Explain yourself! Effects of Explanations in Human-Robot Interaction

Abstract

Recent developments in explainable artificial intelligence promise the potential to transform human-robot interaction: Explanations of robot decisions could affect user perceptions, justify their reliability, and increase trust. However, the effects on human perceptions of robots that explain their decisions have not been studied thoroughly. To analyze the effect of explainable robots, we conduct a study in which two simulated robots play a competitive board game. While one robot explains its moves, the other robot only announces them. Providing explanations for its actions was not sufficient to change the perceived competence, intelligence, likeability or safety ratings of the robot. However, the results show that the robot that explains its moves is perceived as more lively and human-like. This study demonstrates the need for and potential of explainable human-robot interaction and the wider assessment of its effects as a novel research direction.

What’s on Your Mind, NICO?

“…Neural modules can also perform reasoning tasks without directly interacting with the robot's sensors or actuators. For example, models that learn to play a game and decide which next move is ideal [1]. Finally, neural end-to-end models directly transform raw sensory information into actions, such as end-to-end grasp learning [32].…”

“…3) is designed to represent a child-sized humanoid robot that enables research on embodied neurocognitive models. The robotic design and software components are open-source 1 . The design of this robot combines aspects of humanoid and social robotics.…”

“…In [8], users are involved in a personalized dialogue, for which symbolic information is extracted from neural modules and stored in a knowledge base, illustrating an approach for lifting sub-symbolic information to the symbolic space, where information can be expressed verbally. Finally, the effects of verbal explanations on how a robot is perceived are explicitly assessed in the study [1], where the strategy behind the current move in a competitive board game is explained. For details on the experiments, we refer to the original articles; in this paper, we highlight those aspects of the studies relevant for XHRI.…”

“…Explaining Strategies in a Competitive Board Game-Neural Network-based Reasoning. In [1], two virtual NICO robots played a competitive board game against each other. Both robots' actions were based on a neural reinforcement learning approach.…”

“…The XHRI framework covers three stages of explanations: (1) intuitive and fast communication with non-verbal cues like gestures, gaze, facial expressions, and non-speech sounds to establish rapport and signal attention or possible issues, (2) linguistic high-level explanations of the robot's intentions, actions, assumptions as part of its symbolic component, (3) detailed, in-depth explanations of neural and machine learning modules used by the robot that rely on non-symbolic mediums such as visualizations. As an example, consider the scenario shown in Fig.…”