Cummins  argues that etiological considerations are not only insufficient but irrelevant for the determination of function. I argue that his claim of irrelevance rests on a misrepresentation of the use of functions in evolutionary explanations. I go on to suggest how accepting an etiological constraint on functional analysis might help resolve some problems involving the use of functional explanations.
Robert Cummins' "Functional Analysis" [Cummins 1982] gives a straightforward and generally convincing criticism of attempts to "reduce" functional explanation to deductive-nomological explanation and thus eliminate its special character. His central point is one I completely agree with: The primary goal of functional analysis is not the explanation of the presence of the structure analyzed, but is instead the explanation of current complex dispositions and capacities of a system in terms of the simpler capacities and dispositions of its parts.
I have no wish to defend the "traditional" view of functional explanation (of Hempel and Nagel) on which functional analyses are seen as "incomplete" deductive-nomological explanations of the presence of the analyzed item. But along the way, Cummins casts his net too widely by rejecting the relevance of etiology to function in general, and thus arguing for a very strong thesis: That the function of a structure not only need not have played a role in the structure's etiology, but that a structure's function cannot play a role in its etiology. That is, he argues not only against an etiological constraint on functional analysis, but also against the idea that the functional properties of structures might be etiologically relevant.
The acceptance of an etiological constraint on functional analysis is in my view the important germ of truth behind the traditional view, and can play an important role in clarifying some of the general puzzles faced in providing and evaluating functional explanations. I'm far from alone in this inclination, of course. Wimsatt  is perhaps the grandaddy of such arguments; but more recent support for the relevance of etiology to function---particularly in the psychological and social domains, and not just the biological ones---is offered by Dretske [1981 & 1991], Elster , and Sober .
With such a cast of supporters for this view, one might be inclined to let the debate rest with them. But there is still reason for taking Cummins' suggestions seriously and providing a rebuttal to them, since his criticisms focus on the possibility of appealing to function in etiology at all. In suggesting that function must be etiologically irrelevant, he attacks a precondition of these accounts---which themselve focus largely on how to make use of the links between function and etiology. Cummins offers the possibility of undermining a presupposition of any such account, and thus of rendering such accounts irrelevant.
So in the next section, I'll show how (contra Cummins) the functions of structures can play a role in the causal explanation of their presence and character---even for evolved biological systems, where Cummins thinks they cannot. Then in sections 3 and 4, I'll discuss how an etiological constraint on functional analysis helps provide a "selected effects" constraint on functions, and see how such a constrain can avoid his putative counterexamples. I'll then close with some brief mention of further advantages and disadvantages of accepting an etiological constraint on functional analysis.
Cummins' central point against the etiological relevance of function is straightforward. As he puts it: "to `explain' the presence of the heart in vertebrates by appeal to what the heart does is to `explain' its presence by appeal to factors which are causally irrelevant to its presence... ." [Cummins, p. 390] After all, the structure doesn't perform that function at all unless it's already there; and "to explain the presence of a naturally occurring structure or physical process---to explain why it is there, why such a thing exists in the place (system, context) it does---this does require specifying factors which causally determine the appearance of that structure or processes." [Cummins, p. 390] Since explaining the presence of a structure requires specifying its antecedent causes, [One might try to avoid Cummins' conclusions here by denying that explanation of the presence and character of structures must be causal explanation, as some advocates of "functional analysis" in history and anthropology do. See, e.g., [Cohen 1978], esp. chapter 9.] I'm not inclined to do so. If you are, then you write the paper.] and the function can't be antecedent to the structure, the function of a structure is "causally irrelevant to its presence"---i.e., it is etiologically irrelevant. Artifacts are of course an exception to this; as he says, "the question `Why is x there?' can be answered by specifying x's function only if x is or is part of an artifact." [Cummins, p. 391] When we specify the function of an artifact, we're typically specifying the intention with which it was made. If the artifact is made for performing some function, that function is then (through the intentions of the artifact's creator) relevant to a causal explanation of its presence and character. But without the mechanism of explicit design intentions at work, function does not explain presence.
