Cognitive Processing of Humanoid Robot Faces: Empirical Evidence and Factors Influencing Anthropomorphism

In the present work, we explored whether (and when) people cognitively process humanoid robot faces similarly to human faces, i.e., whether they cognitively anthropomorphize them. To investigate this, we conducted three experiments (N = 346) that utilized adapted versions of the face-scrambled task. This cognitive paradigm is commonly used to determine whether specific stimuli elicit configural processing, which is typical for human faces, or analytical processing, which is typical for nonhuman stimuli such as objects. Overall, our experiments revealed that robot faces may elicit configural processing (i.e., cognitive anthropomorphism), but this process is influenced by specific factors. Robot faces with high human-likeness are more strongly anthropomorphized than those with low human-likeness (Experiment 1). Additionally, this form of anthropomorphism emerges only when considering a first-order configural processing, but not when considering a second-order configural processing (Experiment 2). However, and most importantly, when the social categorization between humans and robots is lowered, the anthropomorphism of robot faces also emerges when considering a second-order configural processing (Experiment 3). Theoretical and practical implications of these findings are discussed, with a particular focus on the interaction between higher and lower cognitive processes involved in the social perception of humanoid robots.