Published in Behavioral and
Brain Sciences
Volume 25, Number 4: 489-504 (August 2002)
© 2002 Cambridge University Press
Below is the unedited, uncorrected, unquotable final draft preprint of a BBS target article that was accepted for publication. To order the final published version of this target article, with commentaries and authors’ response, please visit the BBS Homepage at Cambridge Journals Online.
Adaptationism – How to carry out an exaptationist
program
Paul W. Andrews1,*
Steven W. Gangesad2,*
Dan Matthews3,*
1
Department of Biology,
e-mail: pandrews@unm.edu.
2 Department of
Psychology,
e-mail: sgangest@unm.edu
3 Department of
Psychology,
e-mail: danda@unm.edu
The
authors contributed equally to this paper. Order of authorship was determined
alphabetically. Correspondence may be addressed to any of the authors.
Word Counts: (1) Short Abstract
-- 128 words; (2) Long Abstract -- 233 words; (3) Main Text -- 15,002 words;
(4) References -- 2,383 words; (5) Entire Text -- 18,867 words
Keywords: adaptation, ADHD,
brain allometry, constraint, epistemology, evolutionary psychology, exaptation,
female orgasm, optimization, special design, waist-hip ratio
Short Abstract
Adaptationism is a research
strategy that seeks to identify adaptations and the specific selective forces
that drove their evolution in past environments. Adaptationism, especially as
applied towards understanding human behavior and cognition, has been the
subject of attacks by paleontologist Stephen J. Gould and geneticist Richard
Lewontin. Primarily, they argue that adaptationists often use lax standards of
evidence to identify adaptations and that they often fail to consider
alternative hypotheses to adaptation. In this article we discuss the possible
standards of evidence that could be used to identify adaptations and when and
how they may be appropriately used. We also discuss how the testing of
alternative hypotheses implicitly requires the testing of adaptationist
hypotheses. Where possible, we illustrate our points with examples taken from human
behavior and cognition.
Long Abstract
Adaptationism is a research
strategy that seeks to identify adaptations and the specific selective forces
that drove their evolution in past environments. Since the mid-1970’s,
paleontologist Stephen J. Gould and geneticist Richard Lewontin have been
critical of adaptationism, especially as applied towards understanding human
behavior and cognition. Perhaps the most prominent criticism they made was that
adaptationist explanations were analogous to Rudyard Kipling’s
"just-so" stories. Since story telling (through the generation of
hypotheses and the making of inferences) is an inherent part of science, the
criticism refers to the acceptance of stories without sufficient empirical
evidence. In particular, Gould, Lewontin, and their colleagues argue that
adaptationists often use inappropriate evidentiary standards for identifying
adaptations and their functions and that they often
fail to consider alternative hypotheses to adaptation. Playing prominently in
both of these attacks are the concepts of constraint, spandrel, and exaptation.
In this article we discuss the standards of evidence that could be used to
identify adaptations and when and how they may be appropriately used. Moreover,
building an empirical case that certain features of a trait are best explained
by exaptation, spandrel, or constraint requires demonstrating that the trait’s
features cannot be better accounted for by adaptationist hypotheses. Thus, we
argue that the testing of alternatives requires the consideration, testing, and
systematic rejection of adaptationist hypotheses. Where possible, we illustrate
our points with examples taken from human behavior and cognition.
1. Introduction
In the past decade, evolutionary
psychology has emerged as an important theoretical perspective in psychology.
Evolutionary psychology is a methodologically rich field that could be applied
to a variety of interesting questions (e.g., phylogenetic analysis of
psychological and behavioral traits). One approach receiving much attention in
recent years predominantly involves the application of adaptationism to
understanding the evolution and nature of human psychological design (Barkow et
al. 1992; Buss 1995; Ciba Foundation 1997; Pinker 1997a). Adaptationism, as a research
strategy, seeks to identify adaptations and to elucidate the specific selection
pressures that forged them in an organism’s evolutionary past. It has a long
history within evolutionary biology that, in its current form, crystallized in
the 1960’s and ‘70’s (particularly influenced by the writings of George
Williams [1966]) and now dominates the study of animal behavior in biology
(e.g., Krebs & Davies 1993; 1997). Adaptationists sometimes implement
optimization models (formal mathematical theories of selection pressures) to
decide whether a particular design serves some specific function (e.g., Parker
& Maynard Smith 1990). Perhaps as often, however, they use intuitive
arguments for how a particular feature must have served a goal responsible for
its evolution (Williams 1966).
Everyone agrees that organisms
have adaptations. Yet, adaptationism as a research strategy has not enjoyed
consensual affection within evolutionary biology. In the 1970’s, it became the
target of attacks by paleontologist Stephen Jay Gould and geneticist Richard
Lewontin (e.g., Gould & Lewontin 1979; Lewontin 1978; 1979). Perhaps the
most prominent criticism they made was that the explanations that
adaptationists gave for traits were analogous to Rudyard Kipling’s "just-so"
stories. Of course, the criticism is not against story telling in science per
se. The generation of hypotheses and the making of inferences is an inherent
part of science. Rather, the criticism refers to the acceptance of stories
without sufficient empirical evidence. Gould, Lewontin, and their colleagues
have made two important epistemological criticisms of the story telling that
adaptationists do. First, adaptationists often use inappropriate evidentiary
standards for identifying adaptations and their functions. Second,
adaptationists often fail to consider alternative hypotheses to adaptation.
Many have responded to the
criticisms of Gould and Lewontin (e.g., Alcock 1987; 1998; Alexander 1987;
Borgia 1994; Buss et al. 1998; Cronin 1991; Dawkins 1986; Dennett 1995; 1997;
Houston 1997; Maynard Smith 1978; 1995; Mayr 1983; Parker & Maynard Smith
1990; Pinker 1997b; Pinker & Bloom 1992; Reeve & Sherman 1993; Sherman
1988; 1989; Thornhill 1990; Thornhill & Palmer 2000; Tooby & Cosmides
1992; Wright 1997) and Gould has responded to at least some of these arguments
(e.g., Gould 1997a-d). Most recently, the debate between Gould and
adaptationists has been carried out in outlets intended for the lay public,
including exchanges about evolutionary psychology in the New York Review of
Books (Dennett 1997; Gould 1997a-d; Pinker 1997b; Wright 1997). Despite
emerging nearly a quarter-century ago, these debates persist with no consensual
resolution (though each side appears to think matters have resolved in their
favor). Few debates are more central to evolutionary biology and, in
particular, evolutionary psychology — the arena in which skirmishes have most
recently been staged. Our purpose is not to review the entire literature on
this debate. Rather, some confusion about the nature of the debate persists,
and we attempt to clarify the major issues. In particular, the major criticisms
of adaptationism advanced by Gould and Lewontin have been largely
epistemological in nature rather than ontological, a point not always
appreciated.
