I imagine that's one of the benefits of ML as opposed to a program that embodies crystallized human knowledge and skill: when things change, you don't have to get a bunch of humans to rethink the implications and rework the program, you just do more training.
