Keywords

2.1 Introduction

In the new mechanistic literature, authors often distinguish between two different types of mechanistic explanation: etiological mechanistic explanation, in which a phenomenon is explained by its preceding causes, and constitutive mechanistic explanations, which refer to mechanisms that “constitute” the phenomenon-to-be-explained (Craver, 2007a, b; Salmon, 1984a, b). The distinction between these two types of explanation was first introduced by Wesley Salmon (1984a, b). Constitutive explanations, according to Salmon, “account for a given phenomenon by providing a causal analysis of the phenomenon itself” (1984a, p. 297), while an etiological explanation “tells the causal story leading up to its occurrence” (ibid.). Salmon’s focus was on etiological explanation. Carl Craver (2007b) prominently highlighted the relevance of constitutive explanations for the life sciences. He discusses various examples of constitutive mechanistic explanations, such as the explanation of the action potential and spatial memory. Most philosophers of the life sciences and the cognitive sciences now agree that constitutive mechanistic explanation is ubiquitous.

While the notion of causation is extensively discussed in philosophy of science as well as metaphysics, the debate on mechanistic constitution is quite young and there are still many open questions such as whether mechanistic constitution can be spelled out in terms of interventionism, whether approaches to constitution should be singularistic or generalistic, which role time plays in constitution, what the relata of constitution are, and how constitution differs from causation (Baetu, 2012; Baumgartner & Gebharter, 2016; Couch, 2011; Craver, 2007a, b; Craver et al., 2021; Fagan, 2012; Gillett, 2013; Harbecke, 2010; Harinen, 2018; Kästner, 2017; Kirchhoff, 2017; Kistler, 2009; Krickel, 2018b; Romero, 2015; Weinberger, 2019). The new mechanistic account of constitutive mechanistic explanation is not only supposed to capture a common and important explanatory practice in the life sciences. The nature of constitutive mechanistic explanation, according to the new mechanists, has several implications for various ontological questions as well. For example, whether or in which sense mechanistic explanation is reductive, what levels of nature are, and whether there can be causal relationships between these levels directly depends on how the notion of constitutive mechanistic explanation is understood. Thus, the details of the account of constitutive mechanistic explanations are crucial for the evaluation of the ontological implications of the new mechanistic account.

This chapter has two goals: First, I will use ideas developed by Krickel (2018a) and Kaiser and Krickel (2017) to show that there are three different interpretations of what constitutive mechanistic explanation amounts to. I will use these considerations to argue that there are in fact four variants of mechanistic explanation, not just two. Second, I will describe the different ontological implications of the different versions of constitutive mechanistic explanation and outline the different pictures of reduction, levels, and interlevel causation that they suggest.

The paper proceeds as follows: in Sect. 2.2, I will present the general features that are commonly attributed to constitutive mechanistic explanation. In Sects. 2.3 and 2.4, I will summarize the two views of mechanistic constitution and mechanistic phenomena presented by Krickel (2018a)—the functionalist view of constitutive mechanistic phenomena and the behaving entity view of constitutive mechanistic phenomena. I will recap three possible interpretations of the functionalist view as presented in Krickel (2018a) and Kaiser and Krickel (2017). In Sect. 2.5, I will show that the different views on mechanistic constitution and constitutive phenomena suggest that there are indeed four different types of mechanistic explanation—each of which describes a common explanatory practice in the life sciences. In Sect. 2.6, I will outline the different ontological consequences of the different types of mechanistic explanation.

2.2 Constitutive Mechanistic Explanation: Minimal Characterization

What is constitutive mechanistic explanation? In this section, I will provide a minimal characterization in terms of a list of criteria of adequacy that an approach to constitutive mechanistic explanation must satisfy.

It is commonly assumed that mechanistic constitutive explanation is a type of mechanistic explanation where a phenomenon is explained by its underlying mechanism. The relation between the phenomenon and the mechanism is usually called “mechanistic constitution” and the relation between a component of the mechanism and the phenomenon “constitutive relevance”. To get a grasp of what mechanistic constitutive explanation is, let us briefly summarize what the mechanists take mechanisms to be.

The nature of mechanisms has gained a lot of attention in the mechanistic literature. Here is what Glennan (2017) calls the minimal characterization (MC) of a mechanism:

(MC)

A mechanism for a phenomenon consists of entities (or parts) whose activities and interactions are organized so as to be responsible for the phenomenon. (Glennan, 2017, 17; for similar formulations see Craver (2007b) and Illari and Williamson (2012)).

