ArticlePDF Available

What is a mechanism? Thinking about mechanisms across the sciences



After a decade of intense debate about mechanisms, there is still no consensus characterization. In this paper we argue for a characterization that applies widely to mechanisms across the sciences. We examine and defend our disagreements with the major current contenders for characterizations of mechanisms. Ultimately, we indicate that the major contenders can all sign up to our characterization. KeywordsMechanism–Explanation–MDC–Glennan–Bechtel–Astrophysical mechanism
What is a Mechanism? Thinking about mechanisms across the sciences
Phyllis McKay Illari and Jon Williamson
After a decade of intense debate about mechanisms, there is still no consensus characterization. In
this paper we argue for a characterization that applies widely to mechanisms across the sciences. We
examine and defend our disagreements with the major current contenders for characterizations of
mechanisms. Ultimately, we indicate that the major contenders can all sign up to our characterization.
Mechanism, explanation, MDC, Glennan, Bechtel, astrophysical mechanism
1 Introduction
Since Bechtel and Richardson’s 1993 book, there has been nearly two decades of debate
on the right characterisation of a mechanism, intensifying since Machamer, Darden and
Craver’s (MDC’s) controversial 2000 paper. The main contenders are:
Machamer, Darden and Craver: ‘Mechanisms are entities and activities organized such that they are
productive of regular changes from start or set-up to finish or termination conditions.’ (Machamer,
Darden and Craver 2000 p3.)
Glennan: ‘A mechanism for a behavior is a complex system that produces that behavior by the
interaction of a number of parts, where the interactions between parts can be characterized by direct,
invariant, change-relating generalizations.’ (Glennan 2002b pS344.)
Bechtel and Abrahamsen: ‘A mechanism is a structure performing a function in virtue of its
component parts, component operations, and their organization. The orchestrated functioning of the
mechanism is responsible for one or more phenomena.’ (Bechtel and Abrahamsen 2005 p423.)
After small changes of detail (see Bechtel and Richardson’s original 1993, Glennan’s
original 1996, Machamer 2004, Craver 2007, and Glennan 2011), these broad
characterisations remain in use by their original advocate(s), and many others.
In this paper, we will defend a characterization that gives an understanding of what is
common to mechanisms in all fields. We disagree with elements of all of the major
characterizations above, and argue for:
A mechanism for a phenomenon consists of entities and activities organized in
such a way that they are responsible for the phenomenon.
This project is important for two reasons. First, it is important to the broad question of
whether or not scientific method is disunified (see Glennan 2010). Different scientific
disciplines share many methodological concerns, including causal explanation, causal
We lack space to discuss the work of everyone in the debate in detail, but we will also make some
comments on the work of Tabery, Torres, and Woodward on the way through the paper.
inference and causal modelling, which commonly use mechanisms. It is our contention
that we have produced a widely applicable understanding of mechanisms, that is of use in
understanding what these different disciplines share, methodologically. This is
complementary to the alternative project of describing what is distinctive about the kinds
of mechanisms used in a particular domain. (See Steel; and Torres p240 for
methodological disagreement.) Surface differences are methodologically important, but
shouldn't be allowed to obscure what is common. Indeed, we cannot properly understand
the differences without also seeing the similarities. We offer what is common to
mechanisms, which different fields can flesh out with their distinctive methodological
Second, these particular methodological debates and others need a consensus account of
mechanisms. Philosophers and scientists are attempting to use mechanisms to illuminate
causal explanation, inference and modelling, as well as the metaphysics of causality (see
Glennan 1996; Steel 2008; Leuridan and Weber 2011; Broadbent 2011; Gillies 2011).
These debates are impeded by lack of a consensus account, in spite of a great deal of
consensus now existing within the mechanisms literature. To develop an understanding
of the problems of causal explanation, inference and modelling that the sciences share, it
is vital to understand what is common in the use of mechanisms across the sciences. The
problems shared by different fields are just as important to recognise as the
methodological differences (see Glennan 2005 p462). Many mechanistic explanations
are built using components from multiple fields (see Craver 2007, Russo 2009, Illari and
Williamson 2010). Debate on using mechanisms in causal inference includes both
biomedical and social sciences (see for example Steel 2008; Gillies 2011). Such
examples strongly indicate that mechanisms in general share a great deal. Finally, if
there is no widely applicable account of mechanisms, there is no possibility of a widely
applicable mechanistic approach to the metaphysics of causality, so our work is also of
interest to that debate (Williamson 2011). We will assist all these debates by developing
a consensus account of a mechanism that they can use.
We are interested in mechanisms themselves. As Craver claims, there is a sense of ontic
explanation: mechanisms explain phenomena in the sense that their presence produces
the phenomenon (2007 pp27-8). But epistemic explanation is also important, as Bechtel
claims, where the description of the mechanism explains the phenomenon (2008 p16).
But both ontic and epistemic mechanistic explanation require real mechanisms. Bechtel
and Abrahamsen write: ‘mechanisms are real systems in nature’ (2005 p424-5), and
Bechtel agrees that epistemic explanation is parasitic on there being real mechanisms in
the world to describe (private communication). So Bechtel and Craver can hold, with us,
that examining mechanistic explanation tells you about mechanisms themselves, and so
we will move freely between claims about mechanistic explanation and claims about
mechanisms themselves.
Although our characterization is close to MDC's and Craver's, in the next section, S2, we
explain why we do not include certain elements of the current characterizations. In S3
we defend our characterization. We will show that, correctly understood, it applies to the
mechanisms that scientists discover and use in explanation and causal inference. Existing
accounts of mechanisms have been developed in the light of the biomedical sciences
(MDC, 2000) and psychology (Craver, 2007; Bechtel, 2008). We will use astrophysical
mechanisms to demonstrate the wide applicability of our account. In S4 we will take up
the question of what is not a mechanism on this account. In S5 we conclude. While our
characterization of mechanisms differs from those of other accounts, we see our project
as consistent with those of the main contenders, and we briefly indicate why we think
they should have no serious objections to our account.
2 How not to characterize mechanisms
With broad applicability in mind, we do not characterise a mechanism as a structure
(Bechtel and Abrahamsen) or a system (Glennan). Unless read so weakly as to mean
almost nothing, the idea of structure implies some level of inflexibility. This seems at
odds with Bechtel's latest work (Bechtel and Wright 2009; Bechtel 2010, 2008; but
compare Bechtel 2007 p275). A system is more dynamic and more flexible than a
structure, but still implies a level of internal coherence that not all mechanisms show. As
Darden notes (2006 p281), some mechanisms make their own entities as they go, such as
the mechanism of protein synthesis where mRNA is made when needed and broken down
afterwards. Further, many mechanisms are complex, but they can also be simple.
mechanism of thermal dissociation of the diatomic iodine molecule in the vapour phase
seems too simple to be called either a system or a structure. The stretching vibration just
gets more and more energetic until there is enough energy to rupture the bond between
two atoms, and they fly apart.
Unsurprisingly, astrophysical mechanisms are often relatively stable and structured. But
violent sudden change from an existing structure or system to a different one is also
possible, as with supernovae. Thus even for astrophysical mechanisms, it is best to avoid
‘structure’ or ‘system’ in the characterization of a mechanism.
Glennan is initially committed to all mechanisms being systems (2002b p128, p129;
2009a). In Glennan (2009b p323) he allows that there is no 'mechanism qua system' for a
baseball breaking a window. In Glennan (2010, see especially pp260-1) he develops an
account of 'ephemeral mechanisms', where the configuration of parts isn't stable, as it is
in a system. In Glennan (2008, see especially p283) he calls for an account of a possible
third kind, emergent mechanisms, for cases where phenomena produced by mechanisms
depend on the properties of and relations between their parts, but standard mechanistic
strategies such as functional localization are not very successful. We agree that these are
all mechanisms, but are inclined to treat these differences as positions on a continuum,
not differences in kinds of mechanism.
