Piggybacking Robots: Overtrust in Human-Robot Security Dynamics

Abstract

Robots are capable of improving the human experience by performing criticial jobs, ranging from search and rescue to surgical operations to home cleaning. For robots to be successfully integrated into existing systems and institutions, humans must trust their robot collaborators. However, the overtrust of these systems is detrimental. In this thesis, we explore this relationship of overtrust between humans and robots. We examine the vulnerability of an existing physical security system to human-robot overtrust. We do so by positioning a robot around a secure-access building---a student residence---and having the robot ask passersby to assist it by providing passage. We compare the responses of people when the robot asks to exit the building to when the robot asks to enter. We then modify the robot's appearance by disguising it as a food delivery robot, an agent of the fictional start-up Robot Grub. Over 72 experiment trials on 108 participants, we find that study participants are willing to assist the unmodified robot in entering (40% admit rate) or exiting (19%) the building at comparable rates (p = 0.3962). However, we find that groups of participants are more likely to assist the unmodified robot in entering (71%) when compared with individuals (p = 0.0086). We also find that individuals are more likely to assist the Robot Grub robot in entering (76%) than the unmodified robot (p = 0.0008). Groups are as likely to assist the Robot Grub robot in entering (80%) as individuals are (p = 1). We conclude that this study demonstrates overtrust in human-robot interactions, and that the question of robot integration with secure access systems should be addressed.