Introduction: biological proper functions

Every object exhibits a variety of causal properties. My set of Allen keys at home can be used to tighten or loosen certain specific screws, and they also make a loud crashing noise when I open the tool drawer quickly. A polar bear’s thick coat helps keep it warm, and it also makes the bear heavier (Goode and Griffiths 1995, 103). However, some of these causal properties are distinctive in an important way: they are functions of the object. It is a function of my Allen keys to tighten screws, but it is not their function to make crashing noises. It is a function of thick coats to insulate their owners from the cold but not to make them heavier. Functions are different from other causal properties of an object because they connote some kind of purpose: a function is what that object “is for”, and can be used to explain why the object “is there” (Garson 2017, p. 526). I have Allen keys in my house because of their screw-tightening properties, not because of the noise they make. This notion of “what something is for” is often referred to as the proper function of that thing. An object can have a proper function that, as a matter of fact, it does not (or cannot) perform. If one of the Allen keys bends so much I can’t use it to unscrew something, it still has that function. This is why we can say that a badly bent Allen key is dysfunctional. Similarly, when we claim that a function of polar bear coats is to keep polar bears warm, this means that the coats are there to provide warmth, and if a bear’s coat is unable to do this for some reason, it is unfortunately malfunctioning.

There at least appears to be an important difference between our two examples above. We know what the proper function of an Allen key is because it was designed (and purchased) for that purpose. However, in the absence of a clear designer, it isn’t obvious that we have similarly secure grounds for claims about the function of polar bear coats. In fact, given the absence of any explicit intentional design of biological structures, one might be suspicious that function-talk regarding the natural world merely reflects our own interests; that there is no mind-independent basis for the distinction between what some biological structure or behavior (merely) does and what it is for. The fact that thick fur coats are warm is more interesting to us than their weight. It is less clear that there is some independent biological fact which differentiates heat insulation from heaviness above and beyond that salience, such that one is a proper function, with all of the explanatory heft that accompanies this, while the other is not (Ariew and Perlman 2002, p. 2).

One reply to this kind of scepticism is to claim that the process of natural selection introduces at least a kind of design in nature, and so biological functions can arise in an appropriately analogous way to the functions of human artefacts. This design does not occur with the same intent and planning as in the case of human-made objects, but it is design nonetheless (so the thought goes). For example, natural selection favored the presence of thick coats in ancestral polar bears because they retain heat, not because they made the bears weigh more. Or put another way, if a thick coat didn’t retain more warmth, polar bears’ coats wouldn’t be as thick as they are, while if the coat weighed less (and still retained those insulating properties), it would remain just the same. So the reason current-day polar bears have thick coats is because this successfully helped keep their ancestors warm (Godfrey-Smith 1994; Griffiths 1993; Millikan 1989a; Neander 1991).

We can call this account of proper functions in nature the selected effects account, according to which “biological proper functions are effects for which traits were selected by natural selection” (Neander 1991, p. 168). This is by no means the only approach to biological functions or even the only approach to selected functions (Bigelow and Pargetter, 1987; Boorse 2002; Cummins 1975; Godfrey-Smith 1993; Griffiths 2009), but it is a prominent and popular one.Footnote 1

It is highly desirable to have an account of biological proper functions in place that does not refer to our contingent interests. A number of important concepts in biology rely on the idea that there are ways natural systems should be and ways they should not be, and that this distinction can be read off purely natural facts. For example, the concept of function is crucial to a great deal of basic biological description and reasoning. Without this concept, biologists would be unable to differentiate between cases of normal phenotypic variation and dysfunctional states (Matthewson and Griffiths 2017). This is why plants with white or purple flowers can be considered standard variants within a single species but a white panther cannot.

Furthermore, biological proper functions play an important role in a number of derivative notions in philosophy and elsewhere, particularly when we require a naturalized account of how things ought to be. It would be difficult to overstate the extent to which considerations of proper function implicitly and explicitly permeate debates in philosophy of biology and related fields (Garson 2019, 1). Here I will briefly mention two specific ideas that feature prominently in the philosophical literature.

