The Right Kind of Human-Like: Measuring and Testing Human-Likeness Perception of Text-Based Conversational Agents

While conversational agents are designed with human-like features to facilitate interaction, the scholarly understanding of users’ perception of human-likeness as a key concept in human-machine communication remains underexplored. This paper provides a three-legged typology of human-likeness perceptions of text-based conversational agents, including the dimensions of appearance, capacity, and personality. Based on responses of 373 participants, the paper identifies measurement instruments for the different dimensions. Furthermore, it tests to what extent social cues commonly applied in customer service chatbots influence the different dimensions of human-likeness perception. Next to a set of direct measurements, factor analysis confirmed two indirect measurements for appearance-based human-likeness (i.e., more global, and more specific about human-like features), four indirect measurements of capacity-based human-likeness (i.e., capacity to feel and desire, communication capacity, capacity to choose, and capacity to think), and three subdimensions of personality-based human-likeness (i.e., competence, warmth, and morality). The paper discusses what these dimensions reveal about human-likeness attributions in human-machine communication. It concludes that for all three dimensions, direct and indirect measures should complement each other, as they tap into ontological judgments of humanness in relation to different relational attributions.