Résumé
Artificial intelligence is most often used for problems where the data is digital from the outset (internet audience, Netflix, etc.) and where the stakes associated with errors are low. However, the opposite is true for hydrosystems: natural data is fraught with uncertainty and is imperfectly digitised. The systems to be described are complex (non-linear, non-stationary) and associated with high stakes (floods, low water levels). In this context, what precautions should be taken to use AI methods effectively?