Human in the AI loop via xAI and Active Learning for Visual Inspection

Active learning and explainable AI together enable better human-machine collaboration: The paper demonstrates that combining active learning (which identifies the most useful data for AI to learn from) with explainable AI (which makes AI decisions understandable to humans) creates more effective partnerships in industrial visual inspection tasks.

Humans remain essential in Industry 5.0: Rather than replacing workers, the research emphasizes that the future of manufacturing involves reshaping human roles through collaboration with AI systems, where humans provide judgment and oversight while AI handles routine analysis.

Explainability helps workers understand and trust AI decisions: By making machine learning models intelligible to human inspectors, explainable AI enables workers to verify whether the AI's quality inspection decisions are reasonable and to catch potential errors—addressing a key barrier to AI adoption in safety-critical manufacturing environments.

Active learning reduces the data burden on humans: The paper indicates that active learning strategies help AI systems learn effectively by strategically selecting which data points humans should label, rather than requiring humans to annotate massive datasets indiscriminately.

The STAR project demonstrates practical implementation: The EU H2020 STAR project results show these concepts can be applied to real-world visual inspection scenarios, incorporating considerations for human digital twins and cybersecurity alongside AI performance.