Computer Agent’s Advice: Effects of Embodiment and Labelling on People’s Responses to the Agent’s Advice

• In Event: ICA Plenary Interactive Paper/Poster Session I
  In Interactive Paper Session: Communication and Technology Interactive Poster Session

Abstract

With rapid development of technologies, artificial intelligent (AI) agents are expected to grow continuously, including their collaboration with human partners as an attempt to complement human limitations. For a better understanding of the human-agent interaction, the current study examines how humans respond to the AI agent’s advice during a decision-making process when there is disagreement. We conducted an experiment (n = 88) with a 2 (Embodiment: Embodied vs. Disembodied) x 2 (Labeling: Expert vs. Novice) between-subjects design. The results show that embodiment has positive effects on participants’ perception of the AI agent (e.g., perceived attractiveness and evaluation of interaction), as well as advice-taking measured by cooperation intention and compliance behavior. The result of a post-hoc mediation analysis also show that expertise-labeling has an indirect effect on advice-taking via perceived intelligence of the AI agent. Theoretical and practical implications with respect to the decision-making process in human-agent interaction will be discussed.

Authors

• Younbo Jung, Nanyang Techological U

• Taezoon Park, Soongsil U

• Jihyun Kim, Kent State University

• Alvin Wong, A*Star Singapore

• Young-Nam Seo, Kyung Hee University