A call for reflective equilibrium in the methodology of the neurosciences


In one of my recent INSOSCI working papers, I argued that a functional view should take the place of an essentialist understanding of dual systems. I think this principle is foundational for approaching the brain in terms of constitutive explanations. I want to illustrate this with what I regard as one of the most important discoveries in the brain sciences in recent decades, the ‘default mode network’.

Constitutive explanations aim at explaining behavioural phenomena ‘bottom up’. At first sight, this implies reductionism. However, there is a great danger in projecting this directly on empirical research. This danger lies in approaching parts of the brain in an essentialist way. That means, parts that can be clearly identifiable anatomically may be implicitly treated as parts in the explanation, too, just because they are anatomically separate. Throughout the history of the brain sciences, researchers have tried to fix certain parts of the brain that are assumed to be causally responsible for generating certain phenomena. Most of these theories failed, sometimes leading to the backlash claiming that no assignments are possible at all, because the brain is a complex and comprehensive system in which everything connects with everything. Clearly, this is not a satisfying approach, as we have centuries of research collecting observations about the effects of damages of parts of the brain or other ways how causal impacts on parts cause behavioural change.

I suggest a reflective equilibrium between bottom up and top down methods, and this works via the concept of function. This idea was famously proposed by Marr in his three-level methodology of the cognitive sciences, and functionalism is one of the most influential schools of thought in the philosophy of mind. I do not want to delve into these complex matters here but make a very simple point that seems often overlooked, for example, almost universally in debates about dual systems. If we approach the brain via functions that it fulfils, there is no necessary reason why certain functions should be realized by anatomically distinct structures of the brain. One particularly dangerous misconception is that functions that might be assigned to parts for phylogenetic reasons would remain valid in later stages of brain phylogeny (as in some theories about the limbic system). If we adopt a systematic functionalist view on the brain, we should always take into considerations that parts have assumed new functions in the complex division of labour of the brain. Most importantly, functions may be realized by specific assemblages of anatomically distinct parts of the brain in a flexible way, that is, these parts may be active in the realization of different functions, hence would be multifunctional.

To make this idea empirically operational, we need to argue on a ‘meso’ level (what is called ‘modular’ by economist-philosopher Don Ross). Following Marr, we start out from functions that are defined on the level of the interaction between organism and environment. For understanding the way how the brain realizes these functions, the meso level is crucial: These are functions that are necessary for producing the behaviour and which are enabled via assemblages of parts of the brain. This has the important consequence that those assemblages may operate orthogonally way vis à vis the anatomical organization of the brain. In fact, this approach is quite usual in the brain sciences, and Marr’s own treatment of vision is exemplary: It is standard textbook approach to approach a ‘visual system’ which involves many parts of the brain. But most of these standard cases relate to behavioural phenomena which are intuitively straightforward to isolate, such as vision or language.

The discovery and functional role of the default mode network is a case in point for more complex functions, which do not immediately suggest the search for underlying systems. In the late 1990s, neuroscientists discovered a curious fact about the brain: Normally, they would have thought that the brain becomes more active when people take action. But when routinely checking the results of brain scans that coincidently included states of rest, they realized that there are patterns of connectivity in the brain that are more active when people did nothing, and which even reduce activity when people start to act. This ran against all causal intuitions scientists that neuroscientists maintained so far, especially when scanning the brain: External activity must be correlated with increased brain activity in certain parts. Researchers digged deeper and identified the default mode network. This is an assemblage in my sense: a regular pattern of interaction between parts which seems to fulfil a function, and which is orthogonal to preconceived notions of the division of labour in the brain.

The DMN is especially strange, because it is causally relevant for a fact that had been noticed for long: The brain has a very high level of energy consumption independent from the type of activity conducted, even when sleeping, and activity does not add much to this (about 5%). This is difficult to explain in evolutionary terms: Why should a pattern of brain activity survive selective pressures that wastes energy for doing nothing? Well, even the functionalist fallacy and Dr Pangloss’s thinking may be productive here as idea-generators: Because it survived selection, the energy expense must have a function. Today, researchers have identified many functions of the DMN, such as being operative in mentalizing others of reflecting upon oneself. Needless to emphasize, these are very broad and complex functions, and tie up with most fundamental phenomena such as constituting the self.

I think that this example shows the methodological power of asking for functions, both a priori and as emerging in empirical research as potential explanations. A simplistic causal framework that only works bottom-up would entirely miss a central feature of the human brain, since it would always look for changes in brain activity that directly relate with certain observable behaviour. In the case of the DMN, there is no directly observable behavioural change that we can trace back to changes of the activity levels of parts of the brain. Only following functional analysis, we can identify certain aspects of behaviour that may relate causally to the DMN. This is reflective equilibrium in neuroscience methodology.

Raichle, Marcus E., The Brain’s Default Mode Network. Annual Review of Neuroscience 38(1): 433–47.

Funded by: