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).
Evolution does not just target features that construe selective advantages. It targets the underlying heritable physical configuration responsible for possession of the survival-enhancing feature. It is that physical configuration which is primarily heritable hence selectable; the survival enhancing feature is only heritable derivatively, by virtue of heritability of the physical feature that gives rise to it. Moreover, there may be more than one such physical state that gives rise to the feature—that is a form of multiple realizability. However, possession of the physical feature need not necessitate possession of the survival-enhancing one. If membership of the kind goes in part with possession of the physical feature, then membership of the kind will then be compatible with lack of survival-enhancing feature (or performance of the function which defines the functional type). My proposal is the following: functional kinds are associated with sets of properties and particular functions. A token trait t is a member of the functional kind K and has F-ing as its selected function if and only if it:
-
(a) Possesses one of a set S of intrinsic structural properties, \(\{p^{1}, p^{2}, p^{3}, {\ldots },\}\)
-
(b) Stands in a causal historical relationship to at least one token trait which
-
i.
possessed one of a set S of intrinsic structural properties, \(\{p^{1}, p^{2}, p^{3},{\ldots },\}\)
-
ii.
possessed one of a set \(S^{*}\) of properties responsible for that token trait F-ing, \(\{q^{1}, q^{2}, q^{3},{\ldots },\}\)
-
iii.
was selected for F-ing.
Where selection for requires variation, in line with the selected functions theorist’s claim that functions are attributed to traits on the basis of what they did which ‘accounts for [their] presence in the population, as over against historical alternative traits no longer present’ (Millikan 1993a, p. 40). If a token trait meets the conditions set out here, it is a member of a historical trait type (and a functional kind), and can be attributed a function in the usual way, that is, by looking to its history. Note here that I have moved from talking about the selected functional type, as is Davies’s preference, to talking of the historical trait type, as is mine. In the next two sub-sections I will say more about the details of these conditions. Explication of condition (a) will take up the most space here, since the positing of a disjunctive set of properties as required for function possession is a large part of what is novel about my approach.
Condition (a)
Condition (a) picks out a class of intrinsic structural properties. What is required is a set which captures those properties which are possessed by all items to which a particular function would be ascribed, but are not possessed by items which would be ascribed a different function (namely, different traits), or no function at all (namely, severely malformed organic stuff). Here is a rough (and simplified) approximation of one way the historical theorist might go in pinning down the nature of this set (note that I offer this as one way, there are no doubt numerous ways tied in with the production of functional traits which might delineate the set of properties in (a)).
We could appeal to an appropriate range of gene expression associated with recent ancestral items which performed F. By gene expression, I mean to refer to the process through which information contained within a gene is used in the creation of a gene product (e.g. a protein). To qualify as a member of a functional kind K, with the function of F-ing, a token trait t must have been produced by appropriate gene expression, that is, within a range considered a reasonable attempt to produce an item similar to ancestral items which performed F. This set of otherwise heterogeneous intrinsic structural properties cited in (a) can be picked out by an appeal to an appropriate range of allowed departures from the genetic development of ancestral items which performed F. This is just to say that there is some set of properties for each functional kind, members of that set may be rather different from one another, but they can be identified as a set by appeal to an appropriate range of departures from the genetic development of ancestral items which performed F. The idea is that the set of intrinsic structural properties cited in (a) delineate a multiply realizable kind. Past instances of this kind are selected for in virtue of their role in the production of items which were able to perform the function of F-ing, which is adaptive (they might, for example, possess property \(p^{1}\), or \(p^{2}\)). This is compatible with malfunction because not every way of realizing this kind needs to be in an F-ing contributing way (some realizations of the kind might be in virtue of possessing property \(p^{3}\), or \(p^{4}\). These properties are within a range of allowed departures of gene expression associated with ancestral selectively successful items). Evolution chooses between intrinsic properties, function attribution is licensed when these intrinsic properties start to have relational functional properties, that is, when condition (b) is met.
