Extrapolation and AI transparency: Why machine learning models should reveal when they make decisions beyond their training | Synapse