Does judging a robot by its cover lead to worse learning outcomes? Exploring a novel modeling intervention for robot-related fear.

Despite social robots' potential as educational tools, children have demonstrated reduced learning outcomes relative to human teachers - a phenomenon termed the "robot deficit". This study investigated whether peer video modeling could improve children's engagement with and learning from a social robot. Sixty-two typically developing children aged 3-5 years participated in a matched-pair design study, assigned to either a modeling (viewing a peer interacting with NAO robot) or control (peer playing alone) condition. Children subsequently interacted with the robot in a novel word learning task using an eBook format. Children demonstrated above-chance word learning accuracy (63%, p < .001), but learning was not enhanced in the modeling condition. Bayesian analyses suggested the absence of modeling effects rather than reflecting insufficient statistical power. Behavioral engagement (r = .292, p = .022) and story engagement (r = .326, p = .010) positively correlated with learning outcomes. Expected apprehension negatively correlated with emotional engagement (r = -.352, p = .006). While children successfully learned from the robot, and engagement with the robot positively related to learning outcomes, video modeling did not enhance these effects. Future research should target children with higher initial robot apprehension and explore alternative intervention strategies to optimize child-robot educational interactions.