1. Introduction

Theories of function are conventionally divided up into two main categories, historical and ahistorical (or backward looking and forward looking). The selected effects theory (Neander Reference Neander1983, Reference Neander1991; Millikan Reference Millikan1984) is an example of a historical theory, but there are other historical theories, including some versions of the organizational theory (McLaughlin Reference McLaughlin2001). Ahistorical theories include Boorse’s goal-contribution account (Reference Boorse1976, Reference Boorse1977, Reference Boorse, Ariew, Cummins and Perlman2002), the propensity theory (Bigelow and Pargetter Reference Bigelow and Pargetter1987), and the causal role theory (Cummins Reference Cummins1975; Craver Reference Craver2001; Hardcastle Reference Hardcastle, Ariew, Cummins and Perlman2002). In the 1970s and 1980s, it was common to see these two sorts of theories as competing with each other, although more recently, philosophers of biology have generally adopted a pluralistic stance and see them as capturing different aspects of ordinary biological usage (Garson Reference Garson2018). Still, the validity of the basic distinction has never been seriously challenged.

Many proponents of ahistorical theories have argued that we should accept their theories precisely on account of their being ahistorical. In other words, their alleged ahistoricity is often touted as a significant selling point of their theories and as a strong reason to prefer them over historical ones. There are two arguments along these lines. The first argument appeals to bald intuition and says it is just obvious that functions do not always need history. One fanciful variant of this argument appeals to science fiction cases, like swamp creatures, instant lions, and randomly generated worlds (e.g., Boorse Reference Boorse1976, 74; Bigelow and Pargetter Reference Bigelow and Pargetter1987, 188). But one does not have to go as far as science fiction to find plausible cases of ahistorical functions in biology. Many philosophers have a strong intuition that, the very first time a new biological trait emerges and begins to benefit the organism, it has a function even if it never underwent selection (e.g., Bigelow and Pargetter Reference Bigelow and Pargetter1987, 195; Walsh and Ariew Reference Walsh and Ariew1996, 498; Boorse Reference Boorse, Ariew, Cummins and Perlman2002, 66). The second argument, which is closely related, appeals to ordinary biological usage instead of intuition. It says that historical theories run against the way biologists ordinarily think and talk about functions. At least sometimes, when biologists attribute functions to traits, they neither cite nor refer to nor think about history or evolution (e.g., Godfrey-Smith Reference Godfrey-Smith1993, 200; Amundson and Lauder Reference Amundson and Lauder1994, 451; Walsh Reference Walsh1996, 558; Boorse Reference Boorse, Ariew, Cummins and Perlman2002, 73). Hence, ahistorical theories capture important strands of real biology.

In light of the above, my thesis might come as a bit of a shock. I claim that there are no ahistorical theories of function—or, put more precisely, the mainstream versions of the allegedly ahistorical theories on the market are not actually ahistorical. If we poke and prod at those theories a bit, a historical element falls out, like contraband stashed away in a suitcase. In Boorse’s version of the goal-contribution account, history is explicitly embedded in his notion of a statistically typical contribution to fitness. In the propensity account, history is embedded, a little less explicitly, in the idea of a species’ natural habitat. Finally, the only way the causal-role theorist can hope to make sense of dysfunction is to appeal to history.

Before I move on, there is one significant qualification I must get out of the way. One could invent a purely ahistorical theory of function. One could assert, for example, that all of a trait’s effects are its functions. In fact, the biologists Bock and von Wahlert (Reference Bock and Wahlert1965, 274) proposed a theory of function very much along these lines. This theory (panfunctionalism?) would be ahistorical, to be sure, since even if the world were created two seconds ago in pretty much its present form, things would still have effects, and so they would still have functions. In fact, sometimes scientists actually do use the word “function” synonymously with “effect.” They say things like, “climate change is a function of deforestation” or “poor academic performance is a function of malnutrition.” But this is not the ordinary biological use, which the theories I cite above are trying to capture. I come back to this point in the conclusion (sec. 5).

