Improving Artificial Intelligence Through User Feedback in eXplainable Artificial Intelligence (XAI) Systems - Design Recommendations for Sustaining Long-Term Human- XAI Interaction in Industry 4.0

Abstract

This thesis explores Human-XAI interaction and the vital role of user motivation in providing feedback to XAI systems in industrial settings, focusing on the design of Incremental Explanatory Training (IET) components that enable feedback. A welladjusted XAI interface is necessary for feedback provision and in turn, the feedback is used to improve the accuracy of the Artificial Intelligence. Drawing from the interdisciplinary field of XAI, the industry 4.0 context, and User-Centered Design (UCD) practices, it investigates the motivations required for sustained user engagement, identifies interface components that enhance feedback provision, and offers design recommendations for optimizing the user experience and sustaining motivation for feedback provision. This research addresses three key questions: 1) the types of motivations necessary for user feedback, 2) the interface components fostering motivation, and 3) design recommendations for effective feedback mechanisms adhering to motivation. As a result of this research, it was shown that both extrinsic and intrinsic motivations are necessary to encourage sustained feedback provision and that interface components with XAI improvement statistics and effective interaction modalities are vital for fostering motivation. Eight design recommendations were identified that offer generalized insights into important considerations for IET components to contribute to sustained motivation and user engagement for feedback provision.