Malfunction defended

Historical accounts¹ of biological function play an important role in naturalistic theories in the philosophy of mind and language. For example, notions of function are drafted into teleosemantic accounts of mental content (Millikan 1989a; Neander 1995; Papineau 1987). The teleosemantics program provides a naturalistic account of mental representation by appeal to biological function; truth conditions of intentional states are given in terms of what those states are supposed to do (Macdonald and Papineau 2006, p. 1). In putting forward an account of mental content which appeals to biological function, one had better have an independent account of that notion. Equally one had better allow for malfunction in their account of function, to ground misrepresentation. Historical accounts of function have been thought able to do this important work, and are the choice accounts for philosophers seeking naturalistic theories in the philosophy of mind and language.

However, in several places, Davies launches a systematic attack on the historical account of function, calling it a ‘failed theory’, which ‘does not have the virtues that advocates claim for it’ (Davies 2001, p. 32), and has ‘rather little [...] worth preserving’ (Davies 2001, p. 39). More specifically, he objects to approaches which seek to account for malfunction, and suggests that as naturalists, we ought to do away with the notion of malfunction altogether. Here I address the argument against the possibility of historical malfunction, as presented in Davies’s paper ‘Malfunctions’ (2000) and his book Norms of Nature (2001). Davies’s argument teaches us something important about how historical accounts need to be formulated if they are to avoid the ‘internal failing’ of being unable to accommodate malfunction (2001, p. 192). He makes a plea for ‘an argument for the oft-repeated assertion that selected or etiological malfunctions are possible’ (2000, p. 21). I give such an argument (§5). I am not pursuing the more ambitious project of giving an account of malfunction which Davies would accept, since he does not think there is any such phenomenon to give an account of. Rather, I seek to show that an historical approach need not suffer from an internal failing to account for malfunction. I am not addressing Davies’s wider aim which is to take norms out of nature altogether, thus eschewing the notion of malfunction simpliciter. I am addressing the objection that the failure to account for malfunction is an internal failing of historical accounts.

As well as offering a revised historical account to meet Davies’s objection, I make two additional and related points. First, I argue that the scope of Davies’s objection is limited. Davies is wrong to claim that historical accounts cannot accommodate malfunction, it is only malfunction due to congenital factors which is problematic (§4). Second, I identify an inherent and under explored commitment in the work of (at least some) historical theorists (§5.1). It is by bringing this commitment to the fore, and taking a lesson from Davies’s objection, that I offer a historical theory of function which possesses the internal resources required for giving an account of malfunction.

Before getting into the topic of the paper, it is worth reflecting briefly on the appropriate methodology for a philosophical analysis of biological function. Two projects should be distinguished, which are distinct, and may be pursued independently. Each brings different adequacy conditions on accounts of function—an account of function may in principle be adequate for one project whilst being inadequate for the other. This does not call either project or account into question, but rather shows that different accounts of function are needed to subserve different theoretical goals.

The first project is driven by philosophy of biology, and the attempt to understand actual cases of biological research. Such an account will be adequate if and only if it is genuinely illuminative of actual cases in biology. The second project is driven by broadly naturalistic considerations within philosophy of mind and language. An account of function designed to serve this project is a theoretical one, adequate if and only if it fulfils the needs of philosophical accounts in these areas. Since, for example, our best naturalistic accounts of mental representation appeal to biological function, and since these accounts appeal to historical functions, Davies’s attack on historical accounts constitutes a real threat to some naturalistic projects in philosophy.

In an incisive paper on the debate over the correct characterisation of biological function, Arno Wouters notes that the recent revival of the philosophical debate on function can be put down to an interest in developing naturalistic theories like these, and not an interest in biology or biological practice (Wouters 2005, p. 123). Philosophers engaged in such projects differ on how they view their accounts of function with respect to whether they are—or should be—in line with actual biological practice. Karen Neander takes her account to be in step with actual biological practice, defending it as conceptual analysis (Neander 1991a). Ruth Millikan on the other hand is clear that she is not engaging in conceptual analysis (1984, p. 18; 1989b, p. 290). Reflecting on earlier work in which she first laid out her account of function, she notes that the point of her notion of proper function ‘was/is mainly to gather together certain phenomena under a heading or category that can be used in the construction of various explanatory theories’ (Millikan 1989b, p. 289). She intends her notion of function to be a theoretical definition, one which is able to ‘describe a unitary phenomenon that lies behind all the various sorts of cases in which we ascribe purposes or functions to things’ (Millikan 1989b, p. 293). The idea is to offer a theoretical notion of biological function, and then we can decide if it earns its keep by seeing how much theoretical work it can do.