So, I need to amend my thesis slightly. Instead of saying that there are no ahistorical theories of function, I want to say that any theory of function that satisfies two very minimal, very traditional, and largely uncontroversial adequacy conditions is also a historical theory. First, the theory should capture some distinction between functions and accidents (a function of the nose is to help us breathe but not to hold up glasses). Second, the theory should capture the possibility of malfunctioning or dysfunction. If my heart seizes up due to cardiac arrest, it is failing to perform its function, or it is dysfunctional. All of the theorists I engage with in this article purport to satisfy these two adequacy criteria, or something in their vicinity, so I am not begging any questions by insisting on these conditions.

Here is the plan for the rest of the article. The next three sections will examine Boorse’s goal-contribution theory (sec. 2), the propensity theory (sec. 3), and the causal-role theory (sec. 4), in turn. In the conclusion (sec. 5), I draw out some consequences for thinking about functions.

2. Boorse’s Goal-Contribution Account

Boorse’s view (Reference Boorse1976, Reference Boorse1977, Reference Boorse, Ariew, Cummins and Perlman2002), at the most general level, is a goal-contribution account. It holds that a trait’s function is just its contribution to a goal. Here, I focus on the subclass he calls physiological functions. For Boorse, the physiological function of a trait is its species-typical contribution to the survival and reproductive prospects of an organism (Reference Boorse1977, 555; Reference Boorse, Ariew, Cummins and Perlman2002, 72; to be more precise, Boorse carves up species into subgroups based on age and sex; the function of a trait is its typical contribution to fitness among the members of that subgroup). Although he does not define a corresponding notion of dysfunction, he defines a closely related notion of disease: a disease is simply a state that “reduces one or more functional abilities below typical efficiency” (Boorse Reference Boorse1977, 555).

Neander raised a now-famous objection against Boorse. She pointed out that Boorse’s view, as it stands, cannot make sense of pandemic dysfunction: “dysfunction can become widespread within a population. … A statistical definition of biological norms implies that when a trait standardly fails to perform its function, its function ceases to be its function; so that if enough of us are stricken with disease (roughly, are dysfunctional) we cease to be diseased, which is nonsense” (Neander Reference Neander1991, 182). Pandemic dysfunctions, moreover, do not just occupy the realm of science fiction, as in P. D. James’s The Children of Men. Ultraviolet radiation poisoning in anurans is a good example of pandemic dysfunction. Sadly, climate change might create many more pandemic dysfunctions very soon. A good theory of function should not close off the possibility that all, or most, tokens of a certain trait in a certain species are dysfunctional (or, as Boorse prefers, “diseased”).

Intriguingly, Boorse does not deny the possibility of pandemic disease. Instead, he says that in order to make sense of pandemic disease, one has to appreciate function’s historical depth. Specifically, when we consider what is “statistically typical” for a trait, we cannot just look at what is typical right now. We have to examine the trait’s behavior over a slice of time that includes the present moment and reaches far back into the past: “Obviously, some of the species’ history must be included in what is species typical” (Boorse Reference Boorse, Ariew, Cummins and Perlman2002, 99; my emphasis). He tells us that this time slice should be longer than “a lifetime or two” and might include “millennia” (99).

This is an extraordinary admission, given that much of Boorse’s core argument for his view was propped up on the claim that both biology and intuition need purely ahistorical functions, uncluttered by history. His admission implies that his two key arguments for the view do not work. First, by his own lights, it is not the case that biologists do not refer to history; implicitly, when they talk about what is statistically typical, they are talking about history. Second, regardless of whether intuition supports ahistorical functions, Boorse’s theory does not. It is just not true, on Boorse’s account, that if lions popped into being from an unparalleled saltation, their distinctive parts and processes would have functions. They would not, since they do not have the right history (or to be more precise, they have no history at all).