Cummins himself mentions the obvious candidate to take the place of design intentions: natural selection might be what "provides the missing causal link between what something does in a certain type of organism and its presence in that type of organism." That is, we might hold that "by performing their respective functions [those structures] help species incorporating them to survive, and thereby contribute to their own continued presence in organisms of those species, and this might seem to explain the presence of those structures in the organisms incorporating them." [Cummins, pp. 393--394] The function of a structure might then play a role in the causal explanation of the structure's presence by making a causal contribution to the continuing survival of the species. The heart's function as blood circulator contributes to the survival of species that have hearts, and thus is part of the explanation of the presence of hearts in current organisms. Functional analysis is not then underdescribed deductive-nomological explanation for the presence of the analyzed structure, but it may still be a central part of an (evolutionary) explanation of the presence of that structure.
But Cummins claims that this suggestion "involves a subtle yet fundamental misunderstanding of evolutionary theory." [Cummins, p. 394] Whether or not an organism incorporates a part of a given kind depends on whether that part is specified by the genetic "plan" that the organism inherits; and "alterations in the plan are not the effects of the presence or exercise of the structures the plan specifies." [Cummins, p. 394] Change due to random mutation is taken as the clearest example of this. As he puts it:
If a plan is altered so that it specifies s' rather than s, then the organisms inheriting this plan will incorporate s' regardless of the function or survival value of s' in those organisms. If the alteration is advantageous, the number or organisms inheriting that plan may increase, and, if it is disadvantageous, their number may decrease. But this typically has no effect on the plan, and therefore no effect on the occurrence of s' in the organisms in question. [Cummins, p. 394]
Cummins' fundamental mistake lies here. If the mutation underlying s' is advantageous, then a part of the explanation for structure s' now being present in organisms of that kind (i.e., for its having survived and propagated as a phenotypical property of the organism) is the selection advantage conferred by the performance of a function of s'. What makes a genetic mutation advantageous is its causing a change in the phenotype that allows a function to be performed (or performed better), and which then in turn enhances the organism's chances of survival and reproduction.
Certainly particular individual organisms inheriting a genotype specifying s' instead of s will incorporate s' regardless of its function or survival value. But which genotypes (and thus phenotypical features) manage to survive and be transmitted into future generations is in part a causal result of the function and survival value of s'. If, for example, s' is a more efficient heart that allows for greater endurance in the pursuit of prey and the avoidance of predators, then the function of s' will play a significant role in causing the presence of s' in future generations. The functions of parts may play no causal role in determining which alterations get tried out (i.e., in determining which mutations occur), but they do play a critical role in the determination of which genetic plans continue to occur. Thus, the function of a structure under the pressures of natural selection will play a crucial role in the explanation of the presence of those structures in current organisms.
Cummins in fact considers something like this suggestion in noting that "natural selection cannot alter a plan, but it can trim the set. Thus we may be able to explain why a given plan is not a failure by appeal to the functions of the structures it specifies... ." [Cummins, p. 394] So the "non-failure" (i.e., success) of a given type of plan may be explained by appeal to the function(s) of the structure it specifies; and so functions may after all explain why organisms with that kind of plan (and resulting structure) occur.
But then why doesn't the function explain the presence of the structure? Because, according to Cummins,
... this is not to explain why, e.g., contractile vacuoles occur in certain protozoans; it is to explain why the sort of protozoan incorporating contractile vacuoles occurs. Since we cannot appeal to the relative success or failure of these organisms to explain why their genetic plan specifies contractile vacuoles, we cannot appeal to the relative success or failure of these organisms to explain why they incorporate contractile vacuoles. [Cummins, pp. 394--395]
Obviously the function as performed by this particular instance of the structure in a particular organism played no causal role in the production of this particular instance of the structure. [Insofar as structure is being taken as simply a function of genotype---the functions of structures in individuals might also potentially affect structure in that very organism in cases of learning or training.] But the function of that type of structure has played a role in determining the presence of structures of that type---and consequently, of that particular instance. The function of a structure can help explain the structure presence by its past role---implemented by numerically distinct structures of the same type.