This characterization applies to mechanisms in general, i.e., to those that are referred to in etiological and those that are referred to in constitutive mechanistic explanations (Craver, 2007b, p. 22). To illustrate how this characterization applies to constitutive mechanistic explanation, consider the example of the action potential mechanism (Craver, 2007b, Chap. 4). This mechanism consists of various entities such as different types of ions and ion-channels. These ions and ion-channels are engaged in different activities such as opening, closing, or diffusing. Furthermore, these entities and activities are organized in various ways. For example, the ion-channels are spatially organized along the axon, and the location of the ions outside and inside the axon is crucial for the working of the mechanism. The different activities are temporally organized. For example, the temporal order of the opening and closing of the ion-channels is crucial for the action potential to occur. Furthermore, the components of mechanisms are what Craver calls “actively” organized (Craver, 2007b, p. 136). They interact in various ways. For example, the ions diffuse through the ion-channels.

The central idea of a constitutive mechanistic explanation is illustrated in the well-known Craver-diagram (see Fig. 2.1). This diagram has become popular in the new mechanistic literature and most authors take it to be an adequate illustration of a mechanism that constitutes a phenomenon.Footnote 1

Fig. 2.1
The Craver diagram illustrates a mechanism constituting a phenomenon. It consists of a system, S with behavior, psi. X denotes entities and phi denotes activities. Arrows inside the circle represent the interactions between the entities and activities.

The Craver diagram: Illustration of a mechanism constituting a phenomenon. (Adapted from Craver (2007b, p. 6))

Full size image

In Fig. 2.1, the mechanism is located at the bottom (the different Xs stand for the entities and the different ϕ-ings stand for their activities; the arrows inside the lower big circle indicate the interactions between the entities and activities). The phenomenon consists of a system (“S”) that is engaged in a certain behavior (“ψ-ing”). Referring to the phenomenon by “S’s ψ-ing” or “S ψ-ing” has become a common convention in the new mechanistic literature.

The minimal characterization of a mechanism presented above seems to apply well to the example of the action potential mechanism. But what renders the explanation of the action potential a constitutive mechanistic explanation? As indicated in the Craver-diagram, one core feature of mechanistic constitution is that it relates a mechanism and a phenomenon that occur at the same time. Thus, in constitutive mechanistic explanations, the mechanism and the phenomenon do not stand in a causal relationship as a central assumption about causation is that causes and effects occur at different times (Lewis, 1973).

Further features of the relation between mechanisms and phenomena can be inferred from Craver’s approach to constitutive relevance (Craver, 2007a, b) that has become the standard account of constitutive mechanistic explanation (Casini et al., 2011; Gillett, 2013; Illari & Williamson, 2011; Irvine, 2013; Kaplan, 2012; Levy, 2009; van Eck & Looren de Jong, 2016; Zednik, 2015). Craver argues that in constitutive mechanistic explanations the components of a mechanism are not causally but constitutively relevant for the phenomenon. According to Craver’s account, a component X’s ϕ-ing is constitutively relevant for a phenomenon S’s ψ-ing ifFootnote 2:

  1. (i)

    X’s ϕ-ing is a spatiotemporal part of S’s ψ-ing, and

  2. (ii)

    X’s ϕ-ing and S’s ψ-ing are mutually manipulable (Craver, 2007b)

Condition (i) specifies that the entity-component must be a spatiotemporal part of the system whose behavior is to be explained, and the activity-component has to be executed during the system’s ψ-ing. Condition (ii) is spelled out in terms of interventionism (Woodward, 2003, 2015): X’s ϕ-ing and S’s ψ-ing are mutually manipulable if and only if it is possible to ideally intervene into X’s ϕ-ing and thereby change S’s ψ-ing, and it is possible to ideally intervene into S’s ψ-ing and thereby change X’s ϕ-ing. An intuitive understanding of interventions suffices for present purposes. Many authors argue that the mutual manipulability account and interventionism are incompatible (Baumgartner & Gebharter, 2016; Kästner, 2017; Leuridan, 2012; Romero, 2015). Therefore, it remains controversial how to understand the claim that phenomena and mechanisms mutually depend on each other (note that promising attempts have been made to save the combination between constitutive explanation and interventionism; see (Baumgartner & Gebharter, 2016; Craver et al., 2021; Krickel, 2018b; Romero, 2015)). For the sake of argument, I will formulate the mutual manipulability requirement as a general mutual dependency requirement in the sense of “If the phenomenon had been different, the mechanism would have been different; and if the mechanism had been different, the phenomenon would have been different”, while this is supposed to remain silent with regard to how this mutual dependency relation is to be spelled out.

In summary, there are at least six characteristics that are commonly attributed to constitutive mechanistic explanation:

  1. 1.

    Mechanism: Mechanisms are entities and activities organized such that they are responsible for a phenomenon.

  2. 2.

    Phenomenon: The phenomenon consists of a system S showing a behavior ψ-ing.

  3. 3.

    Non-Causal: The mechanism does not cause the phenomenon.