3. The Propensity Theory

Bigelow and Pargetter (Reference Bigelow and Pargetter1987) also developed an influential “ahistorical” theory of function, the propensity theory. They reject the selected effects theory (and etiological accounts more generally) because the selected effects theory gets the modality of functions wrong. In other words, the statement “functions are selected effects,” if true, is contingently true; it might be true on the actual world, but there are possible worlds at which it is false. To illustrate the point, they ask us to consider a world that is pretty much the same as ours except that it randomly popped into being 5 minutes ago. On that world, they claim, there would still be functions, just no selected effects: “we have the intuition that the concept of biological function … [is] not thus contingent upon the acceptance of the theory of evolution by natural selection” (188). This consideration prompts the need for an ahistorical theory.

For Bigelow and Pargetter, functions are propensities, or probabilistic dispositions. We might quibble over what exactly dispositions are, but any good definition will cite three parts: structure, environment, and behavior. Consider the solubility of salt. There is a structure, namely, the polar molecular structure composed of sodium and chloride; there is an environment, namely, water; there is a behavior, namely, dissolving. When we say that salt is disposed to dissolve in water, we are saying that, if you were to take something with this structure and put it in this environment, it would behave this way, all things equal.

Functions, too, are dispositions. Consider “the function of the heart is to circulate blood.” For this statement to be true, there must be a structure (the heart, embedded the right way in the circulatory system), an environment (which they call the creature’s “natural habitat”), and a behavior (conferring a fitness boost on the organism). If one were to put the structure in its natural habitat, it would increase the fitness of the organism (relative, I suppose, to creatures without hearts). The crucial distinction between their view and Boorse’s is that in their view, a trait’s function does not depend on actual frequencies of performance. A trait need not have an actual track record of boosting fitness to have a function; a mere propensity will do.

This raises the thorny question of what a creature’s natural habitat is. For they are clear that a creature’s natural habitat is not just any environment the creature happens to find itself in. Curiously, they refuse to define this crucial notion; instead, they brush it off as vague but unproblematically so: “there may be room for disagreement about what counts as a creature’s ‘natural habitat’; but this sort of variable parameter is a common feature of many useful scientific concepts” (Bigelow and Pargetter Reference Bigelow and Pargetter1987, 192). But one could at least form the suspicion that if one analyzed this unproblematically vague notion, one would find some reference to history tucked away inside of it.

This suspicion is confirmed in the very next paragraph of their paper. There, they tell us that, if a creature’s environment were to change very suddenly, then “natural habitat” will still refer to the old environment and not the new one (Bigelow and Pargetter Reference Bigelow and Pargetter1987, 192). There is a time lag built into the very idea of a natural habitat. So, for example, if climate change melts enough Arctic ice, then, at least for a time, the polar bear’s natural habitat (and by extension, the natural habitat of the trait itself, namely, their thick, water-repellent fur) is the icy habitat of yore and not the contemporary, denuded one. They take that as given, and I agree.

But why would this be? What makes it the case that, in cases of rapid habitat change, “natural habitat,” at least for a time, refers to the old environment and not the new one? What makes it true, I suspect, is that the idea of a natural habitat is an intrinsically historical notion. It is something like “the environment within which the species recently survived and thrived.” And if that is not what a natural habitat is, I would like to know what it is such that, if a creature’s actual habitat shifts suddenly, the natural habitat, for a little while, is still the old one. Just because a concept is vague around the edges, one is not exempt from the obligation to give some sort of analysis. Perhaps one could revise the theory and drop all reference to “natural habitat,” as suggested by Griffiths (Reference Griffiths2009, 27), but that remains to be worked out in a rigorous way. Moreover, it is not clear whether such a theory, when rigorously developed, would hang together with the two adequacy criteria.