Participants in the function debate then take themselves to be up to rather different things. If one is interested in how best to understand function attribution in biology, I agree with Wouters that one ought to focus one’s attention on the ‘actual practice of biological inquiry’ (2005, p. 123). However, if one is interested rather in a program like Millikan’s, then one has no such obligation, though one is advised that getting clear on the correct notion of function in this sense will provide ‘little insight in real biology’ (Wouters 2005, p. 123). I am engaging in the second kind of program, and so those philosophers of biology who think that the philosophical notion of function should be in line with actual biological practice may not find much of interest in this paper. However, for those philosophers who are interested in biological function as an explanatory notion which can play a central role in naturalistic theories in the philosophy of mind and language, the work here can be taken as developing a notion of biological function in this sense. Although I think Davies is right that the best extant historical accounts of biological function cannot accommodate all cases of malfunction, I provide a historical account which can.

Much of the biological function debate has been concerned with accounting for the supposed normativity of functionally characterized items or function statements. An account of function’s ‘main task [...] is to explain how this norm arises in biological contexts’ (Wouters 2005, p. 125), this seems indeed to be the ‘fault line running through [the function] debate’ (McLaughlin 2001, p. 4). Often what the function theorist’s burden comes down to is the requirement to accommodate malfunction. Malfunctions are taken to be ‘ubiquitous’ (Davies 2000, p. 20). Millikan claimed of function categories that it is an ‘obvious fact’ that ‘their members can always be defective—diseased, malformed, injured, broken, dysfunction, etc.,—hence unable to perform the very functions by which they get their names’ (Millikan 1989b, p. 295).² Carolyn Price claims that it cannot be inferred ‘from the fact that an item has a certain function that it will tend to perform that function: the device may be defective, or the environment in which it operates may not cooperate’ (Price 2001, p. 13).

How can we understand the thought that there is something that biological items are supposed to do, that biological items can fail to do what they are supposed to do, that they can malfunction?

Historical accounts look promising. Indeed, it is a ‘persistent boast’ (Davies 2000, p. 19) of this approach, or ‘perhaps the grandest’ of its claims, that it can provide ‘a clear and compelling naturalistic explanation for the occurrence of malfunctions’ (Davies 2001, p. 190). Furthermore, it is claimed by proponents that only a historical account can do this work, Millikan says that this is because non-historical accounts ‘run afoul exactly when they confront the most central issue of all, namely, the problem of what failure of purpose and defectiveness are’ (Millikan 1989b, p. 299). She claims that ‘a crippling defect of any definition that looks for function in current dispositions rather than history is that such definitions cannot ground the notion of malfunction’ (Millikan 1993a, p. 32). This is a serious flaw for non-historical theories since ‘[f]unction categories are essentially categories of things that need not fulfil their functions in order to have them’ (Millikan 1989b, p. 296).

Proponents of historical accounts claim that a certain historical connection to prior tokens of a trait is necessary and sufficient for a trait’s being a member of a functional kind. This connection sets the norms on proper functional performance, that is, ‘being preceded by the right kind of history is sufficient to set the norms that determine purposiveness’ (Millikan 1989b, p. 299). Biological traits have functions ‘whether or not [they are themselves] capable of serving any of these functions’ (Millikan 1993b, p. 56). A trait can make no causal contribution to fitness, but it nevertheless still has something that it is supposed to do (Neander 1991b, p. 459, fn. 7).

Davies claims that from a naturalistic perspective, it is difficult to see ‘how or why we should think that the processes that result in trait preservation confer upon descendent tokens a norm of performance’ (Davies 2001, p. 67, see also 2009, pp. 139–42). I will give an historical account which has the internal resources to lay down this standard, from which token traits can deviate.