  4. 4.

    Temporal Synchrony: The phenomenon and the mechanism occur at the same time.

  5. 5.

    Spatiotemporal Part-Whole Relation: The mechanism’s components are spatiotemporal parts of the phenomenon.

  6. 6.

    Mutual dependency: If the phenomenon had been different, the mechanism would have been different; and if the mechanism had been different, the phenomenon would have been different.

There seems to be a general agreement that constitutive mechanistic explanation has these six features. Still, as Krickel (2018a, Chap. 6) and Kaiser and Krickel (2017) show, there are different ways of how to account for these features. In the following two sections, I will summarize Krickel’s and Kaiser and Krickel’s considerations concerning constitutive mechanistic phenomena and mechanistic constitution that suggest that there are different interpretations of what constitutive mechanistic explanation amounts to.

2.3 The Functionalist View of Constitution and Constitutive Mechanistic Phenomena

Krickel (2018a) and Kaiser and Krickel (2017) discuss the nature of what they call “constitutive mechanistic phenomena”, i.e., phenomena that form the explananda of constitutive mechanistic explanation. One possible interpretation of the nature of constitutive phenomena is what Krickel (2018a) calls the functionalist view of constitutive mechanistic phenomena. According to this view, the system whose behavior is to be explained is the mechanism itself (see Krickel (2018a, Chap. 6)). Hence, the thing that is mechanistically constituted (the “constituee”) is the behaving mechanism. As Krickel (2018a) shows, this idea underlies many discussions in the new mechanistic literature (Bechtel & Abrahamsen, 2005, p. 426; Craver, 2007b, pp. 6–7, 128; Fagan, 2012, p. 467; Fazekas & Kertész, 2011; Illari & Williamson, 2012). In this picture, the behavior that is to be explained is commonly characterized in terms of a complex input-output relation or a causal role. According to the functionalist view, these inputs and outputs are connected by the mechanism (Baetu, 2012; Bechtel, 2008, pp. 201–202; Craver, 2007b, pp. 146, 214; Craver et al., 2021; Fazekas & Kertész, 2011; Kuorikoski, 2012, pp. 146, 375; Soom, 2012). The functionalist view of constitutive mechanistic phenomena can be summarized as follows:

  1. (i)

    The constituee is a behaving mechanism, and

  2. (ii)

    the behavior is characterized in terms of inputs and outputs of the mechanism.

According to the functionalist view of constitutive mechanistic phenomena all there is, is a mechanism that plays a certain causal role, that connects certain inputs with certain outputs. To capture this idea, we can modify the Craver-diagram (presented in Fig. 2.1). Figure 2.2 illustrates what a mechanism constituting a phenomenon looks like according to the functionalist view (a similar picture can be found in (Bechtel, 2017) and Fazekas (2022)).

Fig. 2.2
The Craver diagram illustrates a mechanism constituting a phenomenon according to the functionalist view. It consists of a system, S with behavior, psi. X denotes entities and phi denotes activities. Arrows inside the circle represent the interactions between the entities and activities.

Illustration of a mechanism constituting a phenomenon according to the functionalist view (my illustration based on Bechtel, 2017, fig. 3; and Fazekas, 2022, fig. 2)

Full size image

There are three possible interpretations of this picture. One interpretation can be found in Kaiser and Krickel (2017). It may be called the input-output functionalist view of constitutive mechanistic explanation. According to this view, the phenomenon to be explained just is the inputs plus the outputs of a mechanism. In that sense, the phenomenon is a set of causes and effects. Changing the phenomenon is to change the input or the output. Under this interpretation, Fig. 2.2 must be interpreted accordingly: the phenomenon/S’s ψ-ing is not the circle in the middle but the inputs and the outputs.

The input-output functionalist view, however, is not a valid interpretation of mechanistic constitution. First, it violates Non-Causal: if the phenomenon is the inputs plus the outputs of the mechanism, the two stand in a causal relation. Indeed, the only difference to standard etiological mechanistic explanation is that one cause of the phenomenon is explicitly picked out as the input to the mechanism—which seems to be a difference only in labelling or perspective and not a substantial metaphysical difference. Furthermore, this account violates Temporal Synchrony: the mechanism and the phenomenon do not occur at the same time. Rather, the phenomenon occurs before and after the mechanism occurs. And it violates Spatiotemporal Part-Whole Relation as the mechanism’s components are not parts of the inputs and outputs, and thus, they are not parts of the phenomenon. Hence, the input-output functionalist view should not be considered an account of mechanistic constitution and constitutive mechanistic explanation. Still, as I will argue in Sect. 2.5, the input-output functionalist view gives rise to a valid account of mechanistic explanation (just not constitutive mechanistic explanation).

