The difficulties of reductionism

By Haines Brown, 13 December 2007

A typical definition of reductionism is that the nature of a complex thing reduces to the sum of the natures of its simpler or more fundamental parts. This is what most historians understand by the term.

I will suggest that such a reductionism, however necessary it may be in historiography, betrays explanation and necessarily reduces the truth value of historical knowledge. However, there is another kind of reductionism that I believe can better support historical explanation and increase its potential for truth.

I will suggest that the definition above is too narrow in that, while it may be inevitable and necessary in epistemological terms, it is too restrictive ontologically. To make this distinction clearer, I will label for former a “phenomenological reductionism” and the ontologically broader kind of reductionism that is sometimes employed in the natural sciences a “processual reductionism”. While the latter is contentious, the principle is at least well understood and successfully applied in certain domains of knowledge. I will conclude it would be of great benefit if were employed in historiography.

I will not here discuss ideal types. They are an abstraction of those features that classes of phenomena hold in common. Because there is no objective basis upon which to define those classes in the first place except in terms of features that classes of things hold in common, the reasoning is circular. Defenders of comparative history might argue otherwise, but ideal types do not seem to start from an objective basis, but from the presumptions of daily life. A reductionism here seems merely a reduction of concepts to simpler concepts, in which case it is not germane to our concerns.

Phenomenological reductionism

The conventional definition above implies a relation of parts and wholes. Rather than resurrect the now defunct philosophical debate over parts and wholes, let me instead consider phenomenological reductionism in the more practical terms of an example of our representation of an historical object, say, the modern economic history of Mongolia.

Merely stating the existence of this historical object implies a distinction between what information is relevant and what is not. We gather data relevant to the economic history of Mongolia and also such other information that for good reason we believe relevant to it. For example, a popular rock band in Ulan Bator named after a great historical hero seems quite irrelevant to Mongolian economic history, but a social process such as urbanization is clearly important for market expansion. What's happening in North China seems worth considering, but not what occurs in East Lansing, Michigan.

What is taking place here is that we start with a focus that privileges certain phenomena by considering only those factors that we represent as being of a certain type or falling within a certain range. We also tend to include such other factors that are proximate to the object of study and seem relevant in terms of common sense (such as based on our knowledge of comparable situations). The result is a representation of phenomena in thought as the parts of a whole. The categories of part and whole are unavoidable here, for otherwise the data of observation would have no meaningful relation and significance.

What is this whole? From a phenomenological perspective, it is a manifestation in observable qualities of behaviors of the parts that seem to arise from their their interaction with each other within the network of interactions we call the whole. The whole itself, the Mongolian economy per se, is only inferred, for it is not an observable. On the surface, this procedure does not seem to be a problem, but let us consider more carefully just what we mean by the relation of the parts. We expect it to be causal, for a mere formal relationship is would not give rise to novel features that serve as evidence of the participation of the parts within the whole.

There are various kinds of relationships among parts or factors. For example, there is what is traditionally referred to as constant conjuncture. We find that a rise in the price of lamb in the Mongolian market is usually accompanied by a change in some other economic factor. The totality of these relatively constant conjunctures determines what factors we need to consider in our analysis of the Mongolian economy, the parts of the Mongolian economic system.

However, there is another kind of relation of factors that makes us aware of things that might not normally be considered part of the economy, but in a particular conjuncture apparently do have an economic effect. This is our inference that an event has a casual relation with the economy because of its proximity in space and time with an economic event, and also because it satisfies common sense (our experience that such events might conceivably have effect or have had effect in other situations).

Finally, we also look at the economic factors in functional terms. That is, we employ a counter-factual argument to the effect that were a factor not present, the changes we see in other economic factors would not have occurred. This phenomenological approach depends on our seeing the economy in structural terms, where the behavior of the whole (as manifested in its parts) depends on both the presence and the characteristics of the functional parts.

While these relationships all are part of our mental representation of the Mongolian economic system, a whole that is not directly observable, it nevertheless is empirically based and so is a reasonable inference. It is also to a significant extent (but in important ways not always) operational in that the intervention of the Mongolian government into the economy based on its economic knowledge has fairly predictable results.

Of course this takes us away from historiography, which, to the extent it is the study of the past, might not seem to offer grounds for operationalism. Nevertheless, economic policy that is operational today suggests the Mongolian government has some fairly reliable knowledge of a system that has been in place for a while. Depending on how much we abstract from the specifics of a particular conjuncture, that understanding might reasonably be projected back in time. However, how far and whether it can be projected back through major systemic transformations is a matter of sharp debate (see, for example, the Polanyi thesis and its critics).