Hence, I conclude that, contrary to rumor, the propensity theory is not an ahistorical theory, or not demonstrably so. But if that is right, supporters of the propensity theory lose one of the main virtues of the view, which is to get the modality of functions right. To be fair, there is still a sense in which their view is ahistorical. What they can do, which the selected effects theorist cannot, is to attribute functions to novel traits, so long as that novel trait belongs to the members of a species that has been around long enough to have a natural habitat. Suppose a gene mutation confers a benefit on an organism, say, pesticide resistance in a flour beetle. I suppose they can say that, at the very moment at which it first confers that benefit, the gene mutation has a function, namely, to make the beetle withstand a certain pesticide. This result, they claim, is “intuitively comfortable” (Bigelow and Pargetter Reference Bigelow and Pargetter1987, 195). But they can say that only because flour beetles themselves have a history, and so we can talk meaningfully about their natural habitats. Moreover, I think they will still have a rough time explaining dysfunction (Neander Reference Neander1991, 183), for reasons I point to in the next section. Finally, I think there are good theory-neutral reasons for saying that beneficial traits, on their very first appearance, do not have functions, but rather, whatever benefit they bring is a lucky accident. But I will not argue for that here (see Garson Reference Garson2019, chap. 2).

4. The Causal Role Theory

What about the causal role theory of function? This appears to be a purely ahistorical view. The causal role theory says, roughly, that the function of a component of a system consists in its contribution, in tandem with the other components, to a system-level capacity of interest (Cummins Reference Cummins1975; Craver Reference Craver2001; Hardcastle Reference Hardcastle, Ariew, Cummins and Perlman2002). Craver (Reference Craver2001) helpfully elaborates this view by specifying that the part in question must be a component of a mechanism. All of the basic ingredients of this theory, it seems, are ahistorical: capacities, components, organization, hierarchy, interests. Even if the world were created 5 minutes ago, in pretty much its present form, things would still have causal role functions.

The problem enters when we think about dysfunction. Cummins (Reference Cummins1975, 758) insisted that functions are dispositions, or capacities: “to attribute a function to something is, in part, to attribute a disposition to it.” The function of a trait token, then, consists in its capacity to contribute to a system-level effect. But what if the token in question, through defect or disease, loses the capacity and so cannot contribute to the system-level effect? Then, by Cummins’s analysis, it does not have the relevant function, so it cannot be dysfunctional either.

Causal role theorists have, by and large, been silent about how to make sense of dysfunction from this perspective. Almost everything they have had to say on that score, however, is consistent with the following theme: a trait token is dysfunctional when it cannot do what other trait tokens generally, or typically, do to contribute to the system-level effect of interest. Consider Godfrey-Smith (Reference Godfrey-Smith1993, 200): “Although it is not always appreciated, the distinction between function and malfunction can be made within Cummins’ framework. … If a token of a component of a system is not able to do whatever it is that other tokens do, that plays a distinguished role in the explanation of the capacities of the broader system, then that token component is malfunctional.” Craver (Reference Craver2001, 72) offers the same general line: “the ascription of a function to a malformed or broken part is derivative upon a description of how that type of part (X) fits into a type of higher-level mechanism (S). The malformed and broken part can be identified as an X by the typical properties and activities of Xs.” This is, at root, to rely on a statistical norm for making sense of dysfunction.

This account of dysfunction, like Boorse’s, stumbles when it encounters the problem of pandemic dysfunction. For the modification suggested above implies that, if everyone’s heart seized up at once, nobody’s heart would have a function anymore, so nobody’s heart would be dysfunctional. The best way to solve this problem, and perhaps the only way, is the way Boorse took, namely, to say that the function of a trait is its typical contribution to some system effect, where what is typical is assessed over a chunk of time that stretches back into the past, for at least “a lifetime or two” and perhaps “millennia.” But if causal role theorists take that line, they would have a historical theory.

