Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robot Explanation Identity

Amar Halilovic, Senka Krivic

TL;DR

The paper addresses how robots can justify their actions in socially engaging ways by introducing Explanation Identity, a multidisciplinary framework linking identity concepts with explainable behavior. It proposes a four-phase, context-aware model for generating explanations (What, When, How, How long) and a life-long learning mechanism to adapt to users and settings. It introduces the concept of an explanation identity card with probabilistic features to personalize explanations and maintain consistency across embodiments. The work argues that such adaptive, human-centered explanations can improve trust, engagement, and ethical alignment, paving the way for socially integrated robots.

Abstract

To bring robots into human everyday life, their capacity for social interaction must increase. One way for robots to acquire social skills is by assigning them the concept of identity. This research focuses on the concept of \textit{Explanation Identity} within the broader context of robots' roles in society, particularly their ability to interact socially and explain decisions. Explanation Identity refers to the combination of characteristics and approaches robots use to justify their actions to humans. Drawing from different technical and social disciplines, we introduce Explanation Identity as a multidisciplinary concept and discuss its importance in Human-Robot Interaction. Our theoretical framework highlights the necessity for robots to adapt their explanations to the user's context, demonstrating empathy and ethical integrity. This research emphasizes the dynamic nature of robot identity and guides the integration of explanation capabilities in social robots, aiming to improve user engagement and acceptance.

Robot Explanation Identity

TL;DR

The paper addresses how robots can justify their actions in socially engaging ways by introducing Explanation Identity, a multidisciplinary framework linking identity concepts with explainable behavior. It proposes a four-phase, context-aware model for generating explanations (What, When, How, How long) and a life-long learning mechanism to adapt to users and settings. It introduces the concept of an explanation identity card with probabilistic features to personalize explanations and maintain consistency across embodiments. The work argues that such adaptive, human-centered explanations can improve trust, engagement, and ethical alignment, paving the way for socially integrated robots.

Abstract

To bring robots into human everyday life, their capacity for social interaction must increase. One way for robots to acquire social skills is by assigning them the concept of identity. This research focuses on the concept of \textit{Explanation Identity} within the broader context of robots' roles in society, particularly their ability to interact socially and explain decisions. Explanation Identity refers to the combination of characteristics and approaches robots use to justify their actions to humans. Drawing from different technical and social disciplines, we introduce Explanation Identity as a multidisciplinary concept and discuss its importance in Human-Robot Interaction. Our theoretical framework highlights the necessity for robots to adapt their explanations to the user's context, demonstrating empathy and ethical integrity. This research emphasizes the dynamic nature of robot identity and guides the integration of explanation capabilities in social robots, aiming to improve user engagement and acceptance.
Paper Structure (5 sections, 1 figure)

This paper contains 5 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Identity in Robotics
  3. Explanations and Humans
  4. Explanations in Robotics
  5. Design of Robot Explanation Identity

Figures (1)

  • Figure 1: The robot (a TIAGo from PAL robotics pages2016tiago) has an explanation identity as a part of its personality. Its explanation identity forms an explanation ID card, defined by different explanation-important variables (features), which are estimates of a user's (explainee) preferences at an explanation time. Each variable is accompanied by the percentage (probability, value), which defines the probability of whether that explanation feature will be activated for a specific user at a specific time. The probabilities of features fluctuate over time and are dependent on the environment, context, explainees, and the robot's previous explanation identity state. One approach for adapting these probabilities is through life-long (open-ended) learning, where through interactions the robot continuously adapts its initially learned (usually with supervised learning) explanation identity based on data from sensor inputs and its current explanation identity state.