We also drop MDC’s ‘start or set-up’ or ‘finish or termination conditions’. Craver drops
this without explanation (2007), while Darden (2006) and Machamer (2004) retain it.
This element can be read very lightly, but it is better removed, because 'start' and 'finish'
For this reason, we do not adopt Torres (p247). At 'Mechanisms and Causality' conference, Kent, 2009,
Glennan clearly withdrew ‘complex’ from his characterization.
conditions are not even an aspect of all of our mechanism descriptions, far less of all
mechanisms. They are pragmatic aspects of the descriptions we give of some
mechanisms – but not all. Cell mechanisms such as the Krebs cycle are cyclical. They
are continuing, having no real start or end. Bechtel (2009) notes this for other
mechanisms. For continuing mechanisms, understanding that there is no tidy start or end
is very important. Further, even some mechanisms that are neatly described in terms of
start and finish conditions do not themselves have start and finish conditions. So while
some mechanisms might have a natural descriptive starting point – we might start the
description of the formation of stars with the gravitational accretion of dark matter in a
halo – we should not enforce a start and end-point with a requirement in the
characterization of a mechanism.
We follow MDC and Bechtel in not requiring modularity. Dynamical systems and
systems biology explanations are precisely aimed at describing systems that are largely
non-modular, and we do not wish to rule them out as mechanisms. It may appear that we
disagree with Woodward. However, on closer examination, Woodward is talking only
about representations of mechanisms: ‘(MECH) 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 interventions, and
where (iii) the generalizations governing each component are also independently
changeable, and where (iv) the representation allows us to see how, in virtue of (i), (ii)
and (iii), the overall output of the mechanism will vary under manipulation of the input to
each component and changes in the components themselves.’ (Woodward S375,
emphasis added. Compare Darden 2006 p279.) But Woodward is clear here that he is
concerned with representations or models, not mechanisms themselves. We agree that
our representations or models of mechanisms should be modular as far as possible. Such
a representation will certainly make prediction and intervention easier. But where this is
not possible, a non-modular representation will have to do. Neither all mechanisms
themselves, nor all mechanism descriptions, will be modular. Since Woodward’s
primary concern is representations, not mechanisms themselves, we will put his views
aside for the rest of the paper. Thus, we do not use Woodward's ideas in the
characterization of a mechanism itself.
We will move on now to defending our positive characterization of a mechanism. This is
our characterization of the consensus elements of preceding accounts. Here,
disagreements are more subtle, but they are important if the characterization is to be
widely applicable and so useful to other debates on method or metaphysics.
3 Our characterization of mechanisms
All mechanistic explanations begin with (a) the identification of a phenomenon or some
phenomena to be explained, (b) proceed by decomposition into the entities and activities
relevant to the phenomenon, and (c) give the organization of entities and activities by
which they produce the phenomenon. (See Darden 2006, Bechtel and Abrahamsen 2008.)
Mechanism discovery is often messy and iterative, but always involves finding out about
these three elements.
This is widely known, so all mechanisms share the three elements found in the process of
mechanism discovery. Even astrophysical mechanisms are grouped by the phenomena
they produce. Scientists aim to give a detailed account of how the phenomenon is
produced by entities and activities. Entities include both massive bodies such as stars and
galaxies, and fundamental particles such as quarks, photons, neutrons and neutrinos;
while activities tend to involve movement and energy changes. Organization is vital:
threshold effects are common, and feedback effects, often associated with biological
mechanisms, are not uncommon. Background theory, particularly General Relativity, is
important to astrophysical mechanisms in a way not paralleled by all mechanisms,
because relevant organization can include details of background space-time geometry.
Our favoured characterisation captures the core consensus at the heart of the views of the
main contenders:
A mechanism for a phenomenon consists of entities and activities organized in such a
way that they are responsible for the phenomenon.
In the following subsections, we take each of the three elements here and argue for them:
responsible for the phenomenon (S3a), entities and activities (S3b), and organization
(S3c). We have explored entities, activities and organization in the context of protein
synthesis and natural selection in more detail elsewhere (Illari and Williamson 2010).
3a Responsible for the phenomenon
There are three reasons why we follow Bechtel and Abrahamsen (2005 p422) in saying
mechanisms are 'responsible for a phenomenon'. The first reason is the importance of the
phenomenon for mechanistic explanation. Mechanistic explanation succeeds when the
mechanism discovered and described is the mechanism responsible for the phenomenon.
If no unified mechanism can be found for that phenomenon, the phenomenon is
redescribed to make it susceptible of mechanistic explanation – what Bechtel and
Richardson call ‘reconstituting the phenomenon’ (1993). This is to say that mechanisms
are functionally individuated by their phenomena.
However, we avoid Bechtel and
Abrahamsen’s ‘performing a function’ in our characterization. In wider philosophical
and scientific debate, ‘function’ is a loaded concept, usually involving deliberate design
or natural selection, while the function of a mechanism requires only something like
At least partially. There seem to be other ways to individuate mechanisms that produce the same
phenomenon, such as in terms of the entities or activities involved, and an examination of whether such
ways can always be explained away in terms of functional individuation is a complex issue we reserve for
further work. At 'Mechanisms and Causality' conference, Kent 2009, both Darden and Craver called the
functional individuation of mechanisms ‘Glennan’s Law’, as he was the first to recognise this (see for
example his 1996).
‘characteristic activity’. We therefore avoid using ‘function’ explicitly in the
characterization of the mechanism.
This point is important in the case of astrophysical mechanisms, which involve neither
deliberate design, nor natural selection, but do have characteristic activities. Even in the
absence of natural selection or deliberate design, spectacular phenomena such as
supernovae are typed by the mechanisms that produce them.
In a supernova of Type II
the star explodes but leaves a collapsed black hole, neutron star or white dwarf behind.
The core has little nuclear material left, and is supported by electron degeneracy pressure
(when compressed and cooled, the velocity of all electrons can only fall so low because
two electrons can’t occupy the same quantum state). But the core accumulates mass from
the shell. If it never reaches the Chandrasekhar mass, it will collapse to a white dwarf.
But when the core mass is larger than the Chandrasekhar mass, electron degeneracy
pressure is not enough, and it collapses further. When neutron degeneracy pressure starts
the bounce, many neutrinos escape suddenly, carrying away an enormous amount of
energy, leaving a neutron star behind but blowing away the rest of the mass of the star.
Supernovae of type I are different – they are giant nuclear explosions. In a supernova of
Type Ia (characterized by absence of hydrogen lines in their spectra), the star explodes
completely, leaving nothing behind. The star still has nuclear material, and during
collapse increasing density and pressure rapidly increases nuclear reactions, which
release energy. This stops collapse well before neutron star density, blowing the star
completely apart. Even here we redescribe and regroup phenomena, paying more
attention to some differences than others, when we discover that there is more than one
mechanism for supernovae.
'Responsible for a phenomenon' expresses this. Secondly, it captures the diversity of
things that mechanisms do. Mechanisms carry out tasks, such as regulation or control,
and exhibit behaviours, such as growth. They also maintain stable states. Homeostatic
mechanisms, such as those that maintain human body temperature at 37°, do this. Such a
state might even be a standing capacity of a system. For example, many cells have the
capacity to metabolise lactose, although they do not do so unless glucose is unavailable.
At a higher degree of abstraction, the metabolic mechanism is responsible for more than
one phenomenon: metabolising glucose normally, and metabolising lactose in the
absence of glucose. There is no significant disagreement on this diversity (Darden 2006
p273, 2008 p959; Glennan 2002a p126-7).
Thirdly, 'responsibility' implies something counterfactual. The phenomenon can be
something actual, or something modal – such as the capacity of a cell to metabolise
lactose, even if lactose is never encountered. See Glennan (1997) for a similar view of
capacities. However, the mechanism does not determine the phenomenon, because some
mechanisms may be indeterministic. Nor should a characterization of mechanisms
require that they produce 'regular changes' as MDC do, but Machamer (2004 p37,
Mechanisms are individuated by their phenomena, and phenomena are also individuated by their
mechanisms. This is not circular, because it happens iteratively over time. At the beginning, a mechanism
is not needed to individuate a phenomenon, but the characterization of the phenomenon may be further
refined when a mechanism or mechanisms are discovered. See Darden 2008 p960.
footnote 1) drops. Compare Darden (2008 p964) and Glennan (2010 p257). Some
mechanisms, such as homeostatic mechanisms, might not produce change at all.
may or may not be regular. To give Craver's example, in the mechanism of
neurotransmitter release, only 10-20% of action potentials eventuate in release events.
