DEVELOPMENTAL CONSTRAINTS,
GENERATIVE ENTRENCHMENT
and the
INNATE-ACQUIRED DISTINCTION
William C. Wimsatt
Philosophy and Evolutionary Biology
The University of Chicago
June 26, 1985
This paper was printed (and too often mis-printed) in abbreviated form in W. Bechtel, ed., Integrating Scientific
Disciplines, Dordrecht: Martinus Nijhoff, 1986, pp. 185-208. This version has not been published. copyright, 1985,
by William C. Wimsatt. Address for correspondence: Department of Philosophy, 1050 East 59th St., University of
Chicago, Chigago, Ill., 60637; tel: 312+702-8598.
Introduction:
With mixed feelings and apologies to an old (and dishonored) friend:
The innate-acquired distinction is one of the oldest in our conceptual armamentarium, dating back at least
to the time of Plato. It has also been one of the most broadly applied, finding roots and applications in
conceptions of human nature and knowledge in epistemology, metaphysics, ethics, and philosophy of science,
cognitive psychology, linguistics, neurophysiology, developmental biology, animal ethology, and evolutionary
theory, and in theories of the evolution and nature of human sociality, culture, and morality. As befits any
concept which has cast such a broad and deep shadow, it has at some times and places been one of the most
highly honored, and at others, one of the most deeply dishonored agents of progress and stagnation, of clarity
and confusion, and of liberation and bondage, both conceptual and social.
For all of these reasons, any attempt at analyzing this distinction must be interdisciplinary, tenative, and
humble. It must be interdisciplinary because there are so many disciplines which use or abuse it. I will
draw primarily on philosophy, philosophy of science, evolutionary biology, developmental biology,
ethology, and the psychology of human problem solving for my analysis, and to a lesser extent on perspectives
from linguistics and artificial intelligence. It must be tenative because there are so many sources of possibly
relevant information that some are certain to be missed, and so many conflicting claims that some must be
rejected. It must be humble because, for all of its checkered past, the distinction has permitted many insights
which only the foolish would ignore. In the analysis which follows, I argue that a juxtaposition of elements
from diverse disciplines permits a new view of the phenomena which have motivated that distinction in a way
which is highly productive, though it will be necessarily reconstructive, and therefore, almost certainly
contentious.
This view captures more of these phenomena than any prior analysis, and explains their relevance in a
new way, but necessitates giving up at least two claims which have been dear to advocates of that distinction,
and other assertions which depend on these claims. The first claim, (or presupposition, since it is seldom
claimed explicitly), is that what is innate must be internal to the object in question. This has been a presupposition of making this distinction from the earliest times, and is the basis for claims that innate features are in
some sense independent of or unmodifiable through experience. I will argue that under appropriate
conditions, information which is embodied in the environment, and enters the organism as experience, must be
Wimsatt: Innate-Acquired
page 2
2/15/2016
regarded as innate, or something close to it, and that as a result the environment can in most even moderately
complex cases play a major role in the expression of the trait.
The second claim is a more recent accretion dating from the rise of genetics in the early 20th century.
This is the reanalysis of the classical association between the innate and the hereditary into the more modern
claim that to be innate is to be genetic. Although more recent, this claim is by now hardly less firmly
entrenched than the first. While I will argue that there is a (narrow) sense in which it is correct to say that to
be innate is to be "coded in a genetic program", this is a misleading and unduly narrow way of pointing to the
generative role of innate elements, and leads to an incorrect identification of the innate with the (biologically)
genetic. In particular, it does not follow, on the account I propose, either that if something is coded in a
genetic program, it is "genetically determined" or that the information which determines it is entirely in the
genome. From this it follows that not all things which are biologically genetic are innate, and not all things
which are innate are biologically genetic.
This last association (of the innate with the genetic) has done particular damage:
(1) Some modern sociobiologists have argued for a kind of genetic determinism, not only of biological
traits, but of a variety of cultural traits--selfishness, altruism, agression, xenophobia, dominance hierarchies,
sex roles, incest taboos, moral codes, and practices involving mating and the distribution of resources. Given
that these traits are supposed to be genetically based or innate, it is supposed that they should not, must not,
or cannot be modified, or if they must be modified, require special approaches to counteract supposed
ineliminable biological tendencies.
(2) Some modern nativists have argued that there are innate racial differences in intelligence, drawing on
the first claim to argue not only that the genetic contribution to intelligence is independent of the environment,
but also that social (i.e., environmental) programs to boost school performance (such as project Headstart) are
doomed to failure because the genetic program limits intellectual performance. On the other side, opponents
of these positions have mistakenly assumed that to be a nativist, one must believe in some kind of genetic
determinism, or believe that the environment merely acts as a trigger to release information coded in the
genome, or that environmental manipulations are capable only of realizing only (often limited) genetic
potentialities. (Of course, they believe this in part because some incautious nativists have asserted just this
much!)
It is in part because of abuses like the above, and in part due to the fundamental conceptual errors
embodied in the two preceding claims, that I must recommend radical surgery on the innate-acquired
distinction. With some sorrow (and a sense that I do too little honor to historical insights and thereby join
modern eliminativists with whom I have little sympathy), I suggest that what follows be regarded as an
argument for the elimination of the old distinction, not because it was useless, but because its continued use
would carry with it too many of the old associations, and provide further support for old ideas which have
been both conceptually stultifying and too readily productive of morally reprehensible conclusions. I will
argue that a new distinction in terms of generative entrenchment both preserves what was good in the old
distinction and blocks associations which we should avoid. Under the new analysis proposed here, neither of
the above claims or errors are defensible. For this reason, I will as much as possible avoid speaking of traits
as "innate", except where talking about past analyses. Where I am talking about my analysis, and it is
cumbersome to avoid the term "innate", I will use it in quotes, as in this sentence.
2. Claims about innateness from the philosophical tradition:
Wimsatt: Innate-Acquired
page 3
2/15/2016
The philosophical tradition has provided an enormous wealth of claims about innate knowledge, claims
which, even as the ethologists eschewed talk about knowledge and substituted talk about behavior, have left a
lasting imprint on discussions of the innate-acquired distinction to this day.
Perhaps the claim which has seen the widest range of interpretations is:
(P1) The claim that innate knowledge has been in some sense prior to experience.
(P1a) Innate knowledge exists prior in time to experience.
(P1b) Innate knowledge is a precondition for experience.
Among t
A second claim has been that:
(P2) Innate knowledge is independent of experience.
This also has at least two interpretations:
(P2a) The justification for or origin of innate knowledge is independent of experience.
(P2b) Innate knowledge is independent of any particular experience in that it is
invariant a
(P3) Innate knowledge often arises as an effect of or is "triggered by" experience.
(P4) Innate knowledge is knowledge of general truths.
(P5) Innate knowledge is universal--every member of a given class (usually human
(P6) Innate knowledge has a generative role in producing other knowledge.
produced through the
employment
knowledge.
beings) has
(This could
of
innate
(P7) Innate knowledge is often said to be different from other knowledge in being
These claims have been made, in a variety of different combinations, by a variety of different
philosophers. I must apologize for not documenting their sources and for skirting over the various subtleties
in the positions from which they are drawn. My aim here is only to indicate ideas which have influenced
more recent linguists, psychologists, and biologists who have grappled with the innate-acquired distinction,
and to indicate their general source in the philosophical tradition which has midwifed so many sciences and
scientific concepts.
In moving from the philosophical tradition to the domain of modern ethology, two major moves are
obvious and a third merits special mention:
(1) Although many ethologists ascribed a mental life to at least the more complex animals they studied,
(the fact that the innate-acquired distinction has never been applied to describe the behavior of plants is
revealing), they more frequently steered away from ascribing linguistic or quasi-linguistic knowledge (or its
consequences) to them. This is perhaps one effect of the dominance of scepticism in the tradition of Western
philosophy since Descartes: it was worth doing battle with scepticism for human knowledge, but ethologists
for the most part seemed to have capitulated and dropped at least the 4rd and 7th criteria for animals. (If
animals have any knowledge, it was assumed to be too low grade to be general. It was at best a posteriori and
contingent, and it was hardly analytic. Each of the properties ascribed by criteria 4 and 7 appeared to require
too rich a mental life for animals. (Konrad Lorenz (1941) is possibly the only exception here: he argued that
the biological concept of innateness could be used to give an account of space and time as forms of experience
and the category of causality, thus biologizing Kant.)
analytic, ne
Wimsatt: Innate-Acquired
page 4
2/15/2016
(2) Almost all of the criteria for innateness put forward by ethologists can be seen to be related to one or
the other of the above philosophical claims, but transformed in ways appropriate to the study of animals.
Thus, what was said to be innate was behavior, or sometimes morphological form, rather than knowledge.
Criteria (P1a) and (P2b) were carried across more or less unchanged, and interpreted as comments about
development. (P1b) has an interpretation in the light of results of deprivation experiments in which
deprivation of experience results in loss of a capacity. (P2a) has at least an analogue in criterion (E6), (I am
here referring forward to the list of ethological criteria found in the next section), which suggests different
sources for innate and acquired information. The criterion of universality (P5) was split into two, reflecting
the influence of evolutionary taxonomy: Universality within a species was accepted as a central criterion, and
presence of a trait in phylogentically related species was accepted as a less central and indirect criterion which
was important because it indicated a genetic basis for the trait. (See the discussion in the next section and the
fuller list of ethological criteria in appendix A.)