Natural selection explains the spread and continued presence of mutations which are initially random. The randomness of mutation alone guarantees that selection will not give a complete account of anything. Specifying the function of a structure obviously doesn't give anything like a complete specification of the causal factors leading to its current presence. But functions still might be a critical and ineliminable part of the causal explanation of the presence of particular features in current organisms.
We have yet to see any reason to accept that functions of structures must play a role in their etiology. But we have seen that functions clearly can play such a role---that is, that the appeal to function in explaining the presence and character of a structure needn't involve "appeal to factors which are causally irrelevant to its presence... ."
Let me now to considering the stronger suggestion that we should honor an Etiological Constraint on Functional Analysis: the requirement that any legitimately ascribed function have played some causal role in bringing about the current features of the structure implementing that function. Or to put it in a more schematic form: For F to be the function of S, S's performing F must be part of the causal explanation for S's ability to perform F.
Trying to distinguish between those effects of a structure which should count as its functions and those which should not is a standard problem in functional analysis. A central part of the idea of a function that not all effects of a structure count as its functions; some are simply accidental by-products. The function of the heart is to pump blood, not to act as a slow-burning fuse on lifespan, clogging the coronary arteries slowly but surely until the organism is killed by a heart attack. Heart attacks are breakdowns; they are failures of the heart to perform its function of pumping blood. How are we then to isolate the legitimate functions of a structure from such accidental side effects?
What's needed is a constraint on which of the activities and conditions of the containing system are such that a structure's contribution to those count as performing its function. We need to know which "selected effects" of a structure count as its function or functions, presumably in virtue of contributing to the particular class of activities or conditions of the containing system which count as that system's working right. In short, we need a way to separate the function(s) of a structure from its accidental effects.
Cummins explicitly considers two possible suggestions for such a "selected effects" constraint. [The two possibilities considered by Cummins of entities whose survival is furthered (i.e., organism and species) are only two of a larger class of possibilities. Past contribution to survival might be at these levels (although see next note), but also at other levels of organization---for example, at the level of a group (see Sober ).] The first is roughly Hempel's: To contribute to the right functioning of the system is just to contribute to the maintenance of life and health. This is plausible enough in some standard cases (e.g., the heart's function as circulator, chlorophyll's function in photosynthesis), but it's easy enough to give clear counterexamples: This possible constraint on right functioning fails to be necessary, since proper functioning of a part can be detrimental to continued life and health; e.g., proper functioning of the reproductive system in some species (such as certain salmon) results in the death of the individual organism. And the condition fails to be sufficient, since structures may contribute to life and health without that contribution being the function of the structure; e.g., adrenaline secretion aids in elimination of harmful fat deposits in overweight humans in spite of this not being the function of adrenaline secretion. [An etiological constraint easily handles these cases: The normal (but suicidal) functioning of the salmon's reproductive system has causally contributed to the past survival of the species; and aiding in the elimination of harmful fat deposits isn't the function of adrenaline secretion because it was (likely) no contributor to surviving the selection pressures faced in our evolution---since obesity was likely not a major killer in the evolutionary environment.]