And release events can occur without action potentials (Craver 2007 p26). But dropping
explicit reference to regularity does not imply that mechanisms in general do not have to
exhibit some form of regularity or stability. Some far weaker form of regularity or
stability is already present in the idea of mechanisms being responsible for the
phenomenon. Our formulation captures the importance, diversity and various forms of
stability of what mechanisms do.
3b Entities and activities
There is consensus that mechanistic explanation involves decomposition (see particularly
Bechtel and Richardson), and mechanisms have two distinct kinds of constituents. We
have ‘entities’, ‘parts’ and ‘component parts’ used for the bits and pieces of the
mechanism, and ‘activities’, ‘interactions’ and ‘component operations’ for what those bits
and pieces do. Astrophysical mechanisms have both entities and activities: ‘An
important mechanism for producing X-rays from Solar System objects is charge
exchange, which occurs when a highly ionized atom in the solar wind collides with a
neutral atom (gas or solid) and captures an electron, usually in an excited state. As the ion
relaxes, it radiates an X-ray characteristic of the wind ion. Lines produced by charge
exchange with solar wind ions such as C
, C
, O
, O
and Ne
have all been
detected with Chandra and XMM-Newton [new space observatories]…’. (Santos-Lleo et
al. p998.) Putting this together with mechanisms for supernovae above, entities include:
electron, proton, neutron, neutrino, star, neutron star, white dwarf, black hole, core, gas,
x-ray, ionised atom, solar wind, neutral atom. Activities include: charge exchange,
colliding, relaxing, radiating, collapse, bounce, heating, electron capture, the nuclear
fusion that creates heavier elements in stars, and so on.
For wide applicability, care is needed in understanding entities and activities.
Fascinatingly, astrophysical mechanisms deal simultaneously with the vanishingly small
and the staggeringly enormous. What happens in a supernova depends on properties of
the massive, such as whether the star’s core reaches the Chandrasekhar mass or not –
which is approximately 1.2-1.4 solar masses. On the other hand, it is electron degeneracy
pressure which supports a white dwarf, and this depends on the fact that electrons are
fermions, i.e. they obey Pauli’s Exclusion Principle, which means that there are limits on
the minimum energy that more than two electrons in the same place can have. The end
state of a star depends on the interplay of these very different kinds of factors, so there
can be no a priori restriction according to size on the entities and activities of a
mechanism. Further, mechanistic explanation might not always be in terms of smaller
parts. Darden provides a good example: ‘finding the mechanism of segregation of genes
did not require decomposing genes into their parts but required finding the wholes, the
We reject Tabery’s ‘interactivity’ because it also requires change (Tabery 2004 p12). But see Tabery
(2009) on using mechanisms to explain difference, rather than similarity.
chromosomes, on which the parts, the genes, ride.’ (Darden 2006 p109, see also Darden
2008 p961.) Mechanistic explanation is not always about the little explaining the big.
Finally, the parts of mechanisms vary a great deal in their robustness. Some entities
remain comparatively unchanged over time, but others are more transient, such as the
mRNA that is made from DNA, used as a template to make a protein, and then broken
down again straight away. Activities can also be local and fragile, such as the mutation
or recombination that creates the diversity of strains of HIV that makes it so difficult to
eradicate. Glennan seems committed to a high degree of robustness of parts in earlier
work (2002b, 2009a) – although he notes that the interactions of parts are 'not
exceptionless' – but has relaxed this somewhat now (2010).
MDC have metaphysical arguments for entities and activities. Here, we put these aside to
focus on descriptive reasons for preferring a particular characterization of the
components of a mechanism. We prefer MDC's language of activities and entities for
two main reasons: it offers a powerful resistance to entity-bias, and it allows variability in
the arity of the relation between entities. We take these points in turn.
Many approaches to scientific ontology give entities priority, treating what entities do as
either reducible to entities themselves, or metaphysically dubious. But descriptively,
activities and entities are equally important to mechanisms: neither has priority. MDC
write: ‘There are kinds of changing just as there are kinds of entities. These different
kinds are recognized by science and are basic to the ways that things work.’ (MDC 2000
p5.) Machamer adds: ‘Activities can be abstracted and referred to and identified
independently of any particular entity, and sometimes even without reference to any
entity at all.’ (Machamer 2004 p30. See also Darden 2006 p277.) A bunch of entities
engaging in a certain set of activities will produce something different from the same
bunch of entities engaging in another set of activities. A buyer and seller haggling over
the price may lead to a sale. The same two people chatting about the weather will not.
Further, although entities and activities are always equally important in that they must
both be present to produce the phenomenon, in explaining different kinds of phenomena
entities are sometimes more interesting than the activities, and vice versa. In protein
synthesis, entities are very different from each other and their detailed structure matters a
great deal. But in many dynamical systems and systems biology explanations, the
entities are relatively similar to each other and the activities are vital to produce the
This is consistent with Bechtel's and Glennan’s considered views (Glennan 2009b p321),
but the rhetorical impact of the language matters for scientists and philosophers
elsewhere using an account of mechanisms in other debates. MDC's entities and
activities offer the strongest rhetorical resistance to a default entity-bias.
Our second reason is that variability in the arity of the relation between entities is more
important than has been recognised, and is nicely captured by MDC’s language.
Consider the alternatives: capacities are unary (1-ary) relations since a capacity attaches
to an entity, although one entity can have many capacities (note Darden 2008 p963).
Glennan's 'interaction' implies a relation between at least two entities, so interactions are
binary (2-ary) at least.
Bechtel and Abrahamsen write: ‘Each component operation
involves at least one component part’ (2005 p424), which seems to allow either unary,
binary, 3-ary and so on. The mapping of entities to activities can be unary, as in a bond
breaking, involving no other entity; binary, as in a promoter binding to a strand of DNA;
but it can also be 3-ary, 4-ary and so on (See Darden 2008 p964). The activity of
transcription involves DNA, the newly created mRNA, and various regulation and
control enzymes, while more highly abstract activities such as equilibrating, or osmosis
(Darden 2006 p277) may involve very many entities, of the same or different kinds, or be
such that it hard to decide on any very clearly defined entity that engages in the activity.
Bechtel (2008) examines extensively the importance of mapping entities to activities (his
component parts and component operations) in mechanism discovery, pointing out that it
is often this mapping that allows us to identify the working parts of a mechanism. So we
had better get the arity of the relation right. But Bechtel ties operations closely to parts:
‘We use the term operation rather than activity because we want to draw attention to the
involvement of parts’ (Bechtel and Abrahamsen 2005 p423, footnote 5). The arity of the
relation between entities allowed by activities is unrestricted, covering all this. This is
the best descriptive reason to favour entities and activities.
In summary, mechanisms have two kinds of constituents. We prefer 'entities' and
'activities' because these terms offer rhetorical advantages for avoiding entity-bias, and
there is no limit to the arity of activities. Entities can be of widely varying sizes, in some
cases the big is used to explain the small, and some mechanisms involve comparatively
fragile entities and activities.
3c Organization
Organization is the least controversial element in any characterization of mechanisms,
present in the characterizations of MDC and Bechtel and Abrahamsen, and discussed
explicitly by Glennan elsewhere (see 2005, 2002a). We think it worth the emphasis of
putting it in the characterization, but consider Bechtel and Abrahamsen’s ‘orchestrated
functioning’ too strong. It suggests a tightly integrated form of organization that exists in
highly evolved or designed systems, but not everywhere.