First, a naturalized account of function and dysfunction is central to many analyses of pathology. Authors in the philosophy of medicine are divided regarding whether “objective” or “evaluative” considerations are most important in delineating the boundaries of disease, but there are good reasons to think a distinction between functional and dysfunctional biological systems should feature at least somewhere in the discussion (Murphy 2015; Neander 1983; Wakefield 2003). For example, we might think there ought to be some kind of biological restriction on what can be considered a disease, to definitively exclude cases such as drapetomania from medicalization, regardless of the opinions of society or even of the “sufferer” (Matthewson and Griffiths 2017). One candidate for this biological criterion is that disease can only occur when some structure fails to produce the effect for which it was selected in our ancestors (Neander 1983; Wakefield 1992, 2003). According to such views, disease can only be present when dysfunction is present. That is, it is at least a necessary condition of disease that some structure fails to perform its proper function.

A second notable example of the utility of the proper functions concept is the role it plays in some naturalized accounts of intentionality and content. According to these teleosemantic accounts (in very broad terms), the fact that some physical state constitutes a representation can be linked to the fact that it is produced by a system that has the proper function of generating such representations (Kingsbury 2006; Millikan 1984, 1989b; Papineau 1987; Shea 2013). It is an indispensable part of these accounts that we have a robust understanding of biological function that doesn’t rely on our interests or preferences, since such intentional states are part of what the frameworks set out to demystify.Footnote 2 The selected effects account provides just such an understanding.

Functions and the graded nature of selection

Given the central role played by natural selection in the selected effects account of biological proper functions, reference to these functions brings a number of empirical commitments. For example, to claim that some trait has some proper function is to claim that the entity which bears that trait has ancestors who were part of a population that exhibited variation, fitness differences and heredity with respect to this trait, and that those trait fitnesses manifested in a particular way. Sometimes vindicating these commitments is a relatively straight-forward issue (the gallbladder has been selected to store bile), and sometimes it is quite difficult (is there a function for humans developing “prune-fingers” in damp conditions (Changizi et al. 2011)?)Footnote 3 More broadly, however, a reliance on proper functions brings a responsibility to take the features of natural selection seriously. If there is a disconnect between the way we think about proper functions and the way natural selection actually works, this will undermine the account, which would in turn undermine those views that crucially rely on it.

For the rest of this article I will examine one such potential disconnect, which is this: biological effects are standardly considered to be proper functions or not in a binary manner, while the influence of natural selection in ancestral populations can be graded and partial. We don’t say storing bile is a function of the gall bladder to some degree; we simply say it is a function of the gall bladder. We don’t usually talk about traits having “a little bit” of a function, or refer to middling instances of representation when discussing teleosemantics.

Furthermore, this binary quality of function ascriptions is not just an easy way of speaking. When we consider the tasks we expect such claims to perform, at a first pass at least, it seems that those ascriptions need to be binary. Functions tell us what some structure ought to do or whether it is there to perform some task, and these are standardly thought of as on-or-off notions. It is a function of the redness of raspberries to indicate that they are ripe. It is not clear what it would mean to say that ripe raspberries moderately or partially ought to be red. Similarly, if we want to differentiate between a physiological variant and an abnormal phenotype on the basis of whether the trait is failing to perform its proper function, this demands a distinction between which causal properties are functions and which are not, rather than the degree to which these properties are functions. So it at least appears that functions need to be treated as binary if the concept is able to do what it is standardly deployed to do.

However, natural selection does come in degrees, and arguably admits of marginal cases. These two points therefore generate a tension which needs to be resolved in order for our account of proper functions, and the views that depend on it, to retain their naturalist credentials; credentials that are an essential part of the account’s significance in philosophical theorizing.

The paper proceeds from here as follows: First, I will outline three of the ways in which the influence of natural selection on current phenotypes comes in degrees; This will be followed by the primary claim of the paper—that notwithstanding the details, proponents of the selected effects account of functions owe us some kind of convincing response regarding the graded nature of natural selection; I then consider two of the more promising options for such a response; Finally, I note some of the ways each of these options will affect how we need to treat biological functions. No matter which option we choose, there will be significant alteration to how philosophers standardly think about and deploy the notion of biological proper function.

