INNATENESS AND CANALIZATION
Andre Ariew  
University of Rhode Island

ABSTRACT
Cognitive scientists often employ the notion of innateness without defining it.  The issue is, how is 
innateness defined in biology?  Some critics contend that innateness is not a legitimate concept in 
biology.  In this paper I will argue that it is.  However, neither the concept of high heritability nor 
the concept of flat norm of reaction (two popular accounts in the biology literature) define 
innateness.  An adequate account is found in developmental biology.  I propose that innateness is 
best defined in terms of C. H. Waddington's concept of canalization.



1. Introduction
It is commonplace in the cognitive science literature to ascribe the term "innate" to various 
behavioral capacities.  For two stand-out examples, Chomsky (1988) is famous for contending 
that Universal Language is innate while Murray and Hernnstein (1994) are infamous for boldly 
declaring that IQ is innate.  Critics contend that innateness is not well defined in biology and so 
should be dropped from cognitive science (Johnston (1988), Oyama (1985), Oyama (1988), Gray 
(1992), Lehrman (1953)).  In this paper I'll argue that innateness makes good biological sense.

Where does one turn to acquire at least the rough characterizations to guide our intuitions?  
Commonly, people associate innateness with a process that is "in the genes" or "present at birth".  
But such associations, when taken out of scientific context, are at best hopelessly vague and at 
worst reminiscent of 18th century preformationism (Sober forthcominga).  I think a better place to 
start is with the ethology literature, especially with the work of Konrad Lorenz.  Although 
Lorenz's account of innateness is flawed, his work provides a set of rough characterizations or 
diagnostic features that can be used as benchmarks for alternative proposals.  In this paper I'll 
follow a brief exposition of Lorenz with critiques of two proposals derived from concepts found in 
population geneticsÑheritability and norms of reaction.  I'll show that neither concept provides an 
adequate definition of innateness.  Finally, I'll offer my own proposal that stems from 
developmental biology.  The rough idea is that the degree to which a trait is innate depends on the 
degree to which it is canalized in development.

2. From Ethology
Lorenz (1957) sought to provide natural selection explanations for the origins of certain adaptive 
behavioral capacities, called "instincts", commonly found in non-human animal populations.  For 
instance, Lorenz observed that female mallards raised to reproductive age in exclusive company of 
pintail ducks show no sexual affinity for the pintail drakes.  But upon seeing a male mallard for the 
first time, the female immediately engages in the sexual courtship behavior particular to its species.  
Remarkably, a mallard expresses courtship behavior even if it had no opportunity to learn it.  That 
is, mallards which are naturally or experimentally deprived (via "isolation-rearing experiments") of 
the opportunity to acquire courtship behavior through experience tend to develop it nonetheless.  If 
courtship behavior is not acquired from environmental cues, where does it come from?  Here, 
Lorenz asserts the basic dichotomy from which he defined innateness: what is not learned is innate.  
Accordingly, the mallard's courtship behavior is innate.

But, what does innate mean besides not learned?  Influenced by the theory of natural selection, 
Lorenz reasoned that certain seemingly adaptive species-specific behavior that develop in isolation 
from environmental cues are products of natural selection.   As products of natural selection, such 
behavior is genetically transmitted from parents to their offspring.  Herein lies Lorenz's proposal: 
an innate trait is one that is genetically transmitted as opposed to acquired by cultural transmission 
or individual learning.   To experimentally test for innateness, Lorenz promoted the use of 
isolation-rearing experiments.  If an organism undergoing isolation develops the trait "normally", 
the trait is said to be innate.  

Critics of Lorenz's account assert that the gene/environment dichotomy upon which his notion of 
innateness is defined is false (Lehrman, 1953).  No biological trait  or behavioral capacity develops 
independently of environmental factors.  Development involves complex interactions among genes 
and between genes and environments.  Even Lorenz's deprived organisms developed in some 
(minimal) environment; there is no such thing as a trait that develops independently from an 
environment.  Hence, critics contend, Lorenz's explanation for the origin of the mallard's behavior 
is insufficient; the behavior did not develop in the mallard by genes alone.

