Since the discovery of the structure of DNA by Watson and Crick in 1953, the advances in genetic research have been astounding. It is now common to hear in the press the discovery of the gene for X. Where X is anything from obesity to alcoholism. Probably the most common understanding of innateness is stated nicely by Ned Block. The content of knowledge or various cognitive capacities are said “to be innate just in case the knowledge and capacities are… determined by the genes, rather than learned or otherwise determined by the environment.” (Block 1981, p. 280).

On the flip side of this, people understand an innate trait to be one that was not acquired by experience. Spelke argues “if early knowledge encompasses environmental constraints that are not obvious in the child’s perceptual and motor experience while failing to encompass more obvious constraints, then this knowledge is not likely to have been shaped by the child’s perceptual and motor experience.” (Spelke 1994, p. 438). Spelke characterizes such knowledge as “initial knowledge” and argues that it is innate. So the account we have is that an innate trait is one that is determined by the genes and is not acquired through perceptual or motor experience. In addition to this, it is common to suppose that innate traits are adaptations chosen by natural selection. “If the constraints are highly reliable, moreover, then natural selection may have favored the evolution of mechanisms that give rise to this knowledge.” (ibid.)

In part one I consider the deepest challenge to this account of innateness, Developmental Systems Theory. The best case to be made for the “in the genes” account of innateness is understood by looking at the nature of causal explanation. Contrastive causal explanation as described by Peter Lipton (1991) makes it possible to give a privileged role to genes in developmental explanation, but provides no compelling reason to do so in exclusion of other explanatory strategies. In this section I conclude that not much sense can be made out of this sort of “in the genes” account of innateness.

Another common intuition about innateness is that innate traits are traits we cannot do anything about. That is to say they develop rigidly and robustly in a broad range of normal environmental circumstances. For example, it would take some very unusual and dangerous circumstances for a person not to develop two arms and two legs. In part two I consider two sophisticated “invariance” accounts of innateness and find them inadequate because they conflate two distinct explanatory projects: adaptive evolutionary explanation and developmental explanation. I also consider a recent methodological account of innateness from Richard Samuels (2002). I will show that this account fails for reasons similar to those that account for the failure of invariance accounts.

I will show that there is no natural kind that corresponds to these common intuitions about innateness. Innateness fails to refer in the same way as terms such as phlogiston or superlunary space. Some real biological phenomena bear a small resemblance to nativism as commonly understood. However, if these phenomena are labeled innateness, there is the danger that people will be led to believe that these phenomena have all the properties commonly associated with innateness. Since “innateness” is such a controversial and nothing in nature corresponds to our common intuitions about it, the term should be dropped from the scientific lexicon.



Part I: Developmental Systems Theory and Causal Explanation



Developmental systems theory (DST henceforth) is a relatively new perspective on development that provides a significant challenge to ideas such as there being a “gene for” X or there being such a thing as a “genetic program.” In other words, DST challenges the notion that there is such a thing as innate traits where innateness means that a trait is “in the genes.” DST rejects dichotomous views of development. These are views of development in which there are two classes of developmental resource, genes and the rest. For example, Richard Dawkins divides the developmental resources into what he calls replicators and interactors (Dawkins, 1976). Replicators are entities that are copied from one generation to the next and form lineages. Interactors are entities that interact more or less successfully with the organism’s environment. Genes have been given a privileged role in development in the literature because they are argued to be the unit of inheritance and because they are the only developmental resource that carries information about the developmental outcome. It is for these reasons that many people consider an innate trait to be “in the genes.”



1. Epigenetic Inheritance



According to DST advocates, both of these posits are false. The first thing to point out is epigenetic inheritance. Inheritance of non-genetic factors has been a popular topic in recent biology. There are several levels at which organisms inherit non-genetic factors. Cellular machinery, microorganisms, and various aspects of their environments are stably inherited over evolutionary time. To develop normally an egg cell needs a very complex array of biochemical machinery. Griffiths and Sterelny (1999) give a number of interesting examples of epigenetic inheritance. Some of the non-genetic cellular machinery that is inherited from parent cell includes: basal bodies and microtubule organizing centers, cytoplasm chemical gradients, DNA methylation patterns, membranes and organelles. All of these epigenetic factors can have important influences on development. For example, it has been suggested that methylation patterns may even affect behavior. Methyl groups block transcription of certain genes. One study suggests that females methylate a certain sequence on the X chromosome so that individuals who only get one X chromosome from their mothers cannot transcribe genes in that region. The gene products from this X chromosome are unavailable to males. (Skuse et al. 1997). Methylation patterns are replicated by a specialized copying mechanism that exists in every cell which is descended from the sperm or egg cell.

Many microorganisms are inherited as well. Some castes of the aphid Colophina arma require a large growth spurt as part of their life cycle. These particular castes inherit microorganisms that produce chemicals that make this growth spurt possible (Morgan and Baumann 1994). The growth spurt in this particular caste of aphids is reliably replicated over evolutionary time.

Many parasites rely on the epigenetic factors for development, namely their host species. Some species rely on the same host through host imprinting over evolutionary time. Insects of many kinds lay their eggs on plant species that they fed on as larvae or caterpillars because they imprinted on these species. This imprinting is analogous to birds imprinting on their parents. The epigenetic factors extend to factors in the environment that are reliably replicated over evolutionary time. In his excellent critique of Konrad Lorenz’s conception of genetic of innateness, Daniel S. Lehrman points out that female rats who don’t lick their genitalia during pregnancy will eat their young. Female rats construct a nest and retrieve their young only if they were exposed to temperature variations earlier in their life and had a chance to carry other objects around in their mouths (Lehrman 1953). Developmental systems theorists conclude from cases like this that species do not just inherit genes. They inherit a complex developmental matrix. Those who claim that genes are the unit of selection cannot make their case by arguing that genes are the only developmental resource that is inherited over evolutionary time. They must show that genes play some kind of distinctive or privileged role. Likewise, an “in the genes” account of innateness must also show a privileged role for genes.



