Leveraging Personality to Design Expression for AI Based Embodied Agents

Abstract

Conversational agents empowered by artificial intelligence, for instance Siri, Alexa, and more, are widely adopted and used nowadays. Users tend to perceive certain characters or personalities to those agents, interact with them. These situations lead us to following research questions; Can we leverage personality as a medium to design agents? How can we use it for designing dialogues and other modalities in early-stage of design? We conduct a remote design-sessions with designers who experienced designing dialogues for CA with two different types of pre-defined personality. (N = 16, total 268 sentences) They are asked to write scripts with stage directions based 26 situation units according to five functions on healthcare agents. We find major findings according to open-coding and theme-coding analysis. First, providing certain personality types leads dialogues and other modality-expressions differentiated. Second, utilizing stage directions with scripting enact designers to design not only dialogues but also other modality-expressions were effectively done even without high-fidelity prototypes such as physically-enabled devices. Several kinds of modality-expressions for instance facial expressions, voice tones, head gestures were successfully designed following two different types of personality. All in all, it is effective leveraging personality in early-stage of design process for embodied conversational agents. Moreover, simply scripting with stage directions enables designers to explore and elaborate behaviors of CA without highly-developed prototypes. It is expected to be the next step designing systematic procedures for creating personality for CA and reflecting it to various behaviors.