Up to this point, a phenomenological reductionism does not appear to be problematic. It represents a methodology that is rational, is empirically based, and it often “works” in the sense that we can define a system by ascertaining its parts or factors. However, upon closer inspection the truth it yields is very one-sided and so crippled in its powers of explanation that something better is obviously needed.

One difficulty with a phenomenological reductionism is that it depends on a naive representation of factors, which are treated no differently than we might describe the mass, positions and velocities of billiard balls moving on a table. That is, parts are reduced to their observable qualities, whether they are present initially or are the effect of the other factors (also defined in terms of observables) that impinge on them. In other words, the role played by the parts within a whole, which comes down to the effect of one upon another, is determined by their observable states.

But we know this is simply not true and leaves little room for human freedom. The causal structure of the system, which is not observable, affects the interactions of its parts. This occurs in a variety of ways. For example, one that is naturally of interest to historians is that systems “age”. That is, systems that are “immature” (such as near the beginning of a civilization) behave differently than systems that are “old”. The explanation of this system behavior would divert us from the path we are following here, but suffice it to say that it is well understood in general systems theory and is readily observed.

Another constraint on the interaction of observable factors is the neglect of mediating structures that do not participate in a causal chain, but merely constrain causal relations. For example, were we to pour mixed gravel through a sieve, the resulting ordered pile of gravel could not have been predicted solely by an inspection of the original mixture, and the sieve is not really part of the causal chain because its state does not change (although it necessarily dissipates energy into the environment). Not everything that affects system behavior is an interacting part.

Yet another complication is what we might call contradictory processes. That is, an effect of the development of a contradictory process negates some condition that was necessary for its existence in the first place. For example, if we grow a crystal in a super-saturated solution, its growth reduces the saturation of the solution and therefore reduces the possibility for further growth. While this contradictory behavior can be understood in systemic terms, it cannot have been predicted directly from a description of the initial state of the system that as yet has no crystal. We know from experience that the growing crystal will reduce the saturation of the solution, but this cannot be inferred solely from the initial state. It is true that we can infer a crystal will grow, and while we can also infer that its growth will reduce the saturation, this is an inference of an inference, not an inference from an observable initial state.

What is common to such difficulties of phenomenological reductionism is that they do not really explain anything. This is a contentious point in the philosophy of science, and so I need to tread carefully.

Let us go back and look more carefully at the various kinds of interactions between system parts. What does an observed “constant conjuncture” really do for us? It enables us to predict with some degree of confidence that a particular event will be associated with another at some point in the future. Of course, that association may not actually occur, and this forces us to construct an auxiliary hypothesis to account for the unexpected outcome, but to some extent, this tactic is perfectly legitimate and does not discredit the associative rule. But the point is that a prediction is not an explanation of why one event is frequently associated with another, but merely our observation that they usually are. It only generalizes our experience and does not look to some causal mechanism that makes the association of events necessary. There are some philosophers of science who still maintain that “covering-law explanation” is indeed an explanation as far as they are concerned, but this view is becoming ever more delicate.

Proximity was another way we suggested events are associated. It is true that events that are proximate are more likely to have a causal relation, but this does not at all explain that relation. It lacks any causal mechanism, and so the necessity that is essential to any causal explanation is missing.

Finally, we mentioned a functional relationship among factors that allows us to represent them as parts of a greater whole. Their emergent qualities or behavior cannot be explained except for an assumption that they are causally related. Unfortunately, we can’t turn this negative argument on its head to represent it as really being an explanation, for we are still left only with an inference that the things are in fact related, without any causal mechanism to explain just how or why.

In history, every situation is unique, and so a validation of a reductionism must arise from a sense that the outcome of the interaction of parts is in some way necessary. Only marginally do historians appeal to covering laws, even were they to imply necessity. If the definition parts is reduced to observable phenomena, it is clear that these observables can not in themselves necessitate any specific outcome. A causal interaction depends on a relation of things, but if we take a part or factor in hypothetical isolation, which we do when we describe its state in phenomenological terms, it has no determinant effect on anything else. Of course, we do describe parts as interacting, but if each is defined phenomenologically, there can be no necessary result of their interaction, only a description of its result. A purely phenomenological representation of the factors in the initial state of a system such as the modern Mongolian economy cannot support explanation.

The phenomenological reductionism that is typically employed by historians has the virtue of being empirically based, it generally appeals to common sense based on our experience of daily life, and it is operational if for no other reason than its generation of hypotheses useful for the study of an historical situation. However, it is not at all certain that such a reduction supports explanation, and therefore the truth value of any knowledge based on it is very little. Explanatory truth does not reduce to observational truth when it comes to developing processes.

Processual reductionism

What I am here calling a processual reductionism presumes that unobservables are real. What this means is that the unobservables necessary to account for the observed behavior of things are not just mental constructions to organize our observations, but an aspect of things that is independent of consciousness. For example, a causal relation in this view is not just inferred from its effects, but is a real potential of things to affect other things. For example, we observe that a non-equilibrium system always moves toward an equilibrium state, although the system's non-equilibrium cannot be directly observed.

It is difficult in phenomenological terms to describe the initial state of a system as being in dis-equilibrium and therefore in motion, for it depends on a relation of states rather than the states themselves. It is argued that we can assume that systems are in disequilibrium because each system is heir of another and have their ultimate origin in the cosmological Big Bang, inheriting from it their non-equilibrium state. As a result, a hypothetical temporarily isolated branch system will move toward an equilibrium state without depending on any outside cause. Ultimately, even systems in temporary equilibrium more toward a state of even greater equilibrium that is referred to as the heat death of the universe.

However, scientific realists go further than this, for if our description of an initial state of a system includes unobservables, we can simply state that our experience that all matter is in motion is the nature of matter. An important effort to explain this is offered in David Bohm's, Wholeness and the implicate order (London, 1980). However, Bohm's argument has found acceptance only very slowly because cosmologists feel that he introduces a complication that is insufficiently justified. This may be so in cosmology, but the approach has appeal in more general terms and is particularly useful in historiography, which is concerned with emergent processes for which a probabilistic language is used.

Of concern is not the specific cosmological argument offered by Bohm, but the axiom that matter has unobservable aspects that account for its observed behavior and should be considered real. There's no standard list of these unobservables, and scientists choose those that seem relevant to their particular concerns. However, taken together, these unobservables seem to imply at least two things. One is that matter has a real disposition to change its state, and secondly, these dispositions are probabilistic. That everything is a process is virtually a truism in modern science, and that all processes are probabilistic is widely held.

For example, from a phenomenological viewpoint, if you flip a coin, the outcome will be either heads or tails. Regardless of the number of times you flip it, the outcome is either one or the other (usually spoken of as having a state of 0 or 1, as happening or not happening). This outcome is assumed to be unequivocally determined by the initial state of the system, even if we cannot in principle or practice have the complete knowledge of that initial state necessary to predict the outcome of each toss. Nevertheless, the initial state is assumed to determine the outcome of the toss as it occurs.

It is generally now accepted there often exist statistical laws that determine the probability of possible outcomes without specifying the result of any particular flip of the coin. From a phenomenological viewpoint, these statistical laws simply manifest our ignorance of all the determinations governing any particular outcome, not that the initial determinations were probabilistic.

A realist position takes a radically different approach in that the probability is not simply the result of our observation of many outcomes, but is an intrinsic disposition of the initial state. The tossed coin has a disposition to yield an outcome not only of 0 or 1, but of all values in between, and this probability distribution of possible outcomes collapses to 0 or 1 only when the process is ended and the coin ends up as being either heads or tails. It is the ending of the process that makes the outcome either 0 or 1, not its initial condition.

To suggest that all matter has a disposition to change in a way defined by a hidden inner disposition that results in a probability distribution of possible outcomes must face an old objection to vitalism that the assumption introduces supernatural forces. However, scientific realism insists these forces are quite natural and are entirely compatible with the data of our experience.

Nevertheless, we need to offer some justification for such a step even if we no longer find compelling the empiricist insistence that only the facts of observation are real. In many situations it does not seem to make much difference whether or not we assume unobservables are real, and so arises the objection to Bohm: are we not just unnecessarily complicating things? However, justifications for a processual reductionism do exist.

One is the assumption that all domains of reality are ultimately compatible because they emerge from one universal underlying reality (this is a definition of materialism). This fundamental reality necessarily implies a realist view, for observables are the identities that emerge from it. Indeed, there are domains of scientific knowledge in which treating unobservables as real is found to be necessary, and this implies that it is at least possibly true for all other domains.

For example, the science of statistical mechanics successfully reduces the phenomenal laws of thermodynamics to unobservable particle motions that are represented mathematically because they are unobservable. Also, in quantum mechanics, a particle is a probability distribution of possible outcomes until measured (the Bell Theorem finally put to rest the phenomenological interpretation of quantum particle measurement).