Davies’s argument for the claim that historical malfunctions are impossible is, in summary:

I will look first at Davies’s objection as applied to the selected functions account as this is its primary target; the argument is applied to the weak etiological account only derivatively.

Conditions (i)–(v) refer to recently successful ancestral tokens of T. Finally, Davies stresses that ‘conditions under which a token trait has a selected function definitely do not include possession of the capacity required to actually perform the functional task’ (Davies 2000, p. 25). This is of course granted by, and indeed is one of the insights of, historical accounts.

Davies distinguishes generic trait types from selected functional types. The former includes all traits which are apt for selection. The latter includes only those variants of a given generic trait type which are selectively successful, that is, those which possess the property selected for. In the oft-cited case of Bernard Kettlewell’s (1973) peppered moth³ for example, the targeted trait is wing colouration, and this is the generic trait type. The successful variant of this type is a particular wing colour, and it is this which is the selected functional type.

What kind of trait type does clause (v) concern? This is important: if tokens of T in (v) are members of the generic trait type only, then the selected functions account can accommodate malfunction, because the generic trait type will include instances of the trait which are not selectively successful. So, tokens of T cited in (v) could refer to successful tokens, or unsuccessful (i.e., damaged or diseased) tokens. And the latter, which would qualify as malfunctioning tokens, are nevertheless still part of the functional category. This would be a good result for the selected functions theorist.

However, Davies claims that because tokens of T in (i)–(iv) refer to those tokens which possessed the properties which were selected for and are thus members of both the generic trait type and the selected functional type, so too—on pain of equivocation—must be tokens of T in (v). Tokens of T in (v) must refer not just to members of the generic trait type, but to members of the narrower selected functional type. This means that ‘membership in a selected functional category, per condition (v), requires possession of the property selected for’ (Davies 2000, p. 28). It follows that current incapacitated tokens do not belong to the relevant functional types because they lack the capacity which defines that type, despite being descended from a recently successful lineage.

As a result of this, the selected effects account needs to be revised thus:

On this formulation, the selected functions account cannot accommodate malfunction. This is because if ‘due to some sort of defect, disease, or damage the token of trait T lacks the property selected for, then it is merely a member of the generic type, not the functional type’ (Davies 2000, p. 30). If this is the case, norms of functional performance do not apply to such tokens. Davies concludes that selected malfunctions are impossible.

We can think of Davies as offering a dilemma. On the first horn, if tokens of T in (i)–(v) refer to members of the generic trait type, then the selected functions account is extensionally incorrect. On the second horn, if tokens of T in (i)–(v) refer to members of the selected functional type, then the account cannot allow for malfunction. Either way, the account is inadequate. I will argue later that this is a false dilemma, since there is another notion of type, more inclusive than selected functional type, but less inclusive than generic trait type. Once we appeal to this more inclusive type, the selected functions account can accommodate malfunction without being extensionally incorrect, since membership of this type is compatible with a trait lacking the capacity to perform the relevant function (§5). Before then, it is worth noting that the scope of Davies’s objection is limited, as I will now show.

Despite Davies’s view that his objection undermines all historical attributions of malfunction, the scope of his objection is much narrower. Even the unmodified selected functions account can in fact accommodate certain kinds of malfunction. Only when traits are unable to perform their function as a result of congenital factors does a problem arise.⁴ This is because conditions (i)–(v) refer to the very existence of a given token, and, as we have seen, a trait belongs to a given selected functional type only if it has the success property for being able to perform its proper function.

However, the set of congenitally damaged token traits is not co-extensive with the set of malfunctioning token traits. Malfunction does not require congenital disease or defect.⁵ Davies claims that ‘it is difficult to see what in the theory of selected functions or in the theory of weak etiological functions justifies the claim that functions persist even when the physical capacity is lost’ (Davies 2000, p. 20). Also, as we saw in his summary of the objection, tokens can ‘lack the defining property due to defect’ and they can ‘[lose] the defining property due to disease or damage’ (Davies 2000, p. 19). Here we see that Davies took his objection to apply to traits which malfunctioned not as a result of congenital disease or damage. Indeed, he claims that ‘an incapacitated token of T is no longer a member of the category of Ts’ (Davies 2001, p. 199, n. 6, my emphasis). Even if congenitally damaged traits are not malfunctioning ones because they are excluded from the selected functional category, it does not follow that environmentally damaged traits are excluded from the selected functional category too. It follows only that a necessary condition on function possession is that the trait in question started with the capacity to perform the function. To claim necessity for continued possession of that capacity is too strong, or at least, it is not something we can arrive at from Davies’s argument.

