In this post, I discuss the mechanistic approach to historical (causal) explanation. I argue that while the approach has many virtues, it does not capture some important aspects of historiographical explanatory practices.
In the paper “Ephemeral Mechanisms and Historical Explanation” (2010), Stuart Glennan argues that historical events are often products of processes that can be understood as mechanisms that produce the events. Given that the philosophy of science has produced an extensive analysis of mechanisms and mechanistic explanation in other fields, it is possible to analyze the notion of historical causal explanation within the powerful conceptual framework of mechanistic explanation. Applying the framework of mechanistic explanation also makes it possible to pinpoint the connections between historical and scientific explanations (Glennan 2010, 251).
The mechanistic approach to causal explanation has its roots in the problems that the nomological view of scientific explanation (i.e. the view that general laws do the main explanatory work) faces. Even if we forget about all the well-known counterexamples to the nomological view, the problem with the view is that “that scientists (especially in the biological and social sciences) seldom frame their discoveries and explanations in terms of laws of nature” (256). Instead, according to Glennan, scientists “frame their discoveries and explanations [–] in terms of mechanisms” (ibid.).
But what are mechanisms and mechanistic explanation? “The [–] mechanistic systems approach, explicitly characterizes mechanisms as robust systems or processes consisting of interacting parts- [–] Mechanisms are systems that produce some phenomenon, behavior or function.” (ibid.). “The behavior of mechanisms can be described by [–] ‘mechanistically fragile generalizations’. These are generalizations that are robust and non-accidental, but hold in virtue of the fact that they describe the behavior of the mechanism. That mechanism is a system with a stable spatial, temporal and causal structure in which the parts of the mechanisms act in regular ways to produce the mechanism’s behavior.” A phenomenon is explained once we have identified the mechanism that produces the phenomenon.
A typical example of a mechanism in the literature is chemical neurotransmission where “a presynaptic neuron transmits a signal to a post-synaptic neuron by releasing neurotransmitter molecules that diffuse across the synaptic cleft, bind to receptors, and so depolarize the post-synaptic cell. [–] The neurotransmitter and receptor, two entities, bind, an activity, by virtue of their structural properties and charge distribution” (MDC 2000, 3).
A simpler example is the system that makes a light turn on when we press the switch. The switch is connected to the bulb by an electric wire. When we press the switch, electrons start to flow to the bulb. Here we have a system that has stable spatial, temporal and causal structure and the behavior of which can be described in terms of generalizations that tell us “how changes in properties of one part bring about changes in properties of another part” (Glennan 2010, 258). The simplest of the generalizations that describe the system says that “if the switch is in position A (/B), the light is on (/off)” – a non-accidental and quite robust generalization with well-known exceptions.
The problem in applying the mechanistic approach to historiographical explanation is that “mechanistic system explanations explain regular or type-level phenomena [–]. Mechanists are interested in descriptions of how kinds of mechanisms work: for instance, what is the mechanism by which neurons signal across synapses? Whatever this mechanism is, there are countless such neurons in organisms on this planet, and those neurons fire countless times.” (Glennan 2010, 257.) Historiography, on the other hand, is mostly interested in explaining particular events or states of affairs: the fall of Rome, the discovery of gravitational deflection, the death of a historical agent etc. In fact, Glennan stipulates that he attempts to make sense of historical explanations as explanations of particular events in contrast to explanation of recurrent patterns of phenomena that are ahistorical (2010, 253). However, he also notes that historians (at least sometimes) attempt to find recurrent patterns from the history (2010, 262). We will come back to this issue later. For now, it is enough to realize that the mechanistic approach needs to be modified when applied to historiography due the fact that mechanistic approach was designed to capture type-level explanations while historiography focuses on particular events.
Glennan begins the analysis of historical explanation in mechanistic terms by noting that historians build explanatory narratives. He cites Richards (1992, p. 23) who tells us that ‘‘[n]arratives fix events along a temporal dimension, so that prior events are understood to have given rise to subsequent events and thereby to explain them’’. According to Glennan “The task of the historian is to construct a narrative in which the events described are causally relevant, but the theory of narrative so far described gives us little account of how we establish causal relevance” (2010, 259). Glennan seeks to analyze the notion of causal relevance in mechanistic terms.
In Glennan’s analysis, historical explanation proceeds by describing ephemeral mechanisms. “What seems to distinguish mechanisms in [–] historical sense from mechanistic systems is that the circumstances that bring together the various entities whose interactions constitute [a] narrative are ephemeral” (2010, 259).