(3) Another move which is somewhat surprizing is that the generative role for innate knowledge was
ignored by most ethologists. (Because it is suggested by generative linguistic theories, it was not ignored by
Chomsky, and is in fact central to his analysis: see his debate with Putnam: Chomsky, (1967), and Putnam,
(1967)). Even though ethologists recognized in a variety of ways that some behaviors played an important
role in generating or making possible other behavior, this was not directly connected by them with the concept
of innateness. This is perhaps because innateness was seen as bearing primarily on the origins or causes of
knowledge or behavior, rather than as fundament- ally connected with the effects of having that
knowledge or exhibiting that behav- ior. By contrast, I think that generative role is the most powerful
fulcrum in analyzing the relation between what have been called innate and acquired elements of behavior and
knowledge. It is at the core of the analysis which I will advance below.
3. The ethological debates:
In this section, I plan to do two things: first, to list those criteria which have been most central to those
ethologists (e.g., Lorenz, (1965), or Mayr, (1974)) who have defended the distinction. (These, with the
philosophical criteria listed above are the minimal data for any analysis. A much longer and more complete
list of ethological criteria is found in appendix A.) Secondly, I wish to list the main dissatisfactions which
critics of the distinction (see primarily Lehrman, 1970) have raised.
The following criteria for innateness have been taken as central by those who have chosen to defend the
distinction:
(E1) Innate behavior for a given species is universal among normal members of that
(E1a) On the "genetic" account of innateness, this is taken to indicate that the innate
(E1b) On the "developmental" account of innateness, this is taken to indicate that the
species in th
behavior ha
innate beha
(E2) Innate behavior is that which appears early in development, before it could
possibly ha
(E3) Innate behavior is relatively resistant to evolutionary change.
(E4) Critical periods for learning certain information, or unusually rapid or "one-shot"
learning ind
(E5) Phylogenetic parallels between behavior and morphological traits is evidence for
the innatene
(E6) Innate information is said to be "phylogenetically acquired" (through selection)
and heredita
(E7) Stereotypy of behavior (including insensitivity to environmental influence, cf. P2b
or E1b) is ta
Wimsatt: Innate-Acquired
page 5
2/15/2016
(E8) Relatively major malfunctions occur if innate features don't appear or aren't
allowed to d
In addition, since the rise of genetics and the modern synthetic theory of evolution, two new criteria have
been added, presumably because they are criteria for a trait's having a genetic basis:
(E9) If a trait shows simple (e.g., Mendelian) patterns of inheritance, it is innate.
(E10) If a trait is modifiable through selection, it is innate. (To be modifiable through
selection, a
Those ethologists who have defended the use of the innate-acquired distinction (Lorenz and Tinbergen
are probably the senior statesmen of this group) have also argued strongly for an evolutionary analysis of the
significance of behavior. They have argued that the analysis of behavioral traits required the study of the
organism in its natural environments, or in at least minimal experimental preturbations of these environments,
and have argued that the analysis of the functions of these traits is an essential part of this study. They have
also tended to make use of extensive phylogenetic cross-comparisons. They have tended to stay away from
the exclusive study of organisms raised under laboratory conditions, except under special circumstances
where the aim is to study the effects of deprivations of conditions which are found in the natural environment.
By contrast, another group of students of animal behavior, found primarily in the United States (and
many of whom were influenced by T. C. Schnierla) worked primarily with laboratory animals under
experimental paradigms which were at least partially influenced by the behaviorist tradition. Probably for
this reason, (as good environment- alists), they tended to find the innate-acquired distinction to be of little use,
and argued for its abolition. Although I have little sympathy with their program, I believe that they raised a
number of serious issues worth discussing here. In what follows, I will draw primarily on the critique of D. S.
Lehrman, (cf. Lehrman, 1970), one of the most articulate and influential of these critics.
Lehrman made a number of points which are quite persuasive, even if they were not always right.
(1) He argued that there are two senses of innate, not one, and that equivocations between these senses caused
confusions and increased the apparent defensibility of the innate-acquired distinction. A defensible sense, he
felt was to say that to be innate was to
be genetic, or (what is not
exactly the same thing) to be heritable. I will call this the "genetic" sense of innateness. He argued
(successfully, I think) that this sense had little to do with the other sense. A second sense of innateness (what
I will call the "develop- mental" sense) referred variously to the other criteria given (see, e. g., E1 thru E7
above, but especially E1b and E2). He argued (less successfully I think) that this sense was confused. Given
his preference for the first sense of innateness, he argued that innate and acquired are not strict opposites,
since what was acquired referred to the phenotype, and what was innate referred to the genotype. But the
European ethologists wanted to say that behaviors could be innate or acquired. To Lehrman, this made little
sense, since to him, everything in the phenotype was partially due to the influence of experience, and thus
was at least partially acquired.
I agree with Lehrman that the genetic criteria (E9 and E10) should be treated differently from the others
(with which they in fact conflict), but will draw from this (in section 9 below) the conclusion that they should
be rejected rather than the others.
(2) Partially for the reason of (1) above, Lehrman argued that the innate-acquired distinction as applied to
phenotypic traits had to be treated as a matter of degree. Thus there were no absolutely or purely innate or
absolutely or purely acquired traits. The analogue to innateness, on the analysis I will present, is
generative entrenchment, which is also a degree property.
Wimsatt: Innate-Acquired
page 6
2/15/2016
(3) Lorenz (e.g., 1965) often spoke of innate traits as being "blueprinted" in the genome. Lehrman argued
that the "blueprint" model of gene action was outdated, and led to conceptual errors. Thus, for example, the
isomorphism between a blueprint and the finished product would lead one to expect a 1-1 correspondence
between genes or genetic elements and some behaviors, with things being classified as innate if they are in
the blueprint, and acquired if they were not, and leading to the mistaken idea that innate-acquired is a
dichotomous variable.
Lehrman favored another metaphor for the relation between genes and behavior, regarding the genome as
a program which produces the phenotype. In a program, there are few neat 1-1 mapping between elements of
the program and parts or properties of output of the program, just as there are few neat 1-1 mappings between
genes and characters. (See Wimsatt, 1976.) This is in fact a far better model, one employed by Mayr (1974),
who has offered the best positive defense of the importance of the innate-acquired distinction in the literature.
(Mayr agrees with Lehrman that the old distinction needs to be junked, but disagrees with his claim that it was
useless.) I will also employ the "program" metaphor.
(4) Lehrman (1970) argued that the proper way to study the origins of behavior was through the study of
developmental processes, not through evolutionary arguments and cross-species comparisons. I agree with
Lehrman that the study of development is crucial to t he analysis of behavior (and to analyzing its
status as innate or acquired.) However, I will also argue that phylogenetic comparisons are a powerful
source of information about the character of developmental programs.
Lehrman felt that the evolutionary orientations of the European ethologists blinded them to the role of
experiential determinants in the origin of (to him, virtually all) traits. Thus, he argued against Lorenz's claim
that the pecking behavior of the chick was innate because it appeared as soon as the chick got out of the shell,
and before it could have been learned. He pointed to the studies of Kuo (1932) which suggested that the
pecking behavior developed as a reflex through the pre-natal bobbing of the embryo's head caused by the
heartbeat. This input (as almost all others), Lehrman counted as experience.
(5) Lehrman could do this in part because of an extremely broad definition of experience--one which is not
intuitive, and which is in effect a piece of environmentalist imperialism. (Nor are the nativists without their
imperialist moves. Their corresponding move is to regard environmental inputs as nothing more than
"triggers" which release information which is already included in the genome. See, for example, Fodor,
(1981) or Dawkins (1976, 1978, 1982). The error on their part is to act as if environmental inputs contain at
most neglegible information.)
Lehrman cites with approval Schnierla's "generalized" concept of experience:
"Schnierla...has used the concept of 'experience' to mean all kinds of stimulative
effects from
p.21.)
With this definition, everything not in the genotype (or at least in the zygote at fertilization) is at least
partially a product of experience. One problem with this definition of the innate-acquired distinction is that
nothing in the phenotype is innate. The complementary problem with the nativism of, for example, Fodor, is
that no (or almost no) information is acquired. I think that it is safe to say that a good definitional strategy for
an important distinction is that no definition should be such that one or the other of the two dichotomous
terms has no cases to which it applies. Both of these character- izations violate this condition.
(6) A more specific problem with Lehrman's characterization of experience is that early input through
sensory channels in a developing nervous system has an ambiguous status. Its deprivation often appears not
just to deprive the organism of some information which it has to learn later, but of a capacity for acquiring
experience through that sensory channel. (See, e.g., the series of papers on sensory deprivation collected in
Wimsatt: Innate-Acquired
page 7
2/15/2016
McCleary (1972)). Loss of a capacity, rather than simple loss of some information which could be acquired
later has been a basis for claiming the innateness of the capacity. Early experience may be required for the
development of this capacity, but as such, it performs a function more like food than like information.
Schnierla's or Lehrman's concept of 'experience' would make no distinction between recognizing a possible
meal, and eating it! The nativist would claim that in depriving the organism of this kind of experience, the
system is rendered malfunctional, and is not normal, indicating an innate need for that experience. (One might
sat that early experience is often neither simply food, nor thought, but food for thought!) Related claims are
made for deprivation of information or the right kind of experience during critical periods for learning. The
idea that an organism has periods when it learns certain kinds of information or tasks easily, and after which it
can learn them only with great difficulty or not at all has been an embarrassment for naive empiricists, and is
the basis for Mayr's (1974) distinction between closed and open programs, his substitute for the innateacquired distinction.