2. DNA as Information and the Symmetry Argument

One way to privilege the causal role of genes would be to argue that they are the only entity that carries information about the developmental outcomes. This is done by way of a building metaphor. A dichotomy between the “genetic program” or “genetic blueprint” and the building materials is created. The genes contain the instructions to build the organism and the other developmental resources are mere building materials. This strategy potentially gets around the complaint that resources other than genes are inherited. Advocates of this information argument can argue that while epigenetic inheritance is possible, the developmental resources other than genes that are inherited do not contain information about the developmental outcome. The problem with this is that we have no concept of information that will end up privileging the causal role of genes.

According to Griffiths and Sterelny (1999) there are two types of information: intentional information and causal information. Intentional information is the sort of information that our thoughts have. Our thoughts are about something and it is this “aboutness” that characterizes intentionality. Intentionality is the relationship between a thought and the object of that thought. The idea is that genes carry intentional information about the normal developmental outcome. If that outcome does not come about, the genes were simply misread and the outcome is “abnormal.” If this account is true, then genes are distinct as compared to other developmental resources because they are the only resource that carries intentional information about the outcome. However, if our strategy is to argue that genes contain intentional information, we run into a host of problems. Basically, we run into the same problems familiar from the vast “naturalizing content” literature. One of the major obstacles to a materialist account of mind is to understand how aboutness or intentionality can be a property of a physical system. The literature on “naturalizing content” is notoriously fraught with difficulties.

For example, we could try to apply the teleosemantic theory championed by Ruth Millikan (1989). According to this theory a thought is about things that evolution has designed it to be about. In other words, a thought has a proper function. Suppose a person is walking down a forest path and suddenly they see something that looks like a snake. This person has the thought SNAKE. The proper function of this thought would be to alert the person to the danger of poisonous snakes in their environment when there are snakes around. This theory could be applied to genes. Each gene carries information about developmental outcomes for which it was selected to produce. Griffiths and Sterelny (1999) point out that the problem with this is that epigenetic influences on developmental outcomes have selection histories as well. These influences would carry information about developmental outcomes as well. Thus, under a teleosemantic theory, genes would not have the distinctive informational role that the theory was invoked to show for them.

The other sort of information is causal information. Causal information derives from a mathematical notion developed by Shannon and Weaver (1949). The mathematical theory only measures the quantity of information and not what it is about. However, there is a closely related causal theory of information. Information of this sort flows on a channel between too systems, the receiver and the sender. The receiver is the system that contains the information and the sender is the system that the information is about. There is always a reliable causal channel between the receiver and the sender. This is precisely the way that scientific instruments of measurement work. There is a causal channel connecting a scale with the item it is weighing because the state of the scale readout is reliably caused by the mass of the object placed on the scale and the local gravity. There is a causal chain going from placing the object on the scale all the way to producing the correct readout on the scale. This causal chain makes up the channel conditions. Just like in any transmission system there can be noise. That is to say channel conditions can distort the signal by the time it gets from the sender to the receiver. Based on this idea then, the genome is the signal and all the other developmental resources are channel conditions which make it possible for the life cycle of an organism to receive information from the genome.

At this point it is important to look at one of the most fundamental arguments posited by DST advocates. I will call it the symmetry argument. This argument has been made in a number of places but I will follow Griffiths & Sterelny (1999). It should be noted that Sterelny does not advocate this position. His alternative can be found in Sterelny, Smith, and Dickison (1996). Griffith and Sterelny point out, “It is a fundamental fact of information theory that the role of signal source and channel condition can be reversed.“ (Sterelny & Griffiths 1999, p. 102) Under this conception, information is just causal co-variation. If we hold other developmental influences constant, genes will co-vary with phenotypes and thus can be said to carry information about phenotypes. However, it is also possible to hold all variables constant except some other influence, say DNA methylation patterns. This influence will co-vary with a developmental outcome. This time it is the DNA methylation pattern that carries the information. This can be done for many other developmental resources one might be interested in isolating. Under this conception of information, DNA is not particularly more distinct than many other developmental resources and thus the information argument to privilege genes fails. For the DST advocate, there is causal symmetry between all the developmental resources. Each and every developmental resource is part of the cause of the developmental outcome and forms a causal matrix that is the life cycle of the organism.



3. DST and Causal Explanation

An important complaint leveled against DST is that though it is strictly true that all these resources are a part of the cause of development, it is still possible to privilege genes in explanation. The thesis of DST is that there is nothing metaphysically special about genes. They do not bear the ontological status of master molecule. However, this does not preclude the possibility of privileging genes in explanation with such locutions as a "gene for" X. This is the most significant challenge to DST. In this section, I will discuss the best case that is available for this view using Peter Lipton's (1991) excellent account of contrastive causal explanation. I will show that while this critique forces DST advocates to give up some of their more radical rhetoric, it does not defeat the DST position in any fundamental way. In light of this, I will reassess what we can conclude about the DST challenge to innateness.

Sterelny and Kitcher (1988) argue that genes are difference makers. "We can speak of genes for X if substitutions on a chromosome would lead, in the relevant environments, to a difference in the X-ishness of the phenotype."(p. 348) The "relevant environment" is presumably something like a species typical environment. Whenever we come across terms like "typical" or "standard" our philosophers' rubbish detectors immediately spring into action. "Standard" will have to be cashed out more specifically and this cashing out is notoriously difficult when your audience is filled with highly critical philosophers. I don't intend to delve further into the way in which Sterelny and Kitcher try to cash out the notion "relevant environments" or "typical environments" because I believe a much better version of this can be gleaned by making use of contrastive causal explanation. Furthermore, the arguments presented below will apply to any explanatory strategy in which some causes are taken as background and others are highlighted. But first I need to characterize the problem that Sterelny and Kitcher perceived in more detail.



