Lately I’ve been reading more intellectual histories. For those unaware of the genre, it’s a type of (usually scholarly) book that hunts down the origin of the concept we use, and how they were thought about and framed across time.
Reading a few of these, a curious pattern emerged: there is usually a succession of explanation or metaphors for the concepts, not so much tied to the concept itself, but more to the prevalent and salient memes in the air at the time.
For example, Georges Vigarello identifies four successive Western explanations of tiredness (“fatigue” in the original French):
The representations of the body and their renewal similarly guide the perception of tiredness. The oldest image links the “state” of tiredness with the loss of humors. The tired body is the dried-up body. Exhaustion is the escape of substance, the collapse of density. “Simplistic” image, probably, born from Antiquity and made from an obvious inference: the valuable principles of the bodies are the liquids, the very same which escape the body through wounds, burn with fevers, and disappear with death. […]
In the Enlightenment world, it is not humors but fibers, networks, currents which give meaning to tiredness. New symptoms appear, new aspects are considered: the exhaustion from an overwhelming and unmanaged excitement, the weakness born from repeated or maintained tensions. The lack is not linked with the loss of substance anymore, but with missing stimulation.[…] Hence the research of “tonics”, of particular excitants, and not only of liquid compensations anymore, the search for strengthenings.
The image shifts again when the principle turns to energy, when organic combustion becomes work, following the mechanical model of the XIXth century. Now the loss is the one of fire, the loss of potential conceptualized as “productivity”, the feeling of lost strength, and then the added certainty of chemical waste invading the flesh and causing its suffering. Hence the search for “reconstituants”, the pursuit of energy reserves, the quest for calories, the elimination of toxins.
Tiredness today is perceived in the language of computers, focusing on internal messages, sensations, connection and disconnection. Hence the increased appeal to easing and relaxation.
(Georges Vigarello, Histoire de la fatigue, 2020, p.11-12 , translated by me)
And in his intellectual history of how we think about the brain, Matthew Cobb reveals the following stages in the conceptualization of the brain as a machine:
A key clue to explaining how we have made such amazing progress and yet have still barely scratched the surface of the astonishing organ in our heads is to be found in Steno’s suggestion that we should treat the brain as a machine. ‘Machine’ has meant very different things over the centuries, and each of those meanings has had consequences for how we view the brain. In Steno’s time the only kinds of machine that existed were based on either hydraulic power or clockwork. The insights these mahcines [sic] could provide about the structure and function of the brain soon proved limited, and no one now looks at the brain this way. With the discovery that nerves respond to electrical stimulation, in the nineteenth century the brain was seen first as some kind of telegraph network and then, following the identification of neurons and synapses, as a telephone exchange, allowing for flexible organisation and output (this metaphor is still occasionally used in research articles).
Since the 1950s our ideas have been dominated by concepts that surged into biology from computing — feedback loops, information, codes and computation. But although many of the functions we have identified in the brain generally involve some kind of computation, there are only a few fully understood examples, and some of the most brilliant and influential theoretical intuitions about how nervous systems might ‘compute’ have turned out to be wrong. Above all, as the mid-twentieth-century scientists who first drew the parallel between brain and computer soon realised, the brain is not digital. Even the simplest animal brain is not a computer like anything we have built, nor one we can yet envisage. The brain is not a computer, but it is more like a computer than it is like a clock, and by thinking about the parallels between a computer and a brain we can gain insight into what is going on inside both our heads and those of animals.
Exploring these ideas about the brain — the kinds of machine we have imagined brains to be - makes it clear that, although we are still far from fully understanding the brain, the ways in which we think about it are much richer than in the past, not'simply because of the amazing facts we have discovered, but above all because of how we interpret them.
(Matthew Cobb, The Idea of The Brain, 2020, p.3-4)
What surprised me was the seeming arbitrariness of which metaphor was used to explain the phenomenon. It felt as if they got plucked from the idea current at the time, without thought for adequacy and fitness.
My first instinct was to search for the value of these weird pseudo-mechanisms. Maybe there was something deeper there, some hidden methodological virtue?
But after some time, I came to the conclusion that the explanation was much simpler: there is a deep-rooted human compulsion for explanation, even when the explanations are ungrounded, useless, and weak.
We Need Explanations
Once you look for it, you see this tendency everywhere. Pseudo-mechanisms were postulated in cooking, medicine, astronomy. From old anthropological gods to vapors, humors, forces and energies, I’ve yet to find many examples where people didn’t jump to a pseudo-mechanism, however impotent it was at handling the phenomena.1
I’m guilty of this myself. So when articulating issues of emotional management for example, I grasped at the available mechanism of potential energy (which I was studying) to frame it.
