Look at Me Now: Investigating Delayed Disengagement for Ambiguous Human-Robot Stimuli

Abstract

Human-like appearance has been shown to positively affect perception of and attitudes towards robotic agents. In particular, the more human-like robots look, the more participants are willing to ascribe human-like states to them (i.e., having a mind, emotions, agency). The positive effect of human-likeness on agent ratings, however, does not translate to better performance in human-robot interaction (HRI). Performance first increases as human-likeness increases, then drops dramatically as soon as human-likeness reaches around 70 % to finally reach its maximum at 100 % humanness. The goal of the current paper is to investigate whether attentional mechanisms, in particular delayed disengagement, are responsible for the drop in performance for very human-like, but not perfectly human agents. The idea is that robots with a high degree of human-likeness capture attention and thus make it harder to orient attention away from them towards task-relevant stimuli in the periphery resulting in bad performance. To investigate this question, faces of differing degrees of human-likeness (0 %, 30 %, 70 %, 100 %, non-social control) are presented to participants in an eye-tracking experiment and the time it takes participants to orient towards a peripheral stimulus is measured. Results show significant delayed disengagement for all stimuli, but no stronger delayed disengagement for very human-like agents, making delayed disengagement an unlikely source for the negative effect of human-like appearance on performance in HRI.