How to understand organization is not much discussed, and is far from trivial. What is
organization so that it can reasonably be regarded as an element of all mechanisms?
Here, we examine this, and argue that organization is not confined to complex biological
mechanisms, by showing its importance to astrophysical mechanisms. These exhibit
complex forms of organization requiring investigation by numerical simulation, such as
homeostasis, equilibrium and feedback.
Organization is the final element in the production of the phenomenon. The same entities
and activities organized differently will produce something different. A group of
organisms engaged in feeding, mating and dying will do something different if they are
As Tabery 2004 notes. We thank Glennan for pressing us on this point.
subject to a common selection pressure – a new predator, or bout of cold weather – than
if they are not. Most generally, organization is whatever relations between the entities
and activities discovered produce the phenomenon of interest: when activities and entities
each do something and do something together to produce the phenomenon.
In mechanistic explanation, organization is analogous to initial conditions in laws-based
explanation. Laws and the entities they govern explain nothing until initial conditions are
specified: Newton’s laws do not tell us the movements of the planets until their initial
positions and velocities are specified. In the mechanistic approach organization gives the
ongoing conditions that allow the entities and activities to produce the phenomenon.
‘Ongoing’ is important. Initial conditions for laws matter only at the beginning, while
organization matters throughout the operation of a mechanism. Further, organization is
not independent of the activities and entities and ongoing operation of a mechanism.
Organization might affect which activities and entities are involved, while the operation
of a mechanism might alter the organizational structure. Evolution of a group of
organisms subject to a common selection pressure might alter how widely dispersed those
organisms need to be to be subject to that common selection pressure.
This approach implies, correctly, that it is an empirical question what forms of
organization are important for particular domains, so that the only other informative
thing that can be said about organization is to discuss examples. Organization comes in
many forms, more or less important for different kinds of mechanism. Spatial and
temporal organization is vital to such cases as protein synthesis. (Darden 2006, Craver
2007.) But other forms of organization can be instantiated by spatiotemporally located
mechanisms. Complex forms of organization such as homeostasis, equilibrium, feedback
and self-organization are vital for the production of the phenomena studied by complex
and dynamical systems. (See Bechtel 2006 p33, p39; Mitchell 2003; and possibly
Glennan 2008.) Quantitative description of dynamical organization is often vital. For
example, in simulating supernovae, mass is standardly being lost from the star while
mass is accumulated in the star core. Quantitative simulation over time is needed to see
whether the Chandrasekhar mass is reached. In this way we allow organization to
capture necessary elements of what Bechtel calls 'dynamical mechanistic explanation'.
(See Bechtel 2008, Bechtel and Abrahamsen 2009, 2010). Each of these forms of
organization also lies on a spectrum from less organized to increasingly organized.
Unsurprisingly, then, organization in its most general form – when activities and entities
each do something and do something together to produce the phenomenon – itself comes
in a (multidimensional) spectrum of increasingly complex organization. Whichever form
of organization is most important to the production of a particular phenomenon depends
on the empirical world. Our world seems to involve different forms of organization,
more or less complex, in different cases. In the simplest cases organization might be
simple or trivial, but it is still present.
Use of numerical simulation is a good indicator of complexity of organization, and
simulations are a standard tool for discovering astrophysical mechanisms. They often
We have compared organization in natural selection, and in protein synthesis in Illari and Williamson
reveal complex forms of organization usually associated with biological mechanisms
such as feedback. Simulation of how the first stars formed tend to suggest they formed
on their own, which leads to the question: how did galaxies form? Further simulations
suggest: ‘Some of the feedback processes described above that affect the formation of
individual stars also influence primordial star formation on large scales. The enormous
fluxes of ionizing radiation and H
-dissociating Lyman–Werner radiation emitted by
massive population III stars dramatically influence their surroundings, heating and
ionizing the gas within a few kiloparsecs of the progenitor and destroying the H
within a
somewhat larger region. Moreover, the Lyman–Werner radiation emitted by the first stars
could propagate across cosmological distances, allowing the buildup of a pervasive
Lyman–Werner background radiation field. The effect of radiation from the first stars on
their local surroundings has important implications for the numbers and types of
population III stars that form. The photoheating of gas in the minihaloes hosting
population III.1 stars drives strong outflows, lowering the density of the gas in the
minihaloes and delaying subsequent star formation by up to 100 Myr … . Furthermore,
neighbouring minihaloes may be photoevaporated, delaying star formation in such
systems as well. The photodissociation of molecules by Lyman–Werner photons emitted
from local star-forming regions will, in general, act to delay star formation by destroying
the main coolants that allow the gas to collapse and form stars.’ (Bromm et al. p51.)
Successful simulations are often very difficult: ‘The simulations, starting from
cosmological initial conditions, are just now approaching the resolution and physical
realism required to investigate whether atomic cooling haloes fulfil the criteria for a first
galaxy as defined above. Quite generically, in such models, the first generation of stars
forms before galaxies do, and feedback effects from the first stars are expected to play a
key role in determining the initial conditions for the formation of the first galaxies.’
(Bromm et al. p52.) Astrophysicists want to reproduce phenomena using physically
realistic parameters, and only then do they think they have an empirically significant
result. Investigation of organization by means of simulation is not the sole preserve of
the life sciences.
We have now defended our characterization of a mechanism, argued for its wide
application, including to the case of astrophysical mechanisms. We have indicated where
we disagree with the main contenders while emphasizing that there is a core of agreement
which we capture. Very different scientific work in different fields aims to find and
describe the entities and activities of their domain, their organization, and the phenomena
they are responsible for. This discovery process can be complicated and iterative. It
takes serious empirical work to correctly delimit the phenomena, and that description
determines what activities, entities and organization will be looked for; while what
activities, entities and organization are found affect the description of the phenomena.
We will now show that our characterisation of mechanism is not so broad that it captures
4 Borderline cases
It is important that not everything counts as a mechanism. In this section, we examine
some things produced in this process of discovery that are sometimes called mechanisms
– perhaps erroneously – to further illuminate our account of mechanisms.
Case 1: Description of mechanism is too partial
Sometimes we have a scientific advance, but the description of the mechanism for the
phenomenon is still partial. Consider the various possible forms of memory that have at
some point been phenomenally dissociated: long term versus short term memory,
working memory, episodic versus semantic memory, and non-explicit memory including
various forms of priming.
There may be separate mechanisms producing these
phenomena, but we are not yet even in a position to guess how many mechanisms there
are. (See Bechtel 2008 for extensive discussion.) We have a better description of the
phenomenon to be explained, and only finding the underlying entities will show that there
really are separate mechanisms. Before this point, we have little more than a better
description of the phenomenon to be explained.
Mechanism discovery is gradual, so there will be no sharp line between partial and full
descriptions of mechanisms. The crucial point is where scientists have good reason to
suppose they have got hold of the actual mechanism operating. Before that, the
description might be so partial that it does not pick out a mechanism, and the explanation
might not succeed.
Case 2: Entities without activities – Darden’s stopped clock
Darden writes: ‘The MDC characterization of mechanism points to its operation.
Although someone (perhaps Glennan 1996) might call a stopped clock, for example, a
mechanism, I would not. It is a machine, not a mechanism. The MDC characterization
views mechanisms as inherently active. In the stopped clock, the entities are in place but
not operating, not engaging in time-keeping activities. When appropriate set-up
conditions obtain (e.g., winding a spring, installing a battery), then the clock mechanism
may operate.’ (Darden 2006 p280-1.)
Recall that nothing is a mechanism tout court – mechanisms are mechanisms for
phenomena. A stopped clock is no longer a mechanism for telling the time, but it might
still be a mechanism for something else – for recording a race time. Recall also that for
Darden, as for Machamer and Craver, activities must produce change. The stopped clock
produces no change. But we have argued that some activities and mechanisms, such as
homeostatic ones, exist to prevent change. So the stopped clock, and similar cases such
as chimneys, or pillars supporting roofs, are candidate mechanisms for maintaining
stability of some kind.
