Abstract
Social robots are increasingly being used in educational and tutoring contexts. They enable complex interaction mechanisms resembling that of human teachers and students. This evolution underscores the growing recognition of robots’ potential as social companions and educators but also raises some concerns when they make mistakes. This work aims to understand how robots’ social skills affect learning performance when robot tutors provide incorrect suggestions to human tutees. In particular, we wanted to test whether robots’ sociality could hide the incorrectness of their suggestions and then worsen learning performance. We performed a user study with the humanoid robot iCub, showing social and non-social behaviors and providing correct and wrong suggestions for two cognitive tasks: the Tower of Hanoi and the Fifteen Puzzle. We analyzed behavioral measures and collected participants’ performance regarding the number of moves needed to solve the tasks and completion time and observed that participants interacting with the social iCub achieved the best performance when provided with correct suggestions; on the contrary, those with incorrect suggestions registered the worst performance. However, receiving incorrect suggestions did not impact the performance of participants who interacted with the non-social robot.