I should say something about my use of the terms ‘appropriate range’ and ‘reasonable attempt’. These terms are of course metaphorical insofar as natural selection does not attempt anything. Talk in these terms is intended to be helpful in capturing those traits which come pretty close to being well functioning members of a given kind, but nevertheless go awry. That is, those traits for which something goes awry (as in the case of Holt-Oram syndrome below), but does not go awry enough to disqualify it from the functional kind (as in the case of what might be best described as ‘glob[s] of misplaced organic matter’ (Millikan 1984, p. 25)). For a token trait to qualify as a member of the functional kind heart, we can usefully talk in terms of its being produced within a range considered a ‘reasonable attempt’ to produce something of that kind. How wide is this acceptable range? That will depend on what physical features are selected by natural selection, and that is not something we can pronounce on a priori. The thought is that evolution targets underlying physical structures by virtue of their producing certain traits that confer reproductive advantages. More than one structure or set of properties will be associated with the development of each of these traits. Exactly what counts as the acceptable range is determined by that. This vagueness and imprecision in our language for characterising the set S associated with kind K and function F need not reflect any vagueness or imprecision in actual concrete cases. It is just a by-product of needing a single term to cover the proposal’s structure in abstraction from any given case.Footnote 8
Although admittedly somewhat vague in the abstract, its application in the following case study is hopefully sufficiently clear and precise to alleviate doubts arising from that. Here I run through an example of the genetics of the development of the human heart to show my account at work. The development of the human heart is extremely complicated, and something I cannot do justice to in this short foray into the genetics of this process.Footnote 9 In addition, the genetics of the human heart is a relatively young field, and one which is quickly changing (Pierpont et al. 2007, p. 3015). Let us say that the genetic processes involved in the production of a human heart can produce an array of traits which will fall into one of three categories: working, defective, and what I will call useless (i.e., Millikan’s ‘globs’). Our appeal to the set of intrinsic structural properties demarcates working and defective traits from useless traits, with those in the former two categories meeting condition (a) on membership of the functional kind heart. Such categories will be roughly correlated with the expression of genes which fall into the range considered a reasonable attempt to produce an item similar to ancestral items which pumped blood, and will group together working pumps and defective pumps. And so-called useless traits—excluded from the functional kind—will correlate roughly with the expression of genes associated with the production of ancestral items which pumped blood, but which fall outside of the range of gene expression considered a reasonable attempt to produce such an item.
Take an example of a defective heart, one which my account would classify as a malfunctioning member of the functional kind human heart. The underlying genetic basis for many heart defects is not known, but what has been found is that mutations in those genes encoding core cardiac transcription factors are involved in many forms of congenital heart disease (McCulley and Black 2012, p. 254). In particular, transcription factors NKX2–5, GATA4, and TBX5 have been identified as playing central roles in the development of the heart, and mutations in these genes are involved in congenital heart disease (McCulley and Black 2012, p. 254, see also Pierpont et al. 2007 for analysis of congenital heart defects associated with single gene mutations). For example, in patients with structural malformations of the heart, mutations in NKX2–5 have been found (McCulley and Black 2012, p. 257, see also McElhinney et al. 2003). In cashing out how condition (a) might apply, I will look at the role of the TBX5 gene in Holt-Oram syndrome.
Holt-Oram syndrome, also known as ‘heart-hand syndrome’, occurs in around 0.001 % of the population. It is a condition characterised by deformities in the upper limbs and congenital heart defects (Pierpont et al. 2007, p. 3024). Research on this syndrome has revealed that it is caused by mutations in a single gene: TBX5,Footnote 10 one of the three genes identified above as playing a central role in human heart development (see for example Basson et al. 1997; McDermott et al. 2005; Li et al. 1997). Most of such mutations in this gene prevent the production of the T-box 5 protein which activates genes involved in the normal development of the upper limbs and the heart. Studies on the role of TBX5 in Holt-Oram syndrome have shown that decreased activity of this gene is responsible for septal defects (most commonly in the wall separating the left and right sides of the heart) (Basson et al. 1997, p. 33). I focus on this example for simplicity’s sake, since little genetic heterogeneity is found among patients with Holt-Oram syndromeFootnote 11 (Pierpont et al. 2007, p. 3025). In this syndrome, we have an example of mutations in a single gene causing abnormal hearts. However, in this case we clearly have something fairly close to a functioning heart and fairly far away from a glob of organic matter. To see this consider some of the macro-level consequences of this syndrome. Transcription of TBX5 is Normally required for septal formation. If this gene is defective, this does not completely destroy or eliminate cardiac function, but merely impairs it. The heart is less effective at what is has been designed to do, but it nevertheless continues to function as a heart in broad terms.Footnote 12 Here we have just one gene which goes awry, this is the kind of case I envisage being characterized as within a range considered a reasonable attempt to produce a heart. This means that the defective hearts of patients with Holt-Oram syndrome are malfunctioning hearts, on my account.