However, they still present a puzzle: it seems they must either be mechanisms without
activities, or non-mechanisms due to the lack of activities. The normal explanation for a
pillar supporting a roof involves only its material, spatio-temporal location and forces.
This seems to involve organization and no activities. But this is too quick, as there is no
sharp line between activities and organization. In one explanation, a high-level activity
such as equilibrating might be the activity of a particular group of entities, while in
another it is treated as the organization of the system. Ultimately in such cases there is no
sharp answer to the question of whether these are cases of mechanisms without activities,
and there is no useful purpose in legislating an answer to the question that could constrain
empirical research.
Case 3: Activities without entities
In psychology in particular, capacities like memory are often given a purely functional
description. There are indefinitely many ways the human brain could divide up the task
of remembering things. There may be one part of the brain that remembers events,
including that event from the point of view of all the sensory modalities, and facts.
Alternatively, there could be different parts of the brain that remember these different
kinds of things. For example, perhaps remembering what an event looked like is a sub-
task of the visual system, rather than of a central memory area.
As we have said, finding the parts of the brain that perform these tasks – the working
entities – is what persuades us that we have the right division of the overall task into sub-
tasks. Where this is not possible, we are left with a purely functional explanation, which
seems to consist of postulated activities, without entities. This is not a mechanism,
although it may be a step on the way to a mechanistic explanation (Craver 2006). See
also Bechtel (2008).
Case 4: No organization
There may appear to be no mechanism if there is no apparent organization. In the kinetic
theory of gases, which explains both Boyle’s Law and Charles’ Law, molecules behave
on average randomly.
But in our understanding of organization as when activities and
entities each do something and do something together to produce the phenomenon,
whatever relations amongst the activities and entities produces the phenomenon is the
relevant organization. If the molecules behaving randomly on average produces the
phenomenon, that is the kind of organization present in that mechanism, however trivial
it appears.
This is not the same as the idealization case. If a false assumption is made of average
random behaviour to model a system, that might – or might not – render the model no
longer a model of a mechanism, as we have said above. But if the assumption is not
false, a mechanism is being described.
Case 5: Nothing concrete
Mathematicians sometimes speak of ‘mechanism’, for a technique or schematic method.
These techniques are normally mechanisms for generating derivations or mathematical
We thank Erik Weber for suggesting this example.
entities or structures. For example, forcing is a mechanism for deriving the independence
of the continuum hypothesis, and Foreman and Magidor (1995, p55) write of ‘the
mechanism typically used to show presaturation’.
The ‘mechanisms’ here are purely abstract. They are not causes, and cannot be used in
causal inference or explanation. However, these things are used in explanation,
prediction and control in the particular way appropriate to the abstract realm. There is an
analogous form of explanation in the decomposition to parts, and the understanding of
how parts together produce the overall derivation, entity or structure. They might also be
used to predict a change in the overall result from changing a part – a prediction that
couldn’t be made before the decomposition.
These strong analogies render using the word ‘mechanism’ reasonable. To decide further
whether these things count as mechanisms on our account depends on metaphysical
issues we do not address here. Do entities and activities have to be concrete? If so, then
these are not mechanisms, on most understandings of mathematical entities. However,
even on this view a mathematical Platonist might make a case for these being real
mechanisms. For our purposes it suffices to note that even in that case, there are clearly
no causal mechanisms here.
Case 6: Too much idealization
As Glennan (2005) and Craver (2006) discusses extensively, models of mechanisms are
built using assumptions. These are necessary for enough simplification to build a model.
Sometimes these assumptions are radically false. For example, in the social sciences it is
not uncommon to assume non-communication among people or groups – an assumption
of no organization. In economics, it is standard to assume rationality. Many models in
physics use equilibrium assumptions or no-friction assumptions. Often, these claims are
trivial, merely allowing serious quantitative modelling of a genuine worldly phenomenon.
But once there is too much idealization, these are no longer accurate models of
mechanisms. They are too distant from the system they describe, and their parameters no
longer have plausible physical interpretations. They might be useful predictive tools, or
important explanatory work on the road to mechanism discovery. Such models are often
of further use as accurate descriptions of phenomena to be explained. But scientists using
such models are, as above, not yet in a position to know whether they have got hold of
the actual mechanism.
The level of idealization versus the level of accurate description is a matter of degree, so
there is no particular point where such models cease to be accurate descriptions of
mechanisms. The crucial point is whether they accurately describe anything worldly,
whether their parameters have reasonable physical values (see Bechtel and Wright).
There may still be mechanisms in such cases – but such models have not yet described
them. This extends to many models in science.
5 Conclusion
We have argued for our characterization of mechanisms:
A mechanism for a phenomenon consists of entities and activities organized in such a
way that they are responsible for the phenomenon.
We have examined the various elements – responsibility for the phenomenon; entities and
activities; organization – in some detail, showing how they apply to various fields,
particularly astrophysical mechanisms. However messy the process of mechanism
discovery is, and however important the different challenges faced by different fields, this
characterization lets us see how these elements of mechanisms contribute to a project that
shares a great deal across the sciences. We hope that our account will be useful to
ongoing debates on causal inference and causal modelling.
We believe our account best captures a consensus emerging in the mechanisms literature
by applying very widely to mechanisms while addressing the primary concerns of the
main contenders in the debate.
All the main contenders agree on the functional individuation of mechanisms – Glennan's
Law. Indeed, many have worked on the implications of this (Craver 2007 pp6 ff; Darden
2006 p42, pp289-90; Bechtel 2006 p28). We have used this to frame our account,
spelling out further implications, and there is no obvious reason for the main contenders
to object. We have already explained that our use of MDC's entity-activity language is
not at serious odds with Glennan's or Bechtel's considered views. Finally, there is little
extended discussion of organization, so it is possible for the main contenders to regard
our views on organization as a development of theirs.
Bechtel, Craver, Darden and Machamer do not aim for a widely applicable account of a
mechanism, but they should have no objection to that aim. Craver and Bechtel are
currently extending the applicability of mechanisms, at least to psychology and
neuroscience. They have no reason to object to dropping those elements of their own
characterizations that narrow their applicability.
Glennan does aim for a widely applicable account. He wishes to use an account of
mechanisms to give an account of causation, so his account of mechanisms must apply
anywhere there is causation. But we have argued that Glennan's wish in earlier work for
stability of mechanisms and mechanism parts, and his definition of mechanisms as
'complex systems' narrow the applicability of his account. This creates serious tension in
Glennan’s work. Glennan most of all has excellent reason to alter these elements of his
own characterization in favour of an account like ours, which explains why he is now
moving in that direction (2010, 2009b).
In conclusion, we have offered a characterization of mechanisms that is widely applicable
across the sciences and captures the emerging consensus on mechanisms. It is fit for use
as a framework for ongoing work on causal explanation, inference and modelling.
We wish to thank the Leverhulme Trust for supporting this research. We are also
indebted to colleagues at Kent and in the Causality in the Sciences network for discussion
of many of these issues. The work has been significantly improved due to detailed
comments from Stuart Glennan, Federica Russo and two anonymous referees.
Remaining errors are, of course, our own.
William Bechtel (2010): ‘The Downs and Ups of Mechanistic Research: Circadian Rhythm Research as an
Exemplar’ in Erkenntnis, 73(3): 313-328.
William Bechtel (2009): ‘Looking Down, Around, and Up: Mechanistic Explanation in Psychology’ in
Philosophical Psychology, 22, 543-564.
William Bechtel (2008): Mental Mechanisms, Routledge, Oxford.
William Bechtel (2007): ‘Biological mechanisms: organized to maintain autonomy’ in Boogerd,
Bruggeman, Hofmeyr and Westerhoff (eds.) Systems Biology, 2007, Elsevier.
