When interpolation goes wrong it’s predictably wrong, which is not true of more complex algorithms using AI. If we can predict how it goes wrong we can extract the meaningful information from it.