Mechanisms, Experiments and Memories (Part 3): Interlevel Experiments and Integration

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanation. For an index, see here.

I am currently looking at an article by Craver entitled "Interlevel Experiments and Multilevel Mechanisms in the Neuroscience of Memory".

In Part One, I covered Craver's non-formal account of mechanistic theories and applied it to the LTP mechanism (involved in memory consolidation). In Part Two, I discussed the multilevel nature of mechanisms, once again referring to the LTP example.

In this part, I outline Craver's taxonomy of interlevel experiments and his arguments in favour of theoretic integration (as opposed to reduction).


Interlevel Experiments Taxonomised
As detailed in part two, mechanisms are multilevel, i.e. one mechanism can be broken down into several sub-mechanisms and so on. This property can be exploited in designing experiments that fuel the goal of theoretic integration.

First, we need to grasp the basic tools needed for testing mechanisms. Recall, that mechanisms are organised collections of entities and activities that carry out some task or perform some function. To work out whether your proposed mechanism is the correct one you need to (i) identify a model system that you think has this mechanism (e.g. a strain of mouse); (ii) use an intervention technique (e.g. electrical stimulation); and (iii) use some detection technique (e.g. whole cell recording).

These elements are present in the basic experimental paradigm for mechanisms illustrated below. The image shows a hypothetical mechanism, an intervention technique and a detection technique. The intervention technique is temporally upstream from the detection technique. Note: this only shows one mechanistic level.

If we imagine a two-level mechanism, where the higher level mechanism is simply the role of the lower level mechanism, we can describe two basic types of interlevel experiments. First, there is the bottom-up experiment, where you intervene in the lower-level mechanism and detect the effects this has on the role. Second, there is the top-down experiment, where you intervene in the role and see what effect this has on the lower-level components. These are illustrated below.

To these basic categories, Craver appends three additional experimental strategies.

1. Activation Strategies: these are top-down experiments where you activate some role and detect its effects on its putative components. LTP research began when experimenters recorded from neurons in rat-hippocampi, while the rats were forced to navigate a maze.
2. Interference Strategies: these are bottom-up experiments where you diminish, retard, eliminate or disable some mechanism component and record its effects on some higher level role. In this history of LTP research, various interference methods have been used such as ablation and special breeding ("knockout" mice).
3. Additive Strategies: these are bottom-up experiments where you try to stimulate, augment, hasten or multiply some component of the mechanism. This is often done in mouse genetics, where researchers breed mice that over-express a particular gene.

Theoretic Integration

A central aspect of contemporary neuroscience is its integrative pretensions (this is one of Craver's core arguments - defended in other papers that I will be looking at). In other words, neuroscience does not seek to reduce behavioural theories to biochemical theories; rather, it tries to integrate these theories into a coherent overall picture.

As he describes it, integration has two elements to it: (i) "looking up", i.e. showing how a component is involved in a mechanism; and (ii) "looking down", i.e. showing how an entity can be explained in mechanistic terms.

The interlevel experiments and strategies show how each level of the mechanism is ontologically robust. They do so because independent lines of investigation all converge on the same mechanism.

Turning to LTP, one of the ways in which neuroscientist's decide whether or not they are dealing with a real mechanism is by integrating molecular mechanisms with cellular mechanisms, cellular mechanisms with neural mechanisms, and neural mechanisms with behavioural mechanisms.

Mechanisms, Experiments and Memories (Part 2): The Levels of the LTP Mechanism

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanations. For an index, see here.

I am currently looking at an article by Craver entitled "Interlevel Experiments and Multilevel Mechanisms in the Neuroscience of Memory".

Part One sketched Craver's informal account of mechanistic theories and applied it to the LTP (long-term potentiation) mechanism in memory-neuroscience. This part continues by examining the levels in the LTP mechanism. This paves the way for a taxonomy of the interlevel experiments that are used in neuroscience.


Three Types of Level
One of the crucial features of the LTP-mechanism is that it occurs in a nested hierarchy of mechanisms. This description suggests that we can talk meaningfully of different ontological levels of organisation. If we do so, we must be cautious. Craver identifies three different uses of the term "level".

First, there is the aggregative use of "level". This is the idea that entities are mere aggregates of smaller (i.e. lower-level) entities. This brings with it the idea of aggregative decomposition. You can decompose a ball of wax by cutting it into smaller pieces; you can recompose it by glueing these pieces back together. The aggregative use of "level" is concerned with relations of size, and not with activities and functional organisation.

Second, there is the functional use of "level". This describes relationships between abstract roles. Functional decomposition arises when you break one task down into a number of sub-tasks or sub-routines. This is sometimes done in neuroscience but there is a tendency for it to float-free of ontological reality (they are how-possibly models).

Third, there is the mechanistic use of "level". This arises when one mechanism is broken down into sub-mechanisms, which are in turn broken down into sub-sub-mechanisms. This is the preferred use of "level" in neuroscience and we can see it at work in the case of LTP.


The Levels of the LTP Mechanism
As mentioned in Part One, the LTP mechanism strengthens synaptic connections and is implicated in spatial memory and learning. Viewing it from the perspective of Craver's philosophy, we can discern at least four mechanistic levels.

We begin at the behavioural-organismic level. Suppose we are trying to study spatial memory and learning in the mouse. We will usually do so by placing a mouse in a maze of some sort and then subjecting them to different behavioural tests. From these tests we work out the conditions under which spatial learning takes place.

We then move to the computational-hippocampal level. This is our first glimpse beneath the skin of our model organism. We look to the structure and connectivity of the hippocampus and the various computational processes it performs learning. We investigate this level with direct stimulation, surgical resection, EEG, MRI and computer-modeling.

We next move to the electrical-synpatic level. Here we are examining how neurons are stimulated, the volumes of glutamate they release, the new growth in dendritic spines and so on. We investigate this level with microelectrodes and pharmacological agonists and antagonists.