Even aside from the lackluster explanation power of most of these pseudo-mechanisms (a topic I will explore below), this aching need for an explanation is curious, because it goes against a key methodological tenet I have underscored many times: you want first to find stable phenomenological compressions (patterns in the data) before jumping to mechanistic models, otherwise you have no ways to ground the latter.
Without crisps and manageable handles on the system under study, the quest for a useful and grounded mechanism is quixotic, unmoored to any stable foundation in the data.
Indeed, if we look at some of the most impressive cases of scientific progress throughout history, they usually involve people actively avoiding mechanistic models for a time, or at least cleanly separating them from the phenomenological compressions under study.
For example, Darwin used the black-boxing technique mentioned in a previous post to sidestep the mechanism for inheritance and variability, which was beyond the power level of the science of his day.
Darwin treats inheritance and variation as black boxes. The variability and heritability of certain traits is undeniable. Darwin was clear and explicit about this. Nonetheless, he deliberately set aside and postponed puzzles concerning their nature and structure until more auspicious times. And, when he did try to answer these issues, in later publications, his wrongheaded speculations left his evolutionary analyses unscathed.
(Marco J. Nathan, Black Boxes, 2021, p.55)
Similarly, Mendel captured the most fundamental properties of inheritance without any recourse to mechanisms or explanations, focusing exclusively on phenomenological compressions.
Our key observation is that Mendel did not provide any description of the mechanisms responsible for the transmission of “factors.” In fairness, Mendel was not seeking a mechanistic characterization and, arguably, he did not even have a sophisticated concept of mechanism in mind. In accordance with his training in physics within the Austro-Hungarian school, he was after a mathematical description of laws and their consequences, which could capture and systematize his meticulous data. Furthermore, Mendel predated many new cytological findings, many of which were accomplished in the 1870s and 1880s, years after the publication of his paper.
(Marco J. Nathan, Black Boxes, 2021, p.56)
Other examples come to mind, like the revolutionary work of Milman Parry (covered in a previous post). Through sheer autistic textual analysis of Homeric epics, Parry revealed an intensely formulaic and procedural style, which eventually led him and his student Albert Lord to an exquisitely complex model of oral composition for this and many other poetic traditions.
Similarly, the establishment of Phenomenological Thermodynamics (what we usually called Thermodynamics), one of the most wide-ranging and stable theory in physics, required the efforts of Clausius to remove the weird pseudo-mechanisms (notably the calorific fluid) from Carnot’s insane insights.2
1. When I wrote my First Memoir on the Mechanical Theory of Heat, two different views were entertained relative to the deportment of heat in the production of mechanical work. One was based on the old and widely spread notion, that heat is a peculiar substance, which may be present in greater or less quantity in a body, and thus determine the variations of temperature. […]
Upon this view is based the paper published by S. Carnot, in the year 1824, wherein machines driven by heat are subjected to a general theoretical treatment. […]
The other view above referred to is that heat is not invariable in quantity; but that when mechanical work is produced by heat, heat must be consumed, and that, on the contrary, by the expenditure of work a corresponding quantity of heat can be produced. This view stands in immediate connexion with the new theory respecting the nature of heat, according to which heat is not a substance but a motion. […]
On this account it was thought that one of two alternatives must necessarily be accepted ; either Carnot's theory must be retained and the modern view rejected, accord ing to which heat is consumed in the production of work, or, on the contrary, Carnot's theory must be rejected and the modern view adopted.
2. When at the same period I entered on the investigation of this subject, I did not hesitate to accept the view that heat must be consumed in order to produce work. Nevertheless I did not think that Carnot's theory, which had found in Clapeyron a very expert analytical expositor, required total rejection; on the contrary, it appeared to me that the theorem established by Carnot, after separating one part and properly formulising the rest, might be brought into accordance with the more modern law of the equivalence of heat and work, and thus be employed together with it for the deduction of important conclusions.
(Rudolf Clausius, The Mechanical Theory of Heat, 1867, p.267-269)
And yet, we keep searching and giving explanations, whatever our lack of grounds. If I had to venture an explanation (the compulsion strikes again!), I would say that we just struggle to keep track of and manipulate patterns of data without an underlying story. So we end up making one up, pulling it out of our memetic climate.
Note that this claim also make sense of the recurrent overcompensation in various fields, where some practitioners become allergic to any kind of model or explanation. I expect this is a reaction to both this deep-seated compulsion, and a recent history of of repeated failures of these arbitrary explanations.3
Progress Despite (And Thanks To) This Compulsion
Now, the methodological weaknesses of this compulsion to explain don’t condemn us to never make any progress in modeling and problem solving.
First, these pseudo-mechanism have one clear benefit: they focus our attention, delineating specific enough questions for us to actually investigate without feeling overwhelmed. This investigation, when done well, then yields new phenomenological compressions, which provide the soil for the growth of future, more grounded explanations.