Glennan defines ephemeral mechanism as follows:
“Specifically, I take an ephemeral mechanism to be a collection of interacting parts where:
1. the interactions between parts can be characterized by direct, invariant, change-relating generalizations
2. the configuration of parts may be the product of chance or exogenous factors
3. the configuration of parts is short-lived and non-stable, and is not an instance of a multiply-realized type.” (2010, 260.)
Consider the following example. In the paper ”Distrust and Discovery: The Case of the Heavy Bosons at CERN” (2001), John Krige describes “the microhistorical process whereby different groups of scientific actors [–] came to claim that a new fundamental particle (the W boson) had been discovered at CERN” (2001, 517) The paper illustrates how factors including the theoretical background, the personal qualities of the scientists, the pressure from the funding agencies, and the competition for prestige affected how the W boson was discovered and how the results were announced.
The decision of CERN to search for the W boson was due to a technological advantage over the competitor, Fermilab, and due to problems with the image of CERN (Krige 2001, 522-523). Once that decision was made, the CERN directorate decided to perform two experiments because (i) the most advanced technology was uncertain, because (ii) political situation required the participation of many scientist and (iii) because the directorate did not trust Rubbia (ibid. 525-528). However, two different experiments did not matter much in the end: Rubbia suddenly decided to publish results before adequate scientific work had been done to check those results (ibid. 533-535). Once that decision was made, other people were forces to adapt to the situation due to the political and institutional situation (ibid. 535-537).
The process that led to the announcement of the discovery can (presumably) be understood in mechanistic terms. First, we have a set of parts that interact: The technology used in the experiments, the institution of CERN, the people working in the project, political institutions, and Carlo Rubbia. The interaction between the parts can (probably) be characterized in direct, invariant, and change-relating generalizations. For example, the distrust of CERN towards Rubbia caused them to plan to experiment. “When people do not trust some outcome, they make a back-up plan” seems to be quite invariant, direct, and change relating generalization: It is true in many context and if we change distrust to trust, the need for a back-up plan diminishes.
Secondly, the configuration of the parts was a somewhat chancy process. For example, political institutions do not develop in uniform manner and they do not have their goals, modes of operation, and beliefs are not independent of the contingencies in their development. The institutional practices that were pressuring CERN could have been differently; there probably could have been more “science-liberal” ideology within the institutions.
Thirdly, the configuration of parts was surely non-stable and not multiply-realized type: A similar configuration of political, scientific institutions and a man with Rubbia’s character cannot be find anywhere else and (presumably) it could have easily been broken by the distrust or institutional changes.
It should be clear by now that the applicability of mechanistic analysis of causal explanation to historiography depends on how, exactly, we understand the notions of invariance, stability, a product of chance etc. that are used to explicate the notion of ephemeral mechanism and whether they have similar analysis with respect to type-level mechanisms and ephemeral mechanism. Mechanistic analysis of explanation does have powerful conceptual resources so we might be optimistic that these two issues can be clarified. However, I am still worried.
First, some of the notions used in analysis of ephemeral mechanisms do not seem to have a meaningful reading with respect to type-level mechanisms. For example, what would it mean to say that the configuration of a type-level mechanism is a product of chance/necessity? If it does not make sense to talk about the contingency of (the existence of) type-level mechanisms, then the chancy production of ephemeral mechanisms does not distinguish them from type-level mechanisms. This rises the question why we need to analyze the notion of ephemeral mechanism in terms of chanciness of the configuration. Why not leave it an empirical question whether some historical mechanism is a configuration produced by chance? Glennan himself notes that the condition (2) “is not meant to exclude some historical mechanisms in which there are non-chance factors (2010, 260)”. It is difficult to understand the point behind the condition (2) in the analysis of historical causal explanation.
Secondly, the mechanistic explication of explanation requires that the change-relating generalizations used in them are “counterfactual supporting” (Glennan 2010, 257). The exact meaning of this notion is left unclear. As Woodward (2002) notes, barometer readings and storms are connected by a change-relating and counterfactual supporting generalization. Had the barometer reading dropped, a storm would have appeared. The problem is that the counterfactual that the generalization supports is causally irrelevant. We have to have an analysis of counterfactual supporting generalizations that guarantees that the generalizations used in descriptions of mechanisms support the relevant causally structured counterfactuals. Given that we need to have a notion of causal relevance before we can describe mechanisms, causal relevance cannot be understood in mechanistic terms.