Satisfaction of (a) is not sufficient for membership of a functional kind—this is in line with the historical theorist’s claim that an item’s history is necessary for function ascription. If possession of a property in set S in (a) was sufficient, then, depending on how we pinned down the nature of the set of intrinsic structural properties, we might get function ascription for first traits (those which are not descendants of F-ers with one of the properties causally responsible for the trait F-ing) and SwampFootnote 13 traits (those possessed by the spontaneously coalescing creatures of philosophical thought experiments). Condition (a) can be met by first traits and Swamp traits. When we step back and look at all the traits which were associated with certain sets of gene expression, first traits will be included in this set. As for Swamp traits, they too will meet condition (a) understood in this sense (due to Swampman being a ‘physical replica’ of a biological organism (Davidson 1987, 443)). So these traits may meet condition (a) (again depending on how we spell out the nature of the set of intrinsic structural properties), but in their failing to meet condition (b), they do not qualify as members of a given functional kind. So they do not have functions, and thus they cannot malfunction.
A worry about condition (a) is that the requirement of a structural property may actually be in tension with historical accounts, because proponents hold that history is sufficient for function ascription. It is true that proponents of the selected functions and weak etiological accounts have claimed that history is sufficient to determine purposiveness and norms of performance. However, I think that the appeal to a structural property would not be rejected by historical theorists, since the history which is claimed sufficient is a certain kind of history. A brief aside is needed here to defend the claim that the requirement of possession of one of a set of intrinsic structural properties (condition (a)) is not in tension with historical accounts.
Millikan speaks of devices that are ‘similar to one another—as human hearts are similar to one another’, (Millikan 1984, p. 18, my emphasis). In her definition of a higher-order reproductively established family, as well of talking of ‘similar items’, Millikan talks of traits being ‘in some respects like’ Normal members of a given reproductively established family (Millikan 1984, pp. 24–25, my emphasis). She claims that the use of such terms (‘in some respects like’ and ‘approximates in some degree to a Normal explanation’) reflects the vagueness of the question ‘whether a bit of matter should be called “a malformed eye” or merely “a glob of misplaced organic matter on the forehead”’ (Millikan 1984, p. 25).
How might we to understand what ‘similar’ and ‘in some respects like’ means here? We should not cash out ‘similar’ by appeal to morphological properties, or dispositional properties, because in doing so we cast our net too narrowly—we will exclude malfunctioning traits from our functional categories, as well as traits which simply fail to perform their functions in abNormal conditions. What we can do instead though, is adopt my condition (a) on membership of a functional kind. This would cast the net such that working and malfunctioning traits are distinguished from useless traits. The vagueness Millikan refers to is the kind of vagueness which my condition (a) goes some way to resolve. This is what I am up to when I offer condition (a) on membership of a functional kind. This condition will demarcate working and malformed items of a given trait type, from what I have called useless items.
Earlier I suggested that in cashing out condition (a) we might usefully think in terms of reasonable departures from gene expression. Condition (a) demarcates those working and defective traits from useless traits, with such categories being roughly correlated with gene expression which falls into the range considered a reasonable attempt to produce an item similar to ancestral items which performed F. This story sit very well with Millikan’s claims, bearing ‘some resemblance’ and matching ‘in some relevant respects’ can be cashed out in terms of producing an item within a range considered a reasonable departure from the gene expression associated with selectively successful ancestral tokens of the functional kinds.