William Bechtel (2006): Discovering Cell Mechanisms: the Creation of Modern Cell Biology, 2006, CUP,
William Bechtel and Adele Abrahamsen (2009): ‘Decomposing, Recomposing, and Situating Circadian
Mechanisms: Three Tasks in Developing Mechanistic Explanations’ in Leitgeb, H. and Hieke, A. (Eds.),
Reduction and Elimination in Philosophy of Mind and Philosophy of Neuroscience, Frankfurt: Ontos,
William Bechtel and Adele Abrahamsen (2010): ‘Dynamic mechanistic explanation: Computational
Modeling of Circadian Rhythms as an Exemplar for Cognitive Science’ in Studies in History and
Philosophy of Science, Part A, 1: 321-33.
William Bechtel and Adele Abrahamsen (2008): ‘From Reduction Back to Higher Levels’ In B. C. Love,
K. McRae, & V. M. Sloutsky (Eds.), Proceedings of the 30th Annual Conference of the Cognitive Science
Society (pp. 559-564). Austin, TX: Cognitive Science Society.
William Bechtel and Adele Abrahamsen (2005): ‘Explanation: a mechanist alternative’ in Studies in the
History and Philosophy of the Biological and Biomedical Sciences, 36: 421–441.
William Bechtel and Robert Richardson (1993): Discovering complexity, 1993, Princeton: Princeton
University Press.
William Bechtel and Cory Wright (2009): ‘What is psychological explanation?’ in P. Calvo and J. Symons
(ed.) The Routledge companion to philosophy of psychology, London: Routledge, 2009, 113-30.
Alexander Broadbent (2011): ‘Inferring causation in epidemiology: mechanisms, black boxes, and
contrasts’ in Illari, Russo and Williamson (eds.) Causality in the Sciences, OUP, Oxford, 2011, 45-69.
Volker Bromm, Naoki Yoshida, Lars Hernquist & Christopher F. McKee (2009): ‘The formation of the
first stars and galaxies’ in Nature, Vol 459, 7 May 2009.
Carl Craver (2007): Explaining the Brain, Clarendon Press, Oxford.
Carl Craver (2006): ‘When mechanistic models explain’, Synthese, 153: 355-376.
Lindley Darden (2008): ‘Thinking Again about Biological Mechanisms’ in Philosophy of Science, 75(5):
Lindley Darden (2006): Reasoning in Biological Discoveries, CUP, Cambridge.
M Foreman and M Magidor: ‘Large cardinals and definable counterexamples to the continuum hypothesis’
in Annals of Pure and Applied Logic, 76(1): 47-97.
Donald Gillies (2011): ‘The Russo-Williamson thesis and the question of whether smoking causes heart
disease’ in Illari, Russo and Williamson (eds.) Causality in the Sciences, OUP, Oxford, 2011, 110-25.
Stuart Glennan (2011): ‘Singular and general causal relations: A mechanist perspective’ in Illari, Russo and
Williamson (eds.) Causality in the Sciences. OUP: Oxford, pp 789-817.
Stuart Glennan (2010): 'Ephemeral Mechanisms and Historical Explanation' in Erkenntnis, 72: 251–266.
Stuart Glennan (2009a): ‘Productivity, relevance and natural selection’ in Biology and Philosophy, 24: 325-
Stuart Glennan (2009b): 'Mechanisms' in Helen Beebee, Christopher Hitchcock and Peter Menzies (eds.)
The Oxford Handbook of Causation. OUP: Oxford, 315-25.
Stuart Glennan (2008): ‘Mechanisms’ in Routledge Companion to the Philosophy of Science, Stathis Psillos
and Martin Curd (eds.). Routledge: Oxford.
Stuart Glennan (2005): ‘Modeling Mechanisms’ in Studies in the History and Philosophy of Biology and
Biomedical Sciences, 36: 443-464.
Stuart Glennan (2002a): ‘Contextual unanimity and the units of selection problem’ in Philosophy of
Science 69:118-37.
Stuart Glennan (2002b): ‘Rethinking Mechanistic Explanation’ in Philosophy of Science, 69 (September
2002) ppS342–S353.
Stuart Glennan (1997): 'Capacities, Universality, and Singularity' in Philosophy of Science, 64(4): 605-626.
Stuart Glennan (1996): ‘Mechanisms and the Nature of Causation’ in Erkenntnis, 44(1): 49-71.
Phyllis McKay Illari and Jon Williamson (2010): ‘Function and Organization: Comparing the mechanisms
of protein synthesis and natural selection’ in Studies in the History and Philosophy of the Biological and
Biomedical Sciences, 41: 279-291.
Bert Leuridan and Erik Weber (2011): ‘The IARC and mechanistic evidence’ in Illari, Russo and
Williamson (eds.) Causality in the Sciences, OUP, Oxford, 2011, 91-109.
Peter Machamer, Lindley Darden and Carl Craver (2000): ‘Thinking about Mechanisms’ in Philosophy of
Science 67 (March 2000) pp1-25.
Peter Machamer (2004): ‘Activities and Causation: The Metaphysics and Epistemology of Mechanisms’,
International Studies in the Philosophy of Science, 2004, 18(1), pp27-39.
Sandra D. Mitchell (2003): Biological Complexity and Integrative Pluralism, 2003, CUP, Cambridge.
Federica Russo (2009): Causality and causal modelling in the social sciences: Measuring variations, 2009,
New York: Springer.
Maria Santos-Lleo, Norbert Schartel, Harvey Tenenbaum, Wallace Tucker & Martin C. Weisskopf (2009):
‘The first decade of science with Chandra and XMM-Newton’ in Nature, Vol 462, 24/31, December
Daniel Steel (2008): Across the boundaries: Extrapolation in biology and social science, OUP, 2008.
James Tabery (2009): ‘Difference Mechanisms: Explaining variation with mechanisms’ in Biology and
Philosophy 24:645–664.
James Tabery (2004): ‘Synthesizing activities and interactions in the concept of a mechanism’ in
Philosophy of Science, 71: 1–15.
Phillip J Torres (2009): ‘A modified conception of mechanisms’ in Erkenntnis 71(2): 233-51 (2009).
Jon Williamson (2011): ‘Mechanistic theories of causality’, Philosophy Compass 6(6): 421-447.
James Woodward (2002): ‘What is a mechanism? A counterfactual account’ in Philosophy of Science 69:
... We observed that many of the projects that students work on in makerspaces involve the design and construction of mechanistic systems, such as electronic circuits, mechanical systems, or objects with multiple interlocking parts. Formally, a mechanistic system consists of (a) a phenomenon or phenomena that can be explained or understood by (b) decomposing it into parts or components that (c) are organized in such a way that (d) they give rise to the phenomenon through their causal interactions (Illari & Williamson, 2012). In makerspaces, when students encounter problems with these systems, they engage in a hands-on process of tinkering and debugging that is different from the types of problem solving typically used in schools. ...
Full-text available
This study investigated the impact of participating in a year-long digital-fabrication course on high-school seniors’ problem-solving skills, with a focus on problems involving mechanistic systems. The research questions centered on whether working in a makerspace impacted students’ abilities to solve such problems and whether the process data generated during problem-solving activities could be used to identify the different problem-solving approaches taken by the participants. A novel set of hands-on, mechanistic problems were created to answer these questions, and the results showed that after taking part in the course students performed significantly better on these problems, with the post-course students making more progress towards the solutions than the pre-course students. The process data revealed two distinct problem-solving approaches for each problem, one adopted primarily by experts (the expert approach) and one by pre-course students (the novice approach). The post-course students were more likely to adopt the expert approaches, which were strongly associated with better performance on each problem. The study found that participation in the course made the high-school students better able to “see” the various components and their ways of interacting, making them more like expert engineers.
... Piedāvājumu, kā konceptualizēt efektīvas skolotāju profesionālās pilnveides dizainu un tā elementus, ir bijis daudz, tomēr visaptverošākais piedāvājums, kuru apstiprina statistiski nozīmīgas korelācijas, ir S. Simsa un kolēģu piedāvājums balstīt efektīvas skolotāju profesionālās pilnveides dizainu četros efektīvas skolotāju profesionālās pilnveides principos (redzējuma paplašināšana, motivēšana mācīties, prasmju attīstīšana, ieviešana praksē), kurus skolotāju profesionālajā pilnveidē iespējams realizēt, veicot noteiktas darbības jeb profesionālās pilnveides mehānismus (sk. 1. tabulas 1. aili) (Illari & Williamson, 2012;Michie et al., 2013). ...