In that way, pseudo-mechanisms act as randomizers and symmetry-breakers, where the arbitrary actually jumps us out of the paralysis analysis that a completely open field might cause.
Next, the historical record shows an improvement in the quality of explanations through time, of their “mechanisticness” or “gears-levelness”. As more and more phenomenological regularities accumulate, the first actually mechanistic and powerful explanations emerge, coming notably from the most grounded and regular fields (physics, then chemistry). This creates an improved taste for what a good mechanism looks like, leading to better models all around.
This obviously also brings issues. For example, you can see the developments of fads about what a “real” model looks like. Most fields of science have their youth period where they desperately looked for models like the ones from the physicists, irrespective of whether or not these were adapted. Even in physics, especially early on, some shapes of models ended up sacrosaints for a time.
The reasons why certain formulas and other mathematical formalisms were closer at hand than others can be traced to fundamental, dare I say metaphysical presuppositions regarding the underlying structure of matter. Chief among these was the conviction that all natural phenomena could be reduced to the action at a distance of forces of attraction and repulsion, operating along the lines connecting (microscopic) centers of force. These centers of force were conceived as the smallest particles of matter or of imponderable fluids—traditional notions that were now made rigorously mathematical. […] It also makes clear to what extent the program was stimulated and shaped by the remarkable success of celestial mechanics, embodying what John T. Merz so aptly called the “astronomical view of nature.” On this approach, quantitative methods would not only be applied to previously formalized fields such as optics but also extended to the study of such domains as heat, electricity, and magnetism. The exemplary status of celestial mechanics became apparent even in the construction of specific mathematical tools and constituted the very ideal of precision measurement, for which the phrase “astronomical accuracy” serves as watchword.
[…]
Poisson illustrates both the power of the Laplacean program and its limitations, for within that program, research outlook was shaped not only by the general goals of quantification and mathematical formalization but also by a very specific set of substantive assumptions and its precisely tuned mathematical toolbox. Such tools were suited only to very specific questions, while others were lost to view, and not necessarily because they were deemed uninteresting but simply because there was no clear way to deal with them using existing procedures. Furthermore, the emphasis on mathematical formalization and precise measurement brought with it the disparagement or exclusion of broad-based qualitative experimental research. It is thus no coincidence that the most significant studies of the voltaic pile, which led to equally original and innovative results, took place outside France. The focused quest for mathematical formalization had its price.
(Friedrich Steinle, Exploratory Experiments, 2016, p.25-26,30)
And this also breeds an adversarial tendency to present whatever you have in the trappings of accepted models, so they pass people’s filters. That’s why so many conspiracy theories and pseudo-sciences adopt pseudo-mechanisms that feel scientific, without actually bringing on the right properties for a mechanism.
Last but not least, in exceptional cases, the very compulsion to look for mechanisms can be fruitfully exploited. I know of one particularly salient example: Maxwell’s work on Electromagnetism. While he accepted the core phenomenological compression of lines of forces from Faraday, Maxwell experimented wildly with different methods and explanations, until homing in on what is still one of the most impressive theories in all of human history.4
From the beginning of his study of electromagnetism, Maxwell was committed to Faraday’s conceptual framework of lines of force. The commitment was based on Maxwell’s confidence that the concept of lines of force was the proper way to unify the phenomena in this domain. And indeed Maxwell adhered to this concept—against the dominant concept of action at a distance—but kept changing the discussion by placing it in different methodological contexts from which further consequences were inferred. Thus, in Station 1, the lines of force are imagined to be tubes; in Station 2 they are considered to be physically real, the product of some hypothetical mechanical scheme at the micro-level; in Station 3, they are set in a field; and, finally, in Station 4 they are embedded in a medium which is the seat of energy. The fundamental concept does not change in the course of Maxwell’s journey in electromagnetism, but the settings and the methodologies vary as Maxwell moved from one station to the next.
(Giora Hon and Bernard R. Goldstein, Reflections on the Practice of Physics, 2021, p.209)
Mitigating The Costs Of Pseudo-Mechanisms
Still, realizing that this compulsion exists in each and everyone of us creates new requirements for a good epistemic hygiene.
It would be best if we could simply not follow the compulsion, and stay as much as possible at the level of data patterns and phenomenological compressions, at least until we have a good handle there.
This is one way to interpret my friend Gabe’s point about predicting and interpreting people based on their behavior, not the intents we ascribe to them.5
Personally, when I try to predict the behaviour of people, I start with their past actions. As in, I look at what they have done in the past, and assume they’ll do more of that. When that is not clear enough, I move on to what they have publicly declared.
And finally, begrudgingly, when their past behaviour and public statements are not enough, I move on to analysing their psyche, and try to guess at their intents. This is painful, and the weakest part of my models.
In practice, I often see people doing the opposite. They start with psycho-analysis.