According to some critics, it follows that innateness is not well defined in biology and the term 
should be eliminated from the biological lexicon.  However, this eliminitivist position does not 
follow.  Rather, what follows is that innateness is not grounded in a rigid gene/environment 
dichotomy.  Evidence that "innateness" must refer to something lies in what remains fairly 
uncontroversial about Lorenz's research program.  From Lorenz's work one can highlight a 
number of rough characteristics or diagnostic features that may serve as guidelines for a 
reformulated definition of innateness.  The following are a sample:

(a) On ontogeny: One important aim of ethology is to explain how individual organisms come to 
develop the traits they do.  For Lorenz, innate ascriptions contribute to the determination of the 
significant factors that enters into the development of some traits.

(b) Innateness as an environmentally stable trait: Innateness seems to have something to do with 
what environment doesn't do to influence development in an individual.  Evidence from isolation-
rearing experiments and observations in the wild suggests that some traits develop normally in a 
range of environments, including impoverished and non-normal ones.  In such cases, the 
environment doesn't prevent the trait from being manifested.  Ethologists since Lorenz sometimes 
associate innateness with environmentally stable, as opposed to labile, traits (Hinde, 1982, 86).

(c) Innateness as a product of natural selection:  Recall that Lorenz was interested in applying 
natural selection to explain the prevalence of certain highly adaptive species-specific traits.  This 
project seems promising even though it is an open question whether all such traits are explainable 
via natural selection.  Ever since Lorenz, ethologists interested in the source of a trait's adaptedness 
sometimes call products of natural selection "innate" in contrast with traits that owe their prevalence 
to cultural transmission or individual learning.  

I'll now consider two proposals to ground innateness in concepts drawn from population genetics.  
What makes these proposals worth considering is that they allow us to talk about genetic factors of 
a trait in a way that isn't falsified by the truism that phenotypes always require an interaction 
between genes and environment.  In taking the rough characterizations presented by ethologists as 
a benchmark, I'll demonstrate substantive flaws in both proposals.

3. High Heritability
It follows from the truism about genes and environments that to ask whether genes alone caused a 
particular phenotype to develop is nonsense.  However, we may ask a related question: How much 
of the phenotypic differences in a population of organisms is explainable by genetic differences and 
how much is explainable by environmental differences?  This is the question addressed by a 
heritability study in quantitative genetics.  Accordingly, if the phenotypic differences in a 
population are disproportionately due to genetic differences among members of a population, the 
trait is heritable to that degree.

Heritability in the broad sense is defined as the proportion of phenotypic variance that is due to 
genetic variance.  (Variance is a statistical termÑs2 = the average squared deviation of the 
observations from the mean.)  In a simplified account, phenotypic variance (Vp) decomposes into 
two components, genetic variance (Vg), and environmental variance (Ve) such that heritability 
(H2) is the proportion of the total phenotypic variance of a quantitative trait in a population that is 
due to genetic variation: H2= Vg/Vp.

It might seem as though a plausible account of innateness would be in terms of high heritabilty: a 
trait is innate if and only if it is highly heritable (or, the degree to which a trait is innate is the 
degree to which it is heritable).  This proposal has the virtue of showing how heritability can be 
measured in natural populations, though one must pay attention to the numerous complications in 
trying to establish such measures (for a clear and detailed description of such, see Sober 
(forthcomingb), Block (1995), or Griffiths et. al. (1993)).  A typical test for heritability for some 
quantative character in human populations, e.g. height, is to rely on studies of identical twins 
separated at birth and raised in different environments.   Identical or monozygote twins have 
identical genes, so Vg (mono-twin) = 0.  Thus, if twins are raised apart, phenotypic differences 
between twins are explained solely by environmental differences: Vp (mono-twin) = Ve (mono-
twin).  Notice that this is true only if we grant several key assumptions about gene and gene-
environment interactions (see Sober (forthcomingb)).  Suppose that twins reared apart live in 
environments as varied as any two randomly selected individuals from the general population, that 
is, Ve (mono-twin) = Ve.  Then, Vp = Vg +Ve (mono-twin).  After some combining, we get the 
result that the genetic variance of the population as a whole can be measured by observations of the 
phenotypic variance between members of the general population and phenotypic variance between 
the twins: Vg = Vp - Vp (mono-twin).  Heritability is large if twins are more similar to each other 
with respect to the trait in question than are two randomly-picked individuals from the general 
population: H2 = (Vp - Vp (mono-twin))/Vp. 

Unfortunately, the proposal to define innateness in terms of high heritability suffers from 
substantive problems.  Taking the diagnostic features of innateness provided by ethology as our 
guide, we find that high heritability is neither a necessary nor sufficient condition for innateness.

I'll start with sufficiency.  Recall that ethologists believe that a distinguishing feature of an innate 
trait is its environmental stability.  Innate traits tend to be expressed in a wide range of 
environments.  If the definition in terms of high heritability is to capture these features of 
innateness, highly heritable traits should turn out to be environmentally stable.  But this need not 
be the case.  To see why, consider the hypothetical results of an adoption study on the heritability 
of IQ discussed in Griffiths et al. (1993):

	Biological parents			Children			Adoptive parents
		90				110				118
		92				112				114
		94				114				110
		96				116				120
		98				118				112
		100				120				116
Mean		95				115				115

Here IQ scores between children and their biological parents are perfectly correlated; reading from 
top to bottom there is a 2 point step-up for the IQ scores of both children and biological parents.  
Further, the scores are not at all correlated between children and adoptive parents.  Perfect 
correlation between children and biological parents and non-correlation between children and 
adoptive parents entails that H2 = 1; all of the IQ variation among children is due to variation 
among the biological parents.   But: (i) the mean IQ score of children is identical to that of their 
adoptive parents, and (ii) the mean in both is twenty points higher than that of biological parents.  
This suggests that environmental conditions (manifested in the IQ scores of the adoptive parents) 
has some effect on a child's IQ score.  So while IQ is highly heritable, it is nonetheless 
environmentally plastic.  This example shows that high heritability is not a sufficient condition for 
innateness.

High heritability is also unnecessary for innateness.  Consider what might be considered a 
paradigm human adaptation, the possession of opposable thumbs.  The possession of opposable 
thumbs is so successful an adaptation that all "normal" humans possess human thumbs; it has 
nearly gone to fixation in human populations.  Further, human thumbs are environmentally stable 
traits; like the structure of many limbs and digits, opposable thumbs develop in a wide range of 
viable environments.  But, it turns out, depending on the population under analysis in the 
heritability estimate, the possession of thumbs is not necessarily highly heritable.  To see why, 
recall that high heritability is defined as a ratio that reflects the total amount of phenotypic variation 
that is due to genetic variation (Vg/Vp).  But, for human populations in which the possession of 
opposable thumbs is 100%, there is no variation.  Hence for most human populations, heritability 
is undefined as the denominator in the H2 ratio is 0.  Now, suppose we enlarge our population to 
include one or two individual humans lacking opposable thumbs.  Now there is phenotypic 
variation in the population, so heritability is at least definable.  But consider a case in which the 
individuals lacking opposable thumbs do so because their mothers took particular drugs during 
pregnancy that disrupted fetal development.  Because the variation in the population is due to 
environmental differences, heritability will be very low; hence on the proposal that innateness is 
high heritability, the possession of opposable thumbs is in this case (counter-intuitively) not innate.

To diagnose the problem with the heritability proposal further, innate ascriptions are supposed to 
tell us something about the development of a trait in the individual (see the ontogeny condition 
above).  However, heritability measurements tells us no such thing.  Heritability is a measure of 
the variation of traits in a population; it does not explain why individual members of a population 
have the traits they do  (Sober forthcomingb).  For example, if one were to determine that height in 
humans is 60% heritable, it follows that 60% of the differences in height one sees among humans 
can be associated with genetic differences among them.  It does not follow that 60% of an 
individual's height (say, from mid-thigh up) is due to genes, the rest to environment.  The 
heritability estimate provides no information about how genes and environment interact to express 
height in an individual.