Grāmatā atklāta datos balstīta lēmumu pieņemšana attīstības risinājumiem izglītībā, šī procesa būtība un norise pakāpenisku izmaiņu panākšanai skolu praksē. Piedāvāts datu kompleksās analīzes modelis skolas attīstības risinājumiem, kā arī aplūkota trīsdimensionālā pieeja skolēnu sniegumu analīzei. Analizēti dati par augstu un zemu skolēnu sniegumu valsts pārbaudes darbos matemātikā. Aplūkots optimisms par izaugsmi skolas kopienas un skolēnu snieguma izzināšanai un skaidrošanai, kā arī analizēts, kā gūt pierādījumus par mācīšanu un mācīšanos stundā – ceļā uz mācīšanu iedziļinoties. Apskatīta datos balstītu profesionālās pilnveides risinājumu veidošana, skolotāju profesionālās pilnveides efektivitātes paaugstināšana, datu izmantošana jauninājumu ieviešanai skolas un klases līmenī. Grāmatas noslēgumā autori apraksta datos balstītu lēmumu pieņemšanas nozīmīgumu efektīvai skolu vadībai pašvaldības līmenī. Autori analizē pašreizējo situāciju, salīdzinot to ar citu valstu pieredzi un pētījumu datiem, parāda piemērus, kā šī s pārmaiņas pakāpeniski ienāk skolu praksē, un apzina turpmāk veicamos soļus. Grāmata adresēta izglītības pētniekiem, skolu vadītājiem, skolotājiem, studentiem – lasītājiem, kas ir ieinteresēti iedziļināties būtiskās pārmaiņās, kuras notiek valsts izglītībā.
... Systems tradition is currently the dominant approach to explanations formulated in biology, system research, and cognitive neuroscience, while decomposition is the central heuristic strategy in mechanistic explanations besides the identification of mechanisms (Bechtel & Richardson, 1993/2010cf. Bechtel, 2008;Craver, 2007;Illari & Williamson, 2013). However, the mechanistic view of explanation has met with controversy (cf. ...
Full-text available
The issue of the relationship between predictive processing (PP) and the free energy principle (FEP) remains a subject of debate and controversy within the research community. Many researchers have expressed doubts regarding the actual integration of PP with the FEP, questioning whether the FEP can truly contribute significantly to the mechanistic understanding of PP or even undermine such integration altogether. In this paper, I present an alternative perspective. I argue that, from the viewpoint of the constraint-based mechanisms approach, the FEP imposes an important constraint, namely variational free energy, on the mechanistic architecture proposed by PP. According to the constraint-based mechanisms approach, high-level cognitive mechanisms are integral parts of extensive heterarchical networks that govern the physiology and behavior of agents. Consequently, mechanistic explanations of cognitive phenomena should incorporate constraints and flows of free energy as relevant components, given that the implemented constraints operate as long as free energy is available. Within this framework, I contend that the FEP provides a relevant constraint for explaining at least some biological cognitive mechanisms described in terms of Bayesian generative models that minimize prediction errors.
Full-text available
Although the literature on the issue of pluralism within the philosophy of science is very extensive, this paper focuses on the metaphysical causal pluralism that emerges from the new mechanistic discussion on causality. The main aim is to situate the new mechanistic views on causation within the account of varieties of causal pluralism framed by Psillos (2009). Paying attention to his taxonomy of metaphysical views on causation (i.e., the straightjacket view, the functional view, the two-concept view, the agnostic view and the atheist view) will help clarify differences in opinion and, at the same time, make it possible to elucidate the main metaphysical theses present within the new mechanistic debate. Special attention is given to S. Glennan’s theory of causation, since it is unique in offering an overall metaphysical view of the issue. It is also argued that mechanists are not “atheists” on causation: while all of them are causal realists, most mechanists are “agnostic” on causation, with a few exceptions such as S. Glennan, P. Machamer and J. Bogen.
Full-text available
The existing philosophical views on what is the meaning of causality adequate to medicine are vastly divided. We approach this question and offer two arguments in favor of pluralism regarding concepts of causality. First, we analyze the three main types of research designs (randomized-controlled trials, observational epidemiology and laboratory research). We argue, using examples, that they allow for making causal conclusions that are best understood differently in each case (in agreement with a version of manipulationist, probabilistic and mechanistic definitions, respectively). Second, we analyze clinical practice and argue that these manipulationist, probabilistic and mechanistic causal claims can be used as evidence for different therapeutic decisions. We differentiate among ‘predicting’ that does not change the relata of causal claims, (mechanistic) ‘interferences’, and ‘interventions’ in the strict sense that act on causes to change effects. The central conclusion is that causal claims agreeing with diverse concepts of causality can deliver evidence for different types of therapeutic decisions.
Inspired by recent work in evolutionary, developmental, and systems biology, Systems, Relations, and the Structures of International Societies sketches a robust conception of systems that grounds a new conception of levels (of organization, not merely analysis). Understanding international systems as multi-level multi-actor complex adaptive systems allows explanations of important features of the world that are inaccessible to dominant causal and rationalist explanatory strategies. It also develops a comprehensive critique of IR's dominant conception of systems and structures (narrow, rigid, and unfruitful); presents a novel conception of the interrelationship of the social production of continuities and the social production of change; and sketches models of spatio-political structure that cast new light on the development of international systems, including a distinctive account of the nature of globalization.
Full-text available
Constructing arguments is essential in science subjects like chemistry. For example, students in organic chemistry should learn to argue about the plausibility of competing chemical reactions by including various sources of evidence and justifying the derived information with reasoning. While doing so, students face significant challenges in coherently structuring their arguments and integrating chemical concepts. For this reason, a reliable assessment of students' argumentation is critical. However, as arguments are usually presented in open‐ended tasks, scoring assessments manually is resource‐consuming and conceptually difficult. To augment human diagnostic capabilities, artificial intelligence techniques such as machine learning or natural language processing offer novel possibilities for an in‐depth analysis of students' argumentation. In this study, we extensively evaluated students' written arguments about the plausibility of competing chemical reactions based on a methodological approach called computational grounded theory . By using an unsupervised clustering technique, we sought to evaluate students' argumentation patterns in detail, providing new insights into the modes of reasoning and levels of granularity applied in students' written accounts. Based on this analysis, we developed a holistic 20‐category rubric by combining the data‐driven clusters with a theory‐driven framework to automate the analysis of the identified argumentation patterns. Pre‐trained large language models in conjunction with deep neural networks provided almost perfect machine‐human score agreement and well‐interpretable results, which underpins the potential of the applied state‐of‐the‐art deep learning techniques in analyzing students' argument complexity. The findings demonstrate an approach to combining human and computer‐based analysis in uncovering written argumentation.
Purpose This study aims to further the understanding of multi-level analysis in inter-organisational relationships by investigating the interplay of governance, cooperation and coordination in inter-organisational projects (IOPs) on sub-system and project levels. Design/methodology/approach The authors use the Viable Systems Model as a framework to analyse inter-organisational project governance, cooperation and coordination by adopting a multiple-case study. Findings The findings illustrate how governance and coordination mechanisms exhibit a filter-down effect on lower sub-systems while cooperation influence is confined within each sub-system. While remarking the importance of specific sub-systems on the overall project performance, the interplay of governance, cooperation and coordination across sub-systems appears to be complex, with governance influencing cooperation and coordination, whereas cooperation and coordination influence each other with an incremental effect. Originality/value This study defines two propositions that explain how multiple levels of analysis (project and sub-systems) can support the governance of large inter-organisational projects. The authors elaborate theory on the interplay of inter-organisational project governance, cooperation and coordination.