The selected functions account can accommodate token traits which had the success property, and were able to perform their function, but now, due to damage or disease, are unable to do so⁶ (I will call these cases of environmental malfunction, as opposed to congenital malfunction). These token traits meet the condition put in place by the selected functions account—they hold the appropriate historical connection with previous reproductively successful token traits. Additionally, these token traits meet the condition put in place by Davies—they possessed the capacity to perform their function. They thus count as members of the selected functional type. If they subsequently fail perform their function because, for example, they lose the property which granted them membership into the selected functional type, then they malfunction. This means that traits which start as members of the selected functional type but later are unable to perform their function, count as malfunctioning. These will be items such as hearts which become unable to pump blood or pancreases which become unable to produce sufficient insulin.

Here is a worry with my argument: to say that tokens which had the success property but then lose it due to damage or disease, malfunction, is simply an un-argued denial of Davies’s challenge. For Davies, membership of a selected functional type requires possession of the success property for that type. Just as tokens which do not have that property from the off do not qualify as members of the selected functional type, so too do tokens which had the property and then subsequently lose it not qualify as members of the selected functional type. My response to Davies succeeds only because I assume a less strict condition on selected functional type membership than Davies would accept (namely, that a trait merely start with the success property).⁷

This objection fails. Tokens of ‘T’ in the selected functions theorist’s conditions on function possession had better refer—on pain of equivocation—to the same kind of thing throughout. And those tokens of ‘T’ had better refer to tokens which actually had the capacity to perform the function, or some success property which enabled the function to be performed. However, Davies adds the further claim that if a token was a member of the selected functional type, and had the success property, then if it subsequently loses that property (and the capacity to perform the function) due to damage or disease, that token is no longer a member of the relevant selected functional type. The selected functions theorist need not accept this. She is forced to accept that tokens of ‘T’ in her five conditions had better refer to tokens with the success property (to avoid equivocation), but she need not accept that those tokens need to retain the success property to remain part of the relevant type. This is Davies’s position, but it is not one the selected functions theorist is forced into, and it is not something Davies argues for. Absent positive argument for the more restrictive account, selected functions theorists are free to reject this additional claim. Continued possession of the success property is not something which is required for conditions (i)–(v) to be coherent, and of course, it is something historical theorists are wont to deny.

However, if the selected functions theorist has only this to say, her victory is Pyrrhic. Congenital malfunction is an important kind of malfunction, which a fully adequate account should accommodate. When selected functions theorists discuss malfunction, they often cite examples which are plausibly—at least in some cases—congenital. Millikan talks of members of functional categories which are ‘diseased, malformed, injured, [and] broken’ (Millikan 1989b, p. 295), and Neander talks of items exhibiting ‘pathological deviations from the norm, due to disease, injury or deformity’ (Neander 1991b, p. 467). Millikan and Neander (among others) thus appear to be seeking an account which can accommodate both environmental malfunction and congenital malfunction (there is certainly no evidence to suggest otherwise). Further, it could well be that environmental malfunction is not sufficient for the kind of project that, for example, teleosemanticists are engaged in. Davies’s objection therefore remains. A more robust response is required.

Here I give a historical account which can accommodate malfunction (including congenital malfunction). Davies argues that the tokens of ‘T’ cited in (v) have to refer to members of the selected functional type on pain of equivocation, as tokens of ‘T’ cited in earlier conditions refer to members of the selected functional type. However, we do not have to accept that the only referent of ‘T’ is Davies’s proposed selected functional type, a type characterised by possession of the success property for functional performance. Instead we can have tokens of ‘T’ in conditions (i)–(iv) referring to members of the selected functional type, and have tokens of ‘T’ in condition (v) referring to members of what I will call the historical trait type. What does it take for a token trait to be a member of some historical trait type? I propose placing two necessary and jointly sufficient conditions on function possession and membership of the historic trait type, such that it is possible for the items which meet them to malfunction. To be a member of the historical trait type is to be a member of the functional kind. Membership of a historical trait type (a functional kind), and being ascribable a function, supervene on the same facts (the ones cited in conditions (a) and (b) below).