4. Flat Norms of Reaction
Is there a related concept in genetics that may fare better than the high heritability account?  Here 
we'll consider the merits of defining innateness as flat norm of reaction for a given genotype.  A 
norm of reaction for a genotype is a graph of the pattern of phenotypes produced by a given 
genotype under a range of environmental conditions (Griffith et. al., 794).  The figure depicts a 
hypothetical norm of reaction for body size of two varieties (genotypes) of some hypothetical 
organism across a range of temperature in which the varieties are raised.

FIGURE 

The graph shows that B organisms are less sensitive to environmental changes than As are.  B 
organisms express a consistent body size regardless of the temperature of their developmental 
environment.  This fact is represented by the flat norm of reaction line.  In general, an inflexible 
norm of reaction for a particular genotype indicates that the genotype produces the same phenotype 
across a range of (viable) environments.  We thus have the following proposal: a trait is innate for 
a genotype within a particular environmental range if and only if its norm of reaction is flat across 
the range of environments specified.

Two points are worth mentioning here about the proposed definition.  First, notice that innateness 
on this proposal is relative to particular genotypes.  To illustrate, suppose some genotypes for blue 
eyes in humans exhibit flat norms of reaction across some environmental range.  It does not follow 
that all blue-eye genotypes produce flat norms of reaction lines; it may be that some eye-color 
genotypes are more sensitive to environmental variation thereby expressing blue-eyes only in some 
environments.  Second, norms of reaction are relativized to a specified range of environments.  
Consequently, it would be wrong to extrapolate from the figure that variety B genotypes are innate 
(i.e. produce flat norms of reaction) across, say, all "normal" environments for that variety.  It may 
be that air quality, elevation, or any number of other environmental conditions would produce non-
flat norms of reaction curves for the same genotypes.

The flat norm of reaction account of innateness has merit.  Significantly, it captures the ethologist's 
intuition that innate traits are ones that are environmentally stable rather than labile.  Unfortunately, 
problems plague the account.  First, a flat norm of reaction such as the one shown in figure depicts 
the pattern of adult phenotypes produced by a given genotype under a range of environmental 
conditions that are fixed throughout the course of development.  What's missing is the possible 
effect a fluctuating environment during the course of development has on the phenotype in 
question.  That is, what is needed is to plot individual life histories across a variety of 
environmental conditions, not norms of reactions.  For example, from the figure we have no idea 
how fluctuating temperature throughout development affects body size for each genotype.  In 
short, a flat norm of reaction can mislead one to conclude wrongly that a trait is stable for a 
genotype when in fact it isn't.  Therefore, flat norms of reactions are insufficient to determine 
innateness.

The second problem is that flat norms of reaction are properties of populations, whereas innateness 
is supposed to be a matter of an individual's ontogeny.  A flat reaction norm is a measure of the 
average phenotypic scores for all individuals possessing an instance of the genotype.  (More 
precisely, the norm of reaction is a best fitting regression lineÑthe least square line for all 
individual scores measured.)  Phenotypic scores of individual members of the population are 
represented as dots on the graph, as shown in figure.  No individual member is represented by the 
regression line.  Notice the scatter of dots around the regression lines.  The difference between the 
reaction norm and the individual phenotypic score is what geneticists call "developmental noise" 
which indicates that there are differences in how each individual develops.  The point here is that 
the reaction line does not represent the development of any particular member of the population; the 
dots do.  It would be a category mistake to say that innateness qua property of an individual is 
defined as a flat reaction norm qua property of a population.  Therefore, flat norms of reaction 
don't define innateness.

Nevertheless, under certain conditions, a flat norm of reaction may be a good indicator or predictor 
of innateness in an individual just as a regression line may be a good predictor of the phenotypic 
score of individuals possessing an instance of the genotype.  If a trait is stable in the way that we 
requireÑagainst fluctuations in both the initial environmental conditions and throughout the course 
of developmentÑthen it will produce the same phenotype no matter how the environment varies.  
That is to say that the trait will produce a flat norm of reaction (ideally with little or no scatter 
around the regression mean) for the population of individuals possessing the trait.  So only for 
some traits is a flat norm of reaction a good indicator for innateness.