By way of background, we first
discuss traits and how they evolve (section 2). Next, we discuss the primary
goal of adaptationism — to determine whether traits are adaptations and, if so,
to determine the specific selection pressures that shaped them (section 3). In
this section we also discuss the problems with the different standards that
adaptationists could use (and sometimes have used) to classify traits as
adaptations and make inferences about the specific selective forces that shaped
them especially in light of the criticisms made by Gould and Lewontin. Playing
prominently in these attacks are the concepts of constraint and exaptation. A
constraint opposes the modifying influence of selection on the phenotype
whereas an exaptation is a pre-existing trait that acquires a new beneficial
effect without modification to the phenotype by selection. To Gould and his
colleagues, constraint and exaptation are prevalent and render inferences about
selective history using traditional adaptationist tools problematic. Thus, we
also discuss the ways in which even the best adaptationist evidentiary
standards can fail to identify adaptation (section 4). In the last major
section (section 5), we note that problems with story telling are not unique to
adaptationism. Gould and Lewontin insist that adaptationists consider
alternative hypotheses, but they have not provided any evidentiary criteria for
accepting the alternatives that they ask adaptationists to consider. In the
absence of rigorous evidentiary standards, exaptationist story telling is
just-so story telling. We argue that an adaptationist approach is crucial to
providing empirical support for the alternative hypotheses about trait design
that Gould, Lewontin, and their colleagues insist should be considered. Where
possible, we illustrate our points with examples about human behavior and
cognition.
2. The effects of traits and how
they influence trait evolution.
Biologists use the term
"trait" to refer to aspects of organisms’ phenotypes. The question of
what qualifies as a trait is not so straightforwardly answered as it might
seem, a point emphasized by Gould and Lewontin (1979). Because all aspects of
the organism’s phenotype are integrated with each other, organisms are
"not collections of discrete objects" (Gould & Lewontin 1979).
Genes often have pleiotropic effects (i.e., a single gene may influence many
aspects of the organism’s phenotype) and they often epistatically interact with
each other (i.e., an allele at one locus may influence the phenotypic expression
of an allele at another locus).
Nevertheless, biologists
interested in how an organism’s phenotype evolved are forced to discriminate
between aspects of the phenotype. A liberal definition would allow a trait to
be any aspect of the phenotype that can be discriminated on the basis of any
criterion — its causes, its effects, its appearance, etc. — and would include
dispositional traits (e.g., the disposition to develop callouses with
friction). The subset of such traits that could potentially qualify as adaptations
are those that have effects (a conceptualization that follows from
Williams 1966; see also Gould & Vrba 1982). An effect refers to the way (or
ways) in which an aspect of the phenotype interacts with the environment. This
approach does not imply that traits are completely genetically distinct from
each other, as two traits with very different effects may have common genetic
underpinnings. This is not a problem, however, because adaptationism is
concerned with how traits come into being because of the effects that they
have.
2.1 Should behavioral and
psychological phenomena be considered traits?
Behaviors and psychological
phenomena are often responses of the organism to aspects of the envirnonment.
They are not traits in and of themselves because they are not constructed from
genes or their products. Rather, they are effects of components of the nervous
system interacting with each other (e.g., emotional experience) or effects of
the nervous system interacting with the muscular-skeletal system (e.g.,
behaviors). However, behaviors and psychological processes are like traits in
that they produce effects of their own (e.g., the movement of a hand that
shapes the environment to create a tool), and these effects are often
functional. Throughout the paper we will speak of behaviors and cognitive
processes as if they were traits. But when we do so, we are implicitly
referring to the underlying decision-rules and information processing
algorithms encoded into the structure of the nervous system either through
genetics, learning, or some other process.
2.2 How the effects of traits
influence their evolution.
Traits evolve as the genes from
which they develop evolve. Genes evolve from any one of four evolutionary
forces — mutation (the original source of all genetic variation), migration,
drift or chance, and selection (often partitioned into natural selection and
sexual selection). An effect influences the evolution of a trait if it either
enhances or inhibits the replicative success of the genes from which it
develops. Thus, selection results in modifications to the phenotype by virtue
of the differential effects on replicative success that are generated by
allelic variation.
2.2.1 Some beneficial effects
drive a trait’s evolution whereas others do not.
The word adaptation has
two meanings in evolutionary biology (Gould & Vrba 1982). It refers to the
process by which natural selection modifies the phenotype and generates traits
whose effects facilitate the propagation of genes. It also refers to the
endproducts of that process — i.e., the traits that have been constructed by a
process of phenotypic modification by natural selection for a particular
gene-propagating effect. The effect that causes the trait to evolve is called
the function of the trait.
Gould and Vrba (1982) were the
first to define and discuss the concept of exaptation. An exaptation is
a pre-existing trait (i.e., one that has already evolved) that acquires a new
beneficial effect without being modified by selection for this effect (i.e., it
takes on a new role but was not designed for it by selection). Because the
beneficial effect did not contribute to the trait’s evolution, the effect the
trait is exapted to is not a function but just an effect:
"Adaptations have functions; exaptations have effects" (Gould &
Vrba 1982, p. 6).
Modification of the phenotype is
essential to the concept of adaptation. Natural selection cannot bring about
adaptation (the process or the endproduct) without the changes that new genes
make to the phenotype. However, for a trait to become exapted to a new
beneficial effect, it must have acquired it without being phenotypically
modified by selection for the effect. This point is not always appreciated. For
instance, in an article explaining the differences between adaptations,
exaptations, and spandrels, Buss et al. (1998) stated, "Selection is
necessary ... to explain the process of exaptation itself. Selection
is required to explain the structural changes in an existing mechanism that
enable it to perform the new exapted function" (p. 542). If a trait
undergoes a process of structural modification to facilitate a new beneficial
effect, it has undergone a process of adaptation and the resultant structural
changes are referred to as adaptations. Gould and Vrba (1982) are clear on this
point. They refer to an initially exapted trait as a primary exaptation
and any subsequent adaptive structural modifications as secondary
adaptations.
Some traits are complex, meaning
that subcomponents can be discriminated and interact in ways to produce
effects. The hand is a complex trait, one that has particular effects (e.g.,
grasping) by virtue of the organization of subtraits (e.g., fingers, bone
structure, musculature that permit grasping). Technically, complex features are
probably mixtures of exaptations and secondary adaptations. With regard to the
skeletal structure and musculature of land-living vertebrates, "The order
and arrangement of tetrapod limb bones is an exaptation for walking on land;
many modifications of shape and musculature are secondary adaptations for
terrestrial life" (Gould & Vrba 1982, p. 12). Naturally, one expects
that the finer details of a complex feature that are most subject to secondary
modification are those that do not serve the new exapted effect well: "Any
coopted effect (an exaptation) will probably not arise perfected for its new
effect. It will therefore develop secondary adaptation for the new role. The primary exaptations and secondary adaptations can, in
principle, be distinguished" (Gould & Vrba 1982, p. 13).
2.2.2 The genesis of
exaptations
There are two scenarios under
which a trait may become an exaptation. In the first, the trait initially
evolves as an adaptation for a particular effect, and then subsequently becomes
exapted to another effect (Gould & Vrba 1982). For example, feathers may
have evolved initially for their insulation properties rather than for flight
(Gould & Vrba 1982). Nevertheless, many of the feathers on a bird (such as
wing and tail feathers) have been modified specifically for flight and so
represent at least secondary adaptation for flight. Other feathers on a bird do
not exhibit obvious modification for flight. Contour feathers are surface
feathers covering all parts of the body except for the wings and the tail (Gill
1990). At the proximal end of the feather, close to the skin, they have soft,
plumaceous, fluffy barbs that suggest special design for trapping air that has
been heated by the body and keeping it close to the skin. Toward the distal end
of the feather that is exposed to the air, the barbs form a relatively cohesive
flat surface. This feature may contribute both to heat insulation (by
protecting the underlying thermal layer from being disturbed by wind) and
flight facilitation (by reducing drag while flying). If contour feathers have
not been modified by selection specifically for facilitating flight, they may
be pure exaptations to flight as well as adaptations for insulation.
