A Defense of the Mechanistic Explanation, and a Refutation of the Mechanistic Hegemonist

Foreword

This is the second, and longest paper from Philosophy 951: Larry's Favorite Papers. This paper is split into three parts, a defense of the mechanistic explanation, putnam's peg and micro-explanations, and an argument against the mechanistic hegemonist. I provide sufficient context in the paper, but a little research into those concepts would be helpful. I start by giving defenses of the mechanistic explanation at all, as a handful of peers gave (what I consider to be) poor arguments against the explanation at all (not even against the hegemonist). I then argue against Putnam's argument against the micro-explanation through a creative, but not damning, argument appealing to natural language. Finally, I argue against the hegemonist motivation, and appeal to the practicality of the functional explanation, even if it is just a subclass of the mechanistic explanation.

Body

I. Defense of the Mechanistic Explanation

There exists a dispute between proponents of the functional and the mechanistic explanations. A functional explanation is an explanation that is concerned with the purpose of a component in the system being explained. The mechanistic explanation, on the other hand, focuses on the ‘entities and actions’ of a system. Per Craver, “A distinctive feature of mechanistic explanation is its emphasis on information about the component parts of a system, their activities, and the spatial and temporal constraints on their organization in virtue of which they together produce the system’s behavior.” For example, consider a doorbell. A functionalist may explain the workings by saying something like, “A switch is pushed, which causes an attractor to turn. The attractor pulls a hammer toward a bell. The hammer strikes the bell, which causes it to ring, while at the same time turning off the attractor. Etc.” While the mechanistic explanation would go something like “We can explain the electric bell as having entities: switch, battery, iron core, steel spring, iron armature, screw, hammer, bell, wiring, and activities: switch is pushed, which closes an electrical circuit, which causes electricity from the battery to flow through the wiring, which magnetizes the iron core, …, which causes the hammer to hit the bell until the switch is released.” Craver and Piccinini argue in their paper ‘Integrating psychology and neuroscience: functional analyses as mechanism sketches’ that all functional explanations are just incomplete versions of mechanistic explanations. The intuition is that the mechanistic explanation seems to have all of the same information as the functional one, while also providing more information about the spatial-temporal and material aspects of the mechanism and its components. There were many criticisms of the mechanistic explanation, most I find ungrounded. I will present some arguments in favor of the mechanistic explanation as a whole, and then discuss why it still should remain separate from the functional explanation.

Some people have the intuition that only functional explanations are able to dictate physical constraints or vice versa. If I describe the function of, say, a valve cap, does this limit what possible materials could make up the mechanisms of the valve cap? Or, given the material of some component mechanism, are the possible functions it can realize limited? Some intuit one or the other, I argue it is quite clear that they both restrict each other. Consider again the valve flap. It clearly cannot be realized by gelatin or any comparably flimsy or weak substance. Looking at any gelatin conversely prevents any function that involves it being a valve cap. If, given an original function f1, we can say that a certain material cannot realize it. Conversely, if we iterate through the set of all things that the given material can realize F, we will not find f1. Both dictate each other.

To hammer this intuition, consider the logarithmic function, which we know is undefined for negative numbers. It is clear that the set of numbers that can be used in this function is restricted; it cannot be negative. Thus, we claim that because of how the function is, it restricts the count of elements that can “realize” it. It would be weird to say, though, that if I look at -1, I cannot make any claim about what type of functions it can be used in. I can name many that it can be used in, absolute value, cubic, etc., yet I can also say, “Because -1 is negative, it cannot be used in the logarithmic function,” with absolute certainty.

However, the question of whether the function or the materials define the other, remains open. It seems like the function of something is secondary to the materials that make it up; the first seems like it could be a construct, while the second seems necessary. This implies to me that the latter must take precedence, but arguments can go either way.

There was quite a lot of discussion about the granularity of mechanistic explanations, and how such unclarity of how levels work presents issues for mechanistic explanations. I find this argument quite weak. First, consider the idea of multilevel mechanistic explanations.. Any given mechanism can be made up of smaller component mechanisms. A mechanistic explainer has the option of going into detail on how those component mechanisms work or keeping them abstracted as the component mechanisms. For example, one ought not necessarily be worried about exactly how a battery works to understand how a doorbell works. However, if you are interested in how batteries work, nobody will stop you from adding details about how the battery transmits electricity throughout your mechanism; both are valid explanations, and both remain mechanistic. Furthermore, your interest in the battery’s mechanisms does not obligate you to explain the intricacies of the switch and how that enables electricity to flow. Multilevel explanations are a way of clarifying that a mechanistic explanation can go into greater or less detail to explain how some part of the mechanism works. The term multilevel explanation may be a misnomer, as it may imply that the entire explanation needs to be on the same ‘level’, but any part of the explanation can be at any different level.