(Gabe, Intention-based reasoning, 2025)
Yet this is clearly not sufficient. We can’t just will this fundamental compulsion away — want it or not, we will find ourselves regularly coming up with explanations, with or without grounding. And at some point, this is necessary: never looking for explanation is cowardice, not wisdom.
I see two clear ways of improving our natural explanation-making tendencies:
Methodology
By developing a deeper and more nuanced understanding of different types of explanations, what they do, how they work, we can get better at coming up with good explanations, or at least valuing our explanations only insofar as they are worth it.
Contextualization
Exploring the intellectual history of concepts we use, especially if they are still somewhat ungrounded, gives us some context on our explanation biases. By default we think in the “obvious way” inherited from our various cultures, and intellectual histories help us detect water as a fish.6
Finally Clarifying Phlogiston
On a personal note, understanding this compulsion and how it fits in the messy picture of “technical progress” also clarified a long-standing confusion I had.
Since I first read Hasok Chang’s Is Water H2O?, I’ve been convinced by his argument that there was something to the old chemical concept of phlogiston, which was completely replaced and eradicated by the more compositional chemistry of Lavoisier (in what we now call The Chemical Revolution).
I even wrote a blogpost years ago trying to defend it, since both in scientific circles and rationalist ones, phlogiston is the poster child of a fake explanation.
But it never felt exactly right. Phlogiston obviously smells fake, a curiosity-stopper that just ascribe some behaviours of chemical matter to a substance that is taken out and put back through various processes, notably combustion. It’s quite obvious that the precise weighting and combination of elements advocated by Lavoisier is a better mechanism, and it is indeed the basis of an incredibly successful modern chemistry.
And yet Chang’s point still stand: by completely killing anything related to phlogiston, the Lavoisierians also threw away some of the core questions and observations that the study of phlogiston had revealed, which were not well explained by the new mechanisms.
But, while the phlogistians, on the one hand, were unaware that the burnt product differed from the original combustible otherwise than as ice differs from water, by loss of energy, Lavoisier, on the other hand, disregarded the notion of energy, and showed that the burnt product included not only the stuff of the combustible, but also the stuff of the oxygen it had absorbed in the burning. But, as well observed by Dr. Crum-Brown, we now know "that no compound contains the substances from which it was produced, but that it contains them minus something. We now know what this something is, and can give it the more appropriate name of potential energy; but there can be no doubt that this is what the chemists of the seventeenth century meant when they spoke of phlogiston.
(William Odling, The Revived Theory of Phlogiston, 1876)
Indeed, it is curious now to note that not only their new classification but even their mechanism was essentially correct. It is only in the last few years that we have realized that every process that we call reduction or oxidation is the gain or loss of an almost imponderable substance, which we do not call phlogiston but electrons.
(Gilbert Lewis, The Anatomy of Science, 1926, p.167-168)
My mistake was to try to resolve this tension between fake mechanism and real loss fully on one side, mostly by pushing a Whiggish history that phlogiston was a good mechanism from the beginning.
But now the correct approach jumps at me: the problem with the treatment of phlogiston by the Lavoisierian was not their replacement of a bad mechanism, but their throwing away of the phenomenological compressions revealed by phlogiston: things we now model as potential chemical energy, electron flow, free electrons in metals…
Where I think the Lavoisierian failed was in their totalizing desires, usually rare in the history of chemistry, that made them try to delete every mention of phlogiston from the history books like a modern damnatio memoriae.
In doing so, they fucked up one of the main source of progress, namely the accumulation of phenomenological compressions.
There is a pattern that I can’t clearly see yet, where the attachment to the explanations was inversely proportional to the economic benefits and pragmatism of the field. One gets the impression that chemists working on dyes and blacksmiths forging weapons didn’t care that much about the accuracy of their mechanisms, whereas physicist and fundamental scientists and thinkers of all sorts got obsessed with their specific explanation.
Another amazing example in physics is the invention of the potential. I know of only one great source, in French, but it convincingly argues that the potential was nothing more than a mathematical compression for the manipulation of inverse-square law forces — it’s the incredibly wide applicability and power of the trick that eventual suggested its reification as potential energy, and Lagrangian.
Of the top of my head, I’m thinking of the self-coherence checks of temperature scale by Victor Regnault described in Chang’s Inventing Temperature, the refusal of internal state in Behaviorism, and the fear of ascribing causation anywhere in many modern statistical analyses.
This is a topic I’ve been wanting to write a blogpost about for years. For the curious, I truly recommend the book from which I’m drawing, Reflections on the Practice Of Physics.
Note that the main point of this post is more a decision theoretic one: relying on intents first leads to worse predictions, susceptibility to self and external deception, and generally bad equilibrium of responsibility across society.
Cosmopolitanism in general helps here, because reading and interacting with other cultures reveals your own hidden assumptions.