First example: differing force of selection

Some traits are under stronger selection than others. There has been selective pressure in the past for humans to have hairs in their nostrils, as this helps keep particulate material from entering their airways (Schwab and Zenkel 1998). However, this is nowhere near the level of selective pressure on the ability of the CFTR channel to transport chloride ions across cell membranes. If someone were to lose their nasal hairs they might end up with more respiratory infections than would have otherwise occurred (and this will have some attendant fitness effects), but this is in no way comparable to the fitness loss incurred by someone whose cells are unable to mobilize chloride ions appropriately. This latter phenotype means water cannot be transported effectively in multiple tissues, leading to the multiple health effects of cystic fibrosis, including reduced secretion of digestive hormones, severe respiratory disease, and infertility in males (Welsh and Smith 1993). This shows that the force of natural selection can operate in quantitatively different ways on different traits. Selection has acted on both CFTR cell gates and nose hairs in our ancestors. It has just acted on these traits with differing strengths. So natural selection isn’t merely present or absent; it comes in degrees.

There can even be variation in the strength of selection on different properties of the same structure. The shape of the sclera of our eyes has been under strong selection for the role it plays in vision (McBrien and Gentle 2003), but sclera are also important for cooperative interactions, as the distinctive contrast between their pale color and the iris makes it easier for others to observe what we are attending to (Langton 2010, p. 108, 112). Both of these properties of the sclera have undergone natural selection in our ancestral past, but one has been selected more forcefully than the other. If an ancestral human baby were born with a mutation that altered their sclera in a way that lowered visual acuity, this would be strongly selected against, while a reduction in the ability to convey one’s gaze direction would be selected against to some lesser degree. This is not just the point that a single structure may have multiple proper functions. It is the fact that in such cases, one of these attributes may be under markedly greater selection than the other.Footnote 4 Once again, natural selection isn’t just something that occurs or does not occur—there is a gradient in the strength at which natural selection occurs, even between different causal properties of the same structure.

Second example: changes in selective influence over time

The second instance of graded selection arises in the classic example of natural selection on peppered moth pigmentation.Footnote 5 Consider a population of moths living on and around a particular type of tree. Initially (so the story goes), these trees are pale-colored, so darker moths are more visible to predators, and therefore less likely to survive to produce offspring. This means there is positive selection for light coloration, with a corresponding prevalence of pale-colored moths in the population. However, at some point local industrialization and the attendant pollution leads to darker tree bark due to soot deposition. Selection pressure is therefore progressively reversed, leading to an increase in dark moth coloration within that population. Finally, improvements in environmental protections reduce soot emissions, the trees eventually return to their initial color, and so the population gradually reverts once more to a preponderance of light pigmented moths. Initial positive selection for pale coloration was followed by positive selection for dark coloration, which was in turn followed by selection for pale coloration once more.

This is once again an example of quantitative differences in the selection for biological traits. In this case however, the point is not that the overall force of selection on different traits can vary, but that the force and site of selection can change over time. Indeed, in this example two opposing traits have undergone the same level of selection pressure, but at different times. Light coloration and dark coloration have both been under selection in ancestral moth populations, but each of these traits has waxed and waned over different periods, at one time positively selected, and later selected against. So the different traits haven’t just been under selection or not; each has undergone selection at a particular time, which may be current, recent, far in the past, or any combination of these. Just like selective strength, the temporal distance of selective influence comes in degrees.

Third example: more and less paradigmatic Darwinian populations

A number of authors have recently proposed various accounts of the populations that may undergo natural selection (e.g. Godfrey-Smith 2009; Matthewson 2015; Millstein 2009, 2010, 2015). It is not enough to simply require that some collection of living things exhibits variation, fitness differences and heredity; this would include groups that clearly cannot undergo natural selection, such as those made up of organisms from extremely disparate species or that are completely isolated from one another. This means some other group-level properties must be met as a background for natural selection to occur, and in particular, authors have tended to focus on the idea that the group must be causally interconnected in the right way(s). For example, Roberta Millstein has argued that members of the population must be able to engage in shared Darwinian processes such as survival or reproductive interactions (2009; 2010), while Peter Godfrey-Smith argues that the important interactions are those that involve reproduction and reproductive competition (2009, p. 51). Importantly, each of these kinds of interconnections come in degrees. Some organisms will (positively or negatively) affect the reproductive prospects of each other very strongly, while others compete only minimally. Some populations will be very densely ecologically engaged with one another, while others will be sparse and scattered.