Turning now to Price, a weak etiological theorist. She too indicates that the requirement of a structural property would not be unwelcome. When discussing Andrew Woodfield’s (1976) account of function statements (what she calls an ‘Actual Capacity Account’ (ACT)), she suggests that the ACT theorist should borrow a criterion of Millikan’s, one which ‘focuses on the causal origin of the device: two devices belong to the same group if they were produced by the same genetic mechanisms in (more or less) the same way’ (Price 1995, p. 147, my emphasis). And, in a footnote attached to this claim, Price claims that the ‘phrase “more or less” is needed to allow for malformed devices’, a claim she attributes to Millikan (1984, p. 25) (Price 1995, p. 147, fn. 9). In discussion of her own account, Price claims that ‘two devices will count as belonging to the same type if they were produced by the second system, or its ancestors, in more or less the same way’ (Price 1995, p. 151, my emphasis).
In later work, Price claims that ‘[a]n item will be equivalent to one of these earlier items if it is causally related to [some token device] in much the same way as those earlier items were related to earlier members of [some type device] that assisted them to [perform some function]’ (Price 2001, p. 38, my emphasis). And in a footnote to this claim she states that the ‘processes that produced my heart and my liver were related to each other in more or less the same way as the processes that produced my ancestor’s heart and her liver’ (Price 2001, p. 38, fn. 5, my emphasis).
This ‘in much the same way’ and ‘more or less’ can be understood in my terms. If two token traits were produced within an appropriate range of gene expression associated with recent ancestral items which performed F, they are ‘produced by the same genetic mechanisms in (more or less) the same way’ (Price 1995, p. 147, my emphasis). They can also both be said to be ‘causally related to [some token device] in much the same way as those earlier items were related to earlier members of [some type device]’ (Price 2001, p. 38, my emphasis).
So requiring a structural property for membership of a functional kind would be friendly to both selected functions and weak etiological accounts, and would help pin down more precisely what MillikanFootnote 14 and PriceFootnote 15 are getting at when they talk in terms of items being ‘similar’ or produced ‘in more or less the same way’. It is in bringing out and developing this underexplored commitment in the work of some historical theorists that we can formulate a revised historical account of function which can accommodate malfunction. This is achieved by making explicit the requirement of an intrinsic structural property in the conditions on functional kind membership.
Getting more specific about the kind of intrinsic structural property required is tricky,Footnote 16 and though I have offered a way historical theorists might go with regards to this, it is inevitable that vagueness will arise. Though vagueness of this kind will infect any account of biological function owing to borderline cases, and so it does not speak against the conditions on function ascription I have offered. This is recognised—indeed embraced—by historical theorists. Millikan claims that there ‘are lots of borderline cases of proper function, if not so many in nature, certainly in possible worlds (Millikan 2002, p. 115, fn. 2). When discussing the development of functional terms (‘Cummins biofunction’ and ‘proper biofunction’) she suggests that we should ‘not attempt to give these notions entirely clean boundaries. Nature has many important joints, but these joints are seldom clean’ (Millikan 2002, p. 122, my emphasis). Similarly Price claims that there ‘is nothing wrong with a certain amount of vagueness in our account of functions provided we are clear about what is going on’ (Price 2001, p. 29, fn. 19).
Condition (b)
Let us now look to (b). The three clauses in (b) are each doing important work, they make specific the causal relationship required to hold between a present day token and an ancestral token of a trait type, by requiring that the ancestral token satisfies clauses [i], [ii], and [iii]. Let us first look to (b) [i]. Requiring the ancestral trait in the causal-historical relation to possess one of a set S of intrinsic structural properties, \(\{p^{1}, p^{2}, p^{3},{\ldots },\}\) rules out the possibility of functional kind membership for traits which are descendants of something which performed F but are a different kind of thing, for example, a blood-pumping pseudo-lung (note that I do not call it a lung because biological kinds are typed by their functions). The property in (b) [ii] is the success property, the property which allows the function to be performed, if a trait has this property it is a member of Davies’s selected functional type. This clause is important because there must be historical success for function attribution. Finally, clause [iii] of (b) is present to reflect the selected functions theorist’s commitment to the claim that variation is required for selection. That is, a trait has the function it does because the historical performance of that function accounts for the trait’s presence in the current population, ‘as over against historical alternative traits no longer present’ (Millikan 1993a, p. 40).