Finally, we reach the molecular-kinetic level. We now look at NMDA and AMPA receptors, diffusion rates of Mg2+ and Ca2+, glutamate binding and so on.

These four levels are illustrated in the diagram below.

With this understanding of nested hierarchies and mechanistic levels in place, we can proceed to consider experimental and integrative (non-reductive) strategies in contemporary neuroscience. This will be done in the next part.

Mechanisms, Experiments and Memories (Part 1): The LTP Mechanism

This post is part of my series on the work of the philosopher Carl Craver. The post deals with the nature of neuroscientific explanation. For an index, see here.

Over the next few posts, I will examine the following article:

Craver, C. "Interlevel Experiments and Multilevel Mechanisms in the Neuroscience of Memory" (2002) 69 Philosophy of Science 83-97.

This is one of my favourite of Craver's pieces. It offers a beautiful illustration of the methods and strengths of mechanistic explanations in neuroscience.

Craver's article has three aims:

1. Offer a non-formal account of mechanistic theory structure in neuroscience.
2. Develop a taxonomy of interlevel experiments that are used to explore proposed mechanisms.
3. Use this experimental taxonomy to explore the integrative nature of neuroscientific theories.

As always, to help in achiveing his more abstract philosophical goals, Craver uses a concrete example as a reference point. On this occasion, the example used is that of spatial memory and LTP.

In this part, I will cover Craver's non-formal account of mechanistic theories.

Mechanistic Theories: A Sketch

For people who have been reading other entries in this series, the idea of a mechanism will be old hat. However, in this article Craver offers an excellent summary of the main points. It is worth repeating them here.

According to Craver, mechanisms are collections of entities and activities organised in the production of regular changes from start to finish. There are four important concepts concealed in this definition.

First, mechanisms consist of entities. In neuroscience, the relevant entities are ions, neurotransmitters, membranes, neurons, brain regions and whole animals (e.g. mice).

Second, mechanisms consist of activities. These are the processes or doings in which the entities participate. In neuroscience, relevant activities include: binding, phosphorylation, hydrolysis, firing, processing and so on.

Third, mechanisms are organised. That is: the entities and activities have a very specific spatial and temporal organisation. If they were not organised in this way the mechanism would not work.

Fourth, the mechanism does something: it carries out some task, produces some output or performs some function. We refer to this as the "role" of the mechanism.

A schematic diagram (based on one in Craver's article) is offered below. It illustrates all four features of a mechanism.

The Mechanism of LTP
Long-term potentiation (LTP) is a mechanism for strengthening the synaptic connections between neurons. Behavioural evidence suggests that it is crucial to the process of memory-consolidation. It is frequently studied in the hippocampal region of the mouse brain. The hippocampus has long been identified as a key brain region for memory formation. Many of the studies involve mice learning to navigate through mazes. Thus, spatial memory is the main type of memory being examined.

As with all things in science, there is no complete consensus on the mechanisms underlying LTP. Nonetheless, a great deal has been learned and there is a staggering amount of detail to current theories. If you would like to learn more, I suggest this lecture by Eric Kandel (lecture 4, 2008 is the relevant one). The following is merely a sketch.

LTP increases the effect of a presynaptic neuron on a post-synaptic neuron. Neurons that exhibit LTP use the neurotransmitter glutamate. This is released from the presynaptic neuron and binds to receptors on the postsynaptic neuron. One such receptor is the NMDA receptor. Glutamate changes the conformation (i.e. spatial orientation) of this receptor, thereby opening a pore in the membrane of the postsynaptic neuron.

Most of the time, this pore is blocked by magnesium ions (Mg2+), but when the postsynaptic neuron is depolarised it unblocks and calcium ions diffuse into the cell (Ca2+). These Ca2+ ions set off a cascade of biochemical reactions, which eventually result in new protein-synthesis by the DNA in the postsynaptic neuron. These proteins are used to make new dendritic buds (i.e. new locations at which the presynaptic neuron can exert an influence).

I will provide some diagrams of this process in the next part. For now, the words must paint the picture.

The LTP mechanism has the fourfold structure alluded to earlier:

• It consists of entities: neurons, synapses, glutamate, NMDA, Ca2+, Mg2+ etc.
• It consists of activities: binding, diffusing, changing conformation etc.
• It is spatially and temporally organised: first binding, then changing conformation, then diffusing, then protein synthesis etc.
• It has a role: it strengthens synaptic connections.

There is another crucial feature of the LTP mechanism, but I will discuss that in the next part.

Top-Down Causation (Part 3): Some Problem Cases

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanations. For an index, see here.

I am currently looking at an article written by Craver and William Bechtel entitled "Top-down Causation without Top-down Causes".

In Part One, I detailed how mechanisms occur in nested hierarchies. In Part Two, I showed how this can seem to allow for interlevel causation (i.e. between different levels of mechanisms). I also covered how some popular ideas about causation do not seem to allow for interlevel causation. Finally, I sketched Craver and Bechtel's solution to the problem.

They argue that what is referred to as interlevel causation arises from the constitutive relationship between a mechanism and its parts (including sub-mechanisms). A mechanism is a spatiotemporal organisation of parts. So a change in the parts constitutes a change in the mechanism. And likewise, a change in the mechanism as whole constitutes a change in its parts.

The Craver-Bechtel solution allows for standard interpretations of causation to apply intra-level.

This part covers the application of the Craver-Bechtel solution to some supposed cases of bottom-up and top-down causation.


1. Bottom-Up Causation?
Suppose you are infected by a virus of some kind. Suppose further that the virus kills you. Is this a case of bottom-up causation? The affirmative answer arises from a belief that viruses and humans exist at different "levels".