3.1 DST and Causal Histories

Phenomena in this world have long and wide causal histories. David Lewis puts it well when he says the question is not whether causal histories in this world are short or long, but whether “they are infinite or merely enormous.” (Lewis 1986, p.183) There is this very complex web of causal elements that combine to produce a trait. We can see that this is exactly the sort of thing Lewis' view applies to.

John Stuart Mill argues that we cannot just leave our explanation at a single cause. “It is seldom, if ever, between a consequent and a single antecedent, that this invariable sequence subsists. It is usually between a consequent and the sum of several antecedents; the concurrence of all of them being requisite to produce, that is, to be certain of being followed by, the consequent.” (Mill, Book III Ch. V §5) David Lewis defines a causal History as follows. “The causal history of a particular event includes that event itself, and all events which are part of it. Further, it is closed under causal dependence: anything on which an event in the history depends is itself an event in the history.” (Lewis, 1986 p. 185) According to Mill, “The real Cause, is the whole of these antecedents; and we have, philosophically speaking, no right to give the name of cause to one of them, exclusively of the others.” (Mill, Book III Ch. V §5) Mill would say that the real cause of a phenomenon is its entire causal history. In this regard Mill anticipated DST by a century and a half. The fundamental argument of DST is that all of the developmental resources are causally symmetrical. That is to say that they are all necessary to produce the consequent. This is just a special case of Mill's idea. If we brought Mill to the present day and he was brought up to date on the state of the art in modern biology, he would likely argue that a gene is just a single antecedent and we have no right just to focus on this as the cause of some phenotypic trait.

Imagine that the causal history of a phenotypic trait is a long, wide scroll as depicted below.



fig. 1



On the vertical axis we have a chain of events stretching backwards through time. Unchecked this chain of events would stretch clear back to the big bang. After all, if the big bang had not happened then phenotypic trait P would not exist. Along the horizontal axis are all the concurrent events that contribute causally to the event in question. DST has been criticized as being unmanageably holistic. Sterelny, Dickison, and Smith (1996) ask "Is Elvis Presley part of our developmental systems by virtue of his role in the development of our musical abilities?" (p. 59)¹ There is simply too much causal history to include in explanation. DST advocates are fully aware of this problem and a solution to it will be provided below. What is needed for pragmatic purposes is a principled way to obtain causal focus. In other words, we need an explanatory heuristic that allows us to delimit a subsection from the causal history. The key to this is to find a principled way to leave some parts of the causal history as a constant background while others are highlighted.



3.2 Contrastive Explanation

First, I will discuss the way in which contrastive explanation can provide this causal focus. I will follow this by characterizing the focus already provided by the arguments of DST theorists, and finally I will assess the virtues of the results of this analysis. Contrastive explanation provides the mechanism to select parts of the causal history that are relevant to the question being asked.

This is how it works. You take a fact, say that my computer had an error, and compare it to a foil. The foil eliminates much of the causal history as irrelevant to the question. The foil is the fact that follows the “rather than” locution when the contrast is made explicit. Why did my computer have the illegal protection fault error rather than crashing. The foil in this case is crashing. We are now given two causal histories to compare. All the causes in the causal history of the fact that are different from the causes in the causal history of the foil are possible explanations. We now have eliminated the big bang because both causal histories contain the big bang. Note that picking the foil amounts to picking the question. Suppose you ask me, Why did you go to Lord of the Rings? You might be asking one of many different questions. You might be asking, Why did you go to Lord of the Rings rather than stay at home? Then my answer might be I did not have anything to do at home. You might ask why did you go to Lord of the Rings rather than Harry Potter? I could then answer, I have already seen Harry Potter. One can see that what makes each question different is the foil I choose. Choosing this foil also delimits the explanation space and focuses us in on a part of the causal history. I would not say, I went to Lord of the Rings rather than Harry Potter because I had nothing to do at home. This is an important feature of contrastive explanation because it shows how it is an interest relative heuristic. Based on which foil we choose we can eliminate parts of the causal history as irrelevant to the question at hand.

More needs to be said about the comparison of causal histories alluded to above. Peter Lipton’s difference condition provides us with a principle of causal selection. “To explain why P rather than Q, we must cite a causal difference between P and not-Q, consisting of a cause of P and the absence of a corresponding event in the history of not-Q.” (Lipton, p.217)² Lipton points out that this difference condition has the advantage of picking out actual differences in causal history rather than in counter-factual causal histories, as David Lewis suggested (Lewis 1986), because we compare two real causal histories.

The following example illustrates how the difference condition works. Suppose Fred installed the same program on his computer, but did not get an error as I did after I installed the same program. If I ask why did my computer get the error rather than Fred’s, we might find that Fred has a different video card that causes different settings in the registry. The difference then is the video card, and we will count this as the cause and thus the explanation of this why-question. In this case P is the fact that my computer had the error and not-Q is the fact that Fred’s computer did not have the error. Lipton calls this a compatible contrast because my computer’s having the error does not preclude the possibility of Fred’s computer having the error as well. It is a little more problematic how the difference condition can work for incompatible contrasts. These are contrasts in which P’s being the fact of the matter makes it impossible³ for Q to be a fact.