The second possibility Cummins considers for a "selected effects" account is more promising. At least for evolved organisms, we might see the critical activities and conditions supported by legitimate functions as those that contribute to the survival of the species. ["Contribution to the survival of the lineages posessing the structure" would probably be better here, given the questionable status of species-level selection.] That is, we might try supposing that "the functions of a part or process in an organism are to be identified with those of its effects contribution to activities or conditions of the organism which sustain or increase the organism's capacity to contribute to survival of the species." [Cummins, p. 398]
Although Cummins allows that this "doubtless does capture a great many uses of functional language in biology", he still sees it as "seriously misleading" [Cummins, p. 398]. The problem is that some functions of organisms can cease to contribute to the survival the species (or at least those lineages that possessing the structure) without thereby ceasing to be the legitimate function of the appropriate structure; for example, "if, for some reason, flying ceased to contribute to the capacity of pigeons to maintain their species, or even undermined that capacity to some extent, we would still say that a function of the wings in pigeons is to enable them to fly." [Cummins, pp. 393--399] Flight does not cease to be the function of wings by ceasing to contribute to survival of the species; rather, "only if the wings ceased to function as wings, as in the penguins or ostriches, would we cease to analyze skeletal structure and the like functionally with an eye to explaining flight." [Cummins, p. 399]
This kind of example does seem to weigh heavily against straightforwardly indentifying function with contribution to species survival. But he claims more for it than that:
What this example shows is that functional analysis can properly be carried on in biology quite independently of evolutionary considerations: a complex capacity of an organism (or one of its parts or systems) may be explained by appeal to a functional analysis regardless of how it relates to the organism's capacity to maintain the species. [Cummins, p. 399]
The example hardly supports this more radical conclusion. All that's shown here is that present (or perhaps future) contribution to the survival of the species does not provide the right account of "selected effects". A past function of a structure can continue to be its function without continuing to contribute to the survival of the species; but this harly entails that "functional analysis can properly be carried on in biology quite independently of evolutionary considerations". This sort of example is entirely compatible with the view that past contribution to the survival of the species is critical to the determination of a structure's function. What makes flight the function of the pigeon's wings may be in part the fact that the wings' facilitation of flight has been a critical contribution to the pigeon's success in surviving so far.
Certainly the original presence of proto-wings will not be explained in terms of flight, but most likely in terms of their function as devices for temperature regulation; thus the initial spread of proto-wings would be explained through the advantages of better temperature regulation. But an etiological constraint can still naturally capture the fact that flight is the (or at least one [I take it temperature regulation still is a function of wings, after all.]) function of wings. Once such proto-wings began to be used as an aid in locomotion, selection pressures sensitive to the wings' efficiency in that role became important, and have no doubt played a great role in determining the further structural developments, musculature, size, and other critical properties which current wings display. The function of wings as locomotive devices (first perhaps on the ground, later in true flight) has then surely played a critical role in the survival and spread of the genetic plan specifying the sorts of structures we currently find. Once again, the function of the structure does not explain the fact that certain structures were initially tried out, but does play a critical role in explaining why we presently find structures of the sort that we do.
It is contribution to past success that's critical. The legitimate function of a structure plays a role in the explanation of the structure's presence through its past contribution to the survival of the organisms that implemented that type of structure. The structure's function in past survival allows us both to appeal to the function in explaining the presence of the structure (as we saw in the last section), and to reject the suggestion that examples like that of the pigeons show that "a complex capacity of an organism (or one of its parts or systems) may be explained by appeal to a functional analysis regardless of how it relates to the organism's capacity to maintain the species."
One further kind of putative counterexample offered by Cummins to the use of an etiological constraint is worth noting because it applies particularly to cases of social rather than biological function. Here, Cummins claims that objects may legitimately be ascribed a particular function because of a social role they play even when we know perfectly well that their presence is in no way causally dependent on their playing that role; e.g., "the function of a bowl-like depression in a huge stone may be to hold holy water, but we cannot explain why it is there by appeal to its function if we know it was left there by prehistoric glacial activity." [Cummins, p. 391] Since the function of holding holy water is in no way a part of the explanation of the depression in the stone, it seems that we may have an example of a legitimate function which played no etiological role.
But the presence and use of a rock with such a depression in those religious rituals of the society is no doubt explained in part by the depression's ability to perform the function of holding holy water. The depression is there, in that position in the social structure, because of its ability to perform a given function. [So like the other cases of artifacts, the causal link from function to structure is via intentions.] To take another example: The function of the idle speed adjustment screw on my car's carburetor is (of course) to allow adjustment of the engine's idle speed. I rightly explain the presence of such a structure by appeal to that function, as foreseen and intended by the car's designers and manufacturer. The fact that the performance of that function had nothing to do with determining the number of threads per inch on that particular screw (that having been determined by the company they bought their screws from) does not in any way impugn the explanation of the presence of the screw in terms of its function. The function explains the presence of the screw in that position, but not, of course, why that screw was made with that particular number of threads, or why that particular screw was manufactured at all. Likewise for the depression in the rock: Its presence in that position in the social structure is explained by its ability to perform its function, even though that function was irrelevant in causing that particular rock to have the shape that it does.