Craver and Bechtel think the affirmative answer results from a misunderstanding of what a "level" is. There is nothing about the structure of mechanisms that prevents small things (like viruses) having an effect on a big thing (like a human). For example, I can swat and kill a fly; a spark can ignite a fuel tank; and molecule can exert a gravitational force on a planet. There is nothing problematic about inter-size causation.

A second case might pose more of a problem. Take someone who dies of a heart attack. The heart is very definitely a sub-mechanism within the human body. Is it thus wrong to speak of a heart attack causing someone's death?

Well, if someone dies of a heart attack, it is because the oxygen supply to their brain and other bodily organs is cut-off. This results in the shut-down of all the mechanisms that constitute the human body. We can trace the breakdown of individual sub-mechanisms to intralevel causes and we can trace the breakdown of the overall mechanism to the constitutive relations.

Indeed, the heart attack scenario provides a great illustration of the Craver-Bechtel approach to causation.

2. Top-down Causation?
Let's take a look at Roger Federer. An effortlessly elegant tennis player if ever there was one. However, even Federer, when he starts playing tennis places his body under stress. He needs to metabolise glucose to keep going. This involves blood-borne glucose being taken-up through the cell membrane, then being phosphorylated and being bound to molecules of hexosediphosphate.

The question is whether the tennis-playing causes the increase of glucose metabolism. If it did, this would make tennis-playing a classic instance of top-down causation.

Craver and Bechtel argue that the glucose metabolisation can be given a complete explanation by using their mechanistic model. First, when Federer starts to play, nerve signals are sent from the brain to the muscles. This causes the muscles to metabolise available ATP to ADP, which results in muscle contraction. The increased levels of ADP make that molecule available as a receptor for phosphate in high energy bonds at the end of the glycolytic process. This allows a cascade of reactions earlier in the glycolytic pathway to proceed, part of which involves the glucose reaction described earlier.

This is an incomplete sketch because a full description of the series of biochemical reactions would be incredibly long-winded. The main point is that there is a mechanistic story to be told all the way down and up.

Craver and Bechtel offer another useful example of this drawn from the movie Citizen Kane. However, it deals with memory mechanisms. I will be discussing these in later posts and so will avoid repetition here.

Top-Down Causation (Part 2): Constitutive Relations

This post is part of my series on the work of the philosopher Carl Craver. The series focuses on the nature of neuroscientific explanation. For an index, see here.

I am currently looking at an article Craver wrote with William Bechtel entitled "Top-down Causation without Top-down Causes". In Part One, I covered the nature of mechanisms and the different levels they contain.

In this part, I cover three things. First, why there might appear to be interlevel causes in mechanisms. Second, why there are problems with this idea of interlevel causation. And third, an answer to the problem of interlevel causation.


1. Why might we think there are interlevel causes?
By "interlevel" causation is meant both top-down and bottom-up causation. Recall, that mechanisms are made up of components (activities and entities) that are organised such they perform some function or produce some change. They usually occur in nested hierarchies (i.e. the components of one mechanism are themselves made up of mechanisms).

Top-down causation would arise when either (a) the mechanism as a whole has an effect on its components or (b) a component of a higher level mechanism has an effect on a component in a lower level mechanism. A bottom-up cause would arise when the opposite took place. Neuroscientists and biologists often talk of interlevel causation, why?

Craver and Bechtel suggest that it is because of the experimental methods employed by such scientists. Specifically:

• They sometimes try to interfere with the components of a mechanism to see the effect on the overall function of the mechanism. This creates the impression that the components have causal (bottom-up) effects on the mechanism.
• They sometimes try to interfere with the overall function of the mechanism to see the effect this has on its components. This creates the impression that the mechanism has causal (top-down) effects on the components.

We can give examples of each.

First, bottom-up causation. Lesion methods are often used to study how different regions of the brain effect some function performed by the brain. The most famous patient in the history of neuroscience was a man named Henry Molaison. He had large portions of his hippocampus and temporal lobe removed in order to relieve chronic epilepsy (see image below). This had a profound effect on his ability to form new long-term memories. In other words, an interference with certain components (the hippocampus and temporal lobe) had an effect on the mechanism for memory formation.

Second, top-down causation. Functional MRI is a great tool for imaging brain activity. It would be great to use it in such a way as to figure out which brain regions are active during particular cognitive tasks. The problem is that the brain is always active and performing multiple cognitive tasks. In order to work out which activity is related to which cognitive task, careful experimental design is required.

This is where the method of cognitive subtraction becomes relevant. Suppose we want to know which part of the brain is responsible for thinking about pelicans (you never know...). We need to ensure that we isolate the pelican-region and differentiate from other bird or animal related regions. To do this, experimenters might get subjects to think about owls for one round of imaging and pelicans for another round. By subtracting one image from the other, they will isolate the pelican-region. (This is a very rough idea of cognitive -- for more than you might care to know, I suggest the following series).

In the case of cognitive subtraction, an interference at the functional level of the mechanism seems to have an effect on the components of the mechanism.

2. Why is Interlevel Causation Puzzling?

There is no definitive account of what causation really is. Nonetheless, the idea of interlevel causes seems to conflict with four popular intuitions on the nature of causation.

First, there is the physical transfer intuition. This is the idea that causation always involves the physical transfer of something from one set of events or objects to another set of events or objects. This does not fit with the idea of interlevel causation because the higher levels in a mechanism subsume the physical parts of the lower level. So there is nothing that can be transferred from one level to another.

Second, there is the nonoverlap or nonimplication intuition. This is the idea that causes and effects must be wholly distinct, ontologically separate things. This does not fit with the idea of interlevel causation because the higher levels do overlap and imply the lower levels.

Third, there is the temporal preceding intuition. This is the idea that causes must precede their effects. Again, this fails to fit with the idea of interlevel causation because lower levels are subsumed within the temporal extension of the higher levels.