In addition to the (seeming) problems in the clarity of the definition of ephemeral mechanism, we may wonder whether mechanisms are always needed in historical explanations and whether all interesting historical explanations can be analyzed in mechanical terms. Consider the following example. Why was the Red Army able to push into Finland in the summer 1944? Because the Finnish Army did not have the resources to fight back intensively enough. Had they had more resources, the push would not have happened or would have happened later. We have an explanation here. But what is the mechanism of the defeat? Perhaps we can tell how the fighting – a mechanistic process between troops and equipment – unfolded. The problem is that “the lack of resources” cannot be naturally interpret as a part of a mechanism, as there is no entity corresponding to it. How can the non-existence of something interact directly with something else? Perhaps it could be possible to tell the story in terms of how the Red Army interacted with what was left of the Finnish Army and how during the process they took steps deeper and deeper to Finnish defense lines. However, the explanatory crucial factor was the lack of resources; it underlined the whole process. It is difficult to interpret this crucial factor in mechanistic terms.
Take another example. There was an air raid that took place. We can describe how a city was destroyed by citing the decision of sending the air raid and the movements of the airplanes (P) and working of the bombs. However, if it is true that “had airplanes P not been available, other set of airplanes (P*) would have been sent on the mission”, then the actual configuration of the planes did not make any difference to the outcome. The mechanism that produced the destruction of the city does not seem to explain its destruction. Notice that here the outcomes with P and P* are the same: the destruction of the city at the time t. Therefore, the problem does not go away by analyzing the grain with which the explanandum is described (see Glennan 2010, 264 discussing the grain-of-explanation strategy).
Surely, there could be some way of telling the explanatory story in terms of mechanism that abstracts away the actual planes, i.e. in terms of interaction between army headquarters, the air forces, and the city. However, there are two problems in this approach. First, it seems that abstracting away the details damages the connection between historical mechanisms and the intuitive idea of mechanical explanation in cases like machines or neurons, at least in extreme cases. Secondly, historical narratives do describe the details and there is a decision to be made by the mechanistic approach: Whether to (a) take the similarity between historical narratives and mechanistic explanations as evidence of the applicability of mechanistic analysis of explanation to historiography, or (b) dismiss the details of historiographical works as explanatorily irrelevant. If (a), then the abstracting-away strategy does not work. If (b) the initial motivation stemming from similarities between narratives and mechanisms fades away.
The problems related to the latter example and to the abstracting-away strategy are related to further difficulties in the mechanistic approach when it comes to explicating the relationship between type-level and ephemeral-level explanations. Consider the following example:
“Let us consider how to explain the events in the fall of 2008 that led to the global stock market crash. Arguably there are at least two ways to explain this event. One way would focus on providing a narrative of the particular events that led to panicky sell-offs of particular stocks and the demise of particular companies. Such a narrative would focus on particular decisions of the U.S. Treasury Department, the Federal Reserve and other parties, like the decision not to bail out Lehman Brothers. An explanation of this sort is, according to the analysis I’ve offered, a description of an ephemeral mechanism. The second way of explaining the market crash would be to show how it exemplifies a familiar pattern of economic behavior. At the simplest level, the crash was the result of a speculative market bubble. Market bubbles are a recurrent feature of economic systems throughout modern history—from the speculation in shares of the French Mississippi Company in the 1720s to the dot com bubble of the late 1990s to the real estate bubble of 2008. Viewed from this perspective, the stock market crash of 2008 is not the product of an ephemeral mechanism, but of a mechanism of a more robust and predictable sort.” (Glennan 2010, 262-263.)
Glennan adds that “[e]xplanations of this sort actually fit nicely within the systems conception of mechanism” and that “[w]hile such explanations are general and not historical in the sense of this paper, they still have a narrative structure [–] [b]ut the narratives involved in such explanations are generalized narratives” (Glennan 2010, 263).