Under the second scenario by
which exaptation can occur, the trait is a byproduct of selection for another
trait. The byproduct evolved, not because it was selectively advantageous, but
because it was inextricably linked (either through pleiotropy or linkage
disequilibrium) to another trait that was reproductively advantageous. Such
traits, called spandrels (Gould & Lewontin 1979), can subsequently
become exapted to new beneficial uses. For example, some species of snails have
a space in their shell that they use to brood eggs (Gould 1997e). The space
exists even for snails that do not use the space and is presumably a necessary
consequence of a plan for shell development that was the product of selection.
Those snails that use the space for egg brooding apparently evolved after ones
not using it. The space then appears to qualify as a spandrel that was later
exapted to brooding eggs.
Systematic processes rather than
mere coincidence may lead to exaptation. As argued by Lewontin (1983),
organisms not only adaptively respond to adaptive problems posed by autonomous
environments; they also construct them (see also Laland et al. 2000). One of
the ways by which organisms can adaptively create and successfully move into
new niches is by exapting existing structures–putting old features to new uses.
In this view, birds were able to move into a world of flight precisely because
they possessed structures that could be exapted (and only subsequently,
adapted) for flying. Similarly, modern humans may live in a world very
different in many ways from the ancestral ones that shaped us, but it is far
from coincidence that we fit this new world in interesting ways. Thus, although
we did not evolve in a world in which transportation involved driving vehicles
at high speeds, we would not now live in such a world did we not possess
features that could be exapted for driving. As recognized by Mayr (1963),
shifts into new adaptive zones are often behaviorally led, with secondary
adaptation of exapted morphological and other structures following behind. From
this perspective, exaptations may not be rare fortuitous "accidents,"
but rather regular occurrences. (Parenthetically, however, we emphasize that
organisms may also possess specific features highly maladapted to the new
environments they construct [e.g., Daly & Wilson 1999]. Thus, modern humans
would perhaps survive longer were they to exhibit less of a preference for
fatty foods.)
2.3 Constraints limit the
phenotypic outcomes of selection.
A constraint opposes the
modifying influence of a selective force on the phenotype. In the absence of
constraints, directional selection will continuously modify the phenotype over
evolutionary time and there will be no stable phenotypic outcome of the
selective process. To oppose the effects of selection on phenotypic
modification, a constraint either must limit the phenotypic outcomes that
alleles could produce, or it must be an opposing evolutionary force.
Physical laws are examples of
constraints that limit the possible outcomes that alleles could produce. No
allele could ever arise that will allow an organism to have zero mass or
violate the laws of conservation of mass or conservation of energy. Also, the
epistatic effects of genes may limit the suite of possible phenotypic outcomes.
For example, since a new allele arises in the context of a pre-existing genome,
the range of possible phenotypic modifications that a new allele could produce
may be limited by the prior evolutionary history of the organism.
A selective force on a trait may
constrain other selective forces on the same trait if they have opposing
effects. For example, selection favors large clutch size in birds because
larger clutch size will increase fitness in the absence of an opposing
selective force. But because parents find it difficult to raise all offspring
from a large clutch to weaning, there is an opposing selective force favoring
smaller clutch sizes. Actual clutch sizes should then be influenced by the
tradeoffs between these two selective forces (see e.g., Seger &
Stubblefield 1996).
A selective force on one trait
may also indirectly constrain a selective force on another trait if the traits
are inextricably tied to each other. For instance, when a new mutation arises,
it arises in a genome that has been subject to a long history of selection. As
such, much of the genome will be highly conserved because it results in
advantageous phenotypic effects. It is possible that the only new mutation that
could result in a given beneficial trait also interacts with the existing
genome to produce costly effects that outweigh the beneficial effects.
Selection will then disfavor the evolution of the new trait and the design of
the organism will be constrained. This is referred to as a genetic
constraint because there is no possible mutation that favors the new trait
within the context of the existing genome. A genetic constraint should be
understood as a selective tradeoff between the new mutation and the existing
genome. Because the advantages afforded by the pre-existing genome outweigh the
beneficial effects of the new mutant, the new mutant cannot evolve, and the trait
is constrained from reaching optimal design for its function.
A particular form of a genetic
constraint is a developmental constraint. The construction of an
organism through the developmental process depends on the coordinated action of
many different genes. It is possible that a new mutation could only code for a
beneficial new trait by interfering with the developmental process, thereby
disrupting the development of the rest of the organism. If the costs of
developmental disruption outweigh the advantages provided by the new mutant,
the new mutant will be disfavored and the trait’s design will be constrained.
2.4 The concept of
evolvability
A concept related to genetic
constraint is evolvability. Evolution by natural selection can occur
when new mutations can possibly lead to fitter phenotypes.
"Evolvability" is the genome’s ability to produce adaptive variants,
which depends on the mapping of genotypes to phenotypes. While genetic
constraints entail mappings of genotypes (both actual and potential, through
mutation) to phenotypes that prevent evolution toward certain phenotypic
configurations, evolvable genetic systems are those that allow incremental,
stepwise improvement. A key feature is that further improvements in one part of
the system must not compromise past achievements. Modularity of
genotype-phenotype mapping functions — a relative absence of pleiotropic
effects of gene action — therefore facilitates evolvability (Wagner &
Altenberg 1996).
Kirschner and Gerhart (1998;
Gerhart & Kirschner 1997) argue that developmental systems that display
versatility also have high evolvability. Versatility is the ability of the
development plan to be open to new adaptive possibilities. Kirschner and
Gerhart argue that certain simple developmental processes (e.g., those giving
rise to a common body plan, or Bauplan, within a taxonomic grouping) are
conserved because they allow for versatility while also giving rise to robust
developmental outcomes. Perhaps ironically, these processes themselves
constrain evolutionary outcomes. Kirschner and Gerhart argue that constraint is
inevitable, and that these simple constraining
processes are conserved "because they deconstrain phenotypic variation in
other processes, and hence facilitate evolutionary change" (p. 8426). (See also West-Eberhard 1998.)
Evolvability may not be selected
for at the individual level, for its benefits are in a currency of future
evolutionary adaptation, and selection cannot anticipate what traits are likely
to be beneficial in the future. Within a species, individual selection should
favor increasing specialization of traits because specialized traits will
usually outperform more generalized ones (Symons 1992). Possibly, evolvability
(e.g., in the form of a conserved developmental plan) is selected as a
byproduct of the other, adaptive evolutionary changes it allows. In this view,
conserved developmental plans are not adaptations for evolvability, but their
versatility makes it more likely that they will be exapted to new uses and at
the same time makes it easier for selection to build adaptations for these new
uses. (Interestingly, in their seminal article on exaptation Gould and Vrba
[1982] make a similar claim about repetitive copies of DNA, whose existence
allows for [and is exapted for] a flexible future [evolvability] but whose
existence cannot be due to its role as such.) Alternatively, evolvability may
be subject to clade selection (Williams 1992). That is, even though there
should be an increasing tendency towards specialization for species within
existing niches, those taxa that happen to maintain more versatile, more
evolvable developmental plans may be more effective at entering and exploiting
new niches precisely because they are most open to new adaptive possibilities.
