Believable Robot Compliments: discerning and grounded robot compliments increase trust, intention to use, and satisfaction

We investigate the effects of three different methods for robot compliments: simple compliments, discerning to use simple compliments or not, and discerning compliments with grounded explanations. Robots in customer service positions will need to not only complete functional tasks but also be able to give subjective opinions, such as compliments. There are many factors that contribute to a compliment being effective, and the perception of the compliment is important for building positive rapport. We compare three strategies for robot compliments by building a system that evaluates the suitability of a person and their hat based on expert knowledge, can discern if it should compliment or not, and can ground and explain its opinions. In a 22-person within-participants study, we compare just simply complimenting, discerning whether to compliment or not, and whether to ground those decisions or not. We find that people trust a robot more if it uses discerning and grounded compliments. Further, customers experience higher satisfaction and intention to use the robot, and qualitatively prefer it to their experiences with human shopkeepers. Our findings highlight that more complex and nuanced abilities to provide critical feedback to a person, specifically choosing when and how to compliment someone, is important for successful HRI.