"Reductionism requires a stratified view of nature: that there are various levels to it, the relatively lower being more fundamental and explanatory than the higher. But what would these levels be? how are they defined and what are their boundaries? Is meteorology more fundamental or less so than economics? And are these facts about the world or more to do with our explanatory practices?"This directly connects to Hierarchy Theory from the 1970s which as also related to Bertallanffy's work. This is al a question of systems and how the parts are put together. This also relates to Simon's Science of the Artificial which I still need to read which in turn is connected the Laurillard's current view of learning.
I can also see links to non-extensive entropy. Reductionism depends on everything following equi-partition but space and history make the world heterogeneous and entropy cannot possibly be extensive, it cannot be additive because of the hierarchy of levels making more permutations and combinations of new emergent objects accessible. This is the power of symmetry breaking.
This also fits with network models and the large central component and how to break networks into different levels when they are not homogeneous in their character.
Everything depends on networks and systems and hierarchy. These produce complexity but are stable to change and evolution. It is all incredibly deep with layers within layers and out current understanding is incredibly superficial because we do not make enough connections. We have become so compartmentalised in knowledge that we cannot possibly move to the next level unless we break down these boundaries. They affect how we think, how we learn and how we work.
I cannot believe that there is not a more extensive literature on finding sub-groups in data and showing if they are real or just useful explanations. This is the same argument that Mach was having with Boltzmann 120 years ago and the reason Boltzmann killed himself. Now we consider atoms real and not useful constructions of our intellect but within data and populations can we find REAL subgroups?
I think from experiment that they exist and that there are giant central components of most populations with other groups which can be divided off, but is this just the normal distribution and outliers or do we get a Paretto like distribution as the population is broken into sub-groups? From breaking sequence data into sub-groups I get interesting results but are they real lineages or are they just practically useful?