3. The possible evidentiary
standards for identifying adaptation and function
The goal of adaptationism is to
determine whether traits are adaptations (Mayr 1983). To classify a trait as an
adaptation is to identify its function (Thornhill 1997; Williams 1966). To
identify a trait’s function is to determine the specific selection pressures
(if any) that were at least partially responsible for the evolution of the
trait.
Over the years, Gould, Lewontin,
and their colleagues have argued that adaptationism is not only a flawed
methodology for understanding the outcomes of evolution in general, but even
for understanding the specific outcomes of its core concern, selection (e.g.,
Gould 1984; 1987; 1989a,b; 1991a; Gould & Lewontin 1979; Gould & Vrba
1982; Lewontin 1979; 1983). Their arguments all involve a similar complaint. Adaptationism
is built on a view of evolution that overemphasizes the power of selection and
under-appreciates the constraints on selection and other evolutionary processes.
They do not deny that selection is responsible for workable design. Gould
(1997d) acknowledges that "natural selection is the only known cause of
eminently workable design" and that "adaptive design must be the
product of natural selection" (p. 57). Nor does Gould deny that natural
selection is the primary force responsible for evolutionary change (Gould
1984). Rather, the point is that factors other than selection can lead an
adaptationist to misunderstand the selective processes that gave rise to the
trait. Thus, the explicit object of Gould’s attacks on adaptationism is its
attempt to make inferences about the specific selective forces that shaped a
trait over evolutionary time (Gould 1991a; Gould & Lewontin 1979; Gould
& Vrba 1982).
There are two inferential errors
that adaptationists can make when attempting to identify adaptations and their
functions. First, they can infer that a trait is an adaptation for a proposed
function when it is not. Second, they can infer that a trait is not an
adaptation for a proposed function when in fact it is. Both sides in the debate
agree that the worse error is to classify a trait as an adaptation when in fact
it is not (Gould & Lewontin 1979; Gould & Vrba 1982; Thornhill 1990,
1997; Williams 1966). The point of disagreement centers around
the probative value of the evidentiary standards that adaptationists use to
classify a trait as an adaptation.
In particular, Gould and
Lewontin (1979) have argued that adaptationists use mere consistency
with adaptationist hypotheses as evidence for function. As such, adaptationists
often fail because mere consistency does not test the relative likelihood of
alternatives.
We
would not object so strenuously to the adaptationist programme if its
invocation, in any particular case, could lead in principle to its rejection
for want of evidence. We might still view it as restrictive and object to its
status as an argument of first choice. But if it could be dismissed after
failing some explicit test, then alternatives would get their chance.
Unfortunately, a common procedure among evolutionists does not allow such
definable rejection…. [T]he criteria for acceptance of a story are so loose
that many pass without proper confirmation. Often, evolutionists use
consistency with natural selection as the sole criterion and consider their
work done when they concoct a plausible story. But plausible stories can always
be told. [Gould & Lewontin 1979, pp. 587-588]
Actually, no adaptationist has
ever suggested that mere consistency should be the standard of evidence used to
identify function. Rather, the criticism seems to be that the evidentiary
standards used by adaptationists are, in reality, no better than mere
consistency. Below, we discuss various possible standards and argue that the
last of them, special design, is clearly better than mere consistency.
3.1 Six possible evidentiary
standards for identifying adaptation.
3.1.1 Standard 1: The
comparative approach.
Phylogenetic comparisons seek to
demonstrate a correlation between trait variation and the environment among a
large number of related species in a way predicted by a selective argument
(Leroi et al. 1994; Martins 2000). Some biologists have argued that
phylogenetic comparisons are necessary to any argument for adaptation (Larson
& Losos 1996). In its simplest version, the comparative approach suffers
from the problem of inferring causation from correlation. Various methods have
been suggested to address the causation issue, but phylogenetic analyses, by
themselves, only provide weak evidence of adaptation at best (Martins, 2000).
Still, there appears to be a growing consensus that they are useful when used
in conjunction with other approaches, a point to which we will return (see
section 3.6).
More importantly for our
purposes, use of phylogenetic comparisons is problematic when making inferences
about adaptation within a single species (see also Thornhill 1997). The
existence of a correlation between trait variation and environmental
circumstances for a large number of species cannot be used to conclude that the
trait variant for a particular species in the data set has been
influenced by a particular selective force. Moreover, the comparative approach,
by itself, cannot be used to classify traits that are unique to a single
species because it requires data on a large number of species to make
statistical tests. Thus, it cannot classify any uniquely human trait
(morphological, psychological, or behavioral) as an adaptation. The next five
standards attempt to identify adaptations within a single species.
3.1.2 Standard 2: Fitness
maximization.
One standard that has been
advocated by some adaptationists is that an adaptation is a trait that, among a
suite of variants, maximizes fitness in a particular environment (Reeve &
Sherman 1993). However, there are several problems with this standard (for
reviews, see Symons 1992; Thornhill 1997). First, it requires the scientists to
actually measure the fitness of organisms over time. This is a problem because
selection is a statistical process. Fitness varies temporally such that a measurement
of fitness over a particular period of time may not reflect statistical trends
over evolutionary time.
Second, the standard fails
because it does not incorporate the notion that adaptations only maximize
fitness in the environments in which they evolved. The environment in which an
adaptation evolved and the modern environment that it is currently in may be
very different from each other, although this is not always the case. If the
two environments are meaningfully different from each other, then this standard
would leave scientists without any means for determining whether or not a trait
is an adaptation because it is impossible to directly measure fitness in
ancestral environments.
Even if these other problems
could be solved, the fitness maximization standard does not allow one to
determine the function of a trait. The standard could allow one to determine
which variant that selection is currently favoring, but by itself it gives
little insight into what the trait does such that selection is favoring it.
Many adaptationists have long known that they need some standard that allows
them to make an inference about the specific effects that drove the trait’s
evolution by selection in ancestral environments (e.g., Symons 1992; Thornhill
1990; 1997; Williams 1966).
3.1.3 Standard 3: Beneficial
effects.
One possible standard for making
an inference about function is whether the trait has any effect that would have
been beneficial in ancestral environments. However, this standard fails because
it is also possible that traits that had beneficial effects in ancestral
environments were exapted to those effects. For instance, Singh (1993a,b; 1994a,b; Singh & Luis 1995) has reported that men
find women who exhibit a waist-hip ratio (WHR) of 0.7 or less to be more
attractive than women who exhibit a waist-hip ratio of 0.8 or higher. Whether
WHR is a real component of men’s mating preferences (Tassinary & Hansen
1998), or possibly a contingent preference varying with ecology (Marlowe &
Wetsman, 2001; Wetsman & Marlowe 1999; Yu & Shepard 1998), has recently
been called into question. For instructional purposes only, we assume
throughout the paper that it is a real preference. If so, the preference could
have evolved for any number of reasons: women who have lower WHR’s tend to have
fewer health problems, are young and have greater reproductive value, are more
fertile, are less likely to be pregnant, and may be less likely to have an
infectious disease. It is quite possible that the preference could have evolved
as an adaptation for one of these effects and was exapted to the remaining
effects. Using the beneficial effect standard would lead one to the conclusion
that the trait was an adaptation for each of these effects.