Now, while it might be argued that these probabilistic behaviors are features peculiar to only these two domains, these are so fundamental that all other domains are seen as having emerged from them. That is, the cosmos itself emerged from a quantum soup and therefore everything in it inherits the potential for probabilistic dispositions that characterize quantum particles. Each emergent system is a specification of a parent system that does not reduce to it. That is, each emergent system has its own peculiar mode of behavior that distinguishes it from its parent, although it remains subject to the laws of its parent. We may be conscious biological beings, but that does not free us of the influence of gravity; gravity and the other laws of physics do not explain consciousness, although consciousness cannot violate them.

It is pointed out in the philosophy of science that a great advantage of scientific realism is that it enables one to employ theory (which is understood to be real) to reduce not just observables as in phenomenological reductionism, but unobservables as well to attain an understanding of the deeper and more fundamental level of reality. This opens the way for a unification of the sciences, and, although far from complete, the remarkable history of scientific advance in the 20th century represents progress in our knowledge of the world as a result of this unification. It is generally assumed by scientists that this unification is the principle way that scientific understanding has advanced. Put simply, a significant advance in knowledge depends on an assumption that unobservables are real. This suggests that the assumption is indeed operational.

This kind of reductionism can be called “processual” because the real although unobservable aspects of things defines them essentially as processes. These processes are probabilistic dispositions to have a specific effect on something else (our measurement apparatus, for example). However, given our inability to represent such an object in thought, we artificially deconstruct it into two dimensions, an empirical dimension and a causal relation. This causal relation may not actually connect to anything else, for it represents the dispositions of a thing to have a causal effect. These two dimension are not separable in reality, but we use them to think about processes. Because they are the result of a deconstruction, each has a one-sided truth value.

In the natural sciences, a mathematical formalism often captures the reality of things that we cannot represent in thought. For example, a Klein Bottle is a mathematical object that we can think of only in terms of a very crude approximation. Although the conceptual convenience of distinguishing an empirical and an abstract causal dimension of things might be taken to imply one thing is connected to another within a greater whole, in reality conventional distinction of part and whole collapses, for we have a processual unit of analysis that includes both and cannot actually be reduced without the loss of truth value. This unit is what Arthur Koestler was after when he coined the term “holon”.

Even though a causal relation is a mental abstraction, when dealing with processual objects it seems appropriate to to define them in terms of it rather than in terms of their empirical dimension. This represents is a major methodological distinction between phenomenalism and realism. The phenomenalist insists upon defining things in terms of their empirical qualities because he assumes that only the data of experience are real. The realist, on the other hand, is more concerned with process than static states (which are in any case denied in principle), and so sees things as processes that are potential causes, and this makes their causal relation of greatest significance. Sensibly enough, this significance of something is not self-contained and linked to its empirical characteristics, but derives from its potential relation to all other things.

For example, in Marxism, social “class” is not defined in static empirical terms (such as income level or life-style), but rather as a relation people share in common to what accounts for their development—their “relation of production”. However, we are here not beating a retreat to parts and wholes, for it is not that something external to a class is the cause of its development, but that people employ something external to them as a necessary means for their self development; inside and outside, subject and object, essence and existence, being and becoming, are all conceputal aspects of one unitary reality.

An important implication of the realist position is that causation is not simply inferred from its effect, but is a real disposition of things. This makes possible the explanation of change as the result of a causal mechanism. While this causal mechanism is unobservable, a presumption that it is nevertheless real provides the element of necessity that explains change. This essential connectedness of things, whether it be actual or a potential, also expresses the essential unity of all things at a much deeper level than that of observation, the level of matter itself. Observation has the effect of distinguishing things and obscuring their real interdependence. To only observe that events are proximate, have a constant conjuncture, or seem functionally related is not really an explanation, for it precludes this element of necessity.

The significance for historiography

The importance of being able explain historical phenomena is that it makes clear how we manage to construct our history. To sit back and draw inferences from just our knowledge of phenomena seems inherently passive, leading the historian to reduce the past to mere description fleshed out with short range causal inferences. Without a grasp of causal mechanisms, historic consciousness cannot embrace the unobservable forces for change that we might harness to our purposes.

In nineteenth-century Europe, historiography was honored because of its ideological utility, for it lent historical legitimacy to a set of values and helped define one's relation to them. However, in the twentieth century, historiography lost much of this function as we entered into a far more complex and interdependent world. Nowadays, for historiography to recover a useful social function, it must offer an understanding of the limits and potentials of human agency to shape the course of history, and that requires an understanding of the causal mechanisms at work in the historical process. This concern for the apparent loss of creativity arises as civilizations decline, and it has interested historians ever since the days of Edward Gibbon.