The following example from (Griffiths & Sterelny 1999) is an example of an incompatible contrast. "A certain allele in humans is 'an allele for brown eyes' because, in standard environments, having that allele rather than alternatives typically available in the population means that your eyes will be brown rather than blue." Alleles are alternative forms of genes at one locus on a chromosome. So at one locus one might have the blue eye allele or the green eye allele. In this case P is having brown eyes and not-Q is not having blue eyes. The difference in this case is having a particular allele at a particular locus on the chromosome and thus this allele is selected as 'the cause' of having brown eyes by the difference condition. In this example we have an incompatible contrast because it is impossible for one person to have both a brown-eye allele and a blue-eye allele. The problem in applying the difference condition here is that there is only one causal history and thus there seems to be nothing to compare. P’s causal history is the same as the causal history of not-Q. The following is Peter Lipton’s answer to this problem. “The difference condition does not require that the same event be present in the history of P but absent in the history of not-Q, a condition that could never be satisfied when the two histories are the same, but only that the cited cause of P find no corresponding event in the history of not-Q, where a corresponding event is something that would bear the same relation to Q as the cause of P bears to P.” (Lipton, 1990 p. 217). This idea of a "corresponding event" needs to be worked out in more detail because most of the contrastive why questions that will be asked in the context of innateness will be incompatible contrasts of this sort. The reason for this is that the contrasts of interest will most often be about features of one individual's development. One way that this problem is avoided is by comparing the development of two individuals in cases where their environments are roughly the same. If we accept that these technical problems can be overcome it appears that contrastive explanation provides a principled way to focus on genes as causes of phenotypic outcomes.



3.3 Delimiting Causal Explanation Spaces

Let us now return to the metaphor of the causal history being a long and wide scroll as depicted in figure one above. In doing so we can compare the way in which Griffiths & Gray (1994) delimit the relevant causal space with the various possibilities for doing so provided by contrastive explanation. Griffiths & Gray delimited the causal space in two ways. First, they made the relevant unit of selection the life cycle. "The central theoretical entity in our account is the developmental process, rather than the developmental system. The Developmental process is a series of interactions with developmental resources that exhibits a suitably stable recurrence in the lineage. Its periodicity is unrelated to that of the resources themselves." (Griffiths & Gray 1994, p.292 emphasis from G & G) The distinction between system and process they make is important. While a persistent developmental resource such as the sun stretches way back in time the periodicity of a species-typical interaction between the sun and other developmental resources does not stretch back so far. It recurs over and over during each life cycle. In this way they cut a horizontal slice through the scroll limiting how far back in time the chain of causal events goes. By doing this they avoid having causes going back to the origins of the universe being a part of their developmental story.





Fig. 2



The second way Griffiths and Gray delimit the causal space is to draw some vertical lines in our metaphorical scroll. They do this by selecting only the developmental resources that are responsible for characteristics that are stably replicated from generation to generation.

Another way to draw this distinction is by distinguishing developmental outcomes that have evolutionary explanations from those which do not. The interactions that produce outcomes with evolutionary explanation are part of the developmental system. There is an evolutionary explanation of the fact that the authors of this paper have a thumb on each hand. We have thumbs because of the replication of thumbed ancestors. The thumb is an evolved trait. But the fact that one of us has a scar on his left hand has no such explanation. The scar is an individual trait (we are referring, of course, to the trait of having a scar just thus and so, no the general ability to scar). The resources that produced the thumbs are part of the developmental system. Some of those which produced the scar, such as the surgeon’s knife, are not. (Griffiths and Gray 1994, p. 286)



This will still leave a very broad selection of concurrent causal events within the life cycle but it does reduce the concurrent causal space to something more manageable. So what we get in our metaphor is something like figure three.





Fig. 3



The causal history deemed relevant here is still too large for many purposes. Contrastive causal explanation allows us to carve up the causal space still more. As I showed above some contrasts will lead us to just one cause. In the case of development this one cause might be a gene. To complete the metaphor of the scroll, contrastive explanation allows us to focus in on a single dot on the scroll. This has great pragmatic use in designing experiments to discover the roles of various individual developmental resources. It shows that DST advocates must back off of rhetoric like the following. "The developmental systems view implies that it is not possible to divide traits of organisms into those with a genetic base, which can be explained by biological evolution, and those which are environmentally acquired and are the domain of cultural evolution." (Griffiths & Gray 1994, p. 302 emphasis added). It just does not follow from the DST view that a statement like this is true and it is unnecessarily divisive rhetoric.



3.4 The Compatibility of DST and Contrastive Explanation

Though one cannot conclude that there is no sense to be made of speaking of "gene's for X" or "normal environments," the above critique does not show that DST is fundamentally misguided in some way. There are three important reasons I can see for this. 1) The possibility of privileging genes in explanation does not show that it is possible to privilege them metaphysically. 2) Contrastive explanation is interest-relative. 3) In many cases it is not heuristically desirable to focus causal explanation so narrowly so as to admit just a few possible causes.

In order to see that privileging genes in explanation does not make it possible to privilege them metaphysically, we must see how causal explanation and causal reality come apart. One way one might do this is just to be instrumentalist about explanation. Bas Van Fraasen (1980) has argued that having an explanation of something does not imply that the explanation is true. For example, we can say that Newtonian physics explains all sorts of things about the motion of objects, even though in light of the theory of relativity Newtonian physics is considered to be false. I don't want to pretend to have anything conclusive to say about the realism/anti-realism debate in philosophy of science. I just want to suggest one way in which causation and causal explanation might come apart. The idea here is that though we have highlighted genes using contrastive explanation, Mill's "real cause" for an event is still the actual cause. All that we have done by making a causal contrast is highlighted a part of the causal history that we are particularly interested in.

