Differences in Working-Memory Capacity Modulate Top-down Control of Social Attention

Abstract

Gaze following, or our ability to attend to where others are looking can be top-down controlled by context information about the social relevance of the gaze signal. In particular, it has been shown that gaze signals are followed more strongly when the gazer is believed to have a mind with the ability to show intentional behavior (i.e., human) compared to being pre-programmed (i.e., robot). Perceiving human traits in nonhuman agents (i.e., anthropomorphism) occurs naturally in human-robot interaction, where it has positive effects on performance. It can also attenuate performance, if the robot is designed in a way that makes it hard to categorize as human or nonhuman (e.g., humanoid appearance), and inflicts additional working memory load due to categorical ambiguity. Here, we examine if gaze signals of ambiguous humanoid agents are followed differently than those of unambiguous human or robot agents, and to what extent gaze following is affected by individual differences in working memory capacity (WMC). We assume participants with high versus low WMC to be more capable of suppressing reflexive gaze following behaviors to the cued location in a counterpredictive paradigm (where targets appear with high likelihoods at uncued locations), particularly when being cued by humanoid gazers (top-down control, which requires cognitive resources). While the analysis showed no effect of categorical ambiguity on top-down control abilities overall, it revealed that participants with low WMC had weaker top-down control than participants with high WMC for the most ambiguous humanoid agent. The results are discussed with regard to the design of social agents and human-robot interaction.