The weak etiological account of biological function differs from the selected functions account in that variation during the period in which a trait contributes to organismic fitness is not a necessary condition on function possession. That is, a trait T can have the function to F providing ‘T contributed to the fitness of the ancestors of T’s current bearers by producing an effect E and that this, in turn, contributed to the reproduction of Ts’ (Buller 1998, p. 508). The weak etiological account then is more liberal than the selected functions account, which requires variation for function ascription. On the selected functions account, as we have seen, for a trait t to have the function of F-ing, ‘an item must not just contribute to survival or reproduction, it must contribute to reproduction better than some alternative item, that is, it must contribute to differential reproductive success or fitness’ (McLaughlin 2001, p. 106).

Davies formulates the weak etiological account thus,

Davies suggests that the eschewal of the appeal to selection ‘may give the appearance that [his] argument [...] does not apply to the theory of weak etiological functions’ (Davies 2000, p. 34). However given certain adjustments, the argument applies here also. The adjustment is to cast the argument not in terms of selective success as before, but more generally in terms of properties of historical success (Davies 2000, p. 34).

Even though for the weak etiological theorist variation is not necessary for functional kind membership, for any case, there either was variation among ancestral instances of the trait in question, or there was not variation. Davies argues that whichever is the case, his objection goes through. If there was variation, the weak etiological functional category gets defined by reference to the successful variants that contributed to ancestral fitness by performing F. So, as before, if a present day token does not possess the capacity to perform F due to defect, disease, or damage, that token is not a member of the functional kind, and so it cannot malfunction. If there was not variation, the objection still goes through. Davies supposes that ‘not a single variant was sufficiently defective, diseased, or damaged to prevent it from contributing to fitness’ (Davies 2000, p. 34). Now the weak etiological functional category gets defined by reference to the tokens that contributed to ancestral fitness by doing F, in this case it just so happens to be all ancestral tokens. But again, present day tokens which lack the capacity to perform F, do not count as members of the functional type. So weak etiological malfunctions, as well as selected malfunctions, are impossible (Davies 2000, p. 34).

Again, call the traits which meet conditions (a) and (b) members of the historical trait type. Membership of a functional kind is not defined by reference to tokens which contribute to fitness but rather, as with the selected functions account, is defined by reference to traits which met conditions (a) and (b). The weak etiological account now has it that success of members of the selected functional type—those members which contributed to organismic fitness—resulted in tokens which are members of the historical trait type. And, once again, tokens of the historical trait type can malfunction, because (a) they possess one of a set of intrinsic structural properties, and (b) they are governed by norms of performance bestowed on them by their historical connection to prior tokens of the generic trait type to which they belong which contributed to organismic fitness. If a trait is a member of the historical trait type, then it can malfunction in virtue of there being norms of performance which apply to it. This revised weak etiological account can also accommodate malfunction.

In this paper I have done three things. First, I have shown that the scope of Davies’s objection is limited—some historical malfunctions survive unscathed. Second, I argued that the objection can be answered by my revised historical account. Third, I identified a component of historical theories which have escaped attention. Once that component is brought to the fore, Davies’s challenge can be met by a revised historical account. I proposed a new account according to which there are two conditions on membership of a functional kind. If the historical theorist adopts these conditions, she does not have to accept that functional kinds are defined by reference to successful tokens, it was this which made her account vulnerable to Davies’s objection. Rather she can hold on to the historical component of her account—it is after all by reference to a token trait’s history that a specific function can be specified—and, by adding the requirement of an intrinsic structural property to the conditions on membership of a functional kind, she can distinguish malfunctioning tokens of a given type from so-called useless ones. As I noted at the start of the paper, my aim here has not been to give an account of malfunctions which Davies would accept, since he would not accept any account of malfunction, on the grounds that no account is required. I conclude then that Davies’s challenge can be met by my historical account of biological functions, and that historical malfunctions—of both the selected and weak etiological kind—are possible.