Out of the various proposals and criticisms above we have the makings of a new proposal to define 
innateness: to say that a biological character is innate is to say that an individual's development 
tends to express the biological item as its end-state in a wide range of initial environmental 
conditions and persists regardless of environmental fluctuations during the course of development.  
When such conditions hold, developmental biologists call this canalization.  Let's see how the term 
is used in developmental biology.

5. Canalization
The developmental biologist C. H. Waddington was struck by the fact that developing organisms 
tend to produce a number of distinct and well-defined body-types despite environmental variation.  
To explain the phenomenon, Waddington envisioned development as a branching out of various 
developmental pathways each leading to the production of a distinct end state.  Once development 
starts in the egg, a combination of genetic and environmental factors force the developing mass 
down one or another pathway.  For the development of some traits, once a pathway is chosen it is 
entrenched or bound to produce a particular end state.  It is this entrenchment that Waddington 
called canalization.   Waddington defined canalization as "the capacity to produce a particular 
definite end-result in spite of a certain variability both in the initial situation from which 
development starts and in the conditions met with during its course" (Waddington 1975, 99).  The 
developmental pathways that lead to the development of organs and tissues are canalized such that 
only the severest of environmental fluctuations can force the development of these tissues out of 
their normal path.

For our purposes it can be said that Waddington provided an account of the development of 
environmentally stable traits that occur in individuals (satisfying both the ontogeny and the stability 
conditions above).  This makes Waddington's idea a good candidate for an account of innateness: 
the degree to which a biological trait is innate for a genotype is the degree to which developmental 
pathway for individuals possessing an instance of that genotype is canalized.  The degree to which 
a developmental pathway is canalized is the degree to which it is bound to produce the end-state 
regardless of environmental variation in either (a) its initial state, or (b) during the course of 
development.  Notice that this definition preserves the idea that traits are innate with respect to 
certain genotypes.  It may turn out that, for example, some of the genotypes that typically express 
blue-eyes are canalized while others are more sensitive to environmental fluctuations.  Notice also 
that canalization is a matter of degree.  Limb development in many organisms is highly canalized; 
limbs develop in all but only the harshest environments.  Sexual affinity in mallards may also be 
canalized if the results of isolation-rearing experiments are to be granted as evidence.  However, a 
trait requiring several very specific environmental cue to develop in individuals, e.g., learning 
French (as opposed to learning language simpliciter), is not highly canalized.

In accepting that Waddington's notion of canalization applies well to the idea of innateness, we 
open the doors to a range of parallels between the work of Waddington and the work of ethologists 
like Lorenz.  A striking example will further strengthen the case for thinking that innateness is 
canalization.  In this instance, we find Waddington solving a problem that Lorenz failed to solve: 
how can we invoke natural selection to explain the origins of highly adaptive traits, such as sexual 
affinity in mallards, that are seemingly acquired characteristics but turn out upon further 
investigation (e.g. performing an isolation-rearing experiment) to manifest in isolation?  For 
Waddington, the key is the concept of canalization.

To illustrate the phenomenon, Waddington (1975) managed to induce an extreme environmental 
reaction in the developing embryos of Drosophila.  In response to ether vapor a proportion of 
embryos developed a radical phenotypic change, a second thorax.  At this point in the experiment 
we would say that bithorax isn't innate; it is a kind of chimera induced by an unusual environment.  
But then Waddington continually selected for Drosophila with the developmental capacity to 
respond to the environmental stress.  After about 20 generations of selection, some Drosophila 
were obtained that developed bithorax without being exposed to ether treatment.  What happened, 
according to Waddington, is that selection favored a particular pathway that led to the production of 
the optimal (in this case desired) effect.  Eventually the pathway became canalized, hence the end-
state, bithorax, appeared regardless of environmental conditions.  Waddington thus showed how 
natural selection can, so to speak, install a trait as innate.  In this case natural selection turned what 
was once an acquired traitÑbithorax qua reaction to ether shockÑinto an innate, i.e., canalized, 
trait.

