No one understands reductionism
Including me
For the first 27 years of my life, I was intellectually happy. I spoke with my fellow physicists about topics without cognitive collapse. The meaning of a wavefunction, who should be the next prime minister (answer: try everyone), why are there so many variations of bowling… life was simple, the world made sense, I never banged my head on the table.
Then I started reading health science articles and came across a word that has turned my sunny-side-up neural wiring into a scrambled mess: Reductive.
The word changes meaning every time I set my eyes upon it – it shapeshifts depending on the writer’s intention. I think most use it seriously, but the more literature I read, the more I get the impression that many simply use it as a socially acceptable way to professionally insult a colleague.
Bluntly, I do not know what the word objectively means. I have parsed dictionaries named after upper-class institutions, each vying to be the one true language compendium,1 hoping for a simple answer. I have scoured articles written by philosophers, believing their highly skilled pedantry would provide the desperate clarity my mind desired. Alas, all I managed to do was make things worse. Apparently, there are different kinds of reductionism: methodological, ontological, theoretical, epistemological, and probably more. I am stuck in a reductive labyrinth without a minotaur to put me out of my misery.2
Reductive roundabouts
The first entryway to our semantically challenged maze is a roundabout of contradictory reductive criticisms.
Say, for example, I tell you that planetary orbits can be calculated by assuming planets are point masses – we take all the complexity of a planet and simplify it into a single dot. Without hesitation, many will cry “that’s reductionist!”
But consider the alternative. To fix the reductive model, maybe I start by making the planets spherical, rather than a point. Perhaps I give the planets cores, some magma, a few mountains. Then maybe I model the atmospheres. For life-containing planets like Earth, I add buildings, roads and creatures big and small. If I continue down this road to greater and greater detail, I end up with planets modelled right down to the interactions of atoms. Which many, without hesitation, will also cry “that’s reductionist!”
We are left with nowhere to go, stuck forever circling a reductive roundabout with no exits.
Schrödinger’s stairwells
If we choose the second entryway, a narrow slit in the outer wall of the reducible warren, the passage is split into an infinite superposition of Escherian stairwells.
Before we even get to the philosophical varieties, the ordinary word “reduce” has so many different meanings, it is a wonder any of us are able to understand each other when we use it. Generally, however, a reduction requires us to specify the type intended: a quantity being decreased, a form being reached, or how objects are being compared. For example, “reducing the water” requires us to specify that we intend to decrease the amount (an alternative may be chemical reduction).
So, when I read something like “Can we really reduce the brain down to the interactions of atoms?” my brain swirls into a headache of intellectual dizziness. Reduce down what? Scale? Composition? Explanation? Complexity? Number of doughnuts? When I see the word “reduce” within psychiatric articles, the type of reduction is rarely, if ever, mentioned. Usually, all that will be specified is a vague mention of lower or higher “levels”.
Fractal Reductionisn’t
Our final entryway into reductivist puzzle hell starts with us trapped in the labyrinth’s centre. Instead of being able to find a way out, we end up at the centre of ever larger labyrinths.
When an analysis of a complex system is criticised for being reductive, the proposed remedy is usually to make it more holistic. Sometimes arguing that the analysis should be widened to include the whole environment.
The problem with this solution can be illustrated through the medium of a bad bar joke:
A physicist, biochemist, psychiatrist, World Health Organization official, Buddhist monk and the ghost of Alan Watts walk into a bar.
The physicist starts by arguing that mental illness can be explained through fundamental particles and their interactions. “That’s reductive!” the biochemist decries “Clearly we must consider biochemical reactions as a whole to explain mental illness!” The psychiatrist slams down their pint “That’s reductive! We have to consider the mind and body as a whole to treat mental illness.” Returning from a zoned-out state, the WHO official nearly chokes on their kombucha “That’s reductive! Mental illness is a global phenomenon caused by wider problems like poverty, lack of government funding, lack of community and limited mental health initiatives. We have to consider societal issues as a whole.” The Buddhist monk, free from anger, politely indicates to the WHO official that they are perhaps being reductive “Really, we are all one” he surmises “If we so wish to explain mental illness, we must consider the whole universe. Every human being, plant, rock and star.” The ghost of Alan Watts, levitating slightly while smoking his ghost pipe, chuckles “Isn’t that quite reductive? I think you will find the whole universe is all really nothing. If we search deep into our fundamental nothingness, we can begin to explain the nature of mental illness.” The physicist pipes up again “Well, actually, that is being pretty reductive. The fundamental particles that make up the universe were formed shortly after the Big Bang, which could be argued came from nothing.” The biochemist decries “That’s reductive! Clearly biochemical reactions explain…”.
