powered smart speakers struggled in recognising many emergent users requests. On the other hand, smart speakers powered by human responses were more accurate but slower. In this study, we began by determining, given a choice, emergent users prefer a smart speaker enabled by a human response or an AI response, and what are their preference criteria. We found that they were not completely inclined towards either of those choices. Rather they preferred a smart speaker based on three factors: first, the language of the request, second, the length and complexity of the request, and third, the urgency of response. We developed an integrated version of the smart speaker and evaluated it with emergent user families. From our analysis, it was evident that, when combined, AI and human responses complement each other and provide an elaborate and richer response for emergent users.