I am guessing our approaches are so different, so we might not reach an agreement.. but just to comment.
Sunil said:
Usually this is unproblematic. The observer is simply big and complex enough.
If you by usually, refers to subatomic interactions, from the perspective of a classical lab, then I agree, except for subtle questions of fine tuning that arise during unification.
But generally, I think this "usually" is not good enough for many open questions, and then I think about unification of GR+QM, comsological models and unification.
Sunil said:
And, moreover, it is not even necessary except as an emergent object in statistical theories.
I think the necessity only relates to the desire for increased explanatory value, reduce the number of free parameters and get rid of fine tuning (which is an ugly trait).
Sunil said:
No need for having observers described by the theory, or having a developed psychology or so.
Sunil said:
epistemology is a secondary, non-fundamental human problem.
If you start associating these terms to their meaning for humans, I think we are missing the idea. In a way - everything we talk about is "human problems", but that is missing the point. This is just like people who object to CI and think the human observers are required. (Of course human observers are required in a very superficial sense to write down equations etc, but I think we understand that there is another layer, let's not confuse ourselves).
Sunil said:
The actual standard position is much simpler. So you need a quite serious justification to reject it.
The standard paradigm to me seems improper for an intrinsic inference approach. That is serious enough for me at least. (So the question why I think the optimal approach is the one of intrinsic inference? that is a separate subquestion we need not bring up here, and relates to the philosophy of science and evolutionary learning.)
/Fredrik