(7) Lehrman also accuses the research program of the ethologists of slighting the study of developmental
processes:
"The use of 'explanatory' categories such as 'innate' and 'genetically fixed' obscures
the necessit
To some extent, this is surely true: though ethologists do not regard developmental processes as
irrelevant, their primary focus of concentration is on the behavior of species in their natural environments,
which makes study of developmental processes and their experimental manipulation difficult at best and often
impossible. They have with equal justification claimed that the (primarily American) students of animal
behavior have paid too little attention to the evolution and natural function of behavioral adaptations.
Lehrman is hardly ignorant of genetic and of evolutionary processes, as his 1970 article attests, but his focus
is on the experimental study of the ontogeny of behavior in the laboratory.
It seems natural to say that the two groups are studying different aspects of the same problem, and one
can only feel frustration that they seem more interested in taking potshots at each other than in cooperating.
Happily, things have improved since 1970, and the best modern ethologists draw heavily on results from both
areas, and often combine both kinds of studies in their own work. One advantage of the analysis I will
propose is that it makes studies of both kinds equally relevant (and complementary) to the study of animal
behavior, and the phenomena which the innate-acquired distinction was designed to capture.
(8) Another important feature of the innate-acquired distinction in the work of Lorenz, or the closed/open
program distinction which Mayr wishes to replace it with, is that it is quite clear from their analyses that these
are very important distinctions for the analysis of the evolutionary and ecological dimensions of behavior.
This is perhaps clearest in Mayr's (1974) work. Mayr presents a number of general conditions under which it
is advantageous for a trait to be coded for via a closed or open program and in which the other kind of coding
would not work, and illustrates these with a number of telling and insightful examples. Lehrman suggests
(1970, p.28) that since selection selects only for results that it doesn't matter whether a trait is innate or
acquired, and points out that, for example, in some species, bird song is innate, while in others it is largely
acquired. Mayr's analysis makes it clear that it does matter, and successfully predicts (for the same examples
Lehrman discusses) whether their song should be innate or acquired. The problem for those who would issue
an outright and total denial of nativism and abolish the innate-acquired distinction, leaving nothing in its
place, is that it serves a real function: If the innate-acquired distinction did not exist, it or something
like it would have to be invented by evolutionary biologists to talk about the evolution of the
adaptive design of the phenotype in response to the structure of information in the environment.
(9) One final thing should be noted about the innate-acquired distinction as defended by Lorenz and by
Mayr. Although the ethologists have proposed a large number of criteria for innateness, there is too little
discussion of how they are related. Also, many of them are simply presupposed as conditions of innateness,
Wimsatt: Innate-Acquired
page 8
2/15/2016
and not themselves explained by the analysis. This is understandable if, as I have suggested, most of these
criteria are simply carried over from the philosophical tradition. But if this distinction, or one like it is to play
a central role in a scientific theory, we have a right to demand more: it would be nice to get an integral
explanation of why these criteria are important, and of how they relate to one another.
This is perhaps the greatest strength of the analysis I propose, for the importance of and relation between
almost all of the above criteria (not to mention essentially all of the others in the expanded list of 17 criteria
given in appendix A) is explained by assuming that to be "innate" is simply (with a qualification to be added
later) to be deeply "generatively entrenched". This analysis not only gives reason for affirming the
importance of these criteria, but also gives reason for rejecting the "genetic" criteria (E9 and E10), which,
together with correlative misunderstandings induced by equating the "innate" with the genetic, are responsible
for most or all of the social and scientific misuses of nativism.
Before we can discuss this new nativist analysis, I must first present a model which was first designed for
another purpose--to explain the phenomena of von Baer's Law(s), which assert parallels between phylogenetic
patterns and patterns of development. These regularities turn out to have a fundamental connection with the
innate-acquired distinction.
4. The developmental lock:
To understand the developmental lock, we must first make a detour to consider results from still another
field--the work of H. A. Simon on heuristics of problem-solving. In his classic paper, "The Architecture of
Complexity" (1962, 1981), Herbert Simon uses an analogy of the problems in cracking two different kinds of
combination locks to illustrate the advantage of being able to decompose a complex problem into subproblems which can be solved independently. The following is a slight adaptation of his analogy. Consider
two cylindrical combination locks, each having a row of 10 wheels, each of which has 10 positions. There are
thus 1010 possible combinations, only 1 of which is correct. In the first, or "complex" lock, with no clues for
partial solutions (see figure 1a) the expected number of trials before getting the correct one is one half of the
total, or 5x109. In the second, or "simple" lock (figure 1b), we hear a "click" when each wheel is put into its
right position, so that the expected number of trials is 5 per wheel, or 50 for the lock. The advantage of being
able to get the total combination as an aggregate of independent solutions for the various wheels is the ratio of
the expected number of trials for the two locks, or 108!
Simon argues (1966) that an important problem solving heuristic is being able to "factor" problems into
sub-problems which can be solved independently, and their partial solutions strung together to get a solution
to the original complex problem. His parallel between this and the second lock is not exact, since it is often
the case that these partial solutions don't fit together exactly, but constitute good "first approximations" to the
behavior of components in the context of the whole, or to adequate solutions to the sub-problems in the
context of the complex problem. In this case, the complex problem or system can be treated as "nearlydecomposeable" (Simon, 1962), and the solution is nearly as simple as that suggested by the simpler lock. (It
is more complex through the addition of another stage in which the partial solutions are "fine-tuned" to one
another, producing a consistent and acceptable solution to the composite problem, but this adds much less
complexity than is found with the first lock.)
Indeed, the computational advantages of the second lock is the primary reason for the cost-effectiveness
of reductionistic problem-solving strategies, in which the aim is first to analyze the behavior of the parts in
isolation, and then to attempt to explain the behavior of the whole system in terms of the interactions of these
parts. Simon's "near- decomposeability" heuristic is in effect the "meta-heuristic" for reductionistic or
analytic approaches to the study of system behavior. In my (1980), I described (pp. 230-235) 9 reductionistic
Wimsatt: Innate-Acquired
page 9
2/15/2016
problem-solving heuristics, most of which can be seen as techniques for redescribing or reanalyzing a system
to facilitate this kind of problem-solving approach.
I will now describe a model derived by combining features of Simon's "simple" and "complex" locks,
which I will call the "developmental" lock, for reasons which will soon be apparent. It is a simple lock if
worked from left to right, but a complex lock if worked from right to left. (See figure 1c.) Suppose that in
this third lock, each wheel has its correct position (as in the first two locks), and the correct position is
indicated by a "click" (as in the second lock), but what position is correct for a given wheel is determined by
the actual positions (whether correct or not) of any wheels to the left of it. Thus, resetting a wheel
randomly resets the combinations for any wheels to the right of it.
If this lock is worked from left to right, the solution for the first wheel is a simple problem (with an
expected number of 5 trials before solution) since there are no wheels to the left of it. Given that this position
is now set, and will be left undisturbed, the solution of the next wheel is also a simple problem: although its
solution is a function of the position of the first wheel, that position has already been set. If each wheel is
solved before one proceeds to the next rightmost wheel, the solution to this lock is a simple decomposeable
problem, with an expected number of 50 trials for the whole lock.
The lock is simple because once a wheel is solved (working from left to right) we never need to return to it in
order to get the "compound" solution. Its solution is a serial multi-stage process like that of many (idealized)
mathematical or logical derivations. (I say "idealized" because it is often effective in derivations to work
backwards from the desired conclusion as well as forward from the given premises--a heuristic which Simon
calls "means-end analysis". (Simon, 1966.)) In problems whose structure is like this lock, there is a necessary
order in which the sub-problems must be solved (e.g., because the solution to later sub-problems requires data
from the solution of earlier sub-problems), but if this order is followed, this complex problem is
decomposeable into independently solvable sub-problems.
If the developmental lock is worked from the other end, however, it is similar to the complex lock. One
finds the "correct" solution for the rightmost wheel in 5 trials, but the chance that this represents a correct
solution for the whole lock is but 1 in 109, the number of possible positions of the preceding 9 wheels. If one
now turns to the second wheel from the right, and finds its "correct" position (as determined by the actual
positions of the preceding 8 wheels), there is only 1 chance in 10 that this solution is still correct for the last
wheel, and 1 chance in 108 that it is correct for the preceeding 8 wheels. One has made essentially no
progress! For a wheel m positions from the right, there is 1 chance in 10m-1 that the "right" position for this
wheel is also correct for the wheels to the right of it, and 1 chance in 1010-m that it is right for the wheels to
the left of it.
Wimsatt: Innate-Acquired
page 10
2/15/2016
Wimsatt: Innate-Acquired
page 11
2/15/2016
One more modification must be made to this lock if one is to apply it as a model for development. An nwheel lock as described has but 1 solution out of 10n possible combinations.. (Suppose that there were but
one possible adaptive phenotype--then there would be no adaptive radiations and no diversity of organic
forms! Serial evolution would be possible, by adding additional wheels to simpler locks, but the
evolutionary tree would have but a single branch.) What is required is to modify the lock in a way that
preserves its structure, but allows for a branching multiplicity of adaptive solutions. The simplest way to do
this is to assume that each wheel has 100 positions, 10 of which are solutions (i.e. are adaptive) for that wheel.
When the position of a wheel is changed, the 10 adaptive solutions of each wheel downstream of it are
randomly reassigned to the digits 0 thru 99. In this case, the n-wheel lock has (100)n or 102n possible
solutions, 10n of which are adaptive. We now have a large number of alternative possible solutions, but all of
the other above conclusions are unchanged. (It is still true for example for the 10 wheel lock that 1 in 1010 of
the possible combinations are solutions, and that a random resetting of a wheel m positions from the right has
1 chance in 10m-1 of being correct for the wheels to the right of it.)
