A contribution of the epistemology of models to AI
Résumé
In this short presentation of a work in progress - conducted with Christophe Denis (LIP6, Sorbonne University) - I will show how an epistemology focusing on the plurality of knowledge functions of scientific models, on their distinctive criteria, while emphasizing their reciprocal articulations and determinations, can help to clarify the current demand for explainability and interpretability in AI and machine learning.