I notice that you've just framed most knowledge creation/discovery as a form of gradient descent.

Which it is, of course.