The net effect of this left-to-right dependence in the combinations of the wheels is that if one takes a lock
of this type which has a correct solution to all of its wheels and randomly resets a wheel which is m wheels
from the right, the chance that this resetting is a solution to the lock is 1 in 10m! Thus, as one goes from
right to left, the probability that a change in a wheel does not scramble combinations to the right of
it declines exponentially!
5. The developmental lock and its generalizations as models of development:
Now consider the developmental lock in an evolutionary context. Let the wheels correspond to
developmental stages of an organism, with the earliest stage at the left and successively later stages further to
the right. (I don't want to suppose that development is partitioned neatly into stages. It would be almost
trivial to develop a "continuous" model of this simple lock which would preserve this left-to right dependence,
and permit the same qualitative conclusions, but for present purposes I will continue with the discrete multistage model.) Suppose that expression of a mutation at a given stage of development corresponds to the
random resetting of the appropriate wheel.
The idea that earlier wheels set correct combinations for later wheels corresponds to the idea that the
proper development of features at later stages presupposes or depends upon the occurrence of given features at
earlier stages, and that if these features are not present at earlier stages, this has a high probability of causing a
malfunction at a later stage or stages. Now suppose that the probability of getting a solution which is correct
for that stage is k, rather than 1/10, and that the correctness of a solution at a given stage is independent of its
probability of correctness at later stages. The probability that a mutation will be correct at a stage which is m
stages from the end and also correct at the m-1 later stages is then km.
This last statement presupposes that a single parameter, k, characterizes all single-stage probabilities of
success in a multi-stage adaptive task. This assumption can be dropped, and doing so immensely increases the
generality of the model. Thus, suppose that a given mutation, denoted by #, occurs at a stage which is m
stages from the right, and has a probability k#m of being adaptive at that stage. Suppose that it induces
probabilities k#m,m-1, k#m,m-2, k#m,m-3, . . . that mutation # at stage m will not lead to maladaptive
changes at these stages. The most promising aspect of this generalization is that it allows modelling the
processes of canalization, and the notion of a critical period, which are beyond the scope of the original
Wimsatt: Innate-Acquired
page 12
2/15/2016
model. I will return to the idea of a critical period in section 8 below. The idea of canalization falls directly
out of the model.
If a developmental program was strongly canalized with respect to the effects of mutation # at stage m,
one would expect that the probabilities k#m,m-1, k#m,m-2, k#m,m-3, . . . would all be close to 1 (or close to
the survival probabilities for the normal organism if these are substantially different than 1), rather than fairly
small. Thus the simple addition of a multi-parameter model allows one to express the fundamental
intuition behind the concept of canalization.
The generality of this model could be increased even further by making these probabilities depend upon
the environment, producing an array of probabilities, k#m,Ei for the mutation # at stage m in environment Ei,
which induces effects k#m,m-1,Ei, k#m,m-2,Ei, k#m,m-3,Ei, . . . at subsequent stages. In general one would
have to consider not only the environment Ei in which the mutation occurred at stage m, but also the
environments in which the various subsequent stages develop. This would be required to discuss cases of
phenotypic switching or other cases in which the effect of an adaptation or mutation is strongly dependent on
the environment.
This model can also be specialized or simplified in various ways. (1)If the environments of the mutations
had no effect, we could drop the environmental subsripts, returrning to the first nmulti-parameter model. (2)
If all mutations at a given stage were equally canalized at all later stages, (which seems unlikely, but could
produce a model which was easier to work with), one could drop the # subscripts from the probabilities,
getting km,m-1, km,m-2, km,m-3, . . .. (3) If in addition, all of the probabilities induced at later stages were
identical, and depended only on the stage at which a mutation occurred, the probabilities would all be km, and
their product kmm. (In any of these variants, of course, the probability of the mutation being adaptive is the
product of the probabilities for it being adaptive at that and all later stages.)
It is worth noting a limitation of these multi-parameter extensions of the original model. While it is
possible (as I have indicated) to generalize the model so as to include the effects of canalization, the model
does not explain canalization, but, in this application is merely presupposing it. Other models, like that of
Kauffman in this symposium would be required to explain canalization. Similarly, in the generalization to
cover environ- mental dependencies described above, the model does not explain these dependencies, but only
provides a way of accounting for their effects in a general model of develop- ment. The same remarks apply
to the analysis of the closed/open program distinction in terms of this model which is discussed below in
section 9. While explanation is preferable to mere descriptive adequacy (cf. Lewontin, 1974, chapter 1),
descriptive adequacy can sometimes buy you a great deal.
Thus, if we associate with each mutation which is first expressed at a given stage, a vector whose
components are the probabilities that the changes are adaptive at subsequent stages, this provides a way of
constructing models which include a great deal of detail about developmental events, and which could be
readily included in population genetic models. (This is true because this vector of probabilities has basically
the same form and effects as life history models or population growth models with overlapping generations
which model these processes in terms of a series of probabilities of an organism's surviving from one lifehistory stage to the next.) The fact that these parameters describe the effects of various adaptations without
explaining how they work is not a limitation in this context, for the strategy of population genetic models is to
do an "engineering" analysis of the operation and effects of an adaptation on fitness (see Lewontin, 1978), and
then to plug the fitness estimates from such an analysis into a genetic model to see whether and under what
conditions that design will be selected for. If the genetic model predicts that a given design will be selected
for over the available alternatives, we have an evolutionary explanation for the existence of that design. Why
this model could be important is that, to my knowledge, no population genetic models currently exist which
Wimsatt: Innate-Acquired
page 13
2/15/2016
are capable of handling detailed developmental information about phenotypes. If no such models exist, then
there is no way of talking about the evolution of designs which differ in these respects, so the introduction of
such models could substantially increase the scope of evolutionary explanations. This approach will be
elaborated in a future paper.
6. The developmental lock and von Baer's Law(s):
I will turn now to show how this model, in even its simplest form, can be used to explain an important
class of phenomena in development and in the evolution of diverse forms of life. Assume (1) that k is a
constant, so that km is the probability of a mutation in the m-th wheel from the right being adaptive at that
and at all subsequent developmental stages; (2) that mutations have an equal probability of being first
expressed at any stage of development (an assumption which can also be dropped later); and (3) (for
simplicity) that all adaptive mutations are incorporated. Then we get the following intriguing result: Since k
is between 0 and 1, then ki < kj if and only if i < j. Thus, the proportion of mutations which is adaptive
declines exponentially at earlier stages (i.e., with greater values of m) in development, so we would expect
that evolution should be increasingly conservative at earlier stages of development because features
which are expressed earlier in development (D1) have a higher probability of being required for
features which will appear later, and (D2) will on the average have a larger number of "downstream"
features dependent upon them.
I will hereafter speak of features as being "generatively entrenched" in proportion to the number of
"downstream" features which depend upon them.
In 1828, the German embryologist, Karl Ernst von Baer formulated 4 laws of development, which are
given here (after Ospovat, 1976, p.6):
"(1) The more general characters of a large [taxonomically varied] group of organisms
(2) From the most general forms, the less general are developed, until finally the most
(3) The embryo of a given animal form, instead of passing through the other forms,
(4) Fundamentally, therefore, the embryo of a higher animal form is never identical to
appear earli
special arise
becomes se
any other [a
These laws are often summarized under the general rubric: "Differentiation proceeds from the
general to the particular." Von Baer's Law(s) can be given three distinct formulations depending upon the
interpretation of "generality":
Taxonomic generality: Features which appear earlier in development tend to apply to
broader tax
Morphological generality: Features which appear earlier in development are morph-
ologically m
Functional generality: Features which appear earlier in development are functionally
more genera
Of these three interpretations, von Baer probably intended the first two. The third interpretation is
particular interest for behavioral ontogeny, and also for the ontogeny of conceptual structures. (I will leave
comment on the detailed differences and implications of these interpretations for another occasion. I believe
that all three versions are, under appropriate conditions, true for systems which are appropriately modelled by
the developmental lock.) What should be obvious, is that the truth of von Baer's law under the first
interpretation follows naturally and directly from the developmental lock model. The greater conservatism of
features at earlier developmental stages implies that, on the average, features which are expressed earlier in
development are, probabilistically speaking, (D3) older and (D4) most likely to be more widely taxonomically
distributed than features which are expressed later in development. (Von Baer's law, on this reading comes
Wimsatt: Innate-Acquired
page 14
2/15/2016
out as a probabilistic regularity, rather than as a universal law. This is in accord with other modern
interpretations of it, such as that of Gould, (1977)).
A further consequence which follows naturally from the developmental lock (but not from von Baer's
Law) is (D5) that features which are deeply generatively entrenched are features which, if they are disturbed
or fail to appear, are likely to cause major developmental abnormalities. This follows directly from the
developmental lock and the definition of generative entrenchment because such features indicate points where
changes will induce macro-mutations, which have a very small probability of being adaptive. This
consequence thus reaffirms the neo-Darwinian prejudice against evolution through macro-mutations. (D6)
Given that earlier features have a higher probability of being significantly generatively entrenched, then
mutations which are expressed earlier in development are more likely to have larger, more pervasive, and
more deleterious effects.
