The Imperfectly Relatable Robot: An Interdisciplinary Approach to Failures in Human–Robot Relations

Abstract

Developing social robots that behave in more humanlike ways is widely touted as the key to improving human-robot interactions. 1This includes every thing from making subtle adjustments to physical movements of robot limbs, to developing robots with facial recognition of human emotions, with accompanying adjustments in the responses made by the robots, such as verbal feedback or gaze.As social robots come to play a more prominent role in the human world, ensuring smooth interactions between humans and robots is vitally impor tant.This is particularly true in care contexts where humans may be more vulnerable. 2But what happens when robots and humans fail to relate?Usually, research to improve human-robot interactions in social robotics focuses on producing a "positive" affective experience by exploring how to increase human participants' feelings of trust or safety. 3However, interest is emerging among roboticists in the role played by robotic failures in building more complex and nuanced relationships between humans and social robots.With the advent of machine learning methods, new learning paradigms are emerging.Reinforcement learning (RL), a type of machine learning inspired by the notions of trial and error on one hand and reward and punishment on the other, is intrinsically linked to failure.This is because failures in the context of reinforcement should lead to updated policies about plans and be hav iors.In the case of robotics, the integration of RL methods is being used to determine how robots should act as rational agents.A robot who learns through reinforcement learning will fail, and that failing is an inevitable, if not necessary, part of its learning.In this chapter we ask: how does this use of failure impact human-robot interactions?Focusing on failure as a way to improve human-robot interactions represents a novel approach that calls into question human expectations of robots.Fictional repre sen ta tions of robots (still, for many nonexpert users, 7 THE IMPERFECTLY RELATABLE ROBOT: