Matthews On Calcott_Mechanistic Explanation Without Mechanisms v3-1 Resaltado

8/2/2019 Matthews On Calcott_Mechanistic Explanation Without Mechanisms v3-1 Resaltado

1/21

Mechanistic Explanation without Mechanisms

John Matthewson and Brett Calcott

Dra Version .

Abstract

We provide an account of mechanistic representation and explanation thathasseveraladvantages overpreviousproposals. Inour view, explaining mechanis-tically is not simply giving an explanation of a mechanism. Rather, an expla-nation is mechanistic because of particular relations that hold between a me-chanical representation, or model, and the target of explanation. Under thisinterpretation, mechanistic explanation is possible even when the explanatory

target is not a mechanism. We argue that taking this view is not only coherentand plausible, it gives a more sophisticated view of the relationship betweenmechanical models and their targets. is allows us to address some ambi-guities within the mechanist framework, and delivers a more intuitive way tointerpret scientists use of the term mechanism.

I

is paper offers a new account of the relationships between mechanical models,

their targets, and mechanistic explanation. Central to this account is a clear distinc-

tion between models and what they represent, as well as recognition of the indirect

and flexible nature of this type of representation. A model can be explanatory, evenwhen the properties of the model and what it represents diverge: biologists who

model infinitely large populations are not suggesting that such infinite populations

exist, and chemists who model covalent bonds as springs are not suggesting tiny

springs hold atoms together. Nonetheless, such models represent their target sys-

tems in the right ways to provide explanatory power: they tell us how populations

might evolve, and why organic molecules deform in the way they do. In a similar

manner, we argue that a mechanical model can represent a target system that is not

a mechanism. In many scientific endeavours, the key question is not whether the

target system is a mechanism, but whether it can be usefully represented as though

it were a mechanism.

Our argument involves four central claims. We review what is it that makessomething in the worlda mechanism, clarify the properties of a model that make it

mechanical, and discuss the relation between model and target that makes an ex-

planation mechanistic. Finally, we argue that there are many more possible combi-

nations of target, model type, and explanation type than is usually acknowledged.

We begin by motivating our account, outlining some potential problems in the new

mechanistic philosophy that our framework can address.


2/21

Mechanistic Explanation without Mechanisms Dra Version .

. Background and Motivation

e new mechanistic philosophy describes a particular style of scientific thought

and explanation, particularly prevalent in the biological sciences (Machamer et al.,

; Bechtel and Abrahamsen, ; Glennan, ; Craver, ). Understand-

ing how scientists investigate mechanisms in the world, and the role that these play

in scientific theories and practice has greatly advanced our understanding of mod-

ern biological sciences. Despite this growing body of work on mechanisms, less

attention has been paid to the relationship between mechanisms and their represen-

tations (with some notable exceptions, for example (Glennan, ; Craver, )).

Stuart Glennan provides one plausible reason for this:

Perhaps because of the realist tendencies of the philosophers involved,

most of the literature has focussed on the properties of mechanismsthemselves and hasnot said much about the relationships between mech-

anisms and their models or theoretical representations (Glennan, ,

-).

In fact, it is sometimes difficult to see where in the literature an analysis of mech-

anisms in the world finishes and where (or if) a discussion of the representations

of mechanisms begins. For example, in their paper inking about Mechanisms,

Machamer, Darden and Craver (MDC) discuss the idea that different fields construct

mechanisms differently, according to their purposes (Machamer et al., , p).

Considering that this statement is made in the context of the discovery of biologi-

cal mechanisms, it seems unlikely that MDC literally mean that scientists constructthese mechanisms. Rather, scientists construct different representations of mecha-

nisms, according to their different purposes.

Furthermore, explanation is usually seen as very closely tied to, or even consti-

tuted by, the discovery of mechanisms in the world ( Darden, ; Wimsatt, ).

However, elsewhere in the new mechanist literature, Bechtel and Abrahamsen re-

mark that:

It is crucial to note that offering an explanation is still an epistemic ac-

tivity and that the mechanism in nature does not directly perform the

explanatory work [...] Since explanation is an epistemic activity, what

figures in it are not the mechanisms in the world, but representationsof them. (Bechtel and Abrahamsen, , p)

is ambiguity betweenmechanisms andtheir representations is problematic: when

discussing scientific explanation, it is essential to explicitly differentiate objects in

the world from their representations. Since much of the new mechanist position is

explicitly concerned with offering a view of how scientists explain (For example, see

Bechtel and Abrahamsen (); Glennan (); Craver ()), it is important

that the new mechanists are very clear about how these representations and mech-

anisms in the world are related, what actually does the explanatory work, and how


3/21


this bears on an analysis of mechanistic explanation. Our account makes all of thisexplicit.

e position we outline bears on another controversy: there is a tension be-

tween scientists use of the term mechanism and the core definitions of mecha-

nisms given in the philosophical literature (Skipper and Millstein, ). Scientists

talk of ecological mechanisms, economic mechanisms, and the mechanism of nat-

ural selection. But it is not clear that ecology, economics or natural selection are

composed of mechanisms as defined by the new mechanists.

ere are two obvious ways to remove this tension. We could shoe-horn these

systems into the existing definitions, or we could change the philosophical defini-

tion of mechanism so that it incorporates such systems. However, the first option

seems ad hoc, while the second would take us far from how one would normally un-derstand the word mechanism. Our account provides a way to avoid this dilemma;

it both makes sense of mechanism-talk and maintains a relatively demanding view

of the important properties required for something to be classified a mechanism.

M M

. Defining a Mechanism

We begin with a brief summary of what it takes to be a mechanism. We then turn

to the problem of defining a mechanical model.

Recent definitions for mechanisms come from a number of different authors:

A mechanism is a structure performing a function in virtue of its com-

ponent parts, component operations, and their organization. e or-

chestrated functioning of the mechanismis responsible for one or more

phenomena (Bechtel and Abrahamsen, , p).

Mechanisms are entities and activities organized such that they are pro-

ductive of regular changes from start or set-up to finish or termination

conditions (Machamer et al., , p).

A mechanism for a behavior is a complex system that produces that

behavior by the interaction of a number of parts, where the interactions

betweenparts can be characterized by direct, invariant, change-relating

generalizations (Glennan, , p).

ese definitions differ, but the differences are minor relative to the common ground

shared by the new mechanists. Here we provide a list of requirements that we

consider to be central to, and relatively uncontroversial within, the new mechanist

paradigm. We use two examples to help illustrate these core properties of a mech-

anism, one biological a myocyte (muscle cell), one manufactured the cooling

circuit of a refrigerator (see Figure ).

is was also a hot topic of discussion at the ISHPSSB conference, Exeter.e internal controversies are of course important in their own right (See Tabery()), but we


4/21


(a) Myocyte and Muscle Contraction (b) Fridge Cooling System

F : Two examples of Mechanisms.

. A mechanismdoes somethingof note. It underlies a behavior(Glennan, ),

performs a function (Bechtel and Abrahamsen, ), or produces some reg-

ularity (Machamer et al., ). Our examples both meet this criterion: my-

ocytes contract, shortening themselves in one or more dimensions, while

cooling circuits remove heat from a refrigerator cabinet.

. is behaviour/phenomenon/function is produced in virtue of the mecha-

nisms component parts or entities, and the processes, interactions, or activ-

ities these parts carry out. Myocytes are comprised of (amongst many other

things), actin fibers and myosin fibers, mitochondria and a cell membrane.

ese carry out the activities of interlocking/sliding, respiration and control-ling ion gradients respectively. A cooling circuit has a compressor (which

compresses gas into fluid), an evaporator (turns fluid into gas), tubing (re-

stricts and guides movement), refrigerant fluid (gains and dissipates heat en-

ergy) and fans (move heat away from the system).

. ese components and activities are organised in a particular way in order to

produce the behaviour or regularity. Actin and myosin must be arranged ap-

propriately so they can move relative to each other, while the cell membrane

must surround each of the fiber bundles to ensure rapid propagation of depo-

larisation. If the fan in a cooling circuit is positioned over the wrong section

of tubing, heat will not be removed from the system.

. Mechanisms canbe hierarchically composed (Bechtel andAbrahamsen,;

Craver, , ) A mechanism may be a component in some larger mech-

anism, and the components that make up a mechanism may themselves be

mechanisms. e mitochondria in a myocyte are mechanisms in themselves,

with their own parts, internal processes and organisation. Additionally, my-

ocytes are part of a larger mechanism - an organ such as a voluntary muscle

or the heart. And this organ is itself a part of an even larger mechanism: an

do not think they bear on the points made in this paper.


5/21


animal. Some of the cooling systems parts such as its fans are also mecha-

nisms, and the cooling system is also part of a more substantial mechanism;the refrigerator itself.

. Mechanisms are the kinds of things that scientists are able to make gener-

alisations about. So mechanisms are stable or persistent, producing regular

behaviour over time, rather than simply comprising the causal web of a po-

tentially one-off phenomenon.

An object that exhibits all of these properties is a mechanism in this context. ere

may be other properties necessary for something to paradigmatically be a mecha-

nism, such as a certain level of internal complexity, but we take the above list to form

a sufficient and relatively uncontroversial set of the required properties.

. Defining a Mechanical Model

Given this conception of mechanism, we can now consider the definition of a me-

chanical model. Our starting point for discussion regarding mechanical models is

Stuart Glennans paper Modeling Mechanisms (Glennan, ). Glennan defines

a mechanical model thus:

A mechanical model is (not surprisingly) a model ofa mechanism . . . (Glennan,

, p)

is innocent-sounding claim is problematic. According to Glennan, whether or

not a model is mechanical depends on what it is meant to represent. To understand

why this is a problem, we need to consider the indirect way in which models are

used to represent the world.

Like Glennan, we endorse a view of modeling due to Ronald Giere (), a

view that has recently been expanded by Peter Godfrey-Smith and Michael Weis-

berg, among others (Godfrey-Smith, ; Weisberg, , ). Under this ac-

count, modelsrepresent the worldin an indirect fashion, via two distinct steps. First,

the model is described; this delineates the properties that the model has. Second,

the model is deployed; relevant similarities are found between the properties of the

model and the thing in the worldof interest, the models target system. (SeeFigure .)

It is important to note that in this second step, the relationship between the

model and the target is one ofsimilarity, not one of perfect correspondence (Giere,

; Godfrey-Smith, ). It is always the case that certain properties of the target

are not included in the model just as an architects model of a building doesnt in-

clude all of the plumbing. And it may also be the case that some properties described

by the model do not hold in the world populations do not really have infinite size,

for example. Models are not exact representations of their targets; only similar to

them in particular ways. is flexibility of the similarity relationship is, as Glennan

notes, central to model-based science.


6/21


ModelSystem

ModelDescription

TargetSystem

SimilarityDescribes

Deploying the Model

F : Model-based science, adapted from Giere

. . . [the] relationship between a model and the mechanism it models is

one of similarity rather than isomorphic correspondence. is similar-

ity comes in varying degrees and respects . . . (Glennan, , p)

Together, an understandingof the model and the relevant similarity of the model

to some part of the world, gives the model power, albeit indirect, to describe, predict

or explain its target. Godfrey-Smith calls this approach a model-based strategy for

scientific theorising:

e modelers strategy is to gain understanding of a complex real-worldsystem via an understanding of simpler, hypothetical system that re-

sembles it in relevant respects. (Godfrey-Smith, )

ese two stages describing the model and deploying the model are quite

separate. Once a model has been constructed and its properties determined, one

can choose to apply the same model to different target systems, and the degree and

manner in which the model is similar to a target system is likely to differ from case

to case. A model of ball and spring behaviour can be applied to some actual balls

and actual springs, and here the similarity between the model and the target sys-

tem might be very marked. But demands on similarity must be relaxed if the same

model is intendedto represent chemical bonds, as there are verymanyways in whichchemical bonds are notat all like springs. So the kinds of similarity that make a given

model applicable may differ, depending on the target system and the intentions of

the scientist deploying the model (Giere, ).

Explicitly separating the two stages of model-based science identifies the prob-

lem with Glennans definition. According to Glennan, whether a model is mechan-

ical or not depends on the deployment stage on what target system the model is

intended to represent. Under this definition, the deployment of a model determines

whether or not the model is mechanical. Since the same model can be deployed to-

wards quite different target systems, under Glennans definition it is possible (at least


7/21


F : An orrery: a mechanical model of the solar system.

in principle) for a model to be mechanical when applied to one target, and then not

mechanical when applied to another. Additionally, under Glennans definition, to

call a model mechanical expresses almost no information about the model itself,

only about its current target. We find both of these outcomes to be very counterin-

tuitive. ere is, however, a simple way to avoid this problem.

We claim that mechanical models should be defined via the descriptive stage.

Here, one judges whether a model is mechanical according to the properties of themodel itself. Consider Figure , which depicts an orrery: a device that represents the

movements of the planets in our solar system. An orrery is clearly a mechanism. It

is also a model of our solar system. And both of these facts are true, regardless of

whether or not we think the solar system is a mechanism. Actually, they are true

even though the solar system is most likely nota mechanism under most plausible

definitions. So we can make perfect sense of a model being mechanical purely ac-

cording to the properties it possesses, rather than linking the definition to its target

system. In fact, we contend that this is the more natural way to think about it.

So we reject the idea that a model is mechanical simply because it is used to

represent a mechanism. Instead, a model is mechanical in virtue of the properties

that the model itself has. is raises the question of what properties a model musthave in order to be mechanical.

. Building Mechanisms

I never satisfy myself until I can make a mechanical model of a thing.

If I can make a mechanical model I can understand it (omson, ).

Although we disagree with how Glennan connects a models mechanical status to

its target system, he nonetheless captures something important by requiring that a


8/21


mechanical model must be capable of adequately representing a mechanism. We

can retain this important idea without tying ourselves to the target system as Glen-nan does. In our account, mechanical models are just those models that have the

properties definitive of a mechanism. If a model has the properties of a mechanism,

it will also have the properties required to represent a mechanism.

For example, in order to be a mechanism, an object must have an output be-

haviour and an identifiable structure that produces this behaviour. In the same way,

a mechanical model must have must have both behavioural and structural aspects.

Furthermore, just as in a mechanism, this structure must be comprised of interact-

ing parts that are organised in a particular way, and this organisation must exhibit

some amount of stability. A model that manifests these properties is a mechanical

model.

It may be that the reader finds this a difficult idea to swallow. Perhaps one thinksthat a model just isnt the kind of thing that can exhibit these properties. If this is

the case, then models such as the orrery will require some explaining away. As dis-

cussed above, it is a model, and it does have the above properties. So the orrery is

an existence proof that some models have the properties of a mechanism. Further-

more, scientists oen deal with concrete models, models capable of exhibiting these

kinds of properties, and a general account of modeling must allow for this.

However, we do not wish to limit our definition of mechanical models to in-

clude only those that, like the orrery, are actually realised as a concrete mechanism.

It should also be possible for models that are not realised as concrete objects to be

mechanical models. But now our definition might appear somewhat mysterious.

How can a non-concrete model exhibit such properties? Again, we have a simpleway to think of this. A model is mechanical if, were it to be realised in some con-

crete way, it would meet all the requirements definitive of a mechanism. In this

way, if one has qualms about models having properties such as internal structure or

stability, one can think of the (possible) realisation of such a model as having these

properties.

If the reader still has concerns about such things as a possible realiser of a

model being mechanical, we can put this in a final, more earthy way. A mechanical

model is a model that provides us with adequate resources to build a mechanism.

Roughly, if the model description provides something like the blueprint to build a

mechanism as defined above in Section ., then the model has the right properties

to make it a mechanical model.To summarise: a model is mechanical in virtue of the properties that the model

itself has, rather than the properties of its target. And the properties required for a

model to be mechanical are just those that are definitive of a mechanism. We can

make sense of this even in cases where the model itself is not some concrete object

by asking: if such a model were to be built, would it produce a mechanism?

See, for example, Griesemer () and Downes ().is idea has connections to recent work by Godfrey-Smith, where he argues that models can be

thought of as objects that wouldbe concrete, were they to be realised (Godfrey-Smith, ).


9/21


T R

We have argued that our definition of a mechanical model is a more natural way

to understand the concept. Now we show how this separation of model from tar-

get allows a number of representational possibilities that would not otherwise be

available.

Consider the orrery: it is an example of a mechanical model under our defini-

tion, and it is deployed to represent our solar system. It captures the relationships

between the relative movements of (some) of the planets and moons. So the be-

haviour that the mechanism produces the relative movement of the planets and

moons bears similarities to the target system. However, the similarity between

model and target ends when we examine how this behaviour is produced. Unlike

the orrery, the planets and moons do not move the way they do because of a series ofintermeshed cogs and gears. ere is simply no mapping between the components

in the orrery either their spatio-temporal organisation or their causally relevant

properties and the solar system. e orrery is a mechanical model, but it fails to

represent the target system in virtue of its mechanistic structure. is illustrates

two points. First, it is possible for mechanical models to represent targets that are

not clearly mechanisms. Second, because a mechanical model has both behavioural

and mechanical aspects (according to our list of requirements), it can be similar to

its target behaviourally and/or mechanistically.

Although Glennans definition precludes mechanical models representing non-

mechanical targets, he clearly recognises this second point, and outlines the differ-

ences between behavioural and mechanical similarity in (Glennan, ). He pro-vides a list of criteria by which to assess mechanical models, and divides these into

behavioural and mechanical aspects. A model is behaviourally similar to its target to

the extent that, given the same inputs, the models outputs are similar to the targets

outputs. In contrast, mechanical adequacy requires similarity regarding how the

model and targets input/output profiles are implemented, and here Glennan lists

a number of different features. Most importantly, mechanical adequacy requires

that the model contains analogues of the target mechanisms components. en the

similarity of these components to those in the target are assessed in two ways. First,

they should be organised both spatially and temporally in a similar manner to the

mechanism. Second, the causal roles that these components play within the model

should map to the causal roles they play within the target system.

Under our account, we now have three ways that a mechanical / non-mechanical

distinction can be made, each independent of the others:

Target ings in the world are either mechanisms or not (according to the proper-

ties of a mechanism outlined in section .).

For a similar analysis of the ways in which a model can represent its target, see Craver ()Glennan provides a number of additional requirements that we shall not discuss here for reasons

of space. We take it that these other requirements are not strictly necessary for mechanical adequacy,but may nevertheless improve a mechanical models representational adequacy.


10/21


Model Models can either be mechanical or not (according to the models proper-ties, as argued in the previous section).

Representation relation e similarities between a model and its target system

can be divided into behavioural and mechanical aspects (as noted above).

We differentiate these by saying that that a model can represent its target be-

haviourally and/or mechanistically.

Assuming that we are only interested in mechanical models, these options give four

possible combinations of model, target, and representation relation:

M Mechanical Target; Behavioural representation: A mechanical model that has

only behavioural similarities to a mechanism or class of mechanisms in theworld. A simple wind-up toy car fits this category. e toy car is a mechanism

and represents another mechanism a real car. But it only does this at the

behavioural level, perhaps by moving and turning like a car. Opening up such

a toy car to look at the wind-up mechanism inside will reveal very little about

how a real car engine works.

M Non-Mechanical Target; Behavioural representation: A mechanical model that

has behavioural similarities to a non-mechanical target system. e orrery

is an example of this combination. e solar system is not, or at least not

obviously, a mechanism. Yet we can represent its behaviour with a mechanical

model.

M Mechanical Target; Mechanistic representation: A mechanical model that repre-

sents a mechanism in virtue of its mechanical structure. We take it that the

paradigm cases of mechanistic representation (such as those in cell biology)

are of this type.

M Non-Mechanical Target; Mechanistic representation: A mechanical model that

represents something that is not, or not clearly, a mechanism, but nevertheless

does so in virtue of its mechanical structure. e existence of this category

suggests that just because something can be represented mechanistically, this

is not sufficient reason to infer that it is a mechanism.

e last of these (M) may appear surprising. But consider the model of ecolog-ical interactions in a lake shown in Figure . is is drawn from a standard ecology

textbook, which contains many similar diagrams. is model resembles a circuit

diagram (quite self-consciously, we suspect). And it certainly seems possible that,

given this diagram as a building schematic, we could construct some real-world ob-

ject from it. e resulting object would, like the orrery, produce an output behaviour

Note that these two types of similarity can be combined in the same model, and are graded: themodel isnt similar or not, but similar to a greater or lesser extent. We will treat the type of representa-tion as being binary to help make the relevant points clear, but things are certainly more complicatedthan this.


11/21


F : A model of primary production in a lake from Begon et al. (, p).

(e.g. reflecting the relation between the primary production input and sustainable

fish harvest output). And this behaviour is represented as produced by the vari-

ous components (broad trophic boxes), their organisation (wiring connections) and

the strength of these interactions (flow amounts). e constructed object would bea mechanism. Furthermore, it certainly appears as though these components and

their interactions are intended to be a central part of how the model represents the

lake. But a lake is not itself a mechanism. So M is employed in scientific prac-

tice, and an account of mechanistic representation should accommodate this. e

project for the remainder of the paperis to illustrate a way in which the model-target

interactions in M serve some purpose: they can be used for explanation.

M E

. e Interventionist Account of Explanation

A model might be similar in some interesting respect to its target system, yet fail to

explain that target system. A wind-up toy car is similar to a real car in many ways,

but not in the right ways to explain why a car needs fuel. So we need an account of

what makes a model similar to its target in the right way such that it explains that

target. For this, we adopt the interventionist account of explanation. (For example,

see Woodward (); Woodward andHitchcock (); Hitchcock and Woodward

())

We have several reasons for choosing the interventionist account. First, any ex-

position of mechanistic explanation should be level-agnostic. Since the mechanistic


12/21


sciences examine and explain phenomena at many different levels of organisation,

an account of mechanistic explanation ought to apply at all of these levels, ratherthan privileging only one. It should be able to accommodate explanation at both the

level of the cooling circuit and at the level of the fridge; the interventionist account

has the resources to do this. Second, mechanistically explaining a phenomenon cru-

cially depends on capturing the targets causal structure via the organisation of its

parts, and as we will see, the interventionist account is tailor-made for this. Lastly,

the interventionist framework is explicitly endorsed by at least some of the new

mechanists. Glennan has adopted James Woodwards notion of invariant relation-

ships under intervention into his definition of mechanisms, which had previously

appealed to laws, and Carl Craver also cites Woodwards account in his discussion

of what it takes for a model to be explanatory (Craver, ).

e interventionist account is well covered in the literature, so here we give justa brief reminder of the essential features. Explanations are characterised as patterns

of counterfactual dependence (Woodward and Hitchcock, , p). A target phe-

nomenon is explained by showing what it depends upon, and the structure of the

dependencies involved. is structure consists of a set of change-relating gener-

alisations that are invariant under interventions. ese interventions need not be

actually performed (or even performable in practical terms), but must be possible.

is structure of dependence gives us the resources to answer relevant counterfac-

tuals regarding the target system what Woodward calls what-if-things-had-been-

different questions (Woodward, ). Having these resources is, under the inter-

ventionist account, just what it is to have a causal explanation.

For a model to be explanatory, then, it must be similar (enough) to its target withrespect to the structure of the dependencies it represents, where these dependencies

are understood as generalisations that are invariant under interventions.

. Behavioural and Mechanistic Explanation

Just as a mechanical model can successfully represent its target behaviourally or

mechanistically, a mechanical model can also explain its target in virtue of either

or both of these kinds of representation. e application of a mechanical model to

some target system can therefore have two distinct explanatory goals:

. e model explains the behaviour of its target with respect to its external inputs

or environment. Here we know some of the causal dependencies underlying the

targets behaviour: namely, the external ones. If this is all the explanation gives us,

it might be referred to as a black-box explanation, and many would say it is not

a particularly good explanation. We will stay neutral on this issue, but we note

See Woodward (, p) for a discussion of what kind of possibility is meant.For example,Carl Craverappears to deny that a purely behaviouralmappingfrom model to target

is explanatory at all (Craver, , p). On closer inspection, however, he does not rule out the ideathat behavioural or phenomenal models can explain. Rather, Craver takes himself to be discussinga particular kind of explanation, and he does not think that this kind of explanation is exhaustive:Perhaps not all explanations are mechanistic. (Craver, , p).


13/21


that good or bad, it is not a mechanistic explanation. A mechanical model whose

similarity to the target system is only behavioural does not provide a mechanisticexplanation of its target. is kind of representation andexplanation corresponds

to the categories M and M.

. e model explains the behaviour of the target in virtue of its internal structure.

Here, the model reflects the causal dependencies within the target, and is in-

tended to represent how the target systems behaviour is produced. In this case

true mechanistic explanation is possible. is corresponds to categories M and

M.

How are we to understand the difference between behavioural and mechanistic ex-

planations withinthe interventionist framework? Recall that, according to this view,

an explanation provides us with the resources to know what would happen if the sys-tem were intervened upon. So at least one way to mark a difference between types of

explanation is due to a difference in the types of intervention the explanations deal

in.

Given some system such as a cell, a fridge, or an organism there is a nat-

ural distinction between the kinds of interventions that affect what is external to

the mechanism, and those that affect what is internal to it. Consider the following

two situations. First, we know how a system will behave when its environment is al-

tered in some crucial respects. Second, we know how the system will behave when

we remove, ablate, or otherwise interfere with one or more of its components. In

both cases we have the resources for answering what-if-things-had-been-different

questions, yet we can divide them into two categories: interventions on the external

conditions in which the mechanism functions, or interventions on the components

or internal organisation of the mechanism.

We say that this is a natural distinction, which is not the same as saying it

will always be clear-cut. It is likely that there will be cases where it is not obvious

whetheran intervention is internal or external, as we suspectthe division to be vague

at its boundary, as well as this boundary potentially being perspective-dependant.

However, there will also certainly be cases where it is obvious which category the

intervention falls into, and we think that these will be the majority of cases.

If the model is only similar to the target system with respectto external interven-

tions interventions on such things as environmental inputs or starting conditions

then it only explains the target behaviourally. In contrast, if the model is deployed

to explain how the behaviour of the target system is produced, then the model must

inform us what changes will result from internal interventions changes in the or-

ganisation of components, or in the behaviour of the components of the mechanism.

Since mechanistic explanation is concerned with this second category of interven-

tions, a mechanistic explanation must provide insight into the truth of a particular

kind of counterfactual: those that refer to the internal causal structure of the target

system.


14/21


. e Modularity of Parts

So far we have identified that mechanistic explanation involves knowledge of gener-

alisations that are invariant under interventions internal to the system that produces

the phenomenon of interest. But this is not sufficient. It is distinctive of mechanisms

that their behaviour is produced by the interactions between their parts. So an ac-

count of mechanistic explanation requires a corresponding account of part-hood.

Here we endorse an aspect of Woodwards framework that can be directly associ-

ated with part-hood: the idea ofmodularity. Surprisingly, although Woodward

himself () thinks modularity has a central role to play in an account of mech-

anisms, this aspect of his position has been largely neglected. For although Wood-

wards framework regarding explanation is oen endorsed by the mechanists, less

attention appears to have been paid to his account of mechanisms.

Modularity, rather than referring to the invarianceof aparticulargeneralisation,

refers to invariance between generalisations. A system of related generalisations is

modular if it is possible to change one generalisation without affecting any of the

others. is captures the intuition that, if we have properly identified the parts

underlying the operation of a mechanism, then changes to those parts should affect

the system only through their predefined interactions with other parts. Modularity

ensures that each causal generalisation is atomic: a separable part within some larger

system.

Incorporating modularity onto our framework, a mechanical model explains

mechanistically if it shows how interventions on the targets internal structure will

alter the targets behaviour, and does this by representing the independence of the

mechanisms causally relevant parts in a modular fashion.

is is, we note, very close to Woodwards account of a mechanism. But al-

though the stated goal in his (Woodward, ) is to give an account of what a

mechanism is, his definition is in fact an account of a mechanical model:

...a necessary condition for a representation to be an acceptable model

of a mechanism is that the representation (i) describe an organized or

structured set of parts or components, where (ii) the behavior of each

component is described by a generalization that is invariant under in-

terventions, and where (iii) the generalizations governing each com-

ponent are also independently changeable, and where (iv) the repre-

sentation allows us to see how, in virtue of (i), (ii) and (iii), the overalloutput of the mechanism will vary under manipulation of the input to

each component and changes in the components themselves. [our ital-

ics] (Woodward, , pS)

Given this, we fully agree with Woodward. But again, we emphasize that it is es-

sential to distinguish this from an account of mechanisms in the world. For exam-

ple, the modularity account of part-hood might not be straight-forwardly endorsed

by the new mechanists. Usually, parts are understood in the literature in terms of

See also the discussion of autonomy in Pearl (, p-).


15/21


spatio-temporal delineation and separation. But notice that this is a requirement

regarding mechanisms, not representations. is is where the differences betweenmechanisms, mechanical models, and what is required for mechanistic explanation

are most salient. Having clearly spatio-temporally separable parts is (or at least may

be) important for something to be a mechanism. But under an interventionist ac-

count, explanatory knowledge is knowledge of modular counterfactual invariances,

and counterfactual invariance and modularity do not require spatio-temporal sep-

aration.

If two structures are spatio-temporally separate, it is almost certain that any

causal processes they exert will be modular, but the converse is not true. Causally

separable processes may arise from structures that are spatio-temporally overlap-

ping, or even coincident. Two populations of fish from different species may spatio-

temporally overlap, but it still might be the case that we can interveneon one popula-tion without altering generalisations regarding the other population. So the require-

ments of part-hood needed for something to be a mechanism and for something to

explain mechanistically can come apart.

ere is a final important point to be made regarding how a mechanical model

may represent the modularity of parts. If we wish our model to express modular-

ity, this will occur by showing the causally relevant parts as separate. In the setting

of a mathematical model, this will simply be through independence of the equa-

tions that express the different generalisations. If, on the other hand, the model is

expressed in diagrammatic form, these modular parts will be depicted as spatially

separate. Spatial separation in the modelmay be used to represent genuine causal

independence, rather than representing the clearly delineated location of parts.is is a significant dissimilarity between model and target spatial separation

standing in for causal separation but it is causal separation that is the important

property for mechanistic explanation. So for mechanistic explanation, it need not

be the case that parts in the model map to actual spatio-temporally isolated parts in

the target. It is sufficient that the parts in the model tell us about how the targets

internal causal structure is organised.

In summary, for a mechanical model to explain mechanistically, it must be the

case that the structure of the model the way its parts are organised and connected

reflects the causal structure of the target system with respect to internal interven-

tions. Part-hood is understood in terms of modularity of the represented causal

dependencies, and will oen be expressed in the modelas being spatio-temporallydistinct, even if they arent so in the target. If the mechanical model is similar to its

target in these specific ways, the model explains its target mechanistically.

Glennan also claims that some mechanisms include only functionally defined parts (Glennan,, p).


16/21


Population of O'sVariation and

Heritability of T

DifferentialRepresentation

of form of Tamongst Os

Adaptation ofLineage with

respect to T in E

Predominance ofof Form of Tamongst Os

Os interact withE differently via T

DifferentialSurvival

rates of Os

DifferentialReproductiverates of Os

E change orMutate T

Cycle ofSelection

Key:O = OrganismsT = TraitE = Environment

F :e Mechanism of Natural Selection. Modified from Skipper and Mill-stein ().

M

Our arguments thus far have sought to establish a more complex, but ultimately

more intuitive and useful way to categorise the relationship between mechanical

models and the target systems they represent. In this section we examine a promi-

nent case in the literature on mechanisms in order to motivate the ideas that category

M is employed in the sciences, and that models deployed in this manner can be ex-

planatorily useful. In what follows, we review the arguments given by Skipper andMillstein in their paper inking about Evolutionary Mechanisms: Natural

Selection. is paper questions whether or not natural selection can be classified as

a mechanism according to the current philosophical accounts.

e structure of the paper is as follows: Skipper and Millstein first note that

scientists oen refer to natural selection as a mechanism. ey infer from this that

...there is no question that contemporary evolutionary biology exem-

plifies the view that natural selection is a mechanism(Skipper and Mill-

stein, , p).

ey then construct a mechanistic representation of natural selection, (see Fig-

ure ), that ...sets out the mechanism as a chain of temporal steps or stages layingbare the causal crux of natural selection.(p) e plan is to utilise this model in

order to assess whether the two most prominent accounts of mechanisms due to

MDC and Glennan can endorse natural selection as a mechanism (Machamer et al.,

; Glennan, ). Skipper and Millstein argue that, for a number of reasons,

natural selection does not meet the required criteria for either of these accounts of

mechanisms. ey conclude that neither framework can endorse natural selection

as a mechanism, and thus these philosophical accounts of mechanisms are inade-

quate.