In other words, those advocating for a widened holistic analysis aren’t arguing for a whole without bounds, they are arguing for a larger, or differently bounded, part.
Where Isn’t Waldo
I have been desperately trying to find an explanation, a reductive sat-nav that would be able to guide me through the labyrinth of reductionist critiques. But I remain as lost as when I first spotted the godforsaken word.
Like a Where’s Waldo book where the artist forgot to include the bespectacled main character, I feel like I am searching for a very specific meaning, when, in fact, reductionism is represented by a plethora of other characters. To be honest, I have the funny feeling everyone else is reading from the same book.
We are the ones who draw lines over the canvas of the universe. If we wish to understand anything at all, we have no choice in the matter. It is not possible to measure the entirety of existence to infinite resolution. Consequently, we split the universe into representations of interacting phenomena that we hope best agree with data from our finite resolution measurement instruments. This is a “model”. A system is the domain of the universe we have chosen to represent with a model; its boundary divides the system from the rest of the environment.3
But the representations of interacting phenomena are much more diverse than the popularly assumed Newtonian view of the universe. For instance, at the atomic scale we often don’t use spherical “objects”. We characterise the microscale through bonds, fields, distributions, wavefunctions, spins, angular momenta, potentials, densities; the list goes on. If we want to study interactions at the microscale, accounting for these characteristics makes sense.
The macroscale is a different story. It would be pointless, verging on idiotic, to model an entire brain using atomic interactions. Most pertinently, it’s astronomically computationally expensive. Extrapolating from one of the fastest proposed machine learning molecular dynamics simulations, trained to reproduce ab initio (from first principles) calculations, it would probably take something like 5 trillion years to simulate 1 millimetre cubed of water for 1 millisecond. A brain approximated as just a lump of water would take roughly half a billion times the age of our universe to simulate 1 millisecond of action.4
But say you have a lot of time on your hands and somehow lucked into finding a universe-sized computer just sitting in God’s knapsack. You would still run into the problem of measurement. Because we cannot measure to infinite resolution, every model, no matter its domain, will have assumptions, simplifications and approximations. Every model will be incomplete. If you apply a model outside of the scope it is intended for, the omitted details can become sources of error.
We could potentially end up with a result less representative of our macroscopic brain than a model which purposely averages out fundamental microscopic properties into simpler coarse-grained ones. Why add microscopic complexity onto an already complex macroscopic system, when there is a simpler, more efficient model that does a better job?
Depending on what we want to study, we adjust the model we use. Sometimes it makes sense to represent a planet with a point mass, other times we need a fine-grained gravitational field. The critiques, therefore, should be specific and focused on whether the models and system boundaries are appropriate for the phenomena we are measuring.
Building an unreductive vocabulary
Psychiatric and psychotherapeutic model critiques need a vocabulary upgrade. I think there are better words that more accurately conceptualise things reductionism is trying to describe.5
Oversimplification
I think this is what is most often meant when reductive is used. This is illustrated in the Cambridge Dictionary:
Reductive: “considering or presenting something in a simple way, especially a way that is too simple”
Oversimplification: “the action of describing or explaining something in such a simple way that it is no longer correct or true”
If you are writing a scientific paper on the causal landscape of depression and come up with a model that is three categories called Psychological, Biological & Environmental with causes listed underneath, this would be an oversimplification. Yes, the three categories technically extend over the entire causal landscape, but the list format doesn’t capture the inherent complexity of interactions between causes, and how these interactions may vary from patient to patient.
Ironically, I think the Cambridge Dictionary’s definition of oversimplification is an oversimplification with respect to scientific models. An oversimplified model is one where the simplifications go too far for the intended application of the model.
