My instrumentalist approach, repeated to people ad nauseum, is that the only assumptions one needs is external inputs and the models we build based on them in order to predict the future values of these inputs. To the next standard and obvious question “But where are those inputs coming from, if not from external reality?”, my answer is “This question is meaningless as stated.” It’s no better than the “question of origin” — “where did everything come from?” An instrumentalist is concerned with making accurate predictions (measuring future inputs), which require modeling these inputs based on the past inputs. I have been talking a lot about it on Less Wrong again lately, starting around http://lesswrong.com/lw/h3p/welcome_to_less_wrong_5th_thread_march_2013/8pvb and going deeper and deeper for several days.
Now, what I really wanted to write about is how the models stack up. And how the inputs stack up, as well. I also wanted to trace how models go all the way down to subconsciousness, intuition and insticnt, and how animals, plants and other lifeforms can be described in terms of modeling inputs. However, there is at least one unresolved issue I’m still trying to get a grip on, and it requires working through the mix of hierarchical inputs and feedback loops. Maybe another day.