Using XAI in historical disciplines, it is possible to turn AI into a research partner from which useful information can be learned for conducting historical reconstructions. The presentation will initially focus on some simple examples produced by applying XAI to a dataset obtained from the Sphaera project and concerning the transformation, circulation, and homogenization of astronomical knowledge from the 15th to 17th centuries. The second part discusses the vision and method for using XAI as a historian, capable of conducting abstract inferences useful for complete historical reconstructions concerning the evolution of scientific knowledge. First, possible approaches to formalizing the reasoning of a historical reconstruction will be discussed: the logical one and the one based on network theory. A full example of the second approach will be shown. Second, validation methods as well as the application to different datasets will be discussed. Third, the procedure for improving the results through the extraction of data indicated as relevant by XAI itself and, in conclusion, the applicability and usefulness of a large-scale application in institutions such as libraries and archives will be shown.
Back to Explainable AI for the Sciences: Towards Novel Insights