As in the biological cases earlier, the current presence of type of structure is in part explained by the function performed by that type of structure (and consequently that instance of it). In each case, some more distal facts about the structure's etiology are not in any way explained by its function---facts about the original occurrence of a mutation (as opposed to what caused its persistence) in the biological case; facts about the original production of the object (as opposed to its current position and role) in the social one. But this failure to account for everything does not impugn the significance of function in the explanation of the structure's presence.
My central goal here was to flesh out the logic of a certain kind of fallacy that Cummins offers; and that I think I've done. His arguments all depend on an illegitimate move from the observation that the causal (including functional) effects of a particular instance of a structure can't include the presence of that very structure (a trivial observation about the rejection of backwards causation), to the general (and incorrect) claim that a structure's having a certain function can't be a part of the causal account of the structure's presence. [I think this is motivated in the end by something like the intuition that it can't matter for explanation if it doesn't locally supervene---a general thesis I argue against in McClamrock .] But the routes by which function can influence etiology are straightforward and non-occult. As we've seen, all that's required is a mechanism for (the past instantiations of) the function to produce the current structure. Design intentions will do so in the case of artifacts; and---as we saw in section 2, and contrary to what Cummins says---the mechanism of natural selection can perfectly well play this role in the case of biological systems.
Using etiology to guide functional analysis is now (and should be) widespread, even outside the biological domain. In the philosophy of psychology, its use ranges from David Marr's emphasis (in his landmark book Vision [Marr 1982]) on focusing on the teleology of the visual system---the task it was selected for---in analyzing its internal information processing, to Elliot Sober's [Sober 1985] and Fred Dretske's [Dretske 1991] uses of a "teleological" functionalism to avoid many of the standard puzzles about content that the more standard forms of functionalism in that domain face. [Etiological grounding of function might be through non-evolutionary mechanisms too---in fact, as many as your cognitive ethology presents. Learning is the most obvious candidate here. See, e.g., Dretske 1981, especially part 3.] And in the social sciences, etiology can also constrain the use of "functional explanation" in Marxist political theory and some social anthropology---domains where "functional explanation" for some social structure often simply involves pointing out how it provides some social benefits, even where those benefits play no causal role in bringing about the current state of affairs. [See Jon Elster's [Elster 1983] discussion of using functions to explain social structures only when there is a kind of "feedback loop" providing a causal connection from the function to the presence of the structure.]
Accepting some type of domain-general etiological constraint on functional analysis will have strong implications in many specific domains where functional analysis is a central explanatory device. And nothing we've looked at has shown either that function is "etiologically irrelevant" or that an etiological constraint on functional analysis succumbs to any obvious counterexamples. Such a constraint may in fact hold some significant promise for dealing with problems involving the overly liberal use of functional analysis the special sciences.
As Cummins himself suggests, it is the appeal to "etiologically
irrelevant" factors in explaining the presence of structures which "has
given `functional analysis' a bad name." [Cummins, p. 392] But the route
to avoiding this practice is not, as he would have it, to reject the
idea that a structure's presence is in part explained by its function,
and to thus liberalize the application of functional analysis while
removing some of its explanatory consequences. Rather, we should
constrain the application of functional analysis, and thereby try
to avoid the problems of an overly liberal account while at the same
time preserving the role of functions in explaining the presence of
biological, social, and psychological structures.
[Thanks to Bill Wimsatt, Greg Mikkelson, Dan McShea, Robert Cummins, and the editor and referees of this journal for their comments on earlier drafts.]