Finally, there is the asymmetry intuition. This is the idea effects cannot alter causes, but that causes can alter effects. This does not fit with the idea of interlevel causation, because most of the alleged examples of this phenomenon are symmetrical nature: they involve a lower level having an effect on a higher level and a higher-level having an effect on a lower level. (Read that a few times in case it sounds confusing).

3. Interlevel Constitutive Relations

Craver and Bechtel do not really dispute these intuitions about the nature of causation. They argue instead that the phenomenon of interlevel causation is better understood in terms of the constitutive relations between the levels of a mechanism. This allows for the standard interpretations of causation to remain relevant when assessing the isolated levels of a mechanism.

The idea of constitutive relations is described by the authors as follows:

the mechanism as a whole is fully constituted by the organised activities of its parts: a change in the parts is manifest as a change in the mechanism as a whole, and a change in the mechanism is also a change in at least some of its component parts. There is no need to extend the word "causation" to cover cases of this sort...

That's it for now. In the next part, we will apply this idea of constitutive relations to some classic problem cases.

Top-Down Causation (Part 1): Mechanisms and Their Levels

This post is part of my series on the work of the philosopher Carl Craver. The series looks at the nature of neuroscientific explanation. For an index, see here.

Over the next few posts, I will be looking at the following article:

Craver and Bechtel "Top-down Causation without Top-down Causes" (2007) 22 Biology and Philosophy 547.

The article deals with the problem of top-down causation. This is the idea that the higher-level, abstract properties of a mechanism can causally affect the lower-level components. If you have ever wondered whether your thoughts cause your neurons to fire, or whether your firing neurons cause you to think, then you have wondered about this problem.

The difficulty with top-down causation is that it seems ontologically spooky. Craver and Bechtel offer a qualified defence of the concept. They suggest that what is referred to as top-down causation is a real phenomenon, but that it is best explained in terms of the constitutive relations between the levels of a mechanism.

There is a lot of technical jargon being employed here. In this first post, I will try to cut through the jungle of jargon.



1. What is a Mechanism?
The first thing we need to know is: what exactly is a mechanism? This question has been exhaustively answered elsewhere on this blog. In the interests of brevity, only three properties of mechanisms are described here:

• A mechanism is a spatially and temporally organised collection of entities and activities.
• Mechanisms are affected by and have effects on "things".
• The individual entities and activities of the mechanism do not have the same effects in isolation from the mechanism.

An example might help to flesh-out these ideas. Take an MRI scanner. It maps the spatial and temporal distribution of certain molecules in the human body. So, it is a mechanism with a particular cartographic capability.

An MRI scanner is a long tubular construction consisting of three coils. The first coil generates a magnetic field, the second corrects for inhomogeneities in the magnetic field and produces fields that vary along three dimensions, and the third sends out radiofrequency pulses and records the echoes that are returned. When put together, these coils have certain effects on the hydrogen atoms in the human body. Effects that allow us to build-up a 3-d image of the tissues in the body.

On their own, the components would not have these capabilities; together, they do. Thus, we can begin to see how talk of top-down causation is so common.

2. The Levels of a Mechanism

Of course, to fully appreciate the attraction of top-down causation we need to appreciate how there can be different "levels" in a mechanism. The human eye provides a beautiful illustration of this phenomenon.

The eye is mechanism that transduces (i.e. changes) a photonic signal into a neural or electrical signal. This is the overall change that is brought about by the mechanism. The overall mechanism is itself constituted by separate mechanisms. First, there is a mechanism for collecting the light; then a mechanism for focusing the light; then a mechanism for projecting the light onto the retina; and then a mechanism in rod and cone cells for absorbing the light and converting it into a neural signal.

There is a well-known mechanistic story to be told about what happens within the rod cells as well. The absorbed photons change the conformation of rhodopsin (a protein found in rod cells). This stimulates G-proteins, which in turn set-off a biochemical cascade leading to the hyperpolarisation of the cell. If you would like to get more detail on this story, I suggest the following video (long, but worth the effort):

The eye is thus to be thought of as hierarchical nest of mechanisms. The question to ask is: what distinguishes higher levels from lower levels? The informal way of looking at it is that higher levels are spatially and temporally extended, when compared with lower levels, and that the higher levels are aggregated from the lower levels.

We can turn this into a more formal definition:

 An item (X) is at a lower level than an item (S) if and only if X is a component in the mechanism for some activity (V) of S. And X is a component in a mechanism if and only if it is one of the entities and activities organised such that S Vs.

So, G-proteins are at a lower level than rod cells, because G-proteins are part of the mechanism that allow rod cells to send neural signals. And rod cells are in turn at a lower level to the eye because they are part of the mechanism that allows the eye to transduce photons into neural signals.


That's it for Part One. In Part Two we will consider exactly why interlevel causation might be thought problematic and sketch Craver and Bechtel's solution.

When Mechanistic Models Explain (Part 3): Evaluating Models

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanations. For an index, see here.

I am currently looking at an article by Craver entitled "When Mechanistic Models Explain". Part one covered some distinctions between explanatory models and non-explanatory models. Part Two considered the example of the Hodgkin-Huxley model of the action potential and how it transitioned from being a how-possibly (non-explanatory) model to a how-actually explanatory model.

In this part, I will set-out Craver's evaluative criteria for mechanistic models. The criteria help to distinguish good explanatory models from bad non-explanatory ones. In deriving these criteria, we consider four questions:

1. Does the model cover all the relevant phenomena it purports to explain?
2. Does the model identify all the entities/parts involved in producing the target phenomenon?
3. Does the model identify the activities involved in producing the target phenomenon?
4. Does the model successfully organise the entities and activities?

Before looking at each of these questions in more detail, it is worth refreshing our memories about the nature of a scientific model. A model takes a target phenomenon (T) and represents it in some algorithm or function (S). S can then be implemented in a physical system, computer program, mathematical equation, box-and-arrow diagram or whatever.