The next three standards attempt
to infer a trait’s function by examining its features in relation to its
effects. The idea is that selection often leaves its mark on traits when it
designs them to perform functions. Thus, it should often be possible to
reconstruct the selective history of a trait by examining the features of the
trait in relation to what it does. Arguments from design have been described as
projects of "reverse engineering" (e.g., Dawkins 1982; Dennett 1995).
In building a piece of machinery to solve a particular problem, an engineer thinks
about what kind of design would solve the problem efficiently and economically.
The evolutionary biologist is faced with the reverse task. He or she is looking
at a trait that is the product of evolutionary forces. If the trait was
produced by selection, it has already been "designed" for a special
purpose. The goal of reverse engineering is to figure out what nature designed
the trait for.
3.1.4 Standard 4: Optimal
design.
Adaptationists often use
optimization models to analyze traits. An optimization model quantitatively
models the selection pressures on a particular trait or suite of traits (Seger
& Stubblefield 1996; Winterhalder & Smith, 2000). The model has one or
more actors (e.g., a gene, a plant, two or more individuals engaging in
social interaction, etc.) expressing the phenotypes that the theoretician is
trying to understand. The payoffs of the model are expressed in a currency,
such as actual fitness units or some correlate of fitness (e.g., units of
energy). The goal of the optimization model is to maximize the actor’s
net benefit as measured by the currency. The decision set is the suite
of phenotypic or behavioral options available for pursuing the goal, and the selective
constraints delineate how these options are translated into costs and
benefits. The optimal phenotypic or behavioral option is the one that satisfies
the goal. Complexities can be taken into account, such that optimal strategies
may be contingent on the relative frequencies of each strategy in the
population (frequency-dependent optima) or the condition or phenotype of the
individual (conditional or phenotype-limited optima). (See
Parker & Maynard Smith 1990 for more discussion.)
Optimization models may allow
either a single trait to evolve (atomistic models) or allow multiple
traits to evolve simultaneously (coevolutionary models). Atomistic
models do not necessarily neglect other traits of the organism. Rather, they
may make assumptions about other traits and treat them as inputs in the model,
as in models of sex ratio evolution in which the genetic system itself is a
trait that exerts a selective force on sex ratio (e.g., Charnov 1982). A
coevolutionary model looks at a larger chunk of the organism than an atomistic
model. However, even in coevolutionary models, only a small number of traits
are ever really allowed to coevolve. In coevolutionary models, traits may drive
the evolution of each other, as in some models of signaler-receiver
interactions (e.g., Grafen 1990). Often, traits will exert opposing influences
on each other and force selection to make tradeoffs between them, as in some
life history models (e.g., Stearns 1992).
3.1.4.1 Design arguments
based on atomistic optimization models.
In a sense, atomistic models
presuppose that selection builds traits in the same way that an engineer would
design a piece of machinery to perform a task. They can then be used to make
predictions about how traits should be designed if they were to perform their
functions optimally. A reasonable fit with these expectations is taken as
evidence that selection designed the trait to solve the problem.
However, Gould and Lewontin
(1979) have argued that the underlying premise that selection works like an
engineer is flawed. Evolution does not result in solutions to problems similar
to what an engineer would achieve. Rather, adaptations are jerry-rigged
solutions. Whereas an engineer would be sure to specify the steps of
construction of an optimal piece of machinery to achieve a particular end,
natural selection adds features in unplanned steps. The analogy between human
engineering and organic selection is therefore flawed. In the words of Jacob
(1977), "Selection does not work like an engineer. It works like a
tinkerer" (p. 1163).1
By considering traits in
isolation to each other, adaptationists merely consider how a trait would
optimally perform a particular function. The tinkering process of phenotypic
modification could still yield traits that are optimally designed to perform
their functions in the absence of genetic constraints. But the organism is not
a blank slate on which new traits can be constructed. New alleles arise in the
context of an existing genome. If the new alleles that could
give rise to an optimally designed trait would interact with the existing
genome to produce costly effects, then the tinkering process may actually favor
alleles that produce less costly effects but would build a less than optimally
designed trait. An atomistic approach causes adaptationists to neglect
how genetic constraints force selection to make design tradeoffs between traits
(Gould & Lewontin 1979).
If selection should be thought
of not as an engineer but rather as a tinkerer, the evolutionary biologist
confronted with understanding the outcomes of selection faces a task not of
reverse engineering but rather of "reverse tinkering." The existing
genome may impose genetic constraints on the body plan such that trait design
may be far from the predicted optimum and optimization models will be impotent
to explain phenotypic outcomes (Gould & Lewontin 1979).
For example, Gould (1989a) has
argued that the shell shapes in the West Indian land snail Cerion are
constrained by an allometric relationship between whorl number and whorl size;
the larger the whorls, the fewer there are. This constraint has implications
for relationships between whorl size and shell shape; shells with larger whorls
tend to be squatter. An adaptationist account of why shells with larger whorls
should be designed to be squatter would fail to account for them, for this
association apparently has no adaptive value. Perhaps more importantly, the
constrained relationship between whorl width and shape may prevent optimal
designs (whatever they may be) from evolving. Human brains may be subject to
similar allometric constraint (Finlay & Darlington 1995). Brain features
may be forced to evolve together in developmentally constrained ways, with
their structure nonoptimal as a result.
Predictions about optimal trait
design will often fail to identify adaptations because optimality is too conservative
a standard. Due to genetic constraints, there is probably no adaptation that
(when examined closely enough) exhibits optimal design for its function. For
instance, the vertebrate eye is a marvel of machinery for processing light
information. Yet even the eye exhibits a flaw — the optic nerve attaches to the
front side of the retina such that there is a blind spot in the vertebrate eye
(Dawkins 1986). This is because the wiring from each photocell leaves the cell
from the side that is nearest the light such that the wiring interferes with
the path of light. The blind spot exists at the place where all the wires
aggregate into a bundle to exit the retina. The evolutionary reason for this
peculiar design appears to be an historical accident in which the earliest
photocells randomly oriented in this "backwards" fashion. Now, the
vertebrate eye is constrained from reaching a better design because it would
entail the evolution of intermediate forms that would leave the organism worse
off than it currently is.
3.1.4.2 Design arguments
based on coevolutionary optimization models.
If the scientist is familiar
with all of the precise developmental and genetic constraints imposed on the
possible solutions, he or she may be able to see that the solution is optimal relative
to all other possible solutions given the constraints. In principle,
coevolutionary models have the potential to remedy the concerns of Gould and
Lewontin by incorporating genetic and developmental constraints. Unlike
atomistic models, coevolutionary models allow theoreticians to make predictions
about how selection would optimally make tradeoffs between traits if the actor
were subject only to the selection tradeoffs included in the model. Traits, as
outcomes of coevolutionary models, are not necessarily designed to optimally
perform their functions. Rather, the optimization parameter is how the actor
maximizes fitness by trading off the design features of one trait against those
of another trait.
However, in practice, there are
two problems that limit the incorporation of such constraints into such models.