Craver (Reference Craver2001) and Hardcastle (Reference Hardcastle, Ariew, Cummins and Perlman2002) suggest, all too fleetingly, a different way of thinking about dysfunction, one that depends not on statistics but on our values, that is, the values and goals of people who make function attributions. Craver (Reference Craver2001, 72) suggests that traits are dysfunctional when they cannot do what people want them to do: “the mechanistic role of the broken part only appears against the fixed backdrop of shared assumptions about a type of mechanism within which parts of this type generally (or preferably) make important contributions.” The parenthetical remark alludes to a substantially new doctrine, one that demands our full concentration. It suggests that dysfunction is a mirror of human preferences and goals, of our wishing and wanting. If my heart seizes up, it is dysfunctional, since it is not doing what I want it to do.

Hardcastle (Reference Hardcastle, Ariew, Cummins and Perlman2002) makes remarks along similar lines. She first says that the function of a trait—what it is “supposed to do,” as she puts it—depends on the goals of the scientific discipline that makes the investigation: “The teleological goal for some trait … depends upon the discipline generating the inquiry” (153). The palmomental reflex causes a chin twitch when you stroke an infant’s palm; it is just an accident of cortical wiring with no deep evolutionary rationale. Still, she says, it has the function of indicating the state of brain development in infants because that is how biomedical researchers use it. She then says that something is malfunctional just when it cannot do what it is supposed to do (152). The palmomental reflex is malfunctional when it cannot indicate the state of brain development. Simply put, dysfunction happens when a trait cannot do what we want.

But dysfunctions cannot be reduced to mere preferences in any straightforward way; this is a point that is been taken in the literature for decades (e.g., Boorse Reference Boorse1977, 544; Wakefield Reference Wakefield1992, 375), for reasons that scarcely need to be rehearsed. I would prefer not to need sleep and water; I would prefer if nobody had to go through the pain of childbirth or teething, either. But none of those things are dysfunctions. For that matter, I would prefer if my hands were equipped with retractable adamantium claws. The fact that my hands cannot do what I want them to do does not make them dysfunctional. If one really wanted to run with this value-centered line about dysfunction, one would at least have to add that, in order for a trait to be dysfunctional, it is not enough that it does not do what I prefer, but I must also have a reasonable expectation that it should act in the way that I prefer. But what could possibly ground a reasonable expectation that my hand (say) work in a certain way? Only this: that hands usually do work in the preferred way. But then we are back to statistical norms and long historical slices of time. This value analysis of dysfunction is not a contender to a statistical analysis; instead, the former presupposes the latter.

I have walked through three allegedly ahistorical theories of function and shown that none of them are purely ahistorical; they are infected with history. The conclusion will say what we should do next.

5. Conclusion

There are no ahistorical theories of function, at least among the mainstream theories that are put forward as ahistorical. The first, Boorse’s goal-contribution theory, explicitly refers to what is statistically typical for a trait, where what is typical is assessed over a long historical period of time. The second, the propensity theory, refers to the creature’s natural habitat, which is implicitly historical. And the third, the causal role theory, cannot hope to make sense of dysfunction (or so I argue) without appealing to a statistical norm and thereby (following Boorse) to history. None of these theories will give functions to the parts of swamp creatures, instant lions, or anything on worlds that are similar to ours except for being randomly generated 5 minutes ago. The propensity theory, at least, can give functions to novel traits as soon as those traits begin benefiting their bearers, as long as the population in which the traits emerge has been around for long enough to have something like a natural habitat. But even that theory will probably encounter problems when it comes to making sense of dysfunction, although I have not pushed that line in any detail here.

If my thesis is correct—that there are no ahistorical theories of function—three consequences immediately follow. First, we need to jettison this whole way of dividing up theories of function. The distinction between etiological and nonetiological theories serves us much better. An etiological theory holds that function ascriptions either are or purport to be causal explanations for the existence of traits. Nonetiological theories hold that function ascriptions are not, and they do not purport to be, causal explanations for traits. But the crucial point is that being etiological and being nonetiological are just two different ways of being historical.