Importantly, meeting the three clauses in (b) is not sufficient for function attribution. The historical connection to, for example, ancestral human hearts cannot be enough for functional kind membership for at least two reasons. First, my lungs hold a historical connection to ancestral hearts (since my ancestors had hearts), but that does not mean that my lungs are part of the function kind heart and have pumping blood as their proper function. Second, imagine the case in which something goes so wrong, that the trait produced is a so-called useless one. The resulting product does not come close to being a heart. These items are not malfunctioning items, but rather not functional traits at all. Condition (a), the requirement of an intrinsic structural property, excludes such items from the functional category.
In summing up the presentation of my position, it is important to note how I differ from previous historical accounts of function. My account is very much influenced by Millikan’s, and so I will note the ways in which I depart from her, especially with respect to accommodating malfunction.
In her definitions of higher-order reproductively established families Millikan gives a condition on membership to some functional kind R as its having ‘been produced in accordance with an explanation that approximates in some (undefined) degree to a Normal explanation for production of members of R’ (Millikan 1984, p. 25). She wants to make room for malformed members of R by appeal to this similarity condition (though see fn. 2).
In my account, I make room for malfunction by appeal to a range of gene expression associated with ancestral items which were selectively successful. This range captures a set of otherwise heterogenous intrinsic structural properties, which includes working and defective items, and excludes useless items. Introducing the idea of functional kinds being associated with sets of properties, allows functional kindhood to be multiply realizable by appeal to these sets of properties. Not every way of realizing some functional kind needs to be one which realizes it in virtue of being selectively successful.
Here is a (rather dramatic and admittedly toy) case where my account and Millikan’s license different judgements. Suppose radiation poisoning causes gene expression which produces a heart which cannot perform the function of pumping blood without artificial assistance. If that gene expression were within the range associated with ancestral tokens of the functional kind heart, but the deviation from the ideal were caused by radiation, my account has it that the resulting product would be a malfunctioning heart. For Millikan though the resulting product would not be a malfunctioning heart, but a heart produced under abNormal conditions.Footnote 17
Millikanian Normal conditions and Normal explanations are indexed to a distinctive historical environment. In contrast, I place the emphasis on gene expression in an environment which need not be a Normal one. My account is thus less relational than Millikan’s. Her talk of items being produced in accordance with an explanation that approximates a Normal explanation, raises the question of whether something could be unable to perform its function in a given environment in a way which could properly be described as malfunctioning as opposed to just failure to perform (hence why Millikan is doubtful about putative cases of malfunction, see fn. 2). If one is committed, as Millikan is, to restricting cases of malfunction to ones in which the item is produced in accordance with an explanation which approximates a Normal one, malfunction will be very rare indeed. I am not committed to such a restriction.
The possibility of selected malfunction
If the selected functions theorist adopts my proposal, she can now meet Davies’s objection by reading ‘tokens of T’ in conditions (i)–(v) as follows:
-
(i)
Past instances of tokens of the selected functional type in O performed F in E,
-
(ii)
The selected functional type was heritable,
-
(iii)
Past performances of F caused an increase in O’s relative ability to satisfy demands of E (relative to other organisms in the population which do not have tokens of the selected functional type),
-
(iv)
This increase in O’s ability to satisfy selective demands of E resulted in an increase in O’s long-term relative rate of reproduction,
-
(v)
This increase in relative reproduction resulted in the persistence or proliferation of O and hence tokens of the historical trait type.
Though the referent of ‘T’ changes between conditions (i)–(iv) and condition (v), this is unproblematic. This is because now, the selected functions account has it that the success of members of the selected functional type resulted in tokens of the historical trait type. And 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 the norms of performance bestowed on them by their historical connection to prior successful tokens of the generic trait type to which they belong. As long as a trait is part of the historical trait type, then the norms of performance apply to it, and it can malfunction.
To be clear: I conceded that Davies has shown that the referent of tokens of ‘T’ needs to be such that equivocation is avoided. I suggested that equivocation can be avoided by appealing to the historical trait type, members of which meet my conditions (a) and (b) on function ascription. My response is to the charge that historical accounts do not have the internal resources to account for malfunction. Davies thinks that the referent of ‘T’ throughout conditions (i)–(v) should be the selected functional type. The historical theorist need not accept this proposal. Rather, the success of members of the selected functional type can result in members of the historical trait type without equivocation. This revised selected functions account can accommodate malfunction.