Yates, Scheller, Layng - Premature Robot Apology

Effects of Premature Apology in Robots

Individual Research Topic by Brian Layng

Introduction

My research question deals with perception of premature apology. Specifically, my research question is: How does a human perceives a robot saying, "sorry", before throwing the ball in different rounds of competition. This is a novel concept in robotics. Premature apology could indicate precognition that its actions could cause harm, or self-humiliation. To get a feel for how people perceive apology in robots, I constructed a survey which showed images of various robots, ranging from humanoid to industrial. Each image was accompanied by one question: "How would you feel if this robot apologized to you before throwing you a ball." The survey then listed 6 options and a fill in the blank: pity, fear, empathy, humor, confusion, joy, or other. Following up on this survey, I conducted a second survey which kept the image of the robot constant and adjusted the voice. I chose an image of a PR2, and used text-to-speech audio files to vary cadence.

Related Work

A lot of previous research has been done on robot empathy. In ‘“Sorry, I Can’t Do That”: Developing Mechanisms to Appropriately Reject Directives in Human-Robot Interactions’ by Briggs et al., the authors developed a decision tree method to determine whether or not a high level task was accomplishable. This relates to my research topic because in this paper, the robot apologized after determining whether a task was doable, while I plan to apologize preemptively, regardless of the outcome. In “Gracefully Mitigating Breakdowns in Robotic Services” by Lee et al., the authors discuss expectancy setting in order to mitigate potential breakdowns. They do this by forewarning a human that a given task is difficult, so if it fails, the human is ready. I’m interested in how a person perceives apologies in situations that the robot can accomplish easily. I expect people to cast an insecure, B-Type personality on a robot that would otherwise perform similarly. Finally, The paper “Poor Thing! Would You Feel Sorry for a Simulated Robot?” by Seo et al deals with empathy in simulated robots. While not directly related, it discusses topic which may be relevant to this research question. The gap in this area of research is that there isn’t much research in setting a precedent of insecurity or dominance in a robot. A robot that apologizes, then fails a task could make it seem insecure. On the other hand, a robot that apologizes, then overperforms can appear pompous or cocky. My goal was to determine what kinds of perceptions come from premature apologies.

The Effect of Physical Appearance on Premature Robot Apology

I collected data using Google forms with a sample size of N=8. Overall across all images, fear and confusion were the most widely reported feelings to a robot apologizing before tossing them a ball (see figure below). Contrary to my hypothesis that some will perceive a premature apology as indication of humiliation (pity/empathy), those two feelings came in relatively low. Another experimental manipulation that this survey varied was the appearance of each robot. There were 5 robots shown: PR2, Pepper, my launcher, an industrial loader arm, and a toy demon robot. The PR2 mostly evoked humor and confusion. The demon toy robot understandably evoked fear. The Pepper robot claimed the highest reactions in pity and empathy, perhaps because of its child-like physical features. Our throwing robot evoked mostly confusion; comments indicated 4 that without a mouth or speakers, it wouldn’t be able to "say" sorry. Finally the Industrial robot evoked the most fear.

Effect of Vocal Tone on Premature Robot Apology

In my second survey, I asked participants to listen to 5 different audio files while looking at an image of a robot. These audio files consisted of five different voices: two male, two female, and unisex child. Participants were given the options of describing the voice clip as intimidating, joyful, bored, sad, funny, pitiful, or other accompanied by the image of a PR2 (shown to the right). The results shown in the figure below showed that voice cadence has an effect on perception. Male voices were more often perceived as bored or intimidating while feminine voices were more often evoked pity and joy from participants. The childlike voice predictably the most pitiful. This analysis looked into the vocal cadence of robots. These results are not mutually exclusive and do not contradict each other, but rather build a topic together. The first data set found having innocent and soft features evoke pity, while industrial looking robots make people more uneasy. Vocal patterns, being much easier to manipulate between cooperative and adversarial roles could be a much easier to implement aspect in robotics. Using a childlike voice in cooperative rounds could help the human participant feel at ease and relaxed to improve perform

Discussion, Conclusion and Future Work

My results found that robot image and voice have an influence on how a premature robot apology is perceived. I was unable to run my intended experiment of actually implementing apologies on robots, but the survey I distributed captured much of the same information. If I was going to extend on this topic, I would like to apply intimidating voices to some of the more innocent looking bots to see which has more of an effect: voice or appearance. I hypothesize that whichever medium is taking the intimidating role would dictate the emotion. For example, a Pepper robot with a menacing voice will be seen as intimidating. Otherwise, the juxtaposition of intimidation and innocence may just be perceived as silly.

Works Cited

Seo, Stela H., et al. “Poor Thing! Would You Feel Sorry for a Simulated Robot?” Proceedings of the Tenth Annual ACM/IEEE International Conference on Human-Robot Interaction - HRI ’15.

Lee, Min Kyung, et al. “Gracefully Mitigating Breakdowns in Robotic Services.” 2010 5th ACM/IEEE International Conference on Human-Robot Interaction (HRI), 2010, pp. 203–210.

Briggs, G., and M. Scheutz. “‘Sorry, I Can’t Do That’: Developing Mechanisms to Appropriately Reject Directives in Human-Robot Interactions.” 2015 AAAI Fall Symposium Series, no. 1996, 2015, pp. 32–36.

Google Sites

Report abuse