Of course, a feature may occur very early in development, and have little or nothing which depends upon
it, just as many mutations are "silent", due to synonymy in the genetic code or other higher level functional
equivalencies, redundancies, homeostatic regulations, canalizations, and the like. If we reconceptualize von
Baer's Law in terms of generative entrenchment, rather than simply in terms of earliness in development, then
it is true (as a probabilistic generalization), and the taxonomic data which serves to support the original
formulation supports even more strongly the new formulation, since mutations which are expressed early in
development and are not generatively entrenched would tend to undercut von Baer's law, but not the
reformulated account. (We have perhaps a correlative disadvantage however, since while earliness in
development is a readily observable property, "generative entrenchment is a theoretical construct which is not
directly observable. This is not to say however that we have no grounds for inferring the degree of generative
entrenchment of a feature. The next point bears directly on this issue.)
Perhaps most importantly for the study of developmental programs, (D7) the developmental lock suggests
that cross-phylogenetic analyses of developmental processes can generate significant information about the
structure of developmental programs. On this account, a high degree of taxonomic stability of features earlier
in development implies a high degree of generative entrenchment. Thus, in principle, we can determine
features of the causal structure of developmental programs from doing cross-phylogenetic comparisons,
analyzing the relative stability of taxonomic features, and looking for covariances in changes of features as
clues to which later features may depend upon which earlier ones. (This conclusion, of course requires ruling
out other possible explanations for this taxonomic stability. I will do so on another occasion.) It is time now
to return to the innate-acquired distinction.
7. The innate-acquired distinction revisited:
If we now look back at the various ethological criteria given for innateness, and compare them with the 7
numbered consequences of the developmental lock model given in the preceding section, we see immediately
that most or all of these features are consequences of the developmental lock, or more exactly, of generative
entrenchment. (Indeed, the first 14 of the 17 criteria for innateness given in appendix A are consequences of
generative entrenchment, or of an extended version of this analysis which includes the closed-open program
distinction. I will return to this in the next section.)
The idea that deeply entrenched features should be evolutionarily conservative (D1 and D2) directly
parallels (E3)--that innate behavior is relatively resistant to evolutionary change It indirectly justifies the claim
(E5) that parallels between behavioral and morphological phylogenies is evidence for the innateness of the
behavioral (and also the morphological) traits, since if both morphological and behavioral traits are deeply
entrenched and thus evolutionarily conservative, we would expect substantial phylo- genetic correlations
between them. The same generative entrenchment which predicts evolutionary conservatism also predicts
Wimsatt: Innate-Acquired
page 15
2/15/2016
taxonomic generality (D4), of which universality within normal members of a species (E1) is a special case, as
is the conservatism and resistance to environmental influence of stereotypic behavior (E7). (Stereotypy of
behavior as a criterion actually draws on at least three other criteria, and deserves special treatment, which I
will defer for another occasion.)
The developmental lock model also predicts that modifications in deeply generatively entrenched
features will have major (and probably strongly deleterious) consequences (D5), in this case paralleling (E8)-that major malfunctions occur if innate features don't appear or aren't allowed to develop. Generative
entrenchment itself entails (P6), that innate experience has a generative role, and suggests (P1b), that innate
features are pre- conditions of experience.
Finally, the contingent correlation between earliness in development and a significant degree of
generative entrenchment explains the importance of (E2), though in a different way: On the traditional
account, earliness in development was important as a criterion for innateness because the cause of the
experiential trait was different. This was evidence that (post-natal) experience could have little role in
producing it. On the criterion I propose, earliness in development is important as a criterion because it
suggests that the effects of the trait are different: it is likely to play a generative role in producing a wide
variety of other behaviors or adaptations. On this analysis, Lehrman's (1970) complaint against the analysis of
Lorenz (1965) that Lorenz is ignoring the role of pre-natal experience in the ontogeny of traits present at birth
is irrelevant, since a trait is classified as "innate" on the basis of its effects rather than on the basis of the
source of the information which generates it. Of course, the issue of causes (or rather the source or mode of
acquisition of information which generates the behavior (E6)) remains: is it phylogenetically acquired or
ontogenetically acquired? This, with the discussion of "critical periods" and related issues (E4) will have to
wait until the next section.
Most strikingly, all of these features are seen to issue from a single property--generative entrenchment,
which in turn explains why so many of the criteria should co-vary as neatly as they do. To put it another way,
it explains the relationship among these many diverse criteria, something which no analysis has been able to
do before. Finally, it explains a number of criteria (e.g., E1, E3, E5, and E8) which are simply presupposed by
or which look like ad hoc additions to other analyses. That they should be presupposed by other analyses is
understandable if the source of the ethological criteria is the philosophical tradition, but it is obviously better
to have an analysis which explains, in a unitary way, the relationships among these various criteria.
An important further consequence of the replacement of the concept of innateness with that of generative
entrenchment deserves special notice. On this new account, (D8) environmental information may be
"innate"! This is so because talk of generative entrenchment does not distinguish between what comes from
inside and what comes from outside of the organism. (The idea that innateness implies "insideness" or
isolation from the environment is probably the source of the mistaken idea (rejected by developmental
biologists 70 years ago, but recently independently re-invented by Chomsky and Fodor) that physical or
functional modularity is a good criterion for innateness.)
The analysis presented here leads to an entirely different reading of Dawkins' (1976, 1978, 1982)
concept of the "extended phenotype." Dawkins takes an important step in bringing environmental features
into the extended phenotype. In doing this, he is taking one step further in the tradition of the European
ethologists who argued that behavior could be treated as any other phenotypic feature--it could be selected for
and one could study its taxonomic distribution and phylogenetic development. The problem with Dawkin's
notion is that it is "gene-centered". The "gene's eye view" of the phenotype and of evolution has been a very
productive heuristic for many problems, and Dawkins "extended phenotype" is a major advance in some ways
over prior analyses, but it is not the correct way of formulating the relationship between genes and
environment. I will return to this issue below in section 9.
Wimsatt: Innate-Acquired
page 16
2/15/2016
We should conceive of the phenotype as an unfolding developmental program which has an initially
small number of generative elements (some of the genes, together with some of their somatic environment in
the zygote) which interact with each other and with aspects of the outer environment to produce the
developing phenotype. In this conception, it is quite clear that information acquired from the environment can
have a profound effect if it is deeply generatively entrenched relative to subsequent behavior, and on this
analysis, if it is so generatively entrenched, it is "innate". This explains the ambiguous role of early
experience discussed in item (6) of section 2 above. If the early experience which is withheld in a
deprivation experiment has a generative role with respect to a wide range of subsequent experience in
that sensory modality, its loss will produce such far-reaching consequences that it would readily be
described as a loss of capacity. If in addition, there is a critical period for the acquisition of that early
experience, this loss will be at least partially irreversible after the critical period has passed.
What is required to speak of environmental information as "innate"? I suggest that the following are
minimal requirements:
(1)
(2)
(3)
(4)
The acquisition of that kind of information at that stage of development is deeply
The developmental program is designed to receive information of that sort at that
The information must be of a relatively specific sort.
The environment of the developing organism is a reliable source of the required
generativel
stage of dev
information
A few examples may help here. I will order these with respect to the degree of specificity of the
information, from more to less specific:
(A) On this account, not only the imprinting mechanism of the greylag goose at birth is "innate" (as on the
standard ethological accounts), but also the object of imprinting is "innate". When the infant goose extricates
itself from its shell, it imprints upon and follows any moving object. (Sound may be important, but Lorenz
reports having imprinted greylag geese on himself, his co-workers, his dogs, and even a moving toilet float!)
In their natural environment, however, there is a very high probability that the young goose will properly
imprint on its mother, and will in short order learn to distinguish her cries and her appearance from that of
other female greylag geese nearby. The family structure and behavior of the mother greylag goose at the time
of birth makes it almost a certainty that the baby geese will imprint properly (she stays close to the nest at
hatching time, and imprints upon them as well, and so is readily available during the critical period). Thus the
correct information (that a close moving object first detected at birth is mother) is reliably present in the
environment. This is an important adaptation in colonially nesting birds, who can depend only upon their
parents for food, warmth, and protection, and therefore need to be able to distinguish them from other
conspecifics.
(B) John Maynard Smith (1975) discusses the case of phenotypic switching between migratory and nonmigratory phenotypes in some species of locusts. The relevant variables determining the switch are (1)
humidity and (2) concentration of a pheromone which is secreted by conspecifics. If an appropriately
conditioned larvae experiences low humidity and high concentration of the pheromone, it will metamorphose
into a migratory phenotype, characterized by large wings, thicker chitin, and a behavioral tendency to migrate.
The message it is getting is roughly: "It is dry, so there won't be much food, and there are a lot of others
around to eat it so you'd better make ready and then get out!"
Maynard Smith discusses this because it is also a case of delayed inheritance: this developmental
switching does not happen in the first generation in which larvae experience this stimulus, but only in the
second. (Presumably these stimuli cause the eggs of the first generation to contain chemicals which
unexposed parents do not pass on to their eggs, and the different chemical composition of the eggs plus the
stimuli cause the different (migratory) developmental pathway. The message here is something like: "The
environment is patchy, both in humidity and in concentration of conspecifics, so you may be misinformed by
Wimsatt: Innate-Acquired
page 17
2/15/2016
your first local sample. Furthermore, migration is risky, so be sure the message is reliable before you migrate.
If you get two messages to migrate, it is then sufficiently reliable that you are better off leaving than trying
your luck here!"
Here, there are two alternative messages, either one of which makes the appropriate phenotype for that
environment. Thus we have an open program at the larval stage which has two possible adaptive messages
which the developmental program can act upon, which it does with major implications for morphological,
physiological, and behavioral traits, and ultimately for the fitness of the organism. In this case, it is
appropriate to say that the disjunction of the 2 possible adaptive messages is "innately" given, and which one
actually is given is acquired.