Overemphasis
This is where too much focus is put on one aspect, component, or interaction in a model to explain the entire system. So, for example, saying that all mental disorders are metabolic disorders of the brain. Metabolism is likely one part of the puzzle to varying degrees. But saying it is the main component probably stretches into overemphasis.
Omission
The model does not mention a factor/component/phenomenon that has a significant influence on the system in question. This might hypothetically be a neural pathway not mentioned that was previously shown to have a significant influence on the symptom of anhedonia in depression, for example.
Misapplication
A model is not applicable when it is used beyond the scope of its intended purpose. An extreme example would be using a model meant for train signalling to describe neural signalling in bipolar disorder.
Faulty approximation
Approximation refers to a specific method of simplification. Essentially, it is a substitution of a complex component of a model with an inexact representation. Most of the time this is done to make outputs from the model easier to calculate. An example might be substituting the shape of a vaguely spherical cell in a 2D video with an annotated circle to computationally track its movement. A problem with the approximation may be if you found out that the shape of the cell changed considerably while it moved, causing the centre of the annotated circle to move inaccurately.
Faulty assumption
An assumption is a claim or condition that the model takes as a given. Often it is a claim that an effect is negligible, constant or non-existent.6 For example, a model of genetic predisposition to depression could assume environmental influences are negligible. This assumption could be faulty because environmental conditions may influence gene regulation through epigenetic processes.
The linearity trap
I have often seen claims about psychiatric models that confuse different meanings of “linear” and “nonlinear”. Ladyman et al. (2013) make a similar complaint:
In the popular and philosophical literature on complex systems a lot of heat and very little light is liable to be generated by talk of linearity and non-linearity. For example, Klaus Mainzer claims that “[l]inear thinking and the belief that the whole is only the sum of its parts are evidently obsolete” (33, p. 1). It is not explained what is meant by linear thinking nor what non-linearity has to do with the denial of ontological reductionism. Furthermore, an obvious response to this kind of claim is to point out that it is perfectly possible to think in a linear way about systems that exhibit non-linear dynamics. Unfortunately the discussion of complexity abounds with nonsequiters involving nonlinearity.
When it comes to mathematically modelled systems, linearity and nonlinearity correspond to the relationship between inputs and outputs of the functions that describe the system. Linear functions abide by the superposition principle. Nonlinear functions do not.
The linearity trap is therefore mixing up metaphorical meanings of linear with the mathematical one. For example, a unidirectional chain of events could be metaphorically described as linear, but the relationships between them could be mathematically modelled as linear or nonlinear. Neither of the meanings unilaterally defines whether the system involving the chain of events is complex or not.
It is encouraging to hear more psychiatric researchers advocating greater use of nonlinear dynamical models. However, critiques of previous models must reference mathematical relationships, rather than metaphorical descriptions.
The end of reductionism
No one understands reductionism because the term has so many potential meanings, trying to determine the correct one for each use becomes a near impossible task. There is a crude paradoxical irony that due to the complexity underneath the word’s surface, one could suggest that any argument accusing something of being reductive, is itself reductive.
Quite a few actors in psychiatry have argued for the end of reductionism. I would tend to agree – though not in the way these actors would like. Exploration of the universe through models is here to stay. But, using "reductive" to inaccurately describe this exploration only leads to confusion. Therefore, I don’t think we should use the word at all.
Despite not being a dictionary, clearly the answer is Roget’s Thesaurus.
According to the philosophical definitions, I’m mostly stuck in the methodological reductive labyrinth. The others definitely confuse the heck out of me too, but my brain hurt too much when I tried to include them.
It isn’t possible to completely isolate a system. The boundaries will always be “leaky”. One example that comes to mind is a computer “bit-flip” caused by a cosmic ray.
Assuming a cubic millimetre of water contains 100 quintillion atoms, and assuming the extrapolated theoretical simulations took place on the 45,000 GPU (246 PFLOPS) supercomputer from the Suo et al. preprint.
I’m drawing from the vocabulary I’ve used when writing scientific literature. Keep in mind there may be a more formal list. Also, the criticisms are not mutually exclusive. Models can have multiple overlapping problems.
During my physics exams, writing “assuming no air resistance” almost became an automatic reflex. Air (well, fluid dynamics) is gloriously complicated – it is far easier to pretend air doesn’t interact with things.