Another way to show that contrastive explanation does not privilege a particular cause metaphysically is to see that if we take its explanatory products seriously we are led to apparently contradictory results. Consider the following interesting example from Andy Clark (1998). Phenylketonuria, known as PKU disease, is a horrible disease that causes mental retardation, shortness of stature and lack of pigment. The normal gene at the PKU locus produces a liver enzyme that allows us to metabolize phenylalanine, an amino acid that is common in lots of foods we commonly eat. People with PKU disease cannot produce this important enzyme, phenylalanine hydroxylase. The phenylalanine concentration in the blood gets too high and this interferes with the production of myelin, which serves as a protective sheath around our nerve cells. Without this sheath, the deleterious effects mentioned above occur. Fortunately, these effects are easily avoided. A person with this disease simply has to avoid phenylalanine in their diet. If we are interested in genes the disease will be considered genetic because we will make the following contrast. Why did Mary have the deleterious effects rather than other people from her town? Since everybody in her town eats roughly the same things, this is not the difference that makes a difference in Mary’s having the deleterious effects of PKU. What is different about Mary is that she is homozygous at the PKU locus. For this reason, we attribute the cause of her disease to the PKU gene. However, if we are interested in how to treat this disease the following contrastive question makes more sense. Why do these deleterious effects occur in Mary rather than it being the case that Mary lives a healthy normal life? Whereas the first contrast I gave makes the chemical environment of the gene and the chemicals, such as phenylalanine, coming in from outside the organism irrelevant background information, these things are all causally relevant in the second contrast.

The story above makes two important points. First, it shows how contrastive causal explanation comes apart from the real causes. Under the first contrast I gave, PKU is caused by an abnormal gene. However, under the second contrast I gave it is caused by eating foods that contain phenylalanine. If we were to take each of these explanations as giving the real cause of the disease then we would have a contradiction on our hands.

In fact, we would have something that looks suspiciously like the nature/nurture debate. I would like to suggest that in many nature/nurture debates, the conflict arises from differing contrastive questions rather than any real empirical conflict. While disagreement often spurs scientific progress, conflicts that arise in the way outlined above are counter-productive and will inhibit scientific progress.

Perhaps I have gone a little too fast here. It looks as though I am using the explanations for the two contrasts above to answer a third contrast that has broader scope and might contain both answers as the cause. One might argue that all causal why questions have an implicit contrast built in. If this is true it looks like I have pulled a fast one. It appears I have a question with a very broad scope in mind such as: why do some people get PKU disease rather than others? The answers to the two contrasts above are only contradictory if they are construed as answers to this very broad question. This is right so far as it goes, but I take it that we are interested in the broad question and it is this sort of question that many proponents of nativism ask about particular phenotypic outcomes. When the explanatory scope of a why question is this broad the entire developmental system becomes relevant.

The second point the PKU story makes clear is that contrastive explanation is interest-relative. Remember, in my explication of contrastive explanation above I commented that the foil that is chosen determines the piece of causal history that will be deemed relevant to explanation. In the above story the foil in the first contrast was the population in Mary’s community and the foil in the second contrast was a counter-factual case in which Mary had a normal healthy life. Answers to each question are both interesting and contribute to our understanding of this horrible disease. The point is that explanations generated by contrasts on a given topic differ with our varying interests. In the PKU example above one contrast highlighted the gene in explanation and the other highlighted a chemical from the environment. Contrastive explanation makes it possible to privilege the gene in explanation, but given that genes and phenylalanine are both part of the real cause of PKU disease we are left with no motivation to privilege either cause, other than what our interests dictate as relevant. One might be able to argue that one interest should be privileged over another, but this is a can of worms well beyond the scope of the present discussion. So far as I can tell this argument applies whenever parts of the causal history are relegated to the status of background information while other parts are highlighted in explanation.

The relationship between causal contrastive explanation and DST is not antagonistic. Contrastive explanation works well as a principled heuristic to narrow down the relevant causal space in the interests of designing manageable experiments. It is useful for example to highlight just phenylalanine as a cause of PKU disease when our interest is in immediately stopping the deleterious effects from occurring. The patient can use a special diet which is low in phenylalanine. Highlighting just a gene correlated with a particular trait also has been very fruitful in scientific research. Contrastive explanation is perfectly available to the Developmental systems theorist. "In fact, pragmatic, local simplifications for the purpose of actually doing experiments are equally open to the developmental systems theorist. Developmental systems theorists can study the evolution of particular elements of the life cycle while assuming many other elements play a more or less constant role in reconstructing the life-cycle over relevant stretches of evolutionary time (although sometimes this will prove misleading)." (Griffiths & Gray 1997, p.488) Contrastive explanation complements DST well in providing a principled mechanism for such "local simplifications." It is to the parenthetical comment in the above quote that I now turn.



4 The Relative Values of Explanatory Strategies

Since it is possible to privilege parts of the causal history in explanation for the purposes of designing manageable experiments, it is important to look at what advantages come from taking account of larger portions of the causal history in some explanations. In the quotation above Griffiths and Gray suggest that this sort of causal focus can be misleading. The obvious way in which having too narrow a causal focus might prove misleading is to over-simplify or to prematurely decide that investigation into a phenomenon is complete. The most important advantage to maintaining the broader causal focus of DST is that it captures truly dynamic and interactive processes that simply cannot be described if the causal focus is too narrow.

I will briefly explain what I mean by truly dynamic and interactive processes by using a set of explanatory distinctions made by Andy Clark (1997). Clark uses this set of distinctions to aid explanation of fully developed cognitive systems, but the basic notions can be extended to apply to the development of a trait. Clark describes three types of explanation; componential explanation, "catch and toss" explanation, and emergent explanation. Componential explanation is the most familiar. We explain the functioning of a complex whole by focusing on the individual roles and the overall organization of its parts. Contrastive explanation works nicely to isolate the causal role of individual developmental resources such as genes. In a sense, such narrow causal focus is to engage in a part of the componential explanatory project.