(C) A third kind of case is provided by the visual deprivation experiments discussed in McCleary (1972). In
the series of experiments on interocular transfer in kittens reported there, it was demonstrated that, not only
were kittens which were deprived of visual input from birth for a period of 5 weeks unable to discriminate or
to later learn to discriminate visual patterns in the deprived eye, but that this loss of capacity extended to
deprivation not only of light but of patterns (they tried translucent lenses!), and not only of patterns but of
patterns of a given type. (Kittens which were allowed to see only patterns of vertical bars could not later
discriminate or learn to discriminate patterns of horizontal bars!) In this case, what appears to be "innate" is
the need for receiving an appropriately rich variety of visual stimuli in the first 5 weeks after birth, without
which the capacity to make discriminations in the missing variety was absent.
These three cases presumably all meet conditions 1, 2, and 4, and fall at various points in a continuum of
cases which meet or don't meet condition 3. They thus demonstrate the importance of the third condition. But
when the third condition is not met, this does not mean that nothing is "innate"--it only means that the
disjunction of inputs which is "innately" given is so broad that only a general description (e.g., "visual
information of appropriate variety and richness") will do, and the reception of any input meeting this general
characterization will later provide the capacity for making a wide variety of finer discriminations within this
range. The breadth of this description indicates substantial homeostasis or canalization for the development
of adult visual capacities. Similar remarks would apply to the development of of the capacity for speaking
and understanding language in humans.
8. Mayr's "closed/open" program distinction:
In 1974, Ernst Mayr published a now classic paper on the innate-acquired distinction. In it he broke
away from a number of confining features of more traditional accounts. One of the many virtues of his
analysis is his substitution of the distinction between "closed" and "open" "genetic programs" for the old
innate-acquired distinction. For Mayr, despite appearances, "genetic programs" (whether open or closed) are
not parts of the genome--they are clearly both developmental programs, and as such are properties of the
phenotype. A closed "genetic program" is one which does not require (or permit) informational inputs from
the environment to significantly affect its execution. An open program is one that does require or permit such
information. (Mayr, 1974, p.651.) A "critical period" is simply the temporal window during development or
during a behav- ioral activity when the appropriate open program will recieve relevant information.
Mayr exploited these notions to make clear adaptive sense of the structure of developmental programs for
behavior in terms of the evolutionary and ecological contexts in which the organisms find themselves. He
also has given the best discussion available of the general conditions under which it would be advantageous
for a program to be closed or open, and suggested how this analysis provided a broad kind of classification of
types of behavior which went well beyond that provided by the traditional innate-acquired distinction.
Wimsatt: Innate-Acquired
page 18
2/15/2016
I cannot do justice to the richness of his analysis here, but I wish to argue that the main features of his
distinction--that of a "critical period" and the distinction between "closed" and "open" programs, can also be
captured by the generalized version of the developmental lock model. If my claim is correct, then all of the
resources of his analysis should be deployable through this model.
A critical period is a relatively delimited period during the lifetime of an organism when certain kinds of
input received by the organism have important effects on its subsequent development or behavior. As Mayr
says, it is a "window" in time through which environmental inputs of a delimited sort can enter. Inputs of the
same sort received before or after that time in development have quite different effects, and often little or no
effect at all.
This idea can be easily captured by the multiple-parameter generalized version of the developmental lock
model discussed above in section 5. Suppose that we generalize from the notion of a mutation which is first
expressed at a given stage, m, of develop- ment, to speak of any perturbations of the system, whether genetic
or environmental in origin, acting at that stage of the life cycle. (We have in effect already made this
extension to talk in the preceding section about environmental information as "innate.") To say that
information can get into a "window" in the developmental program during the critical period and not at other
times is to make two claims about the structure and probabilities associated with the developmental lock.
The first claim is simply that an input at stage m, of the appropriate informational sort, is generatively
entrenched with respect to subsequent behavior and development. The probability that a mutation,
purturbation, or environmental input acting at that stage will be adaptive at that stage is simply the probability
that in the normal environment the "correct" input or inputs will be received. If the right input is received,
then the subse- quent probabilities are simply the normal conditional probabilities of the organism surviving
through the various successive stages of its life cycle as are found in life-history or age-structure models in
population genetics. If the wrong input is received, the idea of generative entrenchment requires that one or
more of the probabilities for later stages will almost certainly be substantially lower (such that their net
product is substantially lower in the normal environment.)
The second claim (which introduces the "frame" of the window) is that the same inputs, whether right or
wrong, do not have that effect (or have a much smaller effect) on the probabilities if they are introduced at
earlier or at later stages than the one in question.
This proposed explication of "critical period" ignores some of the complexities of that notion. Thus most
critical periods are not "one-shot" "all-or-nothing" affairs, but a sequen- tial series of overlapping windows
with complex conditional interactions between them. None of these ideas however introduce complexities of
a new kind, and all of them should be expressible in terms of conditional effects on the probability vectors
associated with the generalized developmental lock model.
Similarly, an "open" program is simply a program which has at least one such window, and a closed
program is one which has none--or at least none which is open for the inputs in question during the period in
development being investigated. (It seems likely that there are no totally closed programs in development.
The idea of a closed program must be viewed as a relative one--relative to the period of time of development
under investi- gation, and the class of inputs being investigated, and probably also to the environment and the
prior state of the developing phenotype.)
With this last move, we have now captured all of the criteria of the traditional inate-acquired distinction
worth capturing, at least for ethological uses of that term. Criterion (E4), which refers to critical periods or
unusually rapid or "one-shot" learning now falls within the domain of the developmental lock model.
(Actually, I have not talked about rapidity or ease of learning. This can be further analyzed through another
ethological criterion given in the appendix--the simplicity of innate releasing stimuli.
Wimsatt: Innate-Acquired
page 19
2/15/2016
Even this additional criterion is in the domain of the developmental lock model, but demonstrating that
fact requires a detailed discussion of the application of the developmental lock to the development of
conceptual structures. The developmental lock has independent application within the realm of learning and
cultural evolution, where what is being learned and transmitted are generative cognitive structures. As such
this model has rich applications to the problems of the evolution and development of scien- tific theories and
other cultural structures. Thus, it predicts the relative frequency of scientific revolutions of different sizes,
and the probable places in the structure of scientific theories where changes are most likely to occur, as well
as yielding a naturalistic account of the analytic-synthetic distinction (which exactly parallels that given for
the innate- acquired distinction) and making sense of a variety of experimental and theoretical scientific
practices. Unfortunately, this is a whole other paper (See Wimsatt, 1983) so for the time being, the claim that
I can analyze all parts of criterion (E4) in terms of the developmental lock will have to be taken on faith.
9. The inappropriateness of the "genetic" criteria and the limitations of he "gene's eye" view of
evolution:
There are two criteria from the original list of the ethological criteria in section 3, namely (E9) and
(E10), which I have not tried to analyze in terms of the developmental lock model. This is because I think that
they are flawed and should not be included in the analysis. Lehrman (1970) was the first to point out that
there is a tension among the criteria for innateness--something which he read as suggesting that there are in
fact two distinct senses of innateness. (See section 3, item (1)). Of these two, he rejected the "sense" which
referred to developmental fixity as incoherent. This is the sense which I have analyzed here, and which I
would argue is the primary sense. He argues that another sense of innateness--roughly that to be innate is to
be genetic or heritable--is perfectly coherent, but does not yield the claims which the ethologists wanted to
make about innateness. The situation is worse yet, for I will argue that the two "genetic" criteria actually
conflict with some of the most important of the other criteria on the list. They do so quite independently of
whether or not one accepts the generative entrenchment account I have proposed, but that account structures
and heightens the inconsistency.
The central conflicts are with criteria (E1) and (E3). (E1) states that innate behavior for a given species
is universal among normal members of that species. (E3) states that innate behavior is relatively resistant to
evolutionary change. In addition, there are less striking conflicts with (E5), that parallels between behavioral
and morphological phylogenies is evidence for innateness, and with another criterion found in the expanded
list in the appendix, that presence of a trait in closely related species or higher taxa is evidence for innateness.
The first genetic criterion (one accepted even by Mayr (1974) and Lorenz (1965)) is that if a trait shows
simple patterns of inheritance it is innate (E9). The problem with this criterion is that to demonstrate simple
patterns of inheritance for a trait, one must have genetic variance for that trait--there must be at least two
alternative alleles which show different character states for that trait. (In the simplest case, this might be
simple presence or absence of the trait.) But if this variation is present in natural populations, the
character state can't be universal in the species (criterion(E1)). Naturally, this demands some
qualification. One could be looking at relatively macroscopic mutations, found in the laboratory or (relatively
rarely) in nature, which result in the absence of the trait which is said to be innate. In such cases also, these
macromutations will tend to have a variety of effects, and if the "innate" trait is significantly generatively
entrenched, these effects will tend to be large and highly deleterious. But if this criterion is intended to apply
to natural populations, and not to laboratory mutants, the conflict with (E1) is direct.
The second "genetic" criterion (E10) is that if a trait is modifiable by selection, it is innate. There is
some more justification for this, in the form of (E6), which contrasts the "phylogenetic" acquisition of innate
traits with the "ontogenetic" acquisition of acquired traits, but it produces problems nonetheless. Presumably
Wimsatt: Innate-Acquired
page 20
2/15/2016
the rationale for this criterion, is that if a trait is selectable (over successive generations, not merely in one
generation) then it must be heritable (or "genetic"), and then its origin is "phylogenetic". (See Wimsatt, 1980,
1981). There are two problems with this criterion: (1) If the trait is selectable, then (1) there must exist
genetic variation for it, thus producing the same conflict with (E1) as the last criterion. (2) If it is modifiable
by selection, then given the diversity of environments in which local populations of a species will find
themselves, even contemporaneously, and the much greater diversity of environments available over
evolutionarily significant periods of time, it is quite unlikely that the trait will be highly stable over
evolutionary time (criterion E3). In fact, this same problem exists for the first criterion.