(1) Does the model cover the target phenomenon?

The Hodgkin-Huxley model (S) is a mathematical model of the action potential (T). The action potential is the sudden change in membrane potential in a neuron. It is a multi-faceted phenomenon. It involves the sudden increase in potential, followed by a rapid decrease and recovery period. The first test of a successful model is whether it captures all facets of T.

The second test is whether the model can describe the inhibiting conditions of T. In other words, if we know that action potentials can be prevented or inhibited (e.g. by applying tetrodoxin, which blocks the flow of sodium ions through the ion channel), then our model should be able to account for these inhibiting conditions.

The third test is whether the model can identify the modulating conditions of T. In other words, the model should be able to tell us whether an alteration in certain variables will affect T. So in the case of the action potential, we want our model to tell us if variations in the density of ion channels or the diameter of the cell have an effect on it.

The fourth test is whether the model covers what happens in non-standard conditions. For instance, most laboratory experiments occur in non-standard conditions, the difference between these and standard conditions should be covered by the model.

The fifth test is trickier. One thing that distinguishes a how-possibly model from a how-actually model is that the latter will account for byproducts in the mechanism being modeled. So for example when sodium ion channels are opened there is a slight movement of positive charge to the outside of the cell, this is quickly counteracted by the flow of the positively-charged sodium ions into the cell. The slight movement of positive charge to the outside would not be an essential part of a how-possibly model of the action potential because it is an irrelevant blip. However, it would be an essential part of a how-actually model. More recent models of the action potential cover this blip. It turns out it is caused by a particular component of the ion-channel (the alpha helix, discussed here).

(2) Does the model identify the entities/parts responsible for the phenomenon?

Mechanistic models explain phenomena in terms of entities and activities. A key part to a successful mechanistic explanations is the correct identification of entities. By "correct" is meant actually existent not fictional.

The line between a fictional entity and one that actually exists is indistinct. For instance, one could argue that Mendelian genetics started out with fictional entities (genes), but that this definitely changed through repeated experimentation, manipulation and confirmation of Mendel's model. Despite the blurry lines, we can be confident that the entities actually exist if:

• They constitute a stable cluster of properties: for instance, ion channels were originally fictional, but over time it became possible to sequence their structure and work out how they reacted when in the presence of certain chemical agonists and antagonists. Thus ion channels were found to have a stable cluster of properties.
• They are robust: they can be identified with different techniques and devices. For instance, ion channels can be identified through pharmacological manipulations, X-ray crystallography, and electron microscopy.
• They can be used to intervene in other processes: for instance sodium ion channels can be manipulated so as to open potassium ion channels.
• They are plausible in the circumstances being investigated: this criterion varies from case-to-case. In the case of the action potential, any proposed entities should be plausible given background knowledge about the construction of nerve cells.
• They are relevant to the target phenomenon: in other words, when modeling the action potential we should not include entities that are present in the cell but clearly have no role in changing membrane potential. Examples might include: DNA, RNA, and microtubules.

(3) Does the model identify the relevant activities?

As mentioned above, a mechanistic explanation is built out of entities and activities. The activities are the things that the entities do (e.g. sodium ions diffuse down concentration gradients). It is possible to identify fallacious activities. For example, one could observe that the rooster always crows before the sun rises. But it would be a mistake to think that crowing was what brought about the sun rise. Mistaken causal inferences of this sort are everywhere (astrology, homeopathy etc.).

So how do we know if we have correctly identified the activities? Craver proposes one criterion that can help answer this question: the manipulability criterion. If we propose an activity linking two variables (e.g. roosters and sunrises) then it must be the case that manipulating one of these variables manipulates the other. Thus, silencing the rooster should -- if it were a correctly identified activity -- stop the sunrise. Of course it doesn't, so this is a bad mechanistic model.

(4) Does the model correctly organise the entities and activities?

The final crucial criterion for a good mechanistic model is its organisation. In other words, the model must have the correct spatial and temporal organisation of the entities and activities identified. It cannot be the case that taking out one entity and relocating it produces the same phenomenon.

There is quite a bit to take in here. In the interests of cognitive perspicuity, the following diagram summarises the criteria we need for evaluating mechanisms.

When Mechanistic Models Explain (Part 2): The Hodgkin-Huxley Model

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanations. For an index, see here.

I am currently looking an article by Craver entitled "When Mechanistic Models Explain". In part one, I introduced the concept of a scientific model and differentiated between models that explain and those that do not.

In this part, I will cover the Hodgkin-Huxley model for the action potential. The goal is to show how this model went from being non-explanatory to explanatory.


(1) What is an Action Potential?
A neuron is a cell. Like most cells it has a lipid bilayer membrane. This membrane is semi-permeable: most of the time it keeps what is inside the cell distinct from what is outside the cell, but it does occasionally allow materials to cross the membrane.

In its resting state, the membrane of the neuron separates electrically-charged ions (primarily Ca, K and Na ions) in such a way that the inside of the neuron is negatively-charged when compared to the outside. This difference in charge known as the resting potential of the membrane (roughly -70mV).

When the neuron is stimulated, there is a dramatic change in the distribution of charge. The membrane becomes highly-permeable to positively-charged ions (K, Na), which enter the cell resulting in a sudden increase in membrane potential (up to +35mV). If a certain threshold is reached, this process continues down the length of the neuron. This constitutes the electrical component of neuron-to-neuron communication (there is a chemical component too).

The sudden change in membrane potential (from -70mV to +35mV) is known as the action potential. With appropriate equipment, one can insert an electrode into a neuron and record the action potential. The data from the recording can be graphed as follows.

The graph shows the sudden rise in membrane potential, followed by a sharp decline and a recovery (or refractory) period.