The second interpretation of the functionalist view is in terms of role functionalism (see Krickel (2018a)). According to mechanistic role functionalists, the phenomenon is a causal role, and the relation of mechanistic constitution is that of causal role-playing. However, again, the role functionalist interpretation cannot be regarded as an account of mechanistic constitution because it does not satisfy the criteria listed in Sect. 2.2. According to role functionalism, phenomena are abstract entities (functional second-order properties). Abstract entities are not located in space and time, and they do not have spatiotemporal parts. Hence, the role functionalist view cannot account for the fifth criterion (Spatial Part-Whole Relation).

What is more: if phenomena are causal roles in the role functionalist sense, there cannot be mechanistic explanations of them as this view would either be inconsistent, or it would collapse into the realizer functionalist view (see below). Causal roles in the role functionalist pictures surely have mechanistic realizers (if physicalism is presupposed). But this is an ontological, not an explanatory statement. The whole idea of role functionalism is to provide an explanation that abstracts away from the details of the realizer. The explanatory power comes from the fact that there are different mechanisms that despite their differences play the same causal role. Adding information about the realizer will not improve the explanation as the explanatory generalization is said to be found at the level of the abstract causal role. This reasoning is known from the discussion of non-reductive physicalism and multiple realization. Irrespective of whether role functionalism is a convincing ontological theory, if there are explanations of causal roles in the sense of role functionalism, these will not be constitutive mechanistic ones but, rather, functional explanations. For example, the question “Why do all these chemical substances function as neurotransmitters?” will be answered by citing the causal role that characterize neurotransmitters and by stating that all these chemical substances execute this causal role. If one takes the mechanistic details of the realizer to be explanatory relevant of the causal role, indeed, one is advocating the realizer functionalist interpretation (see below). Thus, the role functionalist view is incompatible with the mechanistic account in general.

A more promising interpretation of the functionalist view that indeed gives rise to a version of mechanistic explanation is the realizer functionalist view of constitutive mechanistic phenomena.Footnote 3 According to this interpretation, in a first step the phenomenon is characterized in terms of an input-output relation, or a causal role. In a second step, the mechanism that plays this causal role is identified. Then, the phenomenon is identified with the mechanism. According to realizer functionalism, the phenomenon just is the mechanism under a functional description. Mechanistic constitution, here, is identity.

The realizer functionalist view has gained some prominence among defenders of so-called causal betweenness accounts of constitutive relevance. A formulation of this account can be found in a recent article by Craver et al. (2021).

Here, ψ-ing is represented as a process beginning with an input, ψin, and terminating with an output, ψout. Between these temporal endpoints, and a mechanistic level down, is a temporally sequenced causal chain of events, involving the Xi and their various ϕi. (…) [T]he problem of constitutive relevance is that of identifying the components of the process bridging ψin and ψout: What lies on the causal path(s) between these phenomenon-defining endpoints? The higher-level activity, ψ-ing, just is an organized collection of Xi ϕi-ing. (Craver et al., 2021, p. 8812)

The realizer functionalist view accounts for all features of constitutive mechanistic explanation listed in Sect. 2.2. The phenomenon, according to this interpretation, just is the mechanism under a functional description. Hence, the relation between the mechanism and the phenomenon is not causation but identity. The mechanism cannot cause itself. Hence, the realizer functionalist view accounts for Non-Causal. Indeed, the realizer functionalist interpretation of mechanistic constitution accounts for all further features on the list—for a rather trivial reason: since the phenomenon just is the mechanism, trivially, the former occurs at the same time as the latter, parts of the latter are parts of the former, and changing one leads to changes in the other, and vice versa.

Other defenders of causal betweenness accounts of constitutive relevance, however, reject the third step, i.e., the identification of the phenomenon with the mechanism. Totte Harinen (2018), for example, is skeptical.

Is S’s ψ-ing something over and above of the organized ϕ-ings of all of the Xs passing the mutual manipulability test, that is, X1,...,n’s ϕ1,...n-ing? Most philosophers and scientist would probably agree that there is some sense in which S’s ψ-ing is indeed more than just the sum of the ϕ-ings of its Xs, but that the relation between the two should not be that of spooky, materialistically inexplicable emergence. At the same time, many would not want to identify S’s ψ-ing with X1,...,n’s ϕ1,...n-ing, and so there is a market for an intermediate type of interlevel relation. (Harinen, 2018, 40)

Defenders of causal betweenness accounts of constitutive relevance who are not convinced by the identity claim, such as Harinen, argue for (a version of) what Krickel (2018a) calls the behaving entity view of constitution—which I summarize and discuss in the next section.

2.4 The Behaving Entity View of Constitution and Constitutive Phenomena

A further possible interpretation of constitutive mechanistic phenomena, according to Krickel (2018a), is the behaving entity view. This view is characterized by two claims:

  1. (i)

    the constituee is the behavior of an object or system that contains the mechanism, and

  2. (ii)

    this behavior is an activity of the object or system.