Both of these criteria run into the same problems because they are reflections of two closely coupled
requirements. For a trait to be selectable,(E10), it must be heritable (E9). And if there is heritable variation
for a trait, (E9), then in different environments, there is a high probability that it will be selected (E10) in
different directions. In the short run, either will produce intra-populational and intra-species variation (contra
(E1)), and in the long run (unless there is strong resistance to evolutionary change for that trait (E3)), this will
be converted into differences among phylogenetic "neighbors" (contra the criterion in appendix A), and
parallels between behavioral and morphological phylogenies would be expected to break down (contra (E5)).
This way of putting it is revealing, because it makes a great deal turn on the evolutionary conservatism of
the trait--not too surprizing, given the centrality of that con- sequence on the present analysis. If a "genetic"
analysis of the innate-acquired distinction also had this consequence, it could be argued that innate traits are
evolutionarily conservative, and the conflicts would be removed by appending (E3) as a qualifier on (E9) and
(E10). This could be argued for within the genetic analysis only by showing either that there are no
spontaneous mutations for the trait, and thus no genetic variation (a course no-one would choose nowadays!)
or by arguing that variants are substantially less fit, and thus rapidly weeded out. But this conclusion can't be
established from the genetics alone, since it is basically a question about the relationship between phenotype
and environment. As we have seen, it can and does follow directly from a model of the generative structure of
developmental programs, which is why this analysis succeeds as well as it does.
The only remaining question, if the "genetic" criteria are inconsistent with the others, is how they should
have come to be accepted as criteria for innateness. I think that the probable source lies in affirming the
consequent with criterion (E6), moving from "to be innate is to be phylogenetically acquired" to "to be
phylogenetically acquired is to be innate." This seems such a simple error that this accusation seems
unreasonable, and it would be but for the complicating factor of the rise of genetics in the 20th century. With
the rise of genetics came a kind of genetic determinism, or at least a genetic imperialism or reductionism
reflected most strongly in the seminal works of G. C. Williams (1966), Richard Dawkins (1976, 1978, 1982),
E. O. Wilson (1975, etc.) and the bulk of modern sociobiology. One of the consequences of this view was to
cloud the present issue considerably.
To be phylogenetically acquired is to be heritable, and to be heritable is in part to be genetic.
Heritability of phenotypic properties requires as well heritability of environments, since the phenotype is a
product of genome and environment. Even more, all evolutionary theory requires (as pointed out clearly by
Lewontin, (1970), is heritability of fitness, which can be relatively stabler than genotype, phenotype, and
environment. A nice example of this is the phenotypic switching between migratory and non-migratory
morphs in locusts discussed above. There, the environment and the phenotype are changing together in such a
way that they thereby reduce changes in fitness below that which would be experienced without the
phenotypic switch. Furthermore, this facultative switching adaptation is reliably generated in the face of
enormous genetic variation from individual to individual, so the genome which produces this adaptation is
probably not stable either under sexual recombination. So the stability of the genotype and the picture of
genetic determinism of character traits or of fitness is undercut by the requirements of evolutionary theory,
which reqire nothing about the stability of genes, but require instead the heritability (i.e., stability! ) of
fitness. To be sure, in a genetically homeogeneous strain (so genetic canalization of phenotypic characters is
Wimsatt: Innate-Acquired
page 21
2/15/2016
undetectable) in a constant environ- ment (so genetic canalization of phenotypic characters is not needed, and
the causal role of the environment is also factored out (Wimsatt, 1981)), genetic determinism seems plausible.
But the real world does not meet these idealized conditions, and phenotype and environment must also be
treated as causal factors in the production of heritable variance in fitness. And heritable variance in fitness, is
for evolution, the name of the game. As Vince Lombardi said in another context, (one simultaneously
exemplifying and caricaturizing the best and worst of Social Darwinism), "It's not just the most important
thing. It's the only thing!"
The focus on the genome as the source of characters has lessened attention to the characters themselves.
At the base sequence level, one cannot see or read off develop- mental pathways, or (with some qualifications)
their evolutionary significance. So if innate information is phylogenetic, and if phylogenetic information is
genetic, and if one cannot tell information about generative entrenchment, or the taxonomic stability of
macroscopic morphological and behavioral traits at the genetic level, then why not treat all genes alike, and
make their information all innate? This is the garden path which leads to the genetic criteria for innateness, to
genetic determinism, and to the view of phenotypes as nothing more than gene-machines. It springs from the
enormously fruitful research program of modern genetics, and suggests an inductive justification for affirming
the consequent in the claim that to be genetic is to be innate. It is the illigitimate offspring of Weissmanism
(for which Romanes first coined the term "neo-Darwinism", and whose misinterpretation has fuelled current
genetic reductionism), and the dominant and largely unrecognized 20th century heresy of genetics and
evolutionary biology.
10. Different strategies for modelling development:
Dick Burian will comment on the relationships between this paper and those of Stuart Kauffman and
Bruce Wallace. Wallace speaks authoritatively as a leading representative of modern population genetics.
While agreeing with much of what he has to say in his paper for this occasion, and with much more of what he
has had to say on almost every other occasion, ultimately our papers are in deep disagreement. I can only
hope that I have provided him with a stronger basis than I did when he heard a distant ancestor of this paper
for changing his mind about the relevance of developmental biology to evolutionary theory, and even more, to
population genetics.
Stuart Kauffman's paper requires more detailed comments, which both the editor and Dick Burian have
encouraged me to provide. I regard it as one of the most important, and probably the most interesting and
original new approach to modelling develop- mental constraints in the literature. (I am tempted to say that it
is the only really new idea in recent literature on the topic.) I hope that I can argue that his approach and mine
are rarely in conflict, and far more frequently are complementary approaches to the modelling of development.
I have been following his work in this area since 1969 but it has only been within the last 2 or 3 years that my
views have taken more substantial shape and been elaborated sufficiently that he and I have again begun to
talk seriously about developmental models. In this discussion, then, I will take his paper as given, and will
draw primarily on conversations we have had about our two approaches. I hope that he would not disagree
with my interpretations of these exchanges.
(1) Kauffman has argued correctly that my approach is fundamentally conservative (no pun intended!)--that I
have basically modelled the neo-Darwinian (indeed, the Darwinian) abhorrence of macro-mutations. In this
model, I have merely systematized the cascading effects possible for earlier developmental changes, and
argued that they would be strongly selected against, producing increasing evolutionary conservatism of earlier
developmental stages. It is a conservative model because developmental constraints are seen as joint products
of the organization of developmental programs and the operation of selection, and the model thus basically
falls within the adaptationist program. No challenge to traditional neo-Darwinism here! Kauffman argues by
contrast that his generic constraints are in some sense prior to the operation of selection, and what's more, that
Wimsatt: Innate-Acquired
page 22
2/15/2016
under quite a broad set of conditions, are capable of overwhelming selection forces which happen to be going
in other directions. His generic constraints are thermodynamic in character (under a generalized informationtheoretic interpretation of thermodynamics) and represent a kind of physical constraint on possible (or highly
probable) modes of genetic organization.
This way of describing the two approaches makes them seem to be in conflict. But this conflict is
largely illusory. The illusions are of two sorts: (1) The assumption that selective forces and generic
constraints will be in opposition is gratuitous: it is primarily a consequence of the fact that one of the most
interesting directions for the development of Kauffman's theory lies in considering models where they are
opposed. (2) The domain of phenomena to which our models apply are only partially overlapping. Each
model has things to say about phenomena which the other does not address. I will address these points in turn
in the next few paragraphs.
(2) If Kauffman's generic constraints are found very early in development (as most of the ones he discusses
are, since they deal largely with the differentiation of different cell types) then we would expect that they
would have a high probability of becoming deeply generatively entrenched as well. They would then be
doubly reinforced as developmental constraints: not only will most mutations not change them (as generic
constraints), but also any mutations which did act to change them would be strongly selected against (because
they are deeply generatively entrenched). In such cases, his theory and mine are obviously complementary
rather than inconsistent.
(3) The description of generic constraints makes it appear if there are no selection forces, only entropic ones
at work. But at this level of organization it may be hard to tell the difference (for real systems) between the
two. It is no accident that Dawkins, for example (1976, 1982) sees the origin of selection in the relative
thermodynamic stability of primitive macromolecules. Generic constraints are a similar kind of stability one
or two levels up, stability in the molar behavioral properties of genetic control networks under random
mutations in the individual control elements.
(4) Kauffman starts with an ensemble of certain types of system. Why this type? As his own earlier analysis
shows (Kauffman, 1969), the parameters of this class may be products of selection. (Kauffman's use of gene
control networks with an average of two connections per element is a conscious decision, based partially on
the properties of the operon model, and partly on his demonstration that randomly connected networks of
binary switching elements have an expected minimum cycle time for about two connections per element. As
he argues, if short generation time is advantageous (as it is), then so is short cycle time. Thus, two
connections per element, a defining property of the ensemble of systems he looks at, is probably a product of
selection!) A somewhat different but related point is that if generative entrenchment produces sufficiently
broad taxonomic generality, the system properties may look (and truly be) generic. (Consider the near
universality of the genetic code, although most people argue that many characteristics of the code are products
of selection, and it is nothing if not deeply generatively entrenched!)
