every time the two systems disagree, it's an opportunity to learn something. both kinds of models can be improved with new information, done through real-world experiments