(2) The Hodgkin-Huxley Model
Recall from part 1, that a scientific model is something that takes a target phenomenon (T) and represents it as an algorithm or function (S). S can be implemented in a physical system, computer program, mathematical equation and so on.

The Hogkin-Huxley model takes the action potential as its target phenomenon and represents it in a mathematical equation. This is illustrated below.

I'm not a huge math person but I think it is relatively easy to get the gist of this equation. It is showing that the total amount of current crossing the membrane is made up of four separate currents (capacitative, potassium, sodium and leakage). These currents are in turn dependent on a number of different variables. The overall equation is derived from certain laws of electricity. In particular, Coulomb's law and Ohm's law.

The question now is whether the model is explanatory or non-explanatory. Interestingly, Hodgkin and Huxley thought it was non-explanatory. They saw it as a how-possibly model. That is, as a model that showed how the changes in membrane potential that were observed could possibly come about. They did not think it accurately represented what took place in the neuron.

They were perhaps slightly unfair on themselves. They were aware that sodium, potassium and other ions crossed the membrane, and they included this in their model. However, in doing so they noted that the ability of these ions to cross the membrane was dependent on three factors (m, n and h in the equation). What these factors were was a mystery.

What Hodgkin and Huxley lacked was an understanding of ion channels. These are assemblages of protein that are embedded in the membrane of the cell. Under certain conditions, they open pores in the membrane (details here) through which ions can pass. The different proteins that make up these channels are the mysterious factors that were unaccounted for by Hodgkin and Huxley.

Experimental work on the structure of these channels, and on the conditions in which they open, was what allowed the Hodgkin-Huxley model to shift from being a how-possibly (non-explanatory) model to a how-actually (explanatory) model.

When Mechanistic Models Explain (Part 1): Models and Explanations

This post is part of my series on the work of the philosopher Carl Craver. The series deals with the nature of neuroscientific explanations. For an index, see here.


Over the next few posts, I am going to take a look the following article:

Craver, C. "When Mechanistic Models Explain" (2006) 153 Synthese 355-376

The article looks at how models are used in science, and at how phenomenal models are distinguished from explanatory models. Craver uses an example to guide his discussion: the Hodgkin-Huxley model of the action potential, derived from experimental work done with the giant axon of the squid (not the axon of the giant squid). The action potential is key to understanding how nerve cells communicate.


When Hodgkin and Huxley first proposed their model, they thought it was non-explanatory (merely phenomenal). It has since become an exemplar of an explanatory model. Craver examines how this happened.


In this first part, we will look at three important distinctions that need to be understood before moving on to consider the Hodgkin-Huxley case: (1) the difference between a phenomenal and an explanatory model; (2) the difference between sketch-models and complete explanations; and (3) the difference between how-possibly models and how-actually models.


(1) Phenomenal Models versus Explanatory Models
Models are used all the time in science. So what is a model? Craver offers a skeletal account: a model is an abstract description of a real system. The scientist will start constructing a model by first identifying a target phenomenon (T). This could be anything. For example, it could be the motion of the planets across the night sky or even how humans learn language.


Having identified T, the scientist will construct an algorithm or function (S) that can reproduce something similar to T. S can be implemented in a physical system, written in a computer program, captured in mathematical equations, or sketched in block-and-arrow diagrams. As an example, an orrery is a model of the solar system.

Models can be explanatory or non-explanatory. A non-explanatory model can be phenomenally accurate without being true (it can save the target phenomenon). For instance, the Ptolemaic model of the solar system is a good phenomenal model, but it is not descriptively accurate. To underline this point, you might enjoy watching Carl Sagan's discussion of Ptolemy.





Now we must ask the question: what differentiates a merely phenomenal model from an explanatory model? Craver puts forward an instrumentalist answer: an explanatory model is more useful for the purposes of control and manipulation. So, if we want to send spacecraft to the outer solar system, we need to work with Newtonian models, not Ptolemaic models.


(2) Sketches versus Complete Descriptions
The next distinction of which we need to aware is that between mechanism sketches and ideally complete mechanistic descriptions. 


Craver wants us to imagine that all mechanistic models are aligned on a spectrum. At one extreme of the spectrum we have vague mechanism sketches. A classic example would be the box models that are used to explain cognitive phenomena. For instance, I might say agency (the ability to act) is the product of a three part mechanism involving a sensor (something that captures information from the environment), a processor and an actuator (something that performs the action). Such a mechanism sketch is just a representation of ignorance. I have no idea what really takes place during the processing stage.


At the other end of the spectrum lie complete descriptive mechanisms. These would faithfully replicate all the details of the target phenomenon. No mechanistic models reach this ideal since they all abstract away from the details somewhat. A more approachable ideal might be the mechanism of chemical neurotransmission, discussed previously.

(3) How-Possibly models versus How-Actually Models
The final distinction is that between models that describe possible mechanisms for producing the target phenomenon and models that describe the actual mechanisms through which the target phenomenon are produced.


The majority of models in artificial intelligence (even connectionist models) are of the how-possibly variety. For example, you might write an algorithm that can do facial or voice recognition. This algorithm is a model, but it may not bear any resemblance to how the human brain manages to do voice and facial recognition.


Okay I'll leave it there for now. In the next post I'll cover the Hodgkin-Huxley model and show how it was transformed from a non-explanatory how-possibly model into an explanatory how-actually model

Thinking About Mechanisms (Part 3): Hierarchies, Schemata and Intelligibility

This post is part of my series on the work of the philosopher Carl Craver. The series focuses on the nature of neuroscientific explanations. For an index, see here.

I am currently looking at an article entitled "Thinking About Mechanisms", which Craver published with two other philosophers in 2000. The article introduces us to the contours of Craver's work.

Part One outlined Craver et al's formal analysis of mechanisms. Part Two cashed-out the value of this formal analysis by taking a detailed look at the depolarisation mechanism in chemical neurotransmission.

