Daniel Burnston & Philipp Haueis, Evolving Concepts of “Hierarchy” in Systems Neuroscience

Daniel Burnston (Tulane University) and Philipp Haueis (Bielefeld University) are the authors of this last post in this book symposium for the edited volume Neural Mechanisms: New Challenges in Philosophy of Neuroscience (Springer 2021).

Concepts in science change over time.  As new results are discovered and incorporated into an existing theoretical framework, the connotations of concepts are often stretched, or their meanings changed altogether.  In recent philosophy of science, these concepts have been referred to as “patchworks,” groupings of connotations and meanings whose evolution is often only implicitly recognized by scientists themselves.

Some have suggested that patchwork concepts are simply an inevitability in science (Wilson 2006), while others have argued that they serve a positive role in engendering scientific progress (Haueis 2018).  We explore the potential benefits, and potential pitfalls, of patchwork concepts via an analysis of the concept of “hierarchy” in systems neuroscience.  Hierarchy is a central organizational principle in biology.  In neuroscience, the traditional view construes hierarchies as functional arrangements of units with a particular kind of signal processing role.  However, we argue that “hierarchy” has evolved over the past twenty-five years, and now comprises multiple distinct connotations whose relationship to each other is unclear.

In a processing hierarchy, units “lower” in the hierarchy process information about simple perceptual features.  This information is fed forward into units “higher” in the hierarchy, which extract more abstract information from combinations of lower-level features.  An example is the traditional picture of the visual cortex, on which V1 cells are selective for particular wavelengths, directions of displacement and orientations of edges.  Only “higher” areas (e.g. MT and V4) integrate this information to represent entire shapes, “patterns” of motion, or categories of color.  At still higher levels (e.g. temporal cortex) cells are selective for categories of objects, such as hands or faces.

This traditional processing view of hierarchy is also applied to the overall organization of the brain.  Visual cortex is but one modal cortex, and others have their own hierarchical organization.  In “association cortex,” information from distinct modalities is combined in task-relevant ways. The frontal cortices process still more abstract mental representations, such as “rules” or “task sets”, which organize modal information to accomplish goal-directed action.  The motor cortex then “unpacks” the dictates of the frontal cortex in a hierarchy which moves from an abstract goal or intention to representations of concrete body movements.

Anyone who studied systems neuroscience in the 1990s or 2000s will probably recognize this overall picture.  However, as network perspectives have become more prominent, the notion of hierarchy has become a patchwork.  On the traditional approach, what distinguishes hierarchical levels is the degree of abstraction of the representations a brain part processes. A graph-theoretic notion of hierarchy replaces the notion of levels of abstraction with the notion of centrality.  Parts of the brain are treated as nodes in a network, and hierarchical levels are defined by how central a node is to the organization of the network.  A node that is very central will connect to a large number of other nodes, and play a central role in distributing information around the network as a whole..

So, the two notions of hierarchy define hierarchical levels in terms of representational abstraction and in terms of centrality.  These two connotations are logically distinct – what makes the current conception of ‘hierarchy’ a patchwork is that these two connotations are operating under the aegis of a single concept.  As evidence for this, we show that different theorists have taken different, and generally unargued, stances as to the relationship between the two connotations.  Some take the two connotations as mutually supporting, some them as conflicting, and some seem to leave the relationship between them open for debate. 

We argue that patchwork concepts can only fulfill a positive function in science if their assorted connotations are eventually made explicit and the relationships between them defined. For the two notions of hierarchy we’ve discussed, three distinct views of the relationship between them are possible – the substantiation view, the conflict view, and a pluralistic view.  The substantiation view argues that graph-theoretical analyses in fact help us understand the anatomical underpinnings of a traditional processing hierarchy.  The conflict view says the opposite, namely that the graph-theoretic construal of hierarchy is in fact a potential replacement for the traditional hierarchy as an organizing principle for the brain.  The pluralist view says that the two notions refer to distinct organizational principles that should be cited as explanations for distinct phenomena. 

Rather than attempting to conclude about which perspective is best, our goal is to use philosophical analysis to clarify one of the most oft-cited organizational principles in neuroscience.  And doing so is no menial task – the different positions on hierarchy lead one in the direction of either a traditional, or a radically revolutionary, view of how the brain is organized.

Wilson, Mark. 2006. Wandering significance: An essay on conceptual behavior. New York: Oxford University Press.

Check the original chapter on the publisher’s webiste: https://link.springer.com/chapter/10.1007/978-3-030-54092-0_3

Back to Top
%d bloggers like this: