It has become commonplace to say that novel predictive success is not epistemically special. Its value over accommodation, if it has any, is taken to be superficial or derivative. We argue that the value of predictive success is indeed instrumental. Nevertheless, it is a powerful instrument that provides significant epistemic assurances at many different levels. Even though these assurances are in principle dispensable, real science is rarely (if ever) in the position to confidently obtain them in other ways. So we argue for a pluralist instrumental predictivism: novel predictive success is important for inferences from data to phenomena, from phenomena to theories, and from theories to frameworks. Ignoring it would deprive science of a crucial tool.
Magnus, P.D. and Douglas, Heather, "State of the field: Why novel prediction matters" (2013). Philosophy Faculty Scholarship. 23.