Motivational Exploration of Explanations in Industrial Analytics*

Abstract

Explainable AI (XAI) provides approaches and techniques for building trust in AI models. This paper presents and explores XAI approaches focusing on user interface concepts in predictive maintenance. The underlying AI model is based on an open dataset for wind turbines. An enhanced multi-class self-conceived labeling strategy improves the model and, thus, supports the XAI approaches. Previous research in user-centered XAI shows that users do not exploit the possibilities of XAI methods and instead rely on their intuition. To counter this tendency, we present user interfaces incorporating gamification elements to enhance understanding of AI outputs. We highlight our approach via two examples, demonstrating a local and a global XAI technique respectively. A preliminary user study was conducted to assess the value added by these gamification aspects. While the findings were inconclusive, they provided an initial insight into the potential of these design elements to foster user engagement in the realm of XAI.