New paper published in Scientific Reports: Explainable AI improves task performance

AI technology holds great promise for supporting human work, but its effectiveness is often limited by the "black-box" nature of many algorithms. These systems make predictions without revealing how they arrive at decisions, making it challenging for professionals to trust and validate AI outputs against their own expertise. We hypothesized that explainable AI, which provides transparent insights into its decision-making processes, could bridge this gap.

To test this, we introduced visual heatmaps—graphical representations highlighting relevant areas of an image—into inspection tasks performed by domain experts. Our study encompassed two large-scale experiments. The first involved factory workers at an electronics plant tasked with identifying defects in products. The second experiment included radiologists who examined chest X-rays to detect lung lesions. Participants were divided into two groups: one assisted by traditional black-box AI and the other by explainable AI featuring heatmaps. The latter group could either follow accurate AI predictions or override them when the AI was incorrect.