Robot Social Identity Performance Facilitates Contextually-Driven Trust Calibration and Accurate Human Assessments of Robot Capabilities

Maria P. Stull, Clare Lohrmann, and Bradley Hayes

People struggle to form accurate expectations of robots because we typically associate behavior (and capability) with the physical entity even when there are clear indicators of different software programs dictating behavior at different times. This is a harmful prior, as commercially available, visibly similar robots do not necessarily share any common ground in terms of capability, safety, or behavior. Prior efforts to calibrate people’s expectations of robots have not extended to anchoring on the robot’s control software rather than its embodiment. In this work, we leverage social participation and flexible identity presentation to facilitate coworkers’ associations of robot capability with the currently running software rather than physical entity itself. By linking each of a robot’s controllers to a social identity, we enable collaborators to more easily differentiate between them. In a human subjects study (n = 30), participants who experienced our social identity signal understood differences between the robot’s two controllers and prevented an unreliable controller from harming perceptions of the robot’s other controller.

The full paper can be accessed here and from our Publications tab.