Designing Gamification Concepts for Expert Explainable Artificial Intelligence Evaluation Tasks: A Problem Space Exploration

Abstract

Artificial intelligence (AI) models are often complex and require additional explanations for use in high-stakes decision-making contexts like healthcare. To this end, explainable AI (XAI) developers must evaluate their explanations with domain experts to ensure understandability. As these evaluations are tedious and repetitive, we look at gamification as a means to motivate and engage experts in XAI evaluation tasks. We explore the problem space associated with gamified expert XAI evaluation. Based on a literature review of 22 relevant studies and seven interviews with experts in XAI evaluation, we elicit knowledge about affected stakeholders, eight needs, eight goals, and seven requirements. Our results help us understand better the problems associated with expert XAI evaluation and paint a broad application potential for gamification to improve XAI expert evaluations. In doing so, we lay the foundation for the design of successful gamification concepts for expert XAI evaluation.