Improving Visual Inspection Accuracy: Explainable AI for Human-centric Quality Control

This study empirically investigates how Explainable Artificial Intelligence (XAI) affects defect detection and balanced accuracy in visual inspection tasks. Using a laboratory experiment, 34 participants, randomly assigned to either AI or XAI conditions, completed 21 visual inspections each. Our results show XAI significantly improves both defect detection and balanced accuracy, supporting the potential of XAI in manufacturing quality control. Further, we explore task-related experience as a moderator for the relationship of AI/XAI and defect detection rate. This study advances our understanding on human-AI collaboration in quality control, highlighting the value of transparency and trust through saliency maps in AI-assisted systems.