6. Conclusion
Taking its cue from critics of Lorenz, I have presented an account of innateness that avoids the 
fallacy of claiming that traits can develop by genetic causes alone.  Innateness, on the canalization 
account, is a property of a developing individual.  At the same time, the proposal captures what is 
thought to be distinguishing features of innateness: satisfying the ontogeny condition, referring to 
the capacity to produce environmentally stable traits, and making sense of the idea that natural 
selection can install innate traits in a natural population.

REFERENCES

Block, N. (1995), "How Heritability Misleads About Race". Cognition 56, pp. 99-128.

Chomsky, N. (1988), Language and Problems of Knowledge. Cambridge, MA: The MIT Press.

Falconer, D. S. (1981), Introduction to Quantitative Genetics. 2nd edition. New York: Longman.

Gray, R (1992), "Death of the Gene" in P. Griffiths (ed.) Trees of Life.  Boston: 
Kluwer Academic Publishers, pp. 165-210.

Griffiths, A. J. F, Miller, J. H., Suzuki, D. J., Lewontin, R. C., Gelbart, W. M. (1993), An 
Introduction to Genetic Analysis.  5th edition. New York: W. H. Freeman and Company.

Hinde, R A. (1982), Ethology.  New York: Oxford University Press.

Johnston, T. D. (1988), "Developmental Explanation and the Ontogeny of Birdsong: 
Nature/Nuture Redux", Behavioral and Brain Sciences 11:  617-629.

Lorenz, K. (1957), "The Nature of Instincts".  In C. H. Schiller (Ed.), Instinctive Behavior.  New
York: International University Press, 129-175.

Lehrman, D. S. (1953), "A Critique of Konrad Lorenz's Theory of Instinctive Behaviour".  
The Quarterly Review of Biology 28: 337-363.

Herrnstein, R. J. and Murray, C. A. (1994), The Bell Curve: Intelligence and Class Struggle in
American Life. New York: Free Press.

Oyama, S. (1985), The Ontogeny of Information: Developmental Systems and 
Evolution. Cambridge: Cambridge University Press.

ÑÑÑÑÑ (1988), "How Do You Transmit A Template?", Behavioral and Brain 
Sciences 11: 664-645.

Richards, R. J. (1974), "The Innate and the Learned: The Evolution of Konrad Lorenz's Theory of 
Instinct".  Philosophy of Social Sciences 4: 111-133.

Sober, Elliott. (forthcominga), "Innate Knowledge" in Routledge Encyclopedia of Philosophy.

ÑÑÑÑÑÑ (forthcomingb), "Separating Nature and Nuture" in D. Wasserman and R. 
Wachbroit (Eds.).

Waddington, C. H. (1975), The Evolution of an Evolutionist.  Ithaca: Cornell U. Press.



FOOTNOTES


  I wish to thank Alan Belasco, Paul Bloom, Denise Cummins, Richard Lewontin, Elliott Sober 
and Denis Walsh for comments and/or extended discussions on earlier drafts.
  For reprints contact the author at: The Department of Philosophy, Chafee, University of Rhode 
Island, Kingston, RI 02881.  Email address: ariew@uriacc.uri.edu.
 Although Lorenz was talking exclusively about behavioral capacities, I take it that his account 
applies to any biological trait whatsoever.
 I'm not sure Lorenz ever said this explicitly, but it captures his idea.  See Richards (1974).
 When I refer to biological traits, items, characters, or features, I simply mean to refer to types of 
biological things like organs, tissues, behaviors, etc.
  Here I follow Sober's exposition in his (forthcomingb).
 See Suzuki et. al. (1993) for an explanation for this.  For a hint, heritability studies among 
relatives measure the ratio of the genetic correlationÑthe chance that two relatives share an 
identical alleleÑbetween relatives over their phenotypic correlations.  
	Innateness And Canalization