Second, given that there are no ahistorical views, the two main arguments that have repeatedly been put forward for those theories—the argument from intuition and the argument from ordinary biological usage—do not actually work. If we took those arguments seriously, they would count as evidence against these allegedly ahistorical theories. That does not mean those theories are wrong. It does mean, however, that we would need to rethink, from the ground up, the motivation for accepting them.

A third consequence is that one popular way of thinking about function pluralism must fail. This sort of pluralist wishes to sort all biological usage under two main umbrella theories, the selected effects theory and the causal role theory. An argument for this sort of pluralism is that it mirrors the two main uses of “function” in biology, the historical sense and the ahistorical sense. If I am right, this incarnation of the pluralist project cannot work either.

True, there are some theories of function I have not addressed here, which fall a bit outside of the mainstream. Might those come to our rescue? In particular, one might wonder how the modal theory of function (Nanay Reference Nanay2010) fares with respect to my analysis. The modal theory holds that a function of a trait token depends on that token’s behavior on nearby possible worlds, where what is “nearby” depends on our explanatory interests. I agree that this is an ahistorical theory through and through, since what function a trait has and whether or not it is dysfunctional depend on what is going on at other possible worlds, rather than the actual past. But it also yields a deeply implausible construal of dysfunction. As Neander and Rosenberg (Reference Neander and Rosenberg2012) point out, if the modal theory is right, then many traits that biologists do not think of as dysfunctional, like the trait of lactose intolerance in most Pacific Islanders, would actually be dysfunctional. So, while the modal theory does not violate the letter of my second adequacy condition—namely, that it should allow for the possibility of dysfunction—it violates the spirit of that condition by carving up functions and dysfunctions in a wildly revisionary way.

Nanay (Reference Nanay2012) argues that the fact that function ascriptions are relative to our explanatory interests can somehow lessen the sting of this counterintuitive consequence, but I do not see how that helps. To illustrate the problem, consider Temitope, an evolutionary geneticist who is interested in how human beings might evolve in the near future. Temitope considers a possible world to be “nearby” if, at that possible world, she has a counterpart, and her counterpart’s genome differs from hers by only a single point mutation, but the rest of the world is largely the same (yielding at least 3 billion nearby worlds). She reasons that, on some of those possible worlds, some of her traits would do things that enhance her inclusive fitness. For example, we might suppose that there is a possible world at which her body’s ability to dissolve arterial plaque is substantially enhanced, one at which she has tetrachromatic vision, and one at which she is resistant to malaria. She realizes, with dismay, that her body’s actual ability to dissolve arterial plaque represents a dysfunction. In fact, she realizes that, relative to her explanatory interests, she has many more dysfunctions than she ever imagined. So even if we agree that function ascriptions are tethered to explanatory interests, we still get deeply revisionary consequences. In my reckoning, a theory that hangs together pretty well with ordinary biological usage is better than a deeply revisionary one, all things equal (see Garson [Reference Garson2016], 105–7, for further discussion).

There is a twist to my story, which I alluded to in the introduction. I think there is a prominent sense of “function” in scientific circles that is ahistorical. Consider that climate change is a function of deforestation, poor academic performance is a function of malnutrition, and wildlife habitat is a function of soil. These notions are ahistorical through and through. “Function,” in this context, means little more than “effect” and, perhaps (as in the last of the three examples), “helpful effect.” But this tepid sense of function is not going to sustain a distinction between function and accident, and it will not give us any sense of dysfunction. This is the sort of “function” that Bock and von Wahlert (Reference Bock and Wahlert1965, 274) were getting at when they equated functions with “all physical and chemical properties arising from [the trait’s] form.” It is also the sort of “function” that Neander (Reference Neander2017) describes in her recent discussion of “minimal functions.” But the proponents of the allegedly ahistorical theories want functions to do much more than that. They are trying to capture the ordinary biological sense (or an ordinary biological sense) of “function,” where functions differ from accidents and sometimes things are dysfunctional. Unfortunately, they cannot have what they want.