According to Krickel, this view can be found in Craver’s discussion of spatial memory (Craver, 2007b), in Stuart Glennan’s work who talks about mechanisms located inside watches, cells, organisms, or toilets (Glennan, 1996, 2002), as well as in Carl Gillett’ interpretation of mechanistic constitution in terms of dimensioned realization (Gillett, 2013, pp. 327–328). According to the behaving entity view, the relation between a mechanism and a phenomenon in constitutive mechanistic explanations can be illustrated as shown in Fig. 2.3.

Fig. 2.3
The Craver diagram illustrates a mechanism constituting a phenomenon according to the behaving entity view. It consists of a system, S with behavior, psi. X denotes entities and phi denotes activities. Arrows inside the circle represent the interactions between the entities and activities.

Illustration of a mechanism constituting a phenomenon according to the behaving entity view

Full size image

As Fig. 2.3 shows, the analysis of the mechanism according to the behaving entity view and the functionalist view are identical (in both cases mechanisms are entities (the Xs) and activities (the ϕ-ings) in a certain organization). The difference between the two views is that, according to the behaving entity view, the phenomenon is not a behavior of the mechanism. Rather, the phenomenon (S’s ψ-ing) contains the mechanism (in Fig. 2.3, the bigger tube that represents the phenomenon contains the circle that stands for the mechanism). The containment relation is a spatial and a temporal one: the mechanism’s entity-components (the Xs) are spatiotemporal parts of S and the different activity-components of the mechanism (the ϕ-ings) occur during S’s ψ-ing.

The behaving entity view accounts for all features listed in Sect. 2.2. Consider the explanation of muscle contraction. According to the behaving entity view, the phenomenon is the muscle contracting (an object showing a certain behavior). The mechanism for muscle contraction consists of various entities, such as actin and myosin filaments and ATP, and activities, such as binding, detaching, energizing and rotating in a certain organization. According to the behaving entity view, the relation between the mechanism for muscle contraction and the muscle’s contracting cannot be causal. The reason is that the muscle’s behavior and the mechanism responsible for this behavior occur at the same time. Furthermore, the muscle’s behavior depends on the behaviors of the filaments. Hence, the phenomenon and the mechanism are not wholly distinct. Distinctness is required for causation. Hence, the behaving entity view accounts for the (Non-Causal) requirement. This shows that (Temporal Synchrony) is accounted for by the behaving entity view as well. The muscle that contains the mechanism for muscle contraction shows the contracting-behavior only during the occurrence of the contracting-mechanism. Furthermore, the mechanism’s components, the actin and myosin filaments and their behaviors, are spatiotemporal parts of the muscle. Hence (Spatial Part-Whole Relation) is accounted for by the behaving entity view.

As mentioned in Sect. 2.2, it is controversial how to account for the last feature (Mutual Dependency). Different authors have argued that the combination of constitutive relationships and interventionism (that provides the framework for mutual manipulability) is problematic. One reason is that interventionism is supposed to be an approach to causation while constitution is supposed to be a non-causal dependency-relation. At least prima facie interventionism is inapplicable to non-causal dependency relations. I will not try to solve this issue here. The applicability of the interventionist framework to mechanistic constitution is a topic of an ongoing debate (Baumgartner & Gebharter, 2016; Craver et al., 2021; Harinen, 2018; Kästner, 2017; Krickel, 2018b; Romero, 2015; Woodward, 2015). Note that the causal betweenness interpretation of constitutive relevance mentioned in Sect. 2.3 that has been put forward as a solution to the problem can be combined with the behaving entity view as well. Along the lines of the causal betweenness account, the mutual dependency between the behaving entity (i.e., the phenomenon) and the mechanism can be interpreted as the causal influence of the input that triggers the entity’s behavior and the mechanism’s components—the top (the top-down intervention) and the causal influence of the mechanism’s components and the output that is produced by the behaving entity (the bottom-up intervention).

The behaving entity view sheds a different light on constitutive explanation than the functionalist view. First, the behaving entity view is not committed to an identity-claim. On that view, the mechanism is a proper spatiotemporal part of the system whose behavior is to be explained. Not all parts of the system are involved in the mechanism (this is why the distinction between relevant and irrelevant parts is so crucial for the new mechanistic approach). For example, the mechanism that explains muscle contraction does not involve all parts of the muscle. Hence, the contracting muscle (the phenomenon) cannot be identical to the mechanism that is responsible for the muscle’s contracting. Similarly, the mechanism that explains the moving of a car does not contain the car’s doors. Hence, the moving car (the phenomenon) cannot be identical to the driving mechanism. Furthermore, following Gillett, the entities that are related constitutively in the way captured by the behaving entity view are what he calls “qualitatively distinct”, they have substantially different features and can enter different causal interactions (2010, p. 172). This qualitative distinctness between the relata blocks an identity-claim.