This final part will consider three philosophical issues: (1) the hierarchical nature of mechanistic explanations; (2) the importance of mechanism schemata in scientific discovery; and (3) the intelligibility of mechanistic explanations.

Before we begin, it is worth recalling the formal definition of a mechanism that is being appealed to:

Mechanisms are entities and activities organised such that they are productive of regular changes from start, or set-up, to finish, or termination.

(1) Hierarchies in Mechanistic Explanations
Mechanisms occur in nested hierarchies. This can be seen in the depolarisation example given in Part Two. The depolarisation mechanism is part of the mechanism for chemical neurotransmission, which is in turn a part of the mechanism for neuron-to-neuron signaling, and neuron-to-neuron signaling is part of virtually every cognitive mechanism.

This implies that mechanistic explanations are not necessarily reductive. Indeed, Craver et al argue that they work by linking together multiple levels of entities and activities. In this sense, mechanistic explanations are constructive, not reductive.

That said, the hierarchies do bottom-out at some point, namely: the lowest level entities and activities that are dealt with by a particular field of inquiry.

In molecular biology, the lowest level entities are the macromolecules, molecules and ions that are involved in biological processes. Molecular biology rarely peers below the atomic layer, but it is not an impassable barrier. As science progresses, the relevance of lower level entities may become more apparent.

As for the lowest level activities, the authors identify four that are of interest to the molecular biologist:

1. Geometrico-mechanical activities: these are familiar from classical mechanics. They include activities such as fitting, turning, colliding, bending, pushing and so on.
2. Electro-chemical activities: these are the attractions, repulsions and bondings that constitute the field of biochemistry.
3. Energetic activities: these are activities that involve thermodynamic processes. For example, diffusion of molecules along a concentration gradient.
4. Electro-magnetic activities: these are only occasionally used in molecular biology, but they are crucial to understanding the conduction of electrical impulses along nerve cells.

The authors argue that the history of science is largely the history of new mechanistic explanations at different levels in a hierarchy. Which brings us to the next point.

2. Mechanism Schemata and their Uses

A mechanism schemata is an abstract description of a type of mechanism. It usually pinpoints the entities and activities that are involved in the mechanism. An example (used by the authors) is Francis Crick's description of the central dogma of molecular biology. The authors attribute this to Jim Watson, but I'm sure it was Crick who originally formulated it. Anyway, it is illustrated below.

Crick's central dogma schema shows the entities and activities involved in protein synthesis (proteins being the essential building-blocks of biological cells). The schema is abstract because it does not contain a lot of detail; but it still has great explanatory scope because it explains a process shared by virtually all biological organisms (viruses are an exception).

Mechanism schemata of this nature are essential to scientific discovery for two reasons:

• They can help to develop predictions. For example, the central dogma schema, with some detail about the DNA code, predict the order of amino acids in a protein.
• They provide blueprints for designing research protocols. The experimenter can try to intervene in some part of the mechanism and observe the changes this results in.

The authors provide a lengthy illustration of these two virtues of mechanistic schemata by looking at the history of molecular biology. I am going to skip this. There will be plenty of time for detailed examples in later posts.

3. Intelligibility

The final point about mechanistic explanations is that they render the universe intelligible. They do so by setting out an elucidative relation between set-up conditions and intermediate activities (explanans) and termination conditions (explanandum). It is important to note that intelligibility has nothing to do with truthfulness: an intelligible explanation may be false. Intelligibility is all about cognitive attractiveness.

At the present moment in the history of molecular biology, the bottom-out activities and entities listed above occupy a privileged explanatory position: they have been used in numerous situations and subjected to numerous experimental manipulations. This could change.

Thinking about Mechanisms (Part 2): The Depolarisation Mechanism

This post is part of my series on the work of the philosopher Carl Craver. The series deals with neuroscientific explanations. For an index, see here.

I am currently looking at an article entitled "Thinking About Mechanisms", published in the journal Philosophy of Science in 2000. Although authored along with two other philosophers of science, this article provides a valuable introduction to the main contours of Craver's work.

In part 1, I introduced two examples of biological mechanisms (Chemical Neurotransmission and DNA replication) and outlined a formal definition of a mechanism. This definition was as follows:

Mechanisms are entities and activities organised such that they are productive of regular changes from start, or set-up, to finish, or termination.

In this part, I take a more detailed look at the mechanism for chemical neurotransmission and see how the formal analysis applies to it.


The Mechanism of Chemical Neurotransmission
Chemical neurotransmission involves the conversion of an electrical signal in a presynaptic neuron (action potential), into a chemical signal in the synapse. This chemical signal is converted back into an electric signal in the postsynaptic neuron.

There are many stages to this process: (i) first there is a depolarisation of the presynaptic neuron; (ii) then an influx of Calcium ions into the presynaptic neuron; (iii) then there is a series of chemical reactions which transport vesicles of neurotransmitter to the membrane of the neuron; (iv) then the neurostransmitter is released into the synaptic cleft; (v) then the neurotransmitter diffuses across the synaptic cleft; and (vi) finally the neurotransmitter binds to receptor proteins in the postsynaptic cell.

This is an incredibly complex sequence of events, about which a great deal is known. It is far more complex than my summary suggests. Indeed, that summary is only an isolated snapshot of what goes on. But isolated snapshots are the essence of mechanistic explanations.

To grasp this point, and to consider in more depth how the formal definition of mechanisms applies to chemical neurotransmission, we will look solely at the first stage in the process outlined above: the depolarisation of the presynaptic neuron. This is its own sub-mechanism within the larger mechanism of chemical neurotransmission.


The Depolarisation Mechanism
In their resting states, neurons are electrically polarised. That is: the fluid inside the neuron is negatively charged with respect to the fluid outside the neuron (-70mV). This is referred to as the resting membrane potential. Depolarisation is a positive change in membrane potential.