First, the mathematics becomes increasingly difficult to solve as the number of
constraints increases. Second, the scientist often has little or no a priori
understanding of how organismal design is integrated to generate constraints on
the evolution of traits. These constraints arose due to the historically
contingent events of the tinkering process (e.g., what mutations happened to
arise, what features evolved first, what evolved traits were later exapted for
other purposes) that the scientist has no clear view of. Thus, coevolutionary
optimization models will usually fail to fully explain trait design because,
even if adaptationists are able to include some genetic constraints in their
models, they will be unable to include all the organism’s traits and all the
genetic constraints acting on them.2
3.1.5 Standard 5: Tight fit.
In part, selection chooses among
variants of a trait on the basis of how well they facilitate a particular
gene-propagating effect. Often, this process generates a tight fit between the
features of a trait and its function. Thus, it is often said that a tight
relationship between a trait’s features and some problem or opportunity in the
environment is demonstrative of function (e.g., Cosmides & Tooby 1995). For
instance, there is a tight fit between the features of the eye and its function
of sight (Williams 1966).
There are two reasons why tight
fit may not be sufficient to establish adaptation. First, as noted earlier, many
organisms have been under selection to modify their environments in ways that
allow them to use their pre-existing traits in novel ways (Dawkins 1982; Laland
et al. 2000; Lewontin 1983). For instance, the hand fits very well inside a
glove, yet this mere fact cannot be taken as evidence that hands evolved to fit
inside the glove. This is precisely the sort of erroneous conclusion that the
tight fit standard could lead us to make. Selection is responsible for the fit
between trait and effect, but selection has not modified the hand to fit inside
the glove. Rather, selection on some other aspect of the phenotype has given
human beings the ability to modify the environment in a way that protects or
insulates the hand.
Second, fit between behavioral
performance and an adaptive problem can arise because of learning. Learning is
a process in which feedback from the environment modifies the neurological
structures that give rise to behavior and cognition. Learning mechanisms are
themselves adaptations that allow the organism to adaptively modulate behavior
with changing environments. As adaptations they have functions (e.g., to learn
a language, to fear a predator, to get along with others, etc.). However, by
their very nature, learning mechanisms are somewhat flexible with respect to
outcome. It is possible that a learning mechanism can be so flexible that it
can develop behavioral and cognitive traits that perform tasks that are not the
function of the mechanism. For instance, being able to drive a car or play the
stock market must in some sense represent the output of learning mechanisms
that evolved for other purposes. Moreover, neural network models suggest that a
single learning mechanism may be able to generate different cognitive
mechanisms each of which exhibits good design for performing a different task
(e.g., Kruschke 1992). For these situations, the learning mechanism has been
exapted to a new problem and so we refer to the trait as the output of an exapted
learning mechanism (ELM).3 Thus, the tight fit standard is
consistent with adaptation, but it is also consistent with the possibility that
the fit between trait and effect was the result of environmental modification
or generated by an ELM (see Gould & Lewontin 1979, for a similar point).
3.1.6 Standard 6: Special
design.
The leading evidentiary standard
for inferring function from the analysis of a trait’s features in relation to
its effects is special design (Symons 1992; Thornhill 1990; 1997; Tooby &
Cosmides 1992; Williams 1966). Sometimes the special design standard appears to
be a pre-specified list of criteria that must be satisfied (e.g., specificity,
proficiency, precision, efficiency, economy, reliability of development,
complexity of design, etc.). Satisfaction of these criteria is surely sufficient
to demonstrate that a trait has been designed by something to perform a
task. But if we are to take the lessons of neural network modeling to heart,
these criteria are also consistent with behavioral and cognitive traits that
develop from an ELM that evolved for another purpose. With sufficient feedback
from the environment, traits that develop from ELM’s can come to exhibit
specificity, proficiency, and even complexity of design for performing a task.
Moreover, a research strategy
that attempts to demonstrate adaptation and function by the satisfaction of a
pre-specified list of criteria misunderstands the burden of proof. Williams
(1966) wrote the leading account of how to demonstrate adaptation and function
from the features of traits and their effects. Rather than proposing a
pre-specified list of criteria, Williams (1966) advocated an approach in which
the scientist makes an inference of adaptation and function only after
demonstrating that all alternative hypotheses to adaptation for a particular
effect are highly unlikely as complete explanations for the trait.
Demonstrating adaptation, Williams argued, carries an onerous burden of proof.
Moreover, "This biological principle [adaptation] should be used only as a
last resort. It should not be used when less onerous principles … are
sufficient for a complete explanation" (p. 11). Williams did suggest
qualities of trait design that could help build a case for adaptation (e.g.,
precision, efficiency, economy) and claimed that formulation of "sets of objective
criteria [of special design]" is a matter of "great importance"
(p. 9). Yet he himself applied only an informal probability standard:
"whether a presumed function is served with sufficient precision, economy,
efficiency, etc., to rule out pure chance [i.e., any possibility other than
adaptation for a particular effect] as an adequate explanation" (p. 10,
brackets added).
There may be no uniform list of
criteria that must be satisfied to demonstrate that a trait has been
specifically designed by selection for a function. Different traits may require
satisfaction of different criteria. Nevertheless, it would be useful to have
some guidelines about the sorts of criteria that can help build a case for
adaptation. For instance, it is difficult to see how the function of a trait
could be elucidated if it did not perform its function with specificity and
proficiency and, hence, these criteria appear to be necessary components of a
special design argument. If a trait’s features produce multiple effects with
equal proficiency, then it will be difficult (if not impossible) to determine
which effect (if any) drove the trait’s evolution. Within evolutionary
psychology, specificity and proficiency interact in the concept of domain
specificity. A cognitive mechanism exhibits domain specificity if it is
good at processing information relevant to certain problems, but not other
problems to which the mechanism might be applied. Domain specificity is often
demonstrated by showing that certain stimuli facilitate the performance of a
cognitive mechanism and other stimuli do not (see, e.g., Cosmides & Tooby
1992).
As argued above, however,
specificity and proficiency merely demonstrate a good (or tight) fit between
the trait’s features and an effect of the trait. By themselves, they do not
test whether the fit could have been caused by modification of the environment
to fit the trait, or whether the trait is the developmental output of an ELM.
The scientist may then want additional evidence demonstrating that the trait’s features
have been phenotypically modified by selection for the proposed function.
For morphological (i.e.,
non-neurological) traits, it is often sufficient to demonstrate that the trait
also exhibits complex design for the proposed function. These traits cannot
have developed from an ELM. For behavioral and cognitive traits, however, the
lesson of neural network modeling is that ELM’s may be able to generate
behavioral and cognitive traits that perform tasks with specificity and
proficiency and these traits may even exhibit complex design for those tasks
(e.g., Kruschke 1992). Fortunately, there are several forms of evidence that
could, along with specificity and proficiency, bolster an argument that the
trait’s features have been constructed by selection for the proposed function.
3.1.6.1 The role of
developmental specificity and biased learning in testing adaptationist
hypotheses.
For behavioral and cognitive
traits, adaptationists sometimes build arguments for adaptation by showing that
the trait is the biased outcome of a developmental or learning mechanism
(Cummins & Cummins 1999). Biased outcomes indicate that the mechanism is
biologically prepared (sensu Cummins & Cummins 1999; Seligman
1971) to produce the trait relative to other traits it could produce. When a
trait exhibits developmental specificity (or learning specificity), it suggests
that the function of the mechanism is the biased outcome. It can be
demonstrated by showing that the trait develops (or is learned) easier, more
reliably, or serves its function with greater proficiency than other traits
that could arise from the same mechanism.