Some functionalists may argue that the idea of ‘levels’ is weak, considering that it is often the case that different levels interact with each other (consider the levels of an ecosystem, you could go down to the level of population in a mechanistic explanation, but that population necessarily interacts with the community level, thus the idea of levels does not exist). This position assumes that, again, going down or up a level in any component of the mechanism obligates you to do so with all other components of the mechanism. This claim is uncharitable, there is no reason why, if you want to discuss the intricacies of a battery, you are then required to discuss the details of the switch. The functionalist may have some sort of argument that this disparity of levels necessarily leads to an incomplete or unintelligible explanation, but that requires much further argumentation: the provided examples show that we are intuitively allowed to do so.

Another argument on why the functional explanation is better was provided during the discussion. It was argued that changing the internal structure of a machine would leave the mechanistic explanation confused, but the functional explanation undamaged. Consider again the doorbell with its functional and mechanistic explanations. Imagine you give a functional explanation to person A, and a mechanistic explanation to person B. Furthermore, you describe to both people that the wires have now been changed to some completely different structure that can still transmit information from the switch. The argument goes that the functional explanation remains undamaged, while the mechanistic explanation has no clue what has happened, nor if the doorbell remains working.

This argument only works as a counterexample if the level of detail of the mechanistic explanation is lower (that is, less detail, or more vague) than what changed in the system. Because the mechanistic explanation can go up and down in granularity for each component, we can describe that component in greater or less detail. Consider a dumbwaiter with a bell that calls it to your room. You ought not to need to know how the doorbell functions to conjure up a reasonable mechanistic explanation of how the dumbwaiter works without needing to go into such details. Furthermore, this means we are justified in making any abstraction from the lowest conceivable explanation to the highest explanation: ‘a doorbell is a doorbell’ (this could be argued to be no explanation at all, but we will assume this to be valid). We discuss this to point out that we can make a perfectly valid mechanistic explanation simply by describing the wire as an ‘information transmitter.’ This abstraction of the component that transmits the information may seem too general at first, but grant that this is the case. Thus, giving someone a mechanistic explanation with this level of granularity, and then describing to them that the wires were switched with some water pipes, no longer confuses the mechanistic explanation, as both are just ways the information transmitter can be realized.

Some may argue that this type of generalization goes against the ideas of the mechanistic explanation. They would argue that the entire point of these explanations is to pay attention to the physical structure and temporal location of components, and this generality loses it. I argue against this. ‘Information Transmitter’ still has spatial, temporal, and physical constraints, though less specific. It still needs to transmit information between the other components, and it can probably not be made out of Sour Patch Kids. For analogy, consider the battery. The battery is essentially just a specific name for something that stores and transmits electricity. This is the same with the “information transmitters”. There exists a type of battery that only starts transmitting electricity once it interacts with water. The water battery and regular alkaline battery can both realize a ‘battery’ despite their clear differences. Regardless, we can still make some spatial distinctions; the battery needs to connect to the wires. The same goes for the other restrictors. Sure, both may not be described with specific physical matter to realize them, but we can make some claims about what type of materials make them up; the materials must help them serve their function. Both can not be made of gelatin, both cannot exist miles away from the rest of the mechanism, and both cannot exist hundreds of years after the rest of the mechanism.

Circling back to the objection, if we assume that the level of granularity of the mechanistic explanation is lower than that of what got replaced (i.e., the wires are described mechanistically), then saying that the mechanistic explanation can’t explain it simply begs the question. Furthermore, ascribing that same restriction to the functional explanation leaves the person given equally confused. If we describe wires in the mechanistic explanation as “a thing that transmits electricity, which is made of a material that conducts electricity well, that is in a position that transmits electricity from the battery to the switch, etc.” then clearly if we switch out the wires for some water pipes, person B will be uninformed. On the other hand, if we have a functional explanation at the same level of abstraction, we would describe the wires as “a thing that transmits electricity from one component to another”. Clearly, if the wires were swapped for something that does not transmit electricity, person A would be just as confused. When our explanations have more detail than the type of thing that could get switched out, regardless of which explanation we give, the person simply would not inherently understand those higher-order concepts.