There are three parts to the description of any mechanism: (i) the set-up conditions; (ii) the intermediate activities; and (iii) the termination conditions. They illustrate the descriptive adequacy of the formal definition of a mechanism that was provided earlier. Let's look at each of these parts in the case of the depolarisation mechanism.

(i) The Set-up Conditions
We begin with an idealised description of the entities (and their properties) that make up the mechanism. The description is idealised in that it is represented as being static. In the depolarisation mechanism, the set-up conditions include:

• The overall structural properties and spatial locations of the entities.
• The differential intra and extra-cellular concentrations of Na+.
• The location of the alpha helix protein in the ion-channel. This protein is a string of evenly-spaced positively charged amino acids.
• The hairpin turn in the protein making up the ion channel. This has a particular configuration of positive and negative charge that becomes important in the intermediate activities.
• Background conditions (held constant) such as temperature, pH, and the presence or absence of pharmacological agonists or antagonists.

These set-up conditions are illustrated in the diagram below (click to enlarge).

(ii) The Intermediate Activities
Having grasped the set-up conditions we can look to see what goes on during the intermediate activities. The entities participate in these activities and they jointly produce the termination conditions of the mechanism. In the case of the depolarisation mechanism, the intermediate activities can be broken down into four parts:

• The action potential, which is an electrical signal traveling down the neuron towards the synapse, spreads some Na+ ions through the interior of the cell. They repel the positive charge in the alpha helix voltage gates.
• The alpha helices are thus rotated about their axes, thereby opening a channel through the membrane.
• These changes in protein structure contort the hairpins such that they now line the channels.
• This makes the channel selective for Na+ ions that are outside the cell and these ions are transported into the neuron.

This stage is illustrated below.

(iii) Termination Conditions

The final part of a mechanistic explanation is an idealised description of the termination condition. In other words: a static description of the changed status of the entities from the set-up conditions.

In the case of the depolarisation mechanism, the termination conditions are easily described: there is an increase in intracellular Na+ concentrations and a corresponding increase in membrane voltage.

This is illustrated below.

So there you have it, a more detailed look at a mechanistic explanation in action. In the next part, we will examine some of the more philosophical questions that arise from these explanations.

Thinking about Mechanisms (Part 1): Introduction to Mechanisms

This post is part of my series on the work of philosopher Carl Craver. For an index, see here.

Craver specialises in fleshing out the nature and status of neuroscientific explanations. His particular interest is in mechanistic explanations. This is also one of my own core research interests.

Today, I am going to take a look at an article that Craver published with Peter Machamer and Lindley Darden (two other philosophers of science). He is not the lead author on this piece but it is an important introduction to some of the themes in his work. The full reference for the article is as follows:

Machamer, Darden and Craver, "Thinking About Mechanisms" (2000) 67 Philosophy of Science 1-25

Those of you with a subscription, can access it via JSTOR.


Mechanistic Explanations are...
Mechanistic explanations are commonplace in science. In particular, in molecular biology and neurobiology. Yet there has been a paucity of serious philosophical analysis on the nature of mechanistic explanations. The authors try to provide this analysis.

To follow them, we need to first look at the types of questions mechanistic explanations are supposed to answer. Mechanisms are usually proposed in answer to questions about how something comes about or how something happens. Here are two questions that cry out for mechanistic answers:

1. How are signals passed from neuron to neuron?
2. How does DNA replicate?

Examples of mechanistic explanations will help to ground the more philosophical analysis that follows, so we will look at rough answers to each of these questions (we do not need detailed versions now, just general outlines).

The answer to the first of these questions runs along the following lines: neurotransmitter molecules are released by the presynaptic neuron into the synaptic cleft (gap between neurons), these neurotransmitters bind to receptor-proteins on the post-synaptic neuron, this depolarises the post-synaptic neuron which can either inhibit or promote further signaling. The image below illustrates this mechanism.

The answer to the second question runs along the following lines: the DNA double helix is unwound by a helicase enzyme, the charged nitrogenous bases (which make up the structure of DNA) are thus exposed and complementary nitrogenous bases bond to them with the help of additional enzymes. After a couple more stages (apologies for the ambiguity), the DNA molecule is replicated. The image below provides an illustration of this.

To reiterate, more detailed accounts of these two mechanisms will be presented later.

A More Formal Look at Mechanisms

With these examples in tow, we can begin to take a more formal look at mechanisms. The authors provide the following definition:

Mechanisms are entities and activities organised such that they are productive of regular changes from start or set-up to finish or termination.

That's about as dry and non-commital as definitions come, but it does contain two important concepts:

1. Activities: these are producers of change. In the neurotransmission example, the activities are "diffusion", "binding", "depolarisation" etc.
2. Entities: these are the things that engage in activities. In the neurotransmission example, the entities are neurotransmitter molecules, neurons, receptor proteins, ion-gated channels etc. To participate in an activity, an entity must have certain properties, e.g. geometric structure or charge.

This approach to mechanisms is deliberately dualistic in that it does not prioritise either activities or entities. Some ontologists do this. For example, substantivalists try to reduce all activities to entities and their transitions; while process theorists try to reduce entities to activities. It important to see that neither dominates: they are interdependent.

Functions

Functions are also central to mechanistic explanations. The most oft-cited example is the heart. It is said that the function of the heart is to pump blood. The authors argue that this description of function is wrong because it reduces a function to a property possessed by an entity (the heart "has" the property of blood-pumping).

A better definition (again, a little dry) would be the following:

A function is a role played by both entities and activities in a mechanism.

On this definition, the correct description of the heart's function would be:

The heart has the function of pumping blood and thereby delivering oxygen and nutrients to the rest of the body.

Okay, that's it for now. The next part will take a more detailed look at the mechanism for chemical neurotransmission and see how the formal analysis applies to it.