However, this does not simply mean that some groups can undergo natural selection and some cannot; it means some populations are more apt to undergo natural selection, or engage in richer kinds of natural selection than others. Perhaps a small number of widely dispersed organisms with plenty of nutrients and space undergo natural selection of a kind, but this is not as intense as the selection we see when the members of a single species fight over some essential resource in a harsh and limited environment. This invites the idea that there can be a spectrum of the Darwinian character of these populations, from “marginal” through to “paradigmatic” (Godfrey-Smith 2009). Once again, we see that the influence of natural selection is graded, depending not only on strength of selective pressure and temporal distance, but also on the properties of the population undergoing the selective process itself.

The fact that Darwinian populations are graded and how this relates to the attribution of functions has already been explicitly noted in the literature. In his article “A generalized selected effects theory of function” (2017), Justin Garson defends a particular variant of the selected effects framework. This still requires selection as in the standard account, but does not require reproduction, so differential retention alone is sufficient to generate functions. For example, Garson convincingly argues that brain plasticity can create new functions through the differential retention of neurons, in spite of the fact that the relevant cells do not have offspring.

Removing the constraint of reproduction leaves Garson’s account potentially vulnerable to a number of counterexamples. For instance, Justine Kingsbury has pointed out that differential retention is quite easy to come by; if some of the rocks on a beach last longer than others due to their differing hardness, this does not seem adequate reason to say rock hardness has the function of maintaining rocks (Kingsbury 2008). Karen Neander has pushed the point further (Garson 2017, p. 537), noting that differential retention can even involve competition of a sort. A set of rocks washing against one another on the seashore will have selective “winners” and “losers” due to their differing hardness. Nevertheless, it still seems questionable to claim this means hardness has a function in this context; that hardness “is for” promoting rock longevity.Footnote 6

Garson proposes a refinement to his account in order to address such cases, incorporating some of the work mentioned above regarding the extent of competition between population members (regardless of whether this competition is to reproduce or merely to persist). The idea here is that rocks of different hardness on a seashore are unlikely to be sufficiently interconnected with one another to form a paradigmatically function-generating population, given the way such a set-up would need to be arranged.Footnote 7

This raises a further question though: how interconnected do populations have to be in order to potentially generate functions? Garson suggests two possible approaches. One is to introduce a measure of interconnectedness and stipulate a value of this measure that delineates the boundary where a group “makes the grade” to be a population of the right sort. The other is to allow a spectrum which corresponds to the extent to which a population is interconnected, and have that reflected in the functional profile of its future members. For example, on the first approach a set of rocks on the seashore might be declared to simply be insufficiently interconnected to form a population capable of generating functions, while on the second approach, it would be considered a poor instance of a function-generating population (p. 538).

The distinction between these two ways to approach this issue and how we might choose one or the other is what will occupy us for much of the rest of this article. But before before addressing this, I want to pause for a moment to reiterate the points made so far and underline why this issue needs to be taken seriously.

A tension, and why it must be addressed

Given that the selected effects account of function relies on the action of natural selection, this account (and any other accounts based on it) must be sensitive to features of natural selection itself. We have seen three ways in which natural selection can be graded and partial, and this is nothing like a complete list. This means that our account of proper functions must be sensitive to the fact that natural selection comes in degrees. However, we standardly treat functions as categorical. We say that the trait “has” (or doesn’t have) this particular function, or that it “is there” due to natural selection acting on that particular property, as though the role of natural selection in the retention of traits isn’t multifaceted, graded, and highly variable.

This is not strictly an inconsistent position to hold; there may well be a way to reconcile these observations. I will shortly turn to explore some of the options we could pursue regarding this issue, but before that, I want to underline a specific point. Even if readers don’t agree with the material following this section, the argument still stands that anyone who wishes to invoke the notion of selected effects functions in their philosophical work needs to have at least something to say about this tension. This is actually the central point of the article—that any account of these functions needs to accommodate the graded nature of natural selection in some way or other, even if it is not one of the solutions I canvass here. Readers may in fact think they have a ready solution. However, as far as I am aware, this has not been seriously addressed in the literature, and we will see that working through the issue generates difficult choices which have potentially serious ramifications for further, derivative notions. So this is a tricky situation to work through systematically, and there is no easy fix, but we do need to properly address the problem.

One further comment before continuing. In what follows, I will sometimes need to refer to “the degree of selection”. I don’t want this to suggest that I think there actually is some kind of precise, quantifiable measure of selection available, such that we could say some trait is under selection “to degree 0.467” (or even degree 0.5). I actually think the prospects for developing such a measure are not at all clear, and this seems particularly conspicuous when we recognize that the three sources of graded selection discussed above are qualitatively distinct from one another. To be under weak selective pressure is not the same as arising in a marginal Darwinian population or having only been under selection far in the past. For example, although there may be some sense in which we could define an in-principle maximum for the degree of interconnectedness within a population undergoing natural selection (i.e. when every member is connected to every other member), it’s not obvious there will be such an upper boundary to selective force. One of these could therefore be represented with a normalized value relatively easily, while this approach may not be possible for the other. This suggests that whatever measures we adopt will likely need to be piecemeal, rather than there being any straightforward, unified approach.

However, I’m certainly not saying there cannot be any useful gradations either. Perhaps it will be sufficient to use relatively course-grained quantification in this context and make do by adopting the broad categories of “marginal selection” / “moderate selection” / “paradigmatic selection”. Perhaps a precise, weighted measure will be forthcoming in the future. In any case, although it will be impossible to talk about how to accommodate the graded nature of selection without using quantity terms, I don’t need there to be a precise method of evaluation to put forward the points I wish to make here. So I’m going to shelve such questions for now (which is not to say they aren’t serious considerations; they certainly are) and continue using a language of degrees of selection to discuss the potential ramifications of this in broad terms, without committing to the existence of a measure.

Outline of some possible approaches to the tension

First I will mention two potential responses for the purpose of discarding them from further consideration here. One is to do away with the problem by simply denying that natural selection is a graded phenomenon. Although it’s true that we often speak as though selection in biological populations is something that simply occurs or does not, I think the discussion above demonstrates that taking such an attitude seriously would simply run contrary to the biological facts. Another possible (but unattractive) response would be to claim that the tension between binary attributions of function and the biological facts of selection is enough reason to jettison the entire idea that there is a close link between functions in nature and the process of natural selection. Once again, this would certainly remove the problem in some sense, but it isn’t a viable approach for anyone invested in the selected effects framework. Besides which, we should at least see whether there is anything more interesting to be done before resorting to such a move.

On the assumption that we accept the graded nature of natural selection and remain committed to the idea that at least some biological functions rely on this process, it seems there are two primary options available to us. The first is to develop an account of how the graded input from the process of natural selection can be converted into a categorical output at the level of biological functions. The second is to alter the selected effects account to allow for biological functions to also come in degrees. I consider each of these in turn.

Degrees of selection, full function

The core proposal of this approach is that there is some threshold where the degree of natural selection is sufficient to generate a function, and anything below this simply will not generate a function. On this view, then, either the combination of selective strength, temporal proximity, and population structure (etc.) is adequate to generate functions, or it falls short and no function arises.

This approach has a great deal to recommend it, not least that it would respect the partial nature of natural selection while allowing us to retain our customary ways of talking about functions, and more importantly, we can retain the standard ways we use the notion, including in our downstream philosophical accounts. As we have seen, many of the theoretical tasks of proper functions do not lend themselves easily to a graded approach.

However, adopting this view will nevertheless still require a substantial revision of the standard framework. Neander’s definition presented above would have to be modified:

A biological proper function is an effect for which a trait was selected by natural selection to a sufficient degree.

How are we to understand the “sufficient degree” threshold in such an account? Even putting aside concerns about the prospects for a precise, unified measurement mentioned above, we are faced with a number of choices regarding how we treat this division between non-function and function, and each choice will have its own benefits and costs. For example, the line could be fixed at some particular level for all scenarios, but it could alternatively be contextual, or even relativized to features specific to the assessor such as their descriptive or explanatory requirements. Unfortunately, no matter which of these options we go with, it turns out to be very difficult to identify a threshold that is both plausible and well-motivated.

First, if the “sufficient degree” is set in a way that is interest-dependent or a matter of convention, this will eliminate any justifiable claim that the distinction between function and non-function is provided by purely natural facts, which was at least one of the primary motivations for the selected-effects view. So regardless of whether it is fixed or contextual, the threshold will need to be determined by some value-free rule, biological or otherwise. With this in mind, some of the more obvious principled, fixed cut-offs turn out to be simply implausible. For example, we might claim that any amount of natural selection is sufficient to generate a function. But hugely many phenotypic features are likely to have at least minimally enhanced fitness some of the time for some ancestor(s), in which case this rule would greatly over-generate a vast array of full proper functions.Footnote 8

Alternatively, we could fix the threshold as allowing only the most strongly or recently selected effects to count as a trait’s function. However, as we learned from the cases of the human eye and the moths, this would under-generate, resulting in too few functions for any structure or behavior that fulfils more than one function or had a recent change in selective environment.

Similar difficulties arise if we consider a contextually-determined threshold, as this would require an account of the contextual features which make that determination. Once again, if we wish to maintain a claim to objectivity, the relevant context will need to be dictated by some well-motivated biological principle(s) rather than our interests. However, it is not at all obvious what those contextual biological principles or rules would be. There’s no clear transition within biology I’m aware of that might mark a change from some degree of selection that won’t produce a function to a degree that will. This is not merely the point that such a boundary would be vague; it is that we need a biological reason to claim this boundary—vaguely specified or otherwise—should be placed on a continuous spectrum of selective force here, rather than there at all, in the absence of a biological distinction on which to base this putative transition.

In any case, I think there are deeper difficulties with any threshold view, whether fixed or contextual, related to the core motivations of the selected effects framework. Let’s take it for granted that we think biological functions arise through the process of natural selection. In that case, what would be the naturalistic incentive for retaining a binary view of functions in light of the recognition that natural selection comes in degrees? If the process that generates functions is graded, then without very strong reasons to think biological functions must be binary, it seems arbitrary and unmotivated for us to impose such a categorical distinction.Footnote 9 Apart from anything else, applying a threshold to the functions generated by a graded natural process would seem to introduce an implausible amount of decisiveness to the transition between non-function and function. If a population is only marginally Darwinian, and selection on the trait is weak, then what principled, naturalistic reason do we have to deny that this scenario might generate a marginal function? Once again, this is not a concern about vagueness; the concern would remain even if we allowed that such a distinction had blurred boundaries.

There is yet a further reason to reject the threshold framework, which is connected to the above points but perhaps more contentious. It seems (to me, at least) that retaining a binary view of biological functions would fail to recognize the lessons of the above examples. Consider the case of the moths when their trees return to a light color. Dark coloration has been under selection quite recently because it meant the bearer of the trait was harder to see against a dark background. However, light coloration also has a history of positive selection, and furthermore, dark coloration currently systematically reduces fitness due to a stable feature of the environment, while light coloration is under positive selection right now. Regardless of how we ended up converting the graded temporal distances in this case into binary outputs, it seems we would be committed to saying one of three things: that both light and dark coloration have functions (not being seen against light or dark backgrounds respectively) in just the same way and to the same extent; that neither has a function; or that one has a function but the other does not. None of these options articulates what really seems to be going on in this example—that light and dark pigmentation both have functions right now, but not in precisely the same way. One of these functions is currently on the wane, as its selective fortunes have been reversed.

The case of the eye is the same. It would be strange to say that aiding gaze detection is not a function of the sclera, and also strange to say that it is a function in just the same way and to the same extent as enabling effective vision. Rather, the lesson of these examples appears to be that in virtue of not all cases of selection being equal, the functions generated are not all equal either. I now turn to an approach that can accommodate this lesson.

Degrees of selection, degrees of function

The second obvious option is to reject the categorical attribution of functions and claim that, just as selection comes in degrees, so do proper functions. One again, this will require a revision of our standard definition:

A biological proper function to a particular degree (either precisely or imprecisely specified) is an effect for which a trait was selected by natural selection to a corresponding degree.

There is much in favor of this proposal in comparison to the threshold approach. First, it seems well-motivated in precisely the ways the threshold view is not. Given that we are committed to the ideas that natural selection underlies biological functions and this selection comes in degrees, then in the absence of good biological motivations to think otherwise, it seems we ought to expect that selected effects functions will similarly come in degrees. A graded view also avoids concerns about arbitrary or interest-dependent boundaries, or about the sharpness (or vagueness) of any such boundary, because there are no boundaries. If we reject an approach that requires a threshold, selected effects functions will be truly graded attributes. Furthermore, it does seem to give us the vocabulary to aptly describe the three types of scenario above: we can say that indicting gaze direction is (or was) less paradigmatically a function of the sclera than facilitating vision; that although dark pigment has the function of making moths difficult to see on dark backgrounds, this functional role is currently on the wane in comparison to that of pale pigmentation; and that marginal Darwinian populations can produce marginal functions. So far, so good.

However, there will be costs. The repercussions of adopting a graded framework will be many, including substantial revisions to how we standardly think and speak about biological functions. It’s difficult to gauge just how much upheaval this would involve, but even some of the more immediate upshots are quite striking.

To begin, our language regarding biological functions would need to be substantially revised. It would no longer be apt (or at least it would be quite imprecise) to simply say “the function of polar bears’ coats is to keep them warm”. Instead, the correct claim would be “keeping polar bears warm is a function of polar bear coats to degree X”, or at least “it is a function of polar bear coats to a high extent.”

Additionally, it would lead to a disanalogy with the functions of human-made objects. Although the function of certain artefacts can be transitional and ambiguous (I could start using my bent Allen keys as paper weights, and most pipe-cleaners no longer have the function their name suggests), in at least a great many circumstances it is entirely appropriate to simply claim that some property is an artefact’s function or not.Footnote 10 This will no longer be the case for the vast majority of biological functions, since almost any function generated by natural selection will be a function to some degree or other. As pointed out to me by Emily Parke, this disanalogy is not necessarily a bad thing, but it is a substantive upshot of the change in framework.

More significantly, however, in a great many cases it would be false to say that some effect is not a function of some biological trait. If functions come in degrees, and the degree of “function-hood” reflects the action of natural selection, then there will be very, very many effects that are functions to some minimal but positive extent. A selective event that enhanced retention of the trait, even if it only slightly influenced fitness, even if it happened in the distant past, and even if it occurred in a population that only minimally met the Darwinian requirements, would result in some non-zero level of function. It’s not entirely clear exactly how far this pervasiveness would extend, but it would be significant. For instance, we would probably need to give up some of our most standard examples of non-functions. It isn’t implausible that there have been instances in our ancestral past when having a quiet-thumping heart was selected against (failing to have a medical problem diagnosed, for example). In this case it would be false to make the seemingly obvious claim that making audible thumping sounds is not a function of heart valves.Footnote 11

Changing the way we talk about proper functions and non-functions might not be such a big deal. We might not be able to say making a thumping noise isn’t a function of heart valves, but we can still say it is only minimally so. This kind of thing is simply what results if we take the idea of graded functions seriously, and if that’s the right way to think about selected effects functions, then so be it. Perhaps some proponents of the selected effects view even consider themselves to have (at least implicitly) always thought about functions in just this way. However, it is still important to engage with the view carefully enough to ask whether a graded account will be able to underwrite the work we usually require from proper functions. Properly committing to thinking about these functions in a graded way will have far-reaching consequences. As discussed earlier, the ascription of proper function plays an extremely important theoretical role in multiple settings, and at least in the first instance, these roles seem best served by a binary notion. When we say the function of X is Z, we are saying that Z is why X is currently present, that X ought to Z, and it is malfunctioning when it doesn’t Z. These are standardly binary concepts which may, or may not, tolerate a shift to being graded, and in some cases it isn’t even clear what a partial reading of them would be.

Further, as noted in the introduction, reference to functions underwrites a great deal of work in philosophy of biology and biology itself. In particular, these changes will have notable impact when we again consider derivative notions such as dysfunction, disease and teleosemantic content. The effects of a graded view filter though to these downstream accounts, and it is not clear how well they can accommodate this change.

First, if there are a vast number of partial functions to be found in nature, this will vastly increase the opportunities for traits to fail to discharge these functions. Once again, this seems a relatively straight-forward corollary of accepting a graded view; that nature is not only full of partial functions, it is therefore going to be full of dysfunctions as well. But this becomes problematic when we consider the role proper function plays in some definitions of disease. Recall that these views claim pathology requires proper dysfunction, and this is intended to place restrictions on what can be classified as disease. If dysfunction is more or less ubiquitous, then this supposed restriction will be largely ineffective. For any putative disease state, there will be a good chance of at least some minimal dysfunction in the vicinity that could justify “pathologizing” that state.