Examples of psychological traits
for which adaptationists have invoked the criterion of developmental
specificity are language facility and intuitive ontologies. Pinker (1994) has
argued that the ease with which children learn new words and the ways by which
they generate syntactic structure indicate a biological preparedness and,
hence, adaptation for language learning. Others have argued that the specific evidence
that Pinker cites does not firmly establish the nature of the developmental
specificity for all aspects of language and suggest that more general learning
capacities play important roles, but appear to accept the general criterion of
developmental specificity (e.g., Gomez & Gerken 2000). The work of
developmental psychologists (e.g., Baillargeon 1987; Spelke 1990) suggests that
humans have an "intuitive physics," a set of expectations about the
physical world that reliably develop, on the basis of work suggesting that
learning from specific instances cannot account for infants’ perceptual
expectations. Keil (1994) has argued for an intuitive biology on the basis of
similar reasoning, though he cautions that the evidence for developmental
specificity is not yet fully compelling (see also Atran 2000).
While intuitive ontologies often
involve learning, they also consider the timing of development of learning
capacities. Most learning bias experiments do not take into account the
development of learning capacities. For this reason, some learning biases could
be the result of prior learning history (e.g., learning algebra first may make
it easier to learn calculus). Still, it is possible to devise experiments that
are difficult for ELM hypotheses to explain. In one important experiment rats
were able to associate a sound with an electric shock but were unable to
associate it with nausea. Similarly, rats were able to associate a taste with
nausea but not with an electric shock (Garcia et al. 1974). This demonstrated
that the learning mechanism involved in the perception of taste is biased
towards providing information about the quality of food, and the learning
mechanism involved in audition is biased towards providing information about
external threats. Because the experimenters used novel stimulus-punishment
associations, it is not immediately clear how any prior learning history could
have caused the biased learning patterns that the rats demonstrated.
3.1.6.2 The trait’s features
exhibit a good fit with a proposed ancestral environment, but exhibit more of a
mismatch in the modern environment in which the trait develops.
Often, it is impossible for
scientists to perform the experiments needed to test developmental specificity.
In that event, it is sometimes possible to make inferences about developmental
specificity from how the features of the trait interact with the environment.
For example, if a particular behavioral or cognitive trait is the output of an
ELM, it will have developed in response to modern environmental input. If such
a trait exhibits specificity and proficiency for a task, it will do so in
modern environments.
However, if the trait is an
adaptation, it will exhibit specificity and proficiency when in its
evolutionary environment. Sometimes the pertinent aspects of the modern
environment are very similar, if not identical, to the environment in which a
trait evolved. If so, one would expect that the trait’s features to match the
modern environment very well if it is an adaptation. If the modern environment
is different in some pertinent way from the proposed evolutionary environment,
the prediction is that the trait will fit better with the evolutionary
environment than the modern environment if it is an adaptation.
For instance, people often
experience a craving for foods that are high in sugar and fat, and these
preferences are particularly robust in small children (for a review, see
Drewnowski 1997). The sweet tooth exhibits specificity and proficiency for
motivating people to seek out such foods or to choose these foods when given a
choice. This evidence by itself, however, is not enough to demonstrate that the
sweet tooth’s evolved function was to motivate people (perhaps particularly
children) to eat foods that are rich with sugar and fat. Food preferences are
modifiable by learning (Drewnowski 1997), and it is possible that the near
universal prevalence of the sweet tooth is an artifact of modern environments
in which everyone develops a preference for sweet foods from an ELM that evolved
for some other purpose.
One of the interesting
characteristics of the sweet tooth is that it motivates us to eat sweets even
when we become obese and our health is endangered. Ironically, this maladaptive
characteristic actually suggests developmental specificity. If preferences for
sweet and fatty foods reflect adaptation, they evolved in response to
calorically limited ancestral environments in which sugar and fat were
sporadically encountered and there was little selection for limiting
consumption. If, however, the preferences are the output of an ELM, presumably
they are reinforced by specific experiences in the modern world. Although
adults often avoid sweet, fatty food because they have been extensively
educated about the health risks of high consumption, they do so despite their
taste preferences for these foods. Similarly, in recent years small children
have been increasingly exposed to adult models who
encourage them to eat the "right" foods, yet still have strong
preferences for sweet foods (Drewnowski 1997). The fact that the features of
the sweet tooth cannot readily be accounted for by adaptive learning in a
modern environment, but exhibit evolutionary adaptation within a calorically
limited ancestral environment, suggests that the sweet tooth is the biased
output of a developmental mechanism and not the output of an ELM.
3.1.6.3 Arguments for
behavioral or cognitive adaptation are bolstered by empirical evidence that
would be difficult to account for by an ELM.
More generally, the case for adaptation
may be strong when it is difficult to see how an ELM could account for the
empirical evidence. Direct evidence for developmental specificity is one
example. In other instances, the evidence against an ELM account need not
directly imply developmental specificity. For instance, women’s preferences for
the scents of men shift over the course of the menstrual cycle such that they
prefer the scents of more symmetrical men at mid-cycle (Gangestad &
Thornhill 1998; Thornhill & Gangestad 1999; Rikowski & Grammer 1999;
Thornhill et al. 2001). It is not clear how the preference could be learned, or
why it would shift over the cycle were it the output of an ELM.
Another instance in which one
could make a case against the trait being the output of an ELM is when there
are several other traits that converge on the same function as the trait in
question. The burden may then shift to ELM advocates to show how an ELM could
plausibly account for the entire pattern. Gangestad and Thornhill (1998;
Thornhill & Gangestad 1999) suggest that the shift in female olfactory
preferences toward the scent of symmetrical men when fertile may be an
adaptation for seeking genetic benefits for offspring in the context of
extra-pair sex, wherein women may pay a cost (e.g., loss of an in-pair mate’s
investment in offspring) and can only reap the genetic benefit when fertile.
Other evidence indicates that: (1) men with more symmetrical faces are
perceived to be healthier (Rhodes et al. 2001); (2) more symmetrical men
sexualize other women more and they invest less time and emotional support in
their primary partner (Gangestad & Thornhill 1997a); (3) more symmetrical
men are more likely to have extra-pair sex partners and are more likely to be
chosen as extra-pair sex partners (Gangestad & Thornhill 1997b); (4) women
are more likely to have extra-pair sex mid-cycle, a pattern not observed for
sex with a primary partner (Bellis & Baker 1990); and (5) women report
greater feelings of sexual attraction to and fantasy about men other than a current
primary partner when fertile, a pattern not observed for feelings about in-pair
partners (Gangestad et al. 2001). These effects are all consistent with the
proposed function of the shift in olfactory preference. Moreover, women also
prefer the scent of men with heterozygous major histocompatibility (MHC)
alleles, a trait that may be particularly valued in a primary mate, as it
should increase the diversity of MHC alleles within a set of offspring and
reduce spread of an infection within a family (Thornhill et al. 2001). As this
preference is purported to have a function different than the preference for
symmetrical men, it should not increase during the fertile phase. It does not
and, in fact, appears to increase during the non-fertile phase (when, according
to this reasoning, selection for long-term mates should dominate). It is
difficult to see how an ELM could plausibly account for the complex and
multiple patterns of evidence (although even more work may be required to
satisfy the onerous standard of special design and convincingly demonstrate the
precise functions of the preference shifts; Thornhill & Gangestad in
press).