Clark's second type of explanation is "catch and toss explanation." Clark has the "catch and toss" occurring between brain and the environment in cognitive explanations. "The world tosses inputs to the brain, which catches them and tosses actions back. The actions may alter or simplify subsequent computations, by causing the world to toss back more easily usable inputs and so on." (Clark 1997, p. 106). This sort of explanation is still a dichotomous approach. If we apply this idea to development, we get something like a division between developmental resources internal to the organism and those from the environment. What you get is a sort of feedback loop between the environment and internal resources. It is important to see just how dynamic "catch and toss" explanation can get. "As the complexity and the dimensionality of crucial interactions increase, it becomes difficult (perhaps impossible) to conceptualize the situation by simply superimposing a notion of feedback loops on top of our standard understanding. Such critical complexity arises when the number of feedback processes increase and when the temporal staging of the various processes goes 'out of synch,' allowing feedback to occur along multiple channels and on multiple, asynchronous time scales." (Clark 1997, p.106). Once this sort of complexity is reached, a broader causal focus provides a better explanation of the phenomena because it is able to capture the interplay between resources. This idea will become clearer with some examples below, but first it is important to discuss emergent explanation.

Emergent explanation occurs when it is too difficult to focus on any one cause.

Clark tells us that we can understanding the notion of emergence as a sort of generalized version of an uncontrolled variable. Controlled variables "track behaviours or properties that can be simply and directly manipulated," while uncontrolled variables "track behaviours or properties that arise from the interaction of multiple parameters and hence tend to resist direct and simple manipulation." (Clark 1997, p. 110). By way of example Clark sites Hofstadter's story about a computer operating system (e.g. Windows) that starts having problems when more than 35 users are on line. Hofstadter notes that it would be silly to ask the programmer to up the "user capacity variable" to get around this problem. Instead, "that number 35 emerges dynamically from a host of strategic decisions made by the designers of the operating system and the computer's hardware and so on. It is not available for twiddling." (Hofstadter 1985, p. 642) Fixing this problem may require changing a number of parameters within the software and perhaps the hardware as well. This problem can only be understood in a complex emergent way. "A phenomenon is emergent if it is best understood by attention to the changing values of a collective variable." (Clark, 1997 p. 112) A collective variable is one that tracks interactions between multiple elements of a system such as a developmental system. An uncontrolled variable as discussed above is a collective variable. Another example of this is bird flocking behaviour. Looking for the cause of their behaviour will lead you to hypothesize that there is a captain bird who gives instructions to the rest of the flock. However, what actually happens is that each bird follows a few simple rules based on its relation to the birds nearest to it. The flocking pattern emerges from a mass of these local interactions. The collective variable tracks the parameters the birds are using in their local interactions.

Now that we have these three types of explanation in hand, it is my contention that the broad causal focus of developmental systems is necessary for the more complicated cases of "catch and toss" explanation and even more so for emergent explanation. As I noted in the bird flocking case, a narrow causal focus would lead to a false hypothesis. Also in the Hofstadter computer case, a solution to the problem requires taking into account a mass of causal factors.

One more example will be helpful here. Consider Thelen and Smith's (1994) research on infant development. A new born infant when held up off the ground performs well-coordinated stepping motions; at around 2 months these motions are lost; the motions reappear between 8 and 10 months as the infant begins to be able to support its weight on its feet; at 12 months, infants can independently walk. Zelazo (1984) has proposed that gradual maturation for reflex-like processes by a higher cognitive centre account for this data. However Thelen and Smith's research suggests that the development of walking ability doesn't proceed according to a linear master plan in this way. Although stepping disappears around two months, nearly kinematically identical motions are still produced by the infant when it is lying on its back. The crucial parameter underlying the disappearance of stepping is leg mass. At two months the mass of the legs is too much for the infants muscles to produce the stepping motion in an upright position. This is supported by a number of experiments. Stepping disappears from infants who normally exhibit stepping behavior when weights are added to their legs. It also reappears in 3 month old infants who are held upright in water, effectively reducing their leg mass. Environmental manipulations work well at 8 to 10 months as well. Non-stepping infants when placed on a treadmill performed coordinated stepping. They were actually able to adjust the rate of their stepping to the differing speeds on the treadmill and were able to adjust to asymmetric constraints when placed on a treadmill with two separate belts going at different speeds. Treadmill stepping was found in infants ranging from 1 month to seven months. These results suggest an important role for mechanical patterning caused by the backward stretching of the infant’s legs. This part of stepping is independent of the normal transitions given above which reflect all kinds of factors such as leg mass. The developmental pattern is not attributable to a simple linear causal account. There are all kinds of causal factors involved in a complete account of walking development. These causes are highly interactive and require reference to one another in explanation. Having too narrow a causal focus would make it impossible to come to this rich understanding. It is this sort of dynamic interactive development that the DST perspective is best able to capture and which an explanatory strategy featuring a more narrow causal focus would fail at explaining.





5. Part One Conclusion

What we have seen above is that it is very difficult to maintain an account of innateness where something is innate if it is “in the genes.” I have addressed two ways this account of innateness might go. First, genes play a metaphysically special role in development. This view is rejected in view of: 1) the fact of epigenetic inheritance and 2) lack of a good account of information such that genes alone could be said to carry information about a developmental outcome. The second “in the genes” account of innateness admits the causal symmetry of the developmental resources, but privileges the role of genes in explanation. This view falls prey to the symmetry argument when it is viewed from a “realist” perspective. That is to say when the explanation highlighted by a causal contrast is taken to be “the cause” of a developmental outcome it is subject to the symmetry argument. When taken as an explanatory heuristic, the contrastive view is quite powerful and rather than being contrary to the DST outlook, it compliments DST quite well. Finally, I characterized DST as creating broader contrast spaces and suggested that this has important heuristic value that should not be ignored in favor of exclusively creating more narrow contrast spaces. Within the framework of these broader contrast spaces “in the genes” accounts of innateness make no sense.



Part II: Invariance Accounts of Innateness



We have seen how difficult it is to develop an “in the genes” account of innateness. A popular way to try to avoid this problem is to back away from development and focus on the outcome of that development. A number of accounts of innateness in recent literature are based on rigid or invariant outcomes of development. These are outcomes in development that are buffered against environmental perturbations. Traits that invariantly develop like this are considered to be innate. Later in the discussion I will be looking at two of these accounts from Sober (1999) and Ariew (1999). I will argue that these accounts of innateness fail for two important reasons. 1) These accounts fail to distinguish between two different types of explanation; evolutionary adaptive explanation and developmental explanation. 2) Invariance accounts encourage us to make the fallacious inference that an invariant or rigid outcome entails an adaptive evolutionary account of a trait and vice versa.



