Abstract (englisch)

This article examines the application of artificial intelligence to one of the controversial attributions in Early Netherlandish painting: the two fifteenth-century panels depicting Saint Francis Receiving the Stigmata, traditionally associated with Jan van Eyck and today in museums in Turin and Philadelphia. Combining art historical inquiry, conservation history, and computational image analysis, the study investigates how AI authenticity assessment can contribute to longstanding questions concerning authorship. In the case of the Turin painting, the AI analysis initially produced a negative likelihood of Van Eyck’s authorship when applied to the painting in its current visible state. However, when areas affected by restoration were digitally excluded from the image, the authenticity probability changed from strongly negative to positive. By contrast, the Philadelphia panel consistently received low authenticity probabilities under all testing conditions, with minimal variation between the analysis of the full visible image and the digitally adjusted version excluding restored areas. However, since the Philadelphia version is painted on parchment that was mounted on oak, these computational results are most likely not the last word in the debate. Our findings highlight the importance of dataset construction, restoration and conservation history, and workshop complexity in the development and interpretation of AI authentication models, suggesting that artificial intelligence may be useful for approximating complex attributional problems when understood not as a source of definitive verdicts, but as an additional interpretative instrument within art historical research.