Full-text available
The relationship between topological explanation and mechanistic explanation is unclear. Most philosophers agree that at least some topological explanations are mechanistic explanations. The crucial question is how to make sense of this claim. Zednik (2019) argues that topological explanations are mechanistic if they (i) describe mechanism sketches that (ii) pick out organizational properties of mechanisms. While we agree with Zednik's conclusion, we critically discuss Zednik's account and show that it fails as a general account of how and when topological explanations are mechanistic. First, if topological explanations were just mechanism sketches, this implies that they could be enriched by replacing topological terms with mechanistic detail. This, however, conflicts how topological explanations are used in scientific practice. Second, Zednik's account fails to show how topological properties can be organizational properties of mechanisms that have a place in mechanistic explanation. The core issue is that Zednik's account ignores that topological properties often are global properties while mechanistic explanantia refer to local properties. We demonstrate how these problems can be solved by a recent account of mechanistic completeness (Craver and Kaplan 2020; Kohár and Krickel 2021) and use a multilayer network model of Alzheimer's Disease to illustrate this.
Full-text available
This chapter explores the idea that causal inference is warranted if and only if the mechanism underlying the inferred causal association is identified. This mechanistic stance is discernible in the epidemiological literature, and in the strategies adopted by epidemiologists seeking to establish causal hypotheses. But the exact opposite methodology is also discernible, the black box stance, which asserts that epidemiologists can and should make causal inferences on the basis of their evidence, without worrying about the mechanisms that might underlie their hypotheses. This chapter argues that the mechanistic stance is indeed a bad methodology for causal inference. However, this chapter detaches and defends a mechanistic interpretation of causal generalisations in epidemiology as existence claims about underlying mechanisms.
Full-text available
Due to the wide array of phenomena that are of interest to them, psychologists offer highly diverse and heterogeneous types of explanations. Initially, this suggests that the question "What is psychological explanation?" has no single answer. To provide appreciation of this diversity, we begin by noting some of the more common types of explanations that psychologists provide, with particular focus on classical examples of explanations advanced in three different areas of psychology: psychophysics, physiological psychology, and information-processing psychology. To analyze what is involved in these types of explanations, we consider the ways in which lawlike representations of regularities and representations of mechanisms factor in psychological explanations. This consideration directs us to certain fundamental questions, e.g., "To what extent are laws necessary for psychological explanations?" and "What do psychologists have in mind when they appeal to mechanisms in explanation?" In answering such questions, it appears that laws do play important roles in psychological explanations, although most explanations in psychology appeal to accounts of mechanisms. Consequently, we provide a unifying account of what psychological explanation is.
The anti-causal prophecies of last century have been disproved. Causality is neither a ‘relic of a bygone’ nor ‘another fetish of modern science’; it still occupies a large part of the current debate in philosophy and the sciences. This investigation into causal modelling presents the rationale of causality, i.e. the notion that guides causal reasoning in causal modelling. It is argued that causal models are regimented by a rationale of variation, not of regularity nor invariance, thus breaking down the dominant Humean paradigm. The notion of variation is shown to be embedded in the scheme of reasoning behind various causal models: e.g. Rubin’s model, contingency tables, and multilevel analysis. It is also shown to be latent—yet fundamental—in many philosophical accounts. Moreover, it has significant consequences for methodological issues: the warranty of the causal interpretation of causal models, the levels of causation, the characterisation of mechanisms, and the interpretation of probability. This book offers a novel philosophical and methodological approach to causal reasoning in causal modelling and provides the reader with the tools to be up to date about various issues causality rises in social science. "Dr. Federica Russo's book is a very valuable addition to a small number of relevant publications on causality and causal modelling in the social sciences viewed from a philosophical approach".(Prof. Guillaume Wunsch, Institute of Demography, University of Louvain, Belgium)
This article explores three issues surrounding the adequacy of the mechanical approach to causation. First, it considers whether the appeal to laws or invariant generalizations in characterizing interactions between parts of mechanisms either makes the mechanical theory circular or reduces it to a regularity, counterfactual, or manipulability theory. Second, the article discusses Machamer, Darden, and Craver's argument that the proper understanding of the causal productivity of mechanisms requires the recognition of the novel metaphysical category of activities. Third, it discusses the relationship between mechanical theories and process theories.
The biological and social sciences often generalize causal conclusions from one context to others that may differ in some relevant respects, as is illustrated by inferences from animal models to humans or from a pilot study to a broader population. Inferences like these are known as extrapolations. How and when extrapolation can be legitimate is a fundamental question for the biological and social sciences that has not received the attention it deserves. This book argues that previous accounts of extrapolation are inadequate and proposes a better approach that is able to answer methodological critiques of extrapolation from animal models to humans.
What distinguishes good explanations in neuroscience from bad? This book constructs and defends standards for evaluating neuroscientific explanations that are grounded in a systematic view of what neuroscientific explanations are: descriptions of multilevel mechanisms. In developing this approach, it draws on a wide range of examples in the history of neuroscience (e.g., Hodgkin and Huxley's model of the action potential and LTP as a putative explanation for different kinds of memory), as well as recent philosophical work on the nature of scientific explanation.
Between 1940 and 1970, pioneers in the new field of cell biology discovered the operative parts of cells and their contributions to cell life. Cell biology was a revolutionary science in its own right, but in this book, it also provides fuel for yet another revolution, one that focuses on the very conception of science itself. Laws have traditionally been regarded as the primary vehicle of explanation, but in the emerging philosophy of science it is mechanisms that do the explanatory work. William Bechtel emphasizes how mechanisms were discovered by cell biologists.
Reasoning in Biological Discoveries brings together a series of essays, which focus on one of the most heavily debated topics of scientific discovery. Collected together and richly illustrated, Darden's essays represent a groundbreaking foray into one of the major problems facing scientists and philosophers of science. Divided into three sections, the essays focus on broad themes, notably historical and philosophical issues at play in discussions of biological mechanism; and the problem of developing and refining reasoning strategies, including interfield relations and anomaly resolution. Darden summarizes the philosophy of discovery and elaborates on the role that mechanisms play in biological discovery. Throughout the book, she uses historical case studies to extract advisory reasoning strategies for discovery. Examples in genetics, molecular biology, biochemistry, immunology, neuroscience and evolutionary biology reveal the process of discovery in action.
The International Agency for Research on Cancer (IARC) is an organization which seeks to identify the causes of human cancer. For each agent, such as betel quid or Human Papillomaviruses, they review the available evidence deriving from epidemiological studies, animal experiments and information about mechanisms (and other data). The evidence of the different groups is combined such that an overall assessment of the carcinogenicity of the agent in question is obtained. This chapter critically reviews the IARC's carcinogenicity evaluations. First it shows that serious objections can be raised against their criteria and procedures - more specifically regarding the role of mechanistic knowledge in establishing causal claims. The chapter's arguments are based on the problem of confounders, of the assessment of the temporal stability of carcinogenic relations, and of the extrapolation from animal experiments. Then the chapter addresses a very important question, viz. how we should treat the carcinogenicity evaluations that were based on the current procedures. After showing that this question is important, the chapter argues that an overall dismissal of the current evaluations would be too radical. Instead, the chapter argues in favour of a stepwise re-evaluation of the current findings.
One of the main problems in establishing causality in medicine is going from a correlation to a causal claim. For example, heavy smoking is strongly correlated with lung cancer, but so is heavy drinking. There is normally held to be a causal link in the former case, but not in the latter. The Russo- Williamson thesis suggests that to establish that A causes B, one needs, in addition to statistical evidence, evidence for the existence of a mechanism connecting A and B. This thesis is examined in the case of the claim that smoking causes heart disease. It is shown that the correlation between smoking and heart disease was established by 1976 before any plausible linking mechanism was known. At that stage, there were doubts about whether a genuine causal connection existed here. Details of the history of research in atherosclerosis from 1979 to the late 1990s are then given, and it is shown that there is now a plausible mechanism connecting smoking and heart disease, and that, correspondingly, most experts now accept that smoking causes heart disease. This historical case study therefore provides support for at least one version of the Russo-Williamson thesis.