1. Adaptive Evolutionary Explanation and Developmental Explanation



Suppose I am telling somebody about language. This person, I will call him Ed, says, “Oh, you are studying language acquisition, how do children learn language so quickly and easily?” I answer him by saying “It’s human nature to learn language quickly and easily.” How would we expect him to respond? He might say, “don’t patronize me, this much is obvious, I wanted to know how they do it, tell me the steps they take along the way to learning the language.” It’s not that what I said was false that was the problem for Ed. The problem was that I was not answering the question Ed was asking. How can we make the distinction between Ed’s question and the question I answered. We ask what contrast each is making. I wish to distinguish, as Ariew does (Ariew 1999), between two explanatory strategies. They are adaptive evolutionary explanation and developmental explanation. At first blush, the answer I gave to Ed is an adaptive evolutionary explanation while the question he was asking was a developmental question. Below I will compare and contrast these two types of explanation and follow by illustrating how these strategies relate to innateness by way of an example.

I start with adaptive evolutionary explanation. An adaptation is a trait that was selected for because it contributes to the survivability of a species in a given habitat. For example, a Giraffe’s long neck allows it to get at a food source high up on a tree. The Giraffe’s neck contributes to its survivability in a habitat with tall trees. This trait is considered adaptive by way of Darwin’s theory of natural selection. Adaptive explanation works at the level of population. It explains the relationship between a trait and the habitat it is adapted too. This relationship can be captured statistically by using the behavioral geneticist notion of heritability. Heritability in a population is described in the following way. “The distribution of the trait in an offspring population is predictable from the distribution of the trait in the parent population and the mating pattern of that population.” (Horvath 2000). The question I answered for Fred was something like why do humans rather than other species acquire language. The contrast space created by heritability questions can also be narrower. For example, it can single out a population segment from within a particular species. It is important to realize that this will always be an explanation at the level of a population.

Developmental explanations work at the level of the individual. A developmental explanation gives us a causal history from the genotype to the phenotype. Alternatively, developmental explanation gives us a causal history from an initial state of not having some trait, knowledge, or capacity to having it. This sort of explanation focuses on a process or a mechanism rather than the static outcome that is the focus of adaptive evolutionary explanation. Adaptive evolutionary explanation black boxes developmental mechanisms. The reason for this is that natural selection does not care, as it were, how you get to an adaptive outcome just so long as you get the right outcome. Developmental explanations are not so concerned with outcomes measurable across a population.

Recall the description of PKU disease above.⁴ In a simplified account of this disease there are two important causal elements. The gene that affects the ability to metabolize phenylalanine and the phenylalanine the patients eat in their diet. If we give an adaptive evolutionary explanation of this disease it will be considered heritable. The contrast question that someone interested in an evolutionary account might use is why did Mary have the deleterious effects rather than other people from her town. Since everybody in her town eats roughly the same things, this is not the difference that makes a difference in Mary’s having the deleterious effects of PKU. What is different about Mary is that she is homozygous at the PKU locus. For this reason, adaptive explanation attributes the cause of her disease to the PKU gene.

What would a developmental explanation for this disease look like? The developmentalist is not making a population level contrast. She is asking something like, why and how do these deleterious effects occur in Mary rather than Mary living a healthy normal life? Whereas the adaptive evolutionary explanation can take the chemical environment of the gene and the chemicals such as phenylalanine coming in from outside the organism as a backdrop, these things are all causally relevant to the developmentalist. They are a part of the developmental system. While it may be obvious that both the gene and phenylalanine are causally relevant, it is not obvious which might be the major contributor and thus the most causally relevant. Suppose we ask how the PKU gene produces PKU symptoms. We will want a detailed account of the processes and mechanisms involved. We see that adaptive evolutionary explanation works at a population level while developmental explanation is at the level of the individual. The developmental account will give us the chemical mechanisms that cause the symptoms of PKU disease, while the adaptive evolutionary account compares the genetic endowment of the population that has PKU disease with the population that does not.

Many scientists in the literature appear to be making developmental claims rather than adaptive claims. However, the distinction between these two types of explanation is lost on some researchers. Here I take a very brief look at a famous innateness claim in the literature and show how it is primarily a developmental claim. Noam Chomsky claims that our knowledge of universal grammar is innate. Chomsky argues that a number of observations lead us to this conclusion. We are capable of producing an infinite variety of sentences, but our lifetime is comparably quite short. The information we get while learning language is very limited. Differences in intelligence don’t seem to effect language acquisition one way or the other. “Such observations lead one from the start, that we are dealing with a species-specific capacity with a largely innate component.” (Chomsky 1975, p.123) Chomsky speaks of a species-specific capacity. This is an observation at the level of population and certainly not at the level of the individual. This seems very much like an evolutionary explanation and indeed Chomsky makes evolutionary claims about language as well as developmental ones. Here we see Chomsky making a developmental claim. “The competence of an adult, or even a young child is such that we must attribute to him a knowledge of language that extends far beyond anything he has learned.” (Chomsky 1975, p. 123) Chomsky is saying something about how people come by or acquire their linguistic competence. Linguistic competence is the outcome of a developmental process and Chomsky claims that this outcome requires more knowledge than we could learn in the normal course of development. “The central problem in designing a language acquisition device is to show how such a system of rules can emerge given the data to which the child is exposed.” (Chomsky 1975, p. 124) Again we see Chomsky making claims about how an outcome “emerges.” What seems confusing in Chomsky’s work and that of others in cognitive science is they do not make it clear whether their arguments that invoke talk about the distribution of outcomes are also arguments about the development of those outcomes.