2.5 Four Variants of Mechanistic Explanation

The forgoing discussion has shown that there are four variants of mechanistic explanation. Table 2.1 provides an overview of these four variants.

Table 2.1 Overview of the four variants of mechanistic explanation
Full size table

As shown in Table 2.1, output mechanistic explanation is what is standardly called “etiological mechanistic explanation”. I chose a different label because, as shown in Sect. 2.3, there is a further version of etiological mechanistic explanation: input-output explanation. The latter two variants—filler mechanistic explanation and dimensioned mechanistic explanation—can plausibly be described as versions of constitutive explanation. The explanantia of all four types of explanations are of the same type—they all refer to mechanisms. As explained in Sects. 2.3 and 2.4, the different variants of mechanistic explanation differ with regard to the nature of their explananda (for an overview see Fig. 2.4) as well as the nature of the relation between the mechanisms and the phenomenon—on which I will say more on below.

Fig. 2.4
An illustration presents the four types of mechanistic explanation, dimensioned, output, filler, and input-output. It consists of a system, S with behavior, psi. X denotes entities and phi denotes activities. Arrows inside the circle represent the interactions between the entities and activities.

The four types of mechanistic explanation grasp different aspects of the same going-on in the world. The grey tubes in the middle represent the mechanism. The arrows point to the different phenomena that are explained by the mechanism

Full size image

Output mechanistic explanations are causal explanations and what is standardly called “etiological mechanistic explanations”. The explanandum refers to an event that is an effect of a mechanism. The relation between mechanism and phenomenon/explanandum and explanans is causation. The relevant question is “How does this effect come about?” For example, when scientists explain neurotransmitter release, they refer to the mechanism that causes neurotransmitter release.

Input-output mechanistic explanation is a type of etiological explanation as well. The phenomenon consists of a set of causes/inputs and effects/outputs that are connected by the mechanism. The relation between the mechanism and the explanandum phenomenon is that of causation. Indeed, input-output mechanistic explanations are common in the life sciences. Explanatory requests asking for input-output mechanistic explanations are, for example, “How does the release of neurotransmitters at the axon terminal change depending on changing inputs to the pre-synaptic neuron?” or “How is neurotransmitter release generated if an action potential reaches the axon terminal?”.

In filler mechanistic explanations, the explanandum is characterized in terms of a functional role. This functional role term functions as a black box or a filler term. The new mechanists talk of “black boxes” that scientists refer to by using “filler terms” that are filled as researchers gain more and more knowledge about the underlying mechanisms (i.e., with details about the mechanism) (Craver, 2007b; Piccinini & Craver, 2011). Scientists often start from “some-process-we-know-not-what” (Craver, 2007b, p. 114) that is specified by an input-output relation in order to find out which mechanism connects the inputs and the outputs. For example, protein synthesis may be characterized as whatever process that starts with DNA molecules being separated into two strands by the RNA polymerase and ends with there being new proteins; then the mechanism that realizes this input-output relation is found, which is then identified with protein synthesis. In other words, “protein synthesis” is a filler term denoting a black box in a mechanistic model involving the existence of newly produced proteins. This black box is filled by gathering more and more knowledge about the entities and activities that are engaged in protein synthesis (the mechanism that leads to there being new proteins). Protein synthesis, then, is identified with the mechanism. The relevant research question here is “What are the components of the process-we-don’t-know-yet?” The process-we-don’t-know-yet is referred to by a filler-term. Explanatory requests that ask for filler mechanistic explanations are “How does *filler term* work?” or “How are the outputs generated given the inputs?”. Given that filler terms are specified in terms of an input-output relation, the two questions are basically the same.

In dimensioned mechanistic explanations, derived from the behaving entity view, the explanandum is the behavior of an object that contains the mechanism. Here, the relevant question is “How does this object do THIS?” This kind of mechanistic explanation comes into play when we explain, for example, how muscles contract, how mice navigate the Morris Water maze, or how neurons fire. They inherit their name “dimensioned mechanistic explanation” from so-called dimensioned accounts of realization. Dimensioned views of realization are standardly contrasted with flat views of realization. The difference between these accounts is whether they take realization to be a relation between properties of one and the same individual (“flat”) or of two different individuals (“dimensioned”) (Endicott, 2011; Gillett, 2010). Since the explananda and the explanantia of dimensioned mechanistic explanation refer to (activities and properties of) different individuals, whereas filler mechanistic explanations identify the explanandum and the explanans, I call this version of mechanistic explanation “dimensioned”.