3.2. The role of optimization
analyses in the adaptationist program.
Complete consistency with
optimization models is too strict a standard for identifying adaptation and
function because optimization models are generally not capable of including all
the constraints that influence the trait’s design. Nevertheless, optimization
models are, in practice, one of the most useful instruments in the
adaptationist toolbox. The primary goal of such models is not to determine
whether a trait is optimally designed, but to determine whether ancestral
selective pressures have acted on the trait in ways predicted by the model. The
approach admits that genetic and developmental constraints may influence trait
design, but gambles that they will not obscure the patterns that selection
would have on trait design in the absence of those constraints. Thus, the
optimization analysis may yield predictions about the modulation of phenotypes
or behavioral decisions with the environment that could not be made in the
absence of a formal mathematical analysis. If these predictions are
particularly novel or non-intuitive, and are empirically borne out, they may
provide powerful special design evidence that the selection tradeoffs included
in the model have, in fact, been operating on the trait.4
For example, optimization models
predict parent-offspring conflict over parental resources (e.g., Trivers 1974;
Parker & McNair 1978). Haig (1993) used this theory to explain why specific
features of the fetus and mother exhibit special design for extracting and
avoiding the extraction of resources by the fetus, respectively. In the absence
of a cost-benefit analysis showing that selection pressures should result in a
conflict over parental resources, no argument from design would have been
possible.
3.3 The role of the
comparative approach in the adaptationist program.
While the comparative approach
cannot demonstrate adaptation by itself, it is another important instrument in
the adaptationist’s toolbox. Examining how a trait functions in one species can
be useful in generating testable hypotheses about how it functions in another
species. The comparative approach can also help determine whether adaptation
has taken placed when used in conjunction with other methods (Leroi et al.
1994; Martins 2000). One important approach is to use comparative data with
optimization models to demonstrate a broad pattern of selection across species
(Charnov 1993; West et al. 1997).
When used in conjunction with
design evidence, the comparative approach can also demonstrate that a trait
within a single species is an adaptation. As we describe in more detail below
(section 4), the special design standard is very conservative and so will
sometimes fail to correctly classify a trait as an adaptation. The comparative
approach can be used to supplement design evidence when the trait cannot be
classified as an adaptation for a particular function based solely on its
design features. For instance, vertebrates have evolved a metabolism that
places greater reliance on the quick mobilization of energy through anaerobic
respiration than invertebrates. However, vertebrate skeletons also dissolve
slightly from the lactic acid that is generated from this process (Ruben &
Bennett 1987). The dissolution process is very detrimental to the organism
because it interferes with the supportive and protective functions of the
skeletal system. If one were to examine the skeletal dissolution process in a
single vertebrate species, one would probably be unable to conclude that the
skeletal system has undergone adaptation for dealing with the problem. Yet, when one examines those invertebrates that have skeletal
systems, one finds that they are almost always composed of calcium carbonate
whereas those of vertebrates are composed of calcium phosphate. Calcium
phosphate appears to be much less soluble than calcium carbonate in salt
solutions and more complex biological media. Since vertebrates generate greater
amounts of lactic acid than invertebrates, their skeletons may be composed of
calcium phosphate because it resists the dissolving effects of lactic acid
better than calcium carbonate (Ruben & Bennett 1987). The comparative
analysis is useful in this instance because it suggests an alternative
substance that vertebrates could have used to construct their skeletal systems.
Comparing the design features of both substances strongly suggests that
vertebrate skeletal systems exhibit adaptation for resisting the dissolving
effects of lactic acid.
3.4 Special design criteria:
Summary.
Gould and Lewontin have helped
highlight many ways in which the evidentiary standards that adaptationists have
used can lead them to erroneously classify a trait as an adaptation for a
proposed function. First, the genes underlying the trait could have evolved by
chance or mutation. Second, the trait could also be the developmental outcome
of novel environmental input. Third, the genetic constraints operating on the
trait may be so strong that selection is incapable of phenotypically modifying
it for the proposed function. Fourth, the trait could be a spandrel that lacks
the proposed effect. Fifth, the trait could be an adaptation that has a different
function. Finally, the trait could be a spandrel or an adaptation for another
function that has been exapted to the proposed function. Because of the ways by
which errors could occur, it is important that the evidentiary standards used
to infer adaptation exceed mere consistency with an adaptationist account. In
many cases, Gould and Lewontin (1979) argued, they don’t. Although this
complaint may apply to certain standards that could and have been used, it
clearly does not apply to the one advocated by Williams, the special design
standard. Indeed, Williams explicitly argued that explanation through
adaptation be used as a "last resort" only after "less
onerous" accounts have been shown to be highly improbable to yield a
pattern of evidence.
4. Ways in which the special
design approach can fail
As noted above, erroneous
inferences about adaptation can be of two sorts: Traits may be misclassified as
adaptations, and traits that are adaptations may not be so classified. Any
trait that satisfies the rigorous evidentiary standards of the special design
approach is highly likely to be an adaptation for the proposed function.
Because the special design approach is very conservative, however, some
adaptations may fail to exhibit special design for their functions. Gould and
Lewontin’s arguments also have implications for the ways in which even the best
evidentiary standards of adaptationism can fail to correctly identify a trait
as an adaptation for a proposed function.
One way in which adaptation
could fail to exhibit special design is when the trait could have been
phenotypically modified by selection for the proposed function, but been so
constrained that it fails to perform its function with sufficient specificity
and proficiency. (See fn. 4.) The calcium phosphate
composition of vertebrate bones may be just such an example. Vertebrate bones
by themselves do not appear to exhibit special design for resisting the
dissolving effects of anaerobic metabolism precisely because they do dissolve
under lactic acid. It is possible that vertebrates are constrained from
designing bones from materials that are even better able to resist dissolution
under lactic acid.
Another way is when it has been
exapted to other beneficial effects in such a way that it lacks specificity for
its function. Earlier, we discussed waist-hip ratio as a possible factor
influencing men’s mating preferences. If so, it could have evolved for any of
the reasons that were discussed (indicator of general health, reproductive
value, fertility, or a lack of pregnancy or infection). But, it could also have
evolved for one of these reasons and subsequently been exapted to the other
effects. If the characteristics of the preference fail to exhibit specificity
for any of these effects, then it will be impossible to determine which effect
is the function of the preference and which (if any) it has been exapted to.
Finally, a trait may undergo
adaptation for one effect, be exapted to a second effect, and then undergo
further adaptation for the second effect. Such a feature may not look well
designed for either the original or the latter function. In other words, the
trait could also have multiple functions that exert opposing influences on its
design such that it lacks special design for either function. If a mixed design
trait doesn’t show some specificity and proficiency in performing its
functions, it will be impossible to reverse engineer. In other instances, a
mixed design trait may exhibit enough specificity and proficiency that it will
be possible to identify its functions even if the trait is not optimally
designed for those functions. The identification of the trait as a mixed design
trait will then depend on being able to identify opposing influences on trait
design such that it is suboptimally designed for its multiple functions.
Consider the human female orgasm
as a possible example of a mixed design trait. (See endnote 5
on Gould’s (1987) own writings on the female orgasm.) One adaptationist
hypothesis for female orgasm is that it functions as a selective sperm
retention