Chomsky is generally not very clear about what he means when he says a trait is innate. Let’s suppose that he means something like a trait with an invariant outcome. That is to say that under a broad variety of environmental conditions language acquisition is reliable and robust. If this is correct, then Chomsky seems to be making some fallacious inferences. I am not claiming this is Chomsky’s view; rather I am just supposing he has this view to show how it is damaging to an invariance view of innateness. This is the second problem with invariance accounts I alluded too. If he is using the fact that language acquisition is specific to humans as evidence that the outcome of language development is invariant or rigid, then he is mistaken. Some evolved traits develop invariantly and some require a rich and highly specific developmental environment. For example, in the case of Rhesus Macaques, the recognition of emotional expressions in conspecifics and the ability to cooperate in agonistic interaction depend on infant social interaction for their development (Mason, 1985). Perhaps Chomsky could use these evolutionary facts about language as a defeasible guide to its innateness. In this case he would likely be making the inference that traits that develop invariantly generally have an adaptive evolutionary story as well. This is the inverse of the inference I have just discussed. However, this inference is fallacious as well. Many traits we acquire invariantly do not have adaptive evolutionary stories. Richard Samuels (2002) has pointed out that we acquire the belief that water is wet rather robustly and invariantly, but there certainly isn’t a gene for this belief.



2. Andre Ariew’s Canalization Account

Andre Ariew (1999) gives potentially the best invariance account of innateness. Ariew recognizes the different explanatory projects above and it is largely his work on this that has inspired my point of view. Despite recognizing these two explanatory strategies, Ariew believes he can somehow heroically save an account of innateness that works in either explanatory strategy. There are three related problems with Ariew’s canalization account of innateness. After summarizing his account I will show how each of these problems comes out of this distinction between explanatory strategies. Furthermore, we will see where Ariew is guilty of making the fallacious inference discussed above.

Ariew revisits the early debates over innateness between Konrad Lorenz and Lehrman. Lorenz characterized the issue as a dichotomy between traits that are genetically determined and those that are acquired or learned. Lehrman (1953) argued that no trait could develop on the basis of genes alone. Every trait requires some interaction between genes and environmental factors. Ariew diagnoses their disagreement as being caused by their different explanatory goals. Lorenz has adaptive explanation in mind and Lehrman has developmental explanation in mind. Ariew believes that Lehrman has made rather damning criticisms of Lorenz’s innateness conception but believes that there are some innocent features of it that are desirable. Ariew in effect is trying to create a middle ground view between the radical view which rejects the distinction between innate and acquired altogether (the DST view) and Lorenz's view that innate traits are genetically determined traits. Ariew derives three desiderata from his review of this debate.



1. “An account of innateness should make it a feature of development.” (Ariew 1999, p.14)

Ariew clearly wants his account to be relevant to a developmental explanation. So far so good, as we saw above, cognitive scientists often seem to be making developmental innateness claims.



2. “Innateness should denote an environmentally stable trait.” (Ariew 1999, p.15)

This also seems quite reasonable. As Ariew points out there are a number of deprivation experiments in ethology in which various species seem to develop traits in impoverished learning environments. Lorenz and others believe these experiments can provide evidence for innateness.



3. “An account of innateness should make clear how natural selection can effect prevalence of some adaptive traits.” (Ariew 1999, p. 15)

It is very unclear why this should be a desideratum given that innateness is supposed to be a developmental concept. It suggests the fallacy I mentioned above. Namely that you could infer from the fact that a trait is “environmentally stable” that there will be an evolutionary account to give on the trait.



Based on these desiderata, Ariew argues that C. H. Waddington’s concept of canalization is innateness. “Canalization denotes a process whereby the endstate (the product of development) is manifested despite environmental perturbations.” (Ariew 1999, p.16) Development is considered an “epigenetic landscape” in which various developmental pathways branch off. Each of these pathways leads to a particular end state. Certain environmental features can induce an organism to follow one of these pathways. Each of these pathways can be more or less canalized. So Ariew equates innateness with canalization. “The degree to which a biological trait is innate for individuals possessing an instance of a genotype (or set of genotypes) is the degree to which the developmental pathway for individuals possessing an instance of that genotype (or set of genotypes) is environmentally canalized. The degree to which a developmental pathway is canalized is he degree to which development of a particular endstate (phenotype) is insensitive to a range of environmental conditions under which the endstate emerges.” (Ariew 1999, pp. 17-18)

It is surprising that Ariew makes the distinction between developmental explanation and adaptive evolutionary explanation and still maintains his third desideratum. The fact that a trait is widespread tells us nothing about the development of the trait. To argue otherwise is to reason from developmental invariance to an adaptive evolutionary account. Trying to combine the two sorts of explanation leads to problems. First, Ariew ends up talking about end states or outcomes again. This is the province of evolutionary explanation. From a developmental perspective such an account is little more than a promissory note to future studies in developmental biology and thus is of little scientific use. Recall the discussion with Ed above. We can see why he was so frustrated with the answer that “it is human nature.” Again the developmental psychologist or biologist is not interested in the fact that a trait is canalized, she is interested in the precise mechanism of that canalization. It appears that, though canalization pretends to be about development, in fact it is not. Ariew thus sacrifices his first desiderata for his third.

How does canalization describe our PKU case? It seems under one construal; PKU isn’t canalized much at all. It is totally contingent upon whether the person includes phenylalanine in their diet. Thus, PKU is highly sensitive to the environment and is not highly canalized. However, PKU is very heritable. It is strange that a trait that is highly heritable is not innate. On another construal, PKU might be thought of as moderately canalized. Given the way I have characterized the disease, the genes and the environment both play relevant causal roles. Nowhere in this characterization do we get anything like my explanatory gloss of the mechanism of PKU. Canalization seems to be just anot