2.6 Constitutive Mechanistic Explanation: Different Ontological Implications

The new mechanistic approach does not only deliver an analysis of scientific explanation. In the context of the mechanistic framework, ontological questions are discussed that are directly related to the nature of mechanistic explanation. I want to highlight three:

  1. 1.

    Are the explanations of higher-level sciences reducible to the explanations of lower-level sciences?

  2. 2.

    What are levels?

  3. 3.

    Can there be causal relations between levels?

Constitutive mechanistic explanation is taken to be the explanatory type that is crucial for addressing these questions because mechanistic constitution is taken to be the relation that holds between levels. To answer the questions, thus, one needs a proper understanding of what constitutive explanation is. Since, as I have shown in the preceding sections, there are two variants of constitutive mechanistic explanation, the discussions of the three questions must pay attention to the different types of mechanistic explanation. In what follows, I highlight some implications of the different explanation types regrading questions (1), (2), and (3).

  1. 1.

    Constitutive mechanistic explanation and reduction

There are various different ways of asking the reduction-question, depending on (a) which relata one is interested in (e.g., disciplines, theories, explanations, entities), (b) what one takes the contrast to be (e.g., reduction vs. autonomy, reduction vs. integration), and (c) what one takes the relevant relation between the relata to be that justifies or blocks reduction (e.g., the impossibility of Nagel-reduction, multiple realization, impossibility of decomposition). Here, I will only briefly outline the different implications of filler and dimensioned mechanistic explanation regarding the question of whether phenomena are reducible to mechanisms.

There are two variants of reduction that need to be kept apart here: ontological reduction and epistemic reduction. The former concerns the question of whether the phenomenon de facto is nothing but the mechanism or not. The latter concerns the question of whether all knowledge about the phenomenon is nothing but or can be derived from knowledge about the mechanism. Regarding ontological reduction, the implications of filler mechanistic explanation and dimensioned mechanistic explanation are already clear. Since filler mechanistic explanation implies that the phenomenon is identical to the mechanism (see Sect. 2.3), and identity is usually taken to be sufficient for ontological reduction, filler mechanistic explanation implies that the phenomenon is ontologically reducible to the mechanism. In contrast to that, according to dimensioned constitutive explanation, phenomena are not ontologically reducible to their underlying mechanisms. Mechanisms, according to this view, are proper parts of phenomena. For example, the navigation mechanism involves only some but not all parts of the navigating mouse. Furthermore, the properties of the phenomenon are “qualitatively distinct” from the properties of the mechanism (see Sect. 2.4). Navigation mechanisms by themselves can’t navigate. Only the system as a whole of which the navigation mechanisms is a part of can do so.

Regarding epistemic reduction, Bechtel—focusing on disciplines rather than phenomena and mechanisms—highlights that higher levels provide information that the corresponding lower levels do not contain: namely organizational and contextual information (Bechtel, 2007, pp. 182–183). This, however, cannot be true for phenomena that are explained by filler mechanistic explanations. As Fazekas and Kertész argue, “[o]nce the required identity statements are in place one is able to infer to higher level processes from lower level knowledge” (2011, pp. 380–381) (see also Krickel (2018a, p. 117)). All there is to know about the phenomenon, based on filler mechanistic explanation, is the respective causal profile. Since the causal profile just is the causal profile of the underlying mechanism, all knowledge about the phenomenon is included in the knowledge about the mechanism. The situation is different for dimensioned mechanistic explanation. Here, the phenomenon is much richer than the mechanism. The phenomenon is a behavior of a system that contains more than just the mechanism. The same holds for the behavior of the system that is to be explained. For example, the mouse’s navigation behavior may de facto involve, say, the mouse being exhausted at a certain time. This is nothing we can derive from knowing the navigation mechanisms. Furthermore, the mouse’s navigation behavior occurs in a certain environment. Again, knowledge about the environment cannot be derived from the knowledge about the mechanism. It may even be that the navigating mechanism does exactly the same in two different environments.

In a nutshell: in filler mechanistic explanation, the phenomenon is ontologically and epistemically reducible to the mechanism. In dimensioned mechanistic explanations, the phenomenon is neither ontologically, nor epistemically reducible to the mechanism.

  1. 2.

    Constitutive mechanistic explanation and levels of mechanisms

The two different variants of constitutive mechanistic explanations have different implications regarding the interpretations of levels of mechanisms. According to the new mechanists, x is at a lower mechanistic level than y if and only if y is a component in the mechanism for y. In other words, the components of the mechanism for a phenomenon P are at a lower mechanistic level than P. If the phenomenon just is the mechanism, it follows that the entities and activities that make up a mechanism are at a lower level than the mechanism itself. That is, the relata of the level-relation, according to filler mechanistic explanation, are the mechanism and its components. For example, according to this view, the hippocampus generating spatial maps is at a lower level than the spatial navigation mechanism. And the opening ion-channel is at a lower level that the mechanism that is responsible for the propagation of the action potential along the axon.Footnote 4 Based on filler mechanistic explanation, mechanistic levels arise due to organization. A bunch of acting entities alone does not give rise to a new mechanistic level. Only if they are put into the right kind of temporal, spatial, and causal organization, they form a mechanism, and thereby create a new level.