For the sake of exposition, we shall accept Skipper and Millsteins arguments


17/21


that the central definitions of mechanisms exclude natural selection. As suggested

in our introduction, this tension between scientists usage and philosophers defini-tions appears to leave the new mechanists with two options. On the one hand, they

can accept the claim that natural selection is a mechanism, from which it follows

that their standard definitions of mechanism are incorrect. On the other, they can

deny that natural selection is a mechanism, in which case evolutionary biologists

referral to natural selection as a mechanism must be accounted for. Faced with this

choice expand the mechanistic picture, or reject natural selection as a mechanism

Skipper and Millstein opt for expansion. Unfortunately, it seems to us that any

definition of mechanisms that admits natural selection in a straightforward manner

is in real danger of subsuming many other things that do not fit well with the usual

understanding of what mechanisms are.

e position we have argued for provides an alternative option. Scientists canusefully treat natural selection as though it is a mechanism, without being commit-

ted to it beinga mechanism. And they do this for the purposes of explanation. is

means that when scientists discuss the mechanism of natural selection, they may

be referring to a mechanical modelof natural selection.

is idea would not be tenable if one held a position such as Glennans, which

does notallowfor mechanicalmodels to represent non-mechanisms. We have shown

that this is not the only option: a mechanical model can be used without the target

in question necessarily being a mechanism. According to our account, the impor-

tant question is not whether natural selection is actually a mechanism, but whether

natural selection can be usefully modeled as though it were a mechanism.

How should we understand a that we can have a mechanical model of naturalselection, where natural selection is not (clearly) a mechanism? Interestingly, this

is preciselySkipper and Millsteins situation. eir model qualifies as mechanical

model under our definition. It has both an output behaviour and an identifiable

underlying causal structure that produces that behaviour. We can imagine building

a natural selection machine based on this model, where the connections between

the stages describe the order and progression of processes on a substrate. Indeed,

there are some clear similarities with the model used to describe the cooling circuit

in Figure (b). It is just that in the case of natural selection, we have a population of

organisms rather than a volume of fluid, and the population undergoes genotypic

and phenotypic change instead of changes in state.

But since their model is mechanical and the target is not, there must be someways in which model and target are dissimilar. ese dissimilarities are, indeed, the

reason whySkipper and Millstein do not think that natural selection itself qualifies

as a mechanism. Notably, the model is dissimilar to the target system in the way it

represents the discreteness, location andspatio-temporal organisation of its parts, so

it is difficult to see how the models parts map to anything spatio-temporally similar

in the target system.

Failures of representational adequacy such as this certainly should be of con-

cern if the model is intended to represent natural selection as a mechanism in some

strict fashion. So Skipper and Millstein may be right to conclude from this (amongst


18/21


other reasons) that natural selection is not a mechanism. However, if the model

is intended to explain (part of) how natural selection works, the requirements onrepresentational adequacy are different. In the previous section, we argued that a

mechanical model does not have to be similar to its target with respect to spatio-

temporal organisation in order to mechanistically explain the behaviour of that tar-

get. Although the model of natural selection does not depict the spatio-temporal

properties of natural selection veridically, it may be that the organisation of the com-

ponents in the model is similar enough, or in the right ways in order for it to be

explanatory.

For this to be thecase, the model in Figure must show howadaptive change in a

lineage is affected by interventions on its components or their organisation. In brief,

it must provide us with the resources for answering particular what-if-things-had-

been-different questions. For example, can we change the generalisation governinghow organisms (O) interact with the environment (E) via the trait (T) without af-

fecting the other components? We think it might but notice that this is an em-

pirical question; it has nothing to do with what kinds of models can represent what

kinds of target. Our account provides a way to assess whether or not this model

explains mechanistically, and this assessment will depend on the biology not our

definitions.

To summarise, in order to make sense of scientists talk of the mechanism of

natural selection, we neither need to expand the concept of what a mechanism is,

nor do we need to squeeze natural selection into the existing framework. What must

be addressed is whether or not natural selection can be represented as though it is

a mechanism and whether or not this has some explanatory payoff. As remarkedearlier, when population geneticists model populations as though they are infinitely

large, we dont need there to be any actual infinite populations in order to make sense

of their talk. Similarly, it need not be the case that natural selection is a mechanism

for this mechanical model to be explanatory.

S

Whether a model is mechanical depends on the models properties, not on those of

its target. Furthermore, mechanistic explanation should be viewed as having three

principle parts: e model and its properties, the thing in the world that is being

explained and its properties, and the similarity relations between this model andthis target. For a model to explain its target mechanistically, there must be sufficient

similarity between the model and targets behavioural profiles and the components

that produce that behaviour. But they dont have to be similar in all ways, only in

the right ways. Localisation in the model might map to localisation in some real

world mechanism, or it might correspond to something less obvious. At minimum,

for the internal structure in the model to be explanatory, it must provide us with

counterfactual information about the internal causal organisation or structure of

its target. e interventionist account, including Woodwards modularity criterion


19/21


provides a way of doing this.

If we are to make sense of mechanism talk by scientists, we think it is impor-tant to recognise that at least some of the time they may be speaking about model

mechanisms. We have shown how adopting our position gives us a way to do this. It

is coherent and useful for scientists to treat some systems as though they are mech-

anisms, without necessarily being committed to them actuallybeing mechanisms.

Although we have explicitly addressed the problem of whether natural selection

can be usefully thought of as though it were a mechanism, we think that the gen-

eral structure of our argument will show that many systems that are not obviously

mechanisms may well be represented by mechanical models for explanatory pur-

poses. Such cases are not hard to find: recall Figure from the ecology textbook,

and there are plenty of cases in the social sciences. All that is required is that the

mechanical details in the model (the components, processes and their causal or-ganisation) map to relevant counterfactuals regarding internal interventions on the

target. is is sufficient for a mechanical model to be mechanistically explanatory,

even when the target is not clearly a mechanism. Finally, this also indicates that

just because something can be represented by a mechanical model is not sufficient

reason to infer that it itself is a mechanism.

All of this raises a further question. Under our framework, what are the mech-

anistic sciences? Only the ones that deal in bona fide mechanisms, or the ones that

deal in mechanical models? According to our account, this becomes a matter of

definition. Perhaps the term mechanistic science should be constrained to those

disciplines that deal in veritable mechanisms-in-the-world. In that case, it will still

be worthwhile for philosophers to pay attention to the scientific projects that dealin as-though-it-were-a-mechanism explanations. And there are certainly many of

these. No matter where the label ends up being applied, we take it that the use of

models, and the indirect way in which they represent the world, allows for a num-

ber of different possible combinations of model, target and representation relation,

each of which might offer some kind of payoff given a particular explanatory goal.

Once we see that these possibilities are available, we can make sense of scientists

mechanism talk without lumbering them with an ontology we suspect they are

neither aware of, nor need commit themselves to, in order to make real explanatory

progress.

R

Bechtel, W. and Abrahamsen, A. (). Explanation: a mechanist alternative. Stud-

ies in History and Philosophy of Science Part C: Studies in History and Philosophy

of Biological and Biomedical Sciences, ():.

Begon, M., Harper, J. L., and Townsend, C. R. (). Ecology : Individuals, Popula-

tions and Communities. Blackwell Science, Oxford.

Craver, C. (). Beyond reduction: Mechanisms, multifield integration and the


20/21


unity of neuroscience. Studies in History and Philosophy of Science Part C: Studies

in History and Philosophy of Biological and Biomedical Sciences, ():.

Craver, C. (). When mechanistic models explain. Synthese, :.

Darden, L. (). Strategies for discovering mechanisms: Schema instantia-

tion, modular subassembly, forward/backward chaining. Philosophy of Science,

():SS.

Downes, S. (). e importance of models in theorizing: A deflationary semantic

view. PSA: Proceedings of the Biennial Meeting of the Philosophy, pages .

Giere, R. (). Explaining Science: A cognitive approach. University of Chicago

Press, Chicago.Giere, R. (). e nature and function of models. Behavioral and Brain Sciences,

():.

Glennan, S. (). Rethinking mechanistic explanation. Philosophy of Science,

:SS.

Glennan, S. (). Modeling mechanisms. Studies in History and Philosophy of

Science Part C: Studies in History and Philosophy of Biological and Biomedical Sci-

ences, ():.

Godfrey-Smith, P. (). e strategy of model-based science. Biology and Philos-

ophy, :.

Griesemer, J. (). Modeling in the museum: On the role of remnant models in

the work of Joseph Grinnell. Biology and Philosophy, :.

Hitchcock, C. and Woodward, J. (). Explanatory generalizations, part II:

Plumbing explanatory depth. Nous, ():.

Machamer, P., Darden, L., and Craver, C. (). inking about mechanisms.

Philosophy of Science, ():.

Pearl, J. (). Causality : Models, Reasoning, and Inference. Cambridge University

Press, Cambridge.

Skipper, R. and Millstein, R. (). inking about evolutionary mechanisms: Nat-

ural selection. Studies in History and Philosophy of Science Part C: Studies in His-

tory and Philosophy of Biological and Biomedical Sciences, ():.

Tabery, J. (). Synthesizing activities and interactions in the concept of a mech-

anism. Philosophy of Science, ():.

omson, W. (). Kelvins Baltimore Lectures and Modern eoretical Physics.

Edited by R. Kargon and P. Achinstein. Cambridge, Mass.: MIT Press.


21/21


Weisberg, M. (). Qualitative theory and chemical explanation. Philosophy of

Science, ():.

Weisberg, M. (). Who is a modeler? British Journal for the Philosophy of Science,

():.

Wimsatt, W. C. (). Complexity and organization. PSA: Proceedings of the Bien-

nial Meeting of the Philosophy of Science Association, :.

Woodward, J. (). What is a mechanism? a counterfactual account. Philosophy

of Science, ():SS.

Woodward, J. (). Making ings Happen: A eory of Causal Explanation . Ox-

ford University Press, New York.Woodward, J. andHitchcock, C. (). Explanatory generalizations, Part I: A coun-

terfactual account. Nous, ():.

Matthews On Calcott_Mechanistic Explanation Without Mechanisms v3-1 Resaltado

Documents