Gaze following occurs automatically in social interactions, but the degree to which we follow gaze strongly depends on whether an agent is believed to have a mind and is therefore socially relevant for the interaction. The current paper investigates whether the social relevance of a robot can be manipulated via its physical appearance and whether there is a linear relationship between appearance and gaze following in a counter-predictive gaze cueing paradigm (i.e., target appears with a high likelihood opposite of the gazed-at location). Results show that while robots are capable of inducing gaze following, the degree to which gaze is passively followed does not linearly decrease with physical human-likeness. Rather, the relationship between appearance and gaze following is best described by an inverted u-shaped pattern, with automatic cueing effects (i.e., attending to the cued location) for agents of mixed human-likeness and reversed cueing effects (i.e., attending to the predicted location) for agents of either full human-likeness (100% human) or full robot-likeness (100% robot). The results are interpreted with regard to cognitive resource theory and design implications are discussed.