What is the relationship between generalized narratives (“mechanisms of a more robust and predictable sort”) and historiographical narratives (ephemeral mechanisms)? Are the two ways of explaining an event complementary or in competition? A natural interpretation would be one that says that a particular ephemeral mechanism is an instance of a generalized mechanism (at least when the latter exists). However, in such cases the ephemeral mechanism would inherit the stability of a generalized mechanism and the mechanism would be multiply-realized, i.e. not ephemeral by definition. And it does not help to argue that a particular historical mechanism is unstable in the sense that its existence is due to historical contingencies because this is true also of systems mechanisms. As Glennan points out, “[s]ystems mechanisms have a historical dimension in the sense that the generalizations describing the behavior of these mechanisms are true only in virtue of the particular organization and interaction of parts, and of the historical processes that brought them about. All of the myriad mechanisms at work in the life of organisms exist only in virtue of those organisms’ evolutionary and developmental history.” (2010, 258.)
The problems become apparent when we consider the problem of overdetermination. Our case of overdetermination was the air raid to destroy a city. In our example, it was the case that if the planes P had not been sent to bomb the city, planes P* would have been sent. We saw that in such cases, we have to say either that (a) the actual mechanism is not explanatorily relevant or (b) abstract away the planes and reformulate the parts of mechanism and the generalizations that describe the mechanism. As (a) is not possible in the mechanistic approach, we are left with (b). The problem with (b) is that the reformulated mechanism seems to be a generalizable mechanism, as it describes the working of a resourceful army in its attempt to destroy a target. Such cases are not rare and they appear to be quite stable in their structure. Moreover, it seems that the actual outcome, the destruction of the city, is an outcome that is achieved precisely because the army headquarters is able to understand the type-level mechanism (i) that describes how planes – no matter what the actual set in the actual raid is – can be used in such operations and (ii) an instance of which is the actual raid. If historical actors plan an operation in terms of a type-level mechanism, it is difficult to understand how the actual history, as long as it follows the plan, can be anything but an instance of that type-level mechanism.
Despite the critical tone of my analysis of the mechanistic approach to historiographical explanation, I find the approach useful. I think it is essentially the correct way to approach the most difficult and complex historical processes because (a) it provides us with tools for understanding how such processes can be explanatory despite the fact that the causal configuration is highly unstable and produced by chance, and (b) it clarifies how such complex processes can be broken into smaller pieces that can be understood causally. However, I think that if we confine historical explanation to the cases that can be described only in terms of ephemeral mechanisms, we leave many important cases of historical explanation outside the scope of the analysis and become unable to tell how generalizability is related to the explaining of particular events. I attempted to make the reasons for this judgement clear above.
There is no reason for me to hide the fact that I take the counterfactual approach to historical explanation to be a better candidate to provide the general analysis of historical explanation that is applicable to a wider range of cases and that tells us how generalizability is related to the explaining of particular events in terms of explanatory depth (see Virmajoki 2020). However, the counterfactual approach is sometimes at the risk of detaching from the actual and concrete history because in a counterfactual scenario we can make all sorts of assumptions that omit the actual historical details. For example, it is true that “had the army B dropped an atomic bomb on the army A, the army A would not have been able to destroy the city C” but it might be far-fetched, historiographically speaking, to assume that the army B was able to drop the bomb. In contrast, the mechanistic approach has the virtue of securing the central role of the actual historical details as an integral part of historiographical explanations. Perhaps it is possible to take the best of the both worlds and analyze the limits of the mechanistic approach in terms of the counterfactual one and vice versa.
Finally, we have to notice that both the mechanistic and the counterfactual approach keep a critical distance towards actual historical explanatory practices. The interesting question is just how much distance we can take. For example, the counterfactual approach that I have defended views historical narratives mainly as a source of explanatory insight and minimizes their role as providers of understanding in historiography (I once suggested that we need to leave the actual world behind altogether in order to understand historiography, see here). For many, this dismissing of the explanatory status of historical narratives is too much and shows that the counterfactual approach cannot be the correct analysis of historiographical explanation as it is conceived by historians. The mechanistic approach does not go quite that far and stays in closer contact with narratives. The danger in this strategy is that dubious historiographical claims – like the idea that historical explanations establish the contingency of explananda – are written into the philosophical analysis rather critically analyzed.
Glennan, Stuart (2010). Ephemeral Mechanisms and Historical Explanation. Erkenntnis 72 (2):251-266.
MDC = Machamer, Peter ; Darden, Lindley & Craver, Carl F. (2000). Thinking about Mechanisms. Philosophy of Science 67 (1):1-25.
Virmajoki, Veli (2020). What Should We Require from an Account of Explanation in Historiography? Journal of the Philosophy of History:1-32.
Woodward, James (2002). What Is a Mechanism? A Counterfactual Account. Philosophy of Science 69 (S3):S366-S377.