It is important to note that the focus here is separate to the issue of clarifying when an organ that partially fails to discharge its (full) function counts as dysfunctional. For example, Peter Schwartz (2007) points out the difficulty of specifying when we should say that a heart which pumps blood suboptimally crosses the threshold from imperfect function to true dysfunction. In cases such as these, we might all agree that circulating blood is the (full) function of the heart, and provide a criterion for what counts as failing to deliver this (Schwartz employs the requirement that there must be negative biological consequences). The existence of partial functions raises a separate issue; namely that an organ might fail to discharge a partial, even minimal, function, either partially or completely. In this case, the question is: are there some functions which are so minimal, even total dysfunction won’t count as sufficient to underwrite pathology?

If we wish to salvage a significant role for selected effects functions here, we are faced with the question of whether attribution of pathology will require the failure of some adequately substantive function, with the return of all of the concomitant questions regarding how the threshold of “substantive enough” will be defined, and whether this could possibly be made precise and independent of our contingent interests. If this is not possible, selected effects functions will no longer underwrite an objective biological criterion of disease, and the primary motivation for its inclusion will be lost. It would be a significant upshot indeed if adopting a graded account of function entirely undermined one of the primary positions in the philosophy of medicine literature. Alternatively, perhaps we will be obliged to allow for a spectrum of pathologies, based on the degree of function that has failed. This would problematize any medical policies usually based on the distinction between disease and normal physiology, such as who merits government assistance and what distinguishes cure from enhancement. Would interventions on some physiological states now count as both, to a certain degree?

Teleosemantics appears to run into similar difficulties. If it becomes very cheap to generate minimal selected effects functions, this seems to threaten that representation will become similarly cheap. Additionally, teleosemantics puts even more pressure on whether the notion of partial, rather than binary, functions can be coherently imported to downstream concepts. Although speaking of “degrees of function” and “partial pathology” feels somewhat strange, I do seem to have some grasp of what these phrases might mean. But what it would mean for some state to exhibit a partial “degree of representational content” is not clear.Footnote 12

I can’t address these extremely interesting questions properly in this article, and besides which, I don’t want to risk the details diluting the main point of this section: namely, that the introduction of degrees of function will ramify widely, perhaps introducing a graded character to more concepts than we might have first anticipated. I think it will be better to simply point out that serious questions such as these will need to be addressed if we wish to adopt a framework that allows for partial selected effects functions, and leave it there for the meanwhile.

Summary

To recap the core claims of this article: First, natural selection does not occur in a binary fashion, as the strength, temporal distance, and population substrate of selection can vary in graded ways; Second, this fact must be addressed in some way or other by proponents of selected effects accounts of biological functions; Third, and most importantly, that such an accommodation looks to be not at all straightforward, no matter which direction we wish to pursue. However, although these points are hopefully by now relatively uncontroversial, I think the downstream ramifications are potentially very significant. In particular, if we wish to retain the features of biological proper functions that make the notion so attractive (differentiating between what some naturally-occurring structure or behavior does and what it should do, while retaining a strong grounding in objective biological facts), it seems we may be forced to substantially revise how we think and speak about these functions and derivative notions. I’m not sure about the best way to go here, although adopting an approach that allows for partial proper functions seems better motivated in spite of the difficulties that may arise.

I’ll finish with one last point, which I owe to a comment from Kim Sterelny. How we ought to understand the notion “function” is a contested issue. In this case, readers may see this article as an argument that selected effects functions cannot underwrite our analysis of disease or naturalized content. Perhaps this is all just ammunition for one of the competing accounts of functions in nature, such as those that refer to statistical normality, or causal contribution to current survival prospects or some other goal. If the selected effects account of functions commits us to hugely many partial functions, hugely many partial dysfunctions, and perhaps even partial content, then this might be considered good reason to prefer a competing view. However, in that case we should be careful to check that we won’t see partial functions occur under these alternative frameworks as well. At first glance at least, neither causal contribution nor statistical normality appear to be binary attributes either, so we may well find that graded functions arise here too. Perhaps partial functions arise in nature no matter what account we choose to adopt. And in that case, we’d better get used to them.