(5) I would argue that generative entrenchment is itself a generic feature of the design of the phenotype,
because complex multiply-connected networks tend to have long causal pathways and many loops of varying
lengths. One could also argue that generative entrenchment is itself deeply generatively entrenched. This is
because generative efficiency (producing a lot from a little) is highly efficient, and strongly hierarchial gene
control systems would have enormous evolutionary advantages over those with no significant hierarchial
organization. (See Simon, 1962, 1981, and also the discussion of the importance of near-decomposeability
and quasi-independence in organic design in (Wimsatt, 1981, pp. 141-142.))
(6) Finally, generative entrenchment applies to some kinds of constraints which generic constraints can't
touch, at least in the present form of Kauffman's model. (Thus, his analysis presently models only the
behavior of gene control networks, and not the expression of structural genes or the changing consequences of
Wimsatt: Innate-Acquired
page 23
2/15/2016
their expression and interaction with changes in the control structure. The same applies to generic constraints,
which appear to be capable of explaining aspects of phenotypic behavior which can only be presupposed in
the developmental lock model. Probably the most striking example of this is Kauffman's demonstration that
his networks are buffered or canalized against major changes in the face of random mutations in control
structure connections.
For all of these reasons, I think it is more fruitful (and more correct) to regard Kauffman's model and my
own as complementary, rather than as competitors. While there might well be room for substantial argument
as to whether a given constraint is due to generative entrenchment, to its generic character, to both, or to
neither, the theoretical structures of our two models support no natural opposition between them.
Acknowledgements:
I first hit on the basic developmental lock model in 1972 or 1973 while looking for a way of explaining Haeckel's
Biogenetic Law in a hierarchial way--an idea in effect suggested by Herbert Simon in his (1962). The paper Steve Gould
gave at a conference at the American Academy of Arts and Sciences in 1974, where I first presented the basic
developmental lock model (a paper which became a major theme of his (1977)) led me to realize that what I had explained
was not Haeckel's Law, but von Baer's Law. Since Steve there argued that von Baer's Law, unlike Haeckel's, was true, I
was encouraged to continue. My interest in applying this model to the innate-acquired distinction, to cognitive
development, and to the elaboration of scientific theories dates from about that time, though I was not then aware of most
of what is discussed in this paper, except in broadest outline. Over the years since, I have benefitted substantially from the
encouragement, elaborations, scepticism and criticism of many people, particularly Stuart Altmann, Bill Bechtel, Bob
Brandon, Dick Burian, Guy Bush, Bob Glassman (with whom I co-authored a paper which gave the developmental lock its
first public exposure, Glassman and Wimsatt, 1984), Jim Griesemer, Joe Hanna, Stuart Kauffman, Bob MacCauley, Jane
Maenschein, Joe Maxwell, Ernst Mayr, Bob Richards, Bob Richardson, Tom Roeper, Marty Sereno, Herb Simon, and
students in many classes, particularly those in my 1983 and 1985 seminars on Evolution and Epistemology. This paper is
most closely related to the content of the Donald Lipkind Memorial Lecture, which I was privileged to give at Chicago in
the spring of 1985.
I would like to dedicate this paper to the memory of two people who have had a largely indirect influence on these
ideas, an influence which undoubtedly would have been much more direct had they lived to see it. The first was Don
Lipkind, who was an excellent and deeply beloved student at Chicago from 1972 to 1977, and who was in the first course I
gave (with Tom Roeper) on philosophy of psychology in 1972, in which I first seriously considered the furor surrounding
the innate-acquired distinction. The second was my father, William Abell Wimsatt, who as a student, teacher, and
researcher in histology, reproductive physiology, and embryology at Cornell gave me paradigms of an information rich
environment, an (often over-)extended phenotype, and above all, an emotionally nurturing and supportive development in
which most of the constraints were highly generative. I only wish that each of them could have lived to argue with me
over this paper.
Wimsatt: Innate-Acquired
page 24
2/15/2016
Appendix A: Criteria for Innateness Arising From the Ethological Literature:
a. developmental fixity/canalization in various environments.
b. universality within species.
c. presence in (phylogenetically) related species.
d. appearance early in development or soon after birth.
e. simplicity of innate releasing stimuli (vs. complexity of conseq. behavior).
f. major malfunctions if innate features don't/aren't allowed to develop.
g. releaser or imprinting object elicits activity of imprinting mechanism.
h. teaching mechanisms are innate.
i. stereotypy of behavior (including insensitivity to experience).
j. parallels between behavior and phylogeny.
k. unusually easy, rapid, or one-shot learning.
l. critical periods for learning.
m. atavism or resistance to evolutionary change.
n. innate information phylogenetically acquired; acquired ontogetically acquired.
o. mendelian or other simple patterns of inheritance.
p. selectability (implies heritability).
q. physical or functional modularity of functional unit.
Bibliography:
Chomsky, Noam, (1967); "Recent Contributions to the Theory of Innate Ideas", Synthese, 17, pp. 2-11.
Fodor, Jerry, (1981); "The Current Status of the Innateness Controversy", in his Representations:
Philosophical Essays on the Foundations of Cognitive Science, Cambridge: M.I.T Press--Bradford Books.
Glassman, R. B. and W. C. Wimsatt, (1984); "Evolutionary Advantages and Limitations of Early Plasticity",
in S. Finger and C. R. Almli, eds., Early Brain Damage, Volume I: Research Orientations and
Clinical Observations, New York: Academic Press, pp. 35-58.
Gould, Stephen J., (1977); Ontogeny and Phylogeny, Cambridge: Harvard University Press.
Kauffman, Stuart A., (1969); "Metabolic Stability and Epigenesis in Randomly Constructed Genetic
Networks", Journal for Theoretical Biology, 22: 437-467.
Kuo, Zing-Yang, (1932); "Ontogeny of Embryonic Behavior in Aves, IV. The Influence of Embryonic
Movements Upon the Behavior after Hatching", Journal of Comparative Psychology, 14:109-122.
Lehrman, D. S., (1953); "A Critique of Konrad Lorenz's theory of Instinctive Behavior", Quarterly Review
of Biology, 28: 337-363.
Lehrman, D. S., (1970); "Semantic and Conceptual Issues in the Nature-Nurture Controversy", in L. R.
Aronson, et. al., eds. Evolution and Development of Behavior (a Festschrift for T. C. Schnierla), San
Francisco: Freeman, pp. 17-52.
Lewontin, R. C., (1970); "The Units of Selection", Annual Review of Ecology and Systematics, 1: 1-18.
Lewontin, R. C., (1974); The Genetic Basis of Evolutionary Change, New York: Columbia University
Press.
Wimsatt: Innate-Acquired
page 25
2/15/2016
Lewontin, R. C., (1978); "Adaptation", Scientific American, (September).
Lorenz, Konrad Z., (1941), "Kant's Categories in Light of Contemporary Biology", reprinted in English in
Yearbook of the Society for General Systems Research, vol. 7, (1959).
Lorenz, Konrad Z., (1965); Evolution and Modification of Behavior, Chicago: University of
Chicago Pre
Maynard-Smith, John, (1975); The Theory of Evolution, 3rd. ed., London: Pelican.
Mayr, Ernst, (1974); "Behavior Programs and Evolutionary Strategies", American Scientist, 62: 650-659.
McCleary, R. A., ed., (1970); Genetic and Experiential Factors in Perception, Glenview: Scott-Foresman.
Ospovat, D., (1976); "The Influence of Karl Ernst von Baer's Embryology, 1828-1859", Journal of the
History of Biology, 9: 1-28.
Putnam, Hilary, (1967); "The 'Innateness Hypothesis' and Explanatory Models in Linguistics", Synthese, 17,
pp. 12-22.
Simon, H. A., (1962); "The Architecture of Complexity", Proceedings of the American Philosophical
Society, reprinted as chapter 7 of Simon, (1981), pp. 193-229.
Simon, H. A., (1966); "Thinking by Computers", in R. G. Colodny, ed., Mind and Cosmos, Pittsburgh:
University of Pittsburgh Press, pp. 3-20.
Simon, H. A., (1981); The Sciences of the Artificial, 2nd. ed., Cambridge: M.I.T. Press.
Sober, Elliott, ed., (1984); Conceptual Issues in Evolutionary Biology, Cambridge: MIT Press--Bradford
Books.
Wimsatt, W. C., (1976); "Reductive Explanation: A Functional Account", in R. S. Cohen, C. A. Hooker, A. C.
Michalos, and J, van Evra, eds., PSA-1974, Boston Studies in the Philosophy of Science, vol. 32,
Dordrecht: D. Reidel, pp. 671-710. reprinted in Sober, 1984.
Wimsatt, W. C., (1980); "Reductionistic Research Strategies and their Biases in the Units of Selection
Controversy", in T. Nickles, ed., Scientific Discovery, v. II: Case Studies, Dordrecht: D. Reidel, pp. 213259. Reprinted in Sober, 1984.
Wimsatt, W. C., (1981); "Units of Selection and the Structure of the Multi-Level Genome", in R. Giere and P.
Asquith, eds, PSA-1980, v.II, East Lansing, Mich.: The Philosophy of Science Association, pp. 121-183.
Wimsatt, W. C., (1983), "Von Baer's Law of Development, Generative Entrenchment, and Scientific Change",
currently under revision for publication in Biology and Philosophy, (52 page typescript).