A functionalist may still want to argue that people may be able to deduce these higher-order concepts. They may want to further argue that functional explanations would give people better tools to deduce those concepts. This needs further argumentation, as it seems in most scenarios where a mechanistic explanation can be given, and assuming they are at the same level, that the mechanistic explanation does not simply encompass the functional one.

A functionalist may have one final argument. They may say that the more general the mechanistic explanation, the more it seems like it just becomes a functional explanation. However, I don’t see this as an issue. Look no further than the battery in the doorbell. We can think of many different ways for a battery to be realized, through some standard flow of cathodes and anodes, or some thermo-neuclear reaction in the battery, or through solar panels from the outdoors, or from a windmill. Insofar as the functions of the battery are realized, this abstraction is perfectly valid. If we allow this type of abstraction for the battery, there is no reason why we shouldn’t allow that abstraction for other parts of the mechanism.

All of this discussion may lead one to believe that I agree with Piccini and Craver: that all functional explanations are just incomplete mechanistic ones. I have so far argued that if a mechanistic and functional explanation exists for the same mechanism, and that they both reside at the same level for the components, the mechanistic explanation is at least as good. I also believe that it has more information. It is hard to say that this necessarily makes it better; more information does not always make an explanation better. However, I do find the idea of functional explanations being mechanistic appealing. Consider Strenberg’s task analysis of how the brain searches for numbers using memory. The task analysis diagram drawn up does look functional. However, it is not hard to imagine that there are some mechanistic limitations that we are yet to understand. Limitations that could very well be added to the diagram. Though this is hedging on speculation. My intuition is that maybe, instead of describing the functional explanation as an incomplete mechanistic explanation, it would be better to think of the functional explanation as a type of mechanistic explanation. I am not subscribed to this position, though, as I understand that some things may seem to be functional explanations and cannot be mechanistic.

II. Putnam’s Peg and the Micro-Explanation

We now must spend time analyzing the intuitions people have for the reducibility of these two concepts. I think a good place to start is Putnam’s peg. As Sober has argued, both the macro and micro descriptions explain why the board couldn’t fit through the peg. I wish to argue that the micro explanation is rather the only thing doing the explaining, while the macro explanation is what is conventionally useful.

Consider an alternative version of Putnam's peg. Imagine there is a board, with a square cutout that is one inch by one inch. And further, a peg that is one inch by one inch, by six inches, that fits through the board. If it fits through the board, then Putnam would argue that it is because of the macro properties. I wish to provide an argument about the degrees of precision that I hope goes against Putnam’s intuition.

When I say, “This peg is one by one by six”, I am making a claim about the dimensionality of the board. Consider, though, what the board actually looks like. Wood often has many irregularities, grooves, splinters, etc., that make it not take up exactly inches of space. Then, strictly speaking, my first statement is false because of these imperfections. These are the degrees of precision in language. When I say a table is flat, I do not mean that there is not a single molecule that is higher in space; I mean roughly, it is flat. Thus, when I make the aforementioned claim, what I mean is something like “This peg is one by one by six, plus or minus epsilon”, where epsilon is the implied degree of precision error margin. The presented issue is that we can now imagine scenarios in which some peg is described to be able to go through the hole, but it is not necessarily the case that it will. Imagine that the hole is perfectly cut with the best technology we have available to only be able to fit things that are exactly one inch by one inch; any amount more, even by the tiniest factor possible, will no longer allow something to pass through. Thus, assuming we take the definition given before, something colloquially described as being ‘one inch by one inch’ no longer passes through in virtue of that macro explanation.

The issue with this is that, assuming some non-zero degree of precision, the following counterfactual becomes untrue: “If the particles in the peg and board had been different, the peg still would have passed through, as long as the macro-properties were still as described.” There are now examples of macro-property descriptions in which different micro-property configurations no longer fit through the hole.

This discussion supposes some type of degree of precision in ordinary language. There are a few possible lines a dissenter may take. Initially, one may argue that the degree of precision we are talking about is zero: the item must be exactly those dimensions. This has a lot of merit, it seems like descriptions should be the type of thing that has no margin of error when speaking epistemically. Though again, imagine a piece of wood. Even if we stipulate that we have no margin of error, wood is not a completely flat material. Even to a microscopic level, it is not possible for each grain or particle to be completely flat on any board. However, if we want to say that these measurements are absolute, we must claim that they take up exactly that much space. Thus, the claim that “this peg is one by one by six” is impossible: there will always be space between particles. Thus, we either need some nonzero degree of precision, or the causal reason that the peg fits through the board is because it has the property “takes up less space than a one-by-one hole”. This fits neatly into what Putnam is claiming. I may stipulate that the claim that ‘a peg is 15/16ths of an inch by 15/16ths of an inch’ is not possible, but that ‘a peg is less than 15/16ths of an inch by 15/16ths of an inch’ would be a causal claim that can guarantee that the peg can fit through the board. However, we will discuss how this conclusion is not the most preferable.

It is important to note that, in ordinary language, this translation is not what one means when giving measurements of a peg. When we say something has certain measurements, we are giving an estimate of the size, not an upper bound. This leads us to conclude that, in the sense of ordinary language, it is not the macro-level description that does the explaining, but rather the micro-level.

The last idea I wish to push against is Putnam’s description of the micro-explanation as “terrible.” As my discussion has hopefully shown, it is not clear whatsoever that the macro-explanation is as simple and useful as Putnam initially attempts to portray. Much more nuance is required for his description. I think a large intuition behind how the micro-level explanation is the only explanatory one is that the micro-level explanation instantiates the macro one, not the other way around. The converse of this is the idea that any macro-description can be realized by a multitude of micro-level structures. This issue is that many macro-descriptions can do the same explanatory work. Because our definition of something that fits is now “a peg fits because it is smaller than the hole”, it is realized by an infinite number of macro-explanations. Such as “this peg is less than 15/16ths of an inch by 15/16ths”, “this peg is less than 14/16ths of an inch by 14/16ths”. Are these infinite explanations all amazing, compared to the single ‘terrible’ micro-explanation of “this is how the material is made and this is what would happen if you attempted to put it through the board?” Considering a similar counterfactual to the one that Putnam provided, it seems like these infinite macro-explanations are bad, in comparison to the most general “a peg fits because it is smaller than the hole”. I am sure this is not what Putnam would want to conclude. That explanation simply begs the question; it is not doing any explanatory work.

To me, it seems that macro-explanation purely depends on how the micro-explanation is. Macro-properties are simply conventions of instantiated micro-properties. This would imply to me that micro-properties are former to the macro. If this is true, macro-properties don’t really exist.

Furthermore, I wish to distinguish between the epistemology and usefulness of reduction. Even if one still thinks the macro-explanation is significantly better than the micro, it is pretty clear that we at least can reduce Putnam’s peg to be explained purely on the basis of the micro explanation, and, if we accept my argument, this is actually the only explanatory aspect of why the peg can go through. This isn’t a claim against the usefulness of these generalizations. Just as the reductionist would not get much use out of studying particles firing in the brain as compared to the psychologist, it is hard to see how we would even begin to understand micro-properties and how they interact with the world. The opposite is true, these macro-generalizations are more practical for doing things. Unfortunately, practicality and causality differ.

III. An Argument Against the Mechanistic Hegemonist

In my other paper, I argued that the mechanistic hegemonist still must ascribe useful information from psychology. However, I wish to argue that any functional explanation may still be useful.

The hegemonist may finally wish to argue that explanations with greater detail serve a greater purpose. That, because the functional explanation lacks more information, and it is fundamentally worse. This assumes the premise that “explanations with greater detail are better.” This is clearly problematic, even for their position. This adheres to the idea that a lower-level mechanistic explanation is also always better than a higher-level mechanistic explanation. This makes them claim that the physical level is the best explanation. Again, it still may be the case that such a lowest-level explanation is the only one doing the explaining, but if the hegemonist ever thinks that a more abstract explanation provides use, they also ought to claim that this reductio must lead them away from the idea that greater detail is more useful. It is clear, at least currently, that economic explanations are more practical for explaining currency exchange than physics. Thus, in an explanatory sense, these things may not be autonomous, but for any useful part of our society, we should confidently claim that this autonomy does nothing to undermine the usefulness of the functional explanation.