XAI Design Goals and Evaluation Metrics for Space Exploration: A Survey of Human Spaceflight Domain Experts

Abstract

The National Aeronautics and Space Administration (NASA) and its international partners are beginning a new era of human exploration by sustainably returning humans to the Moon and preparing for missions to Mars. Due to the great distances these mission will travel from Earth, reliance on Earth-based support will be diminished and greater levels of autonomy will be required. Recent advances in Artificial Intelligence (AI) such as Machine Learning (ML) and robotics are beginning to make highly autonomous systems a reality. However, these systems are highly complex and obscure to both their developers and users which inhibits trust and degrades human-autonomy teaming. Explainable AI (XAI) is a new and vibrant field exploring solutions to these issues with AI. The objective of this study is to determine user goals and metrics by which to design and evaluate XAI systems for crews and remote operators of AI systems in deep space exploration. This involved a survey-based study of scientists, engineers, and operators of spaceflight systems to determine the desired XAI outcomes, i.e. answering "what to explain?" and how to evaluate a user's level of satisfaction with an explanation. Our results will inform future XAI system development for autonomy in deep space exploration.