Dimensioned mechanistic explanation provides a different picture. Here, the relata are the behaving system and the mechanism’s components. For example, according to this view, the hippocampus generating spatial maps is at a lower level than the navigating mouse. The ion channel opening is at a lower mechanistic level than the firing neuron. According to this picture, it is the containment-relation plus organization that gives rise to new level of mechanisms. A bunch of acting entities put into a certain organization and put into the context of a larger system (e.g., an organism) creates a new mechanistic level—because the larger system will show a new behavior that it would not be able to perform without the acting entities in that specific organization and that the acting entities in that organization could not do by themselves.

  1. 3.

    Constitutive mechanistic explanation and interlevel causation

The different pictures of mechanistic levels suggested by filler mechanistic explanation and dimensioned mechanistic explanation have different implications for the possibility and nature of interlevel causation. According to filler mechanistic explanation, interlevel causation would require that the mechanism as a whole would causally interact with its components. According to dimensioned mechanistic explanation, interlevel causation would hold between the behaving larger system and the mechanistic components.

At a first glance, interlevel causation between mechanistic levels is excluded for almost trivial reasons. A standard assumption about causation is that it relates events that occur at different times—i.e., causes are standardly assumed to precede their effects. On both views, however, the relata of mechanistic levels are wholes (mechanisms, larger behaving systems) and their parts (mechanistic components). If, however, one thinks of mechanisms and behaving systems as temporally extended things that have different temporal phases (Krickel, 2017), then this worry can be solved. One could think of interlevel causation as holding between the mechanism or the behaving system at tm and a mechanistic component at tn (where m ≠ n). On this assumption, interlevel causation in the filler mechanistic picture would, however, be identical to same-level causation. The reason is that each temporal phase of the mechanism just is the interaction of the mechanism’s components at the given time. For example, whatever the navigation mechanism does at ti—it just is what the components that make up the navigation mechanism at ti are doing at ti. Thus, the claim that the navigation mechanism at ti causes, say, the hippocampus’s activity at tj would be simply translated to the claim that the mechanistic components at ti cause the hippocampus’s activity. The different components of the navigation mechanism, however, are not at different levels of mechanisms. Thus, on the filler mechanistic picture, there is no interlevel causation in mechanism.

Again, dimensioned mechanistic explanation provides a different picture. Remember that we described causation between different level of mechanisms as a causal relation between a temporal phase of the higher-level at tm and temporal phase of the lower-level at tn. On this picture, this would be, for example, the navigation behavior of the mouse at tm and the hippocampus’s activity at tn. The temporal phases of the mouse’s navigation behavior are not just the interactions of the mechanistic components. These temporal phases are, for example, the mouse’s turning left, the mouse’s stopping, or the mouse’s running faster. It is in line with the overall picture to say that the mouse’s turning left at tm is at a higher level than the hippocampus activity at tn (for an argument in that direction see Krickel (2017)). Furthermore, prima facie, it makes sense to say that the mouse’s turning left at tm is a cause of the hippocampus’s activity at tn. Whether this is indeed true depends on what exactly one takes causation to be and, of course, on empirical facts.

In a nutshell: there cannot be interlevel causation between the phenomena and the mechanistic components as understood in filler mechanistic explanations. Dimensioned mechanistic explanation provides a picture of phenomena and mechanistic components that in principle could be causally related.

2.7 Conclusion

I have argued that there are indeed four different types of mechanistic explanation, two of which could be summarized under the label “etiological mechanistic explanation” and the other two as “constitutive mechanistic explanation”. This insight follows from taking a closer look at the different assumptions that have been made in the mechanistic literature on constitutive mechanistic explanation. Based on Krickel (2018a) and Kaiser and Krickel (2017), I have shown that there are three different types of explanation that might (mistakenly) all be subsumed under the label of constitutive mechanistic explanation: input-output mechanistic explanation, filler mechanistic explanation, and dimensioned mechanistic explanation. While all of these types of explanation can indeed be found in the life sciences, only the latter two exemplify the features that are standardly attributed to constitutive mechanistic explanation. Input-output mechanistic explanation, indeed, is a variant of etiological mechanistic explanation. Furthermore, I have shown that the two acceptable views of constitutive mechanistic explanation have different implications regarding reduction, levels of mechanism, and interlevel causation.