Fostering Human Agency: A Process for the Design of User-Centric XAI Systems
Maximilian Förster, Mathias Klier, Kilian Kluge, Irina Sigler
Abstract
The emerging research field of Explainable Artificial Intelligence (XAI) addresses the problem that users do not trust or blindly follow AI systems that act as black boxes. XAI research to date is often criticized for not putting the user at the center of attention. Against this background, we design a process to systematically guide the instantiation, calibration, and quality control of XAI systems such that they foster human agency and enable appropriate trust in AI systems. The process can be applied independent of the XAI method, application domain, and target user group. It incorporates the principles of user-centric design, insights into explanations from the social sciences, and established XAI evaluation scenarios. Following the Design Science methodology, we demonstrate the practical applicability of our artifact and evaluate its efficacy in a realistic setting. Our work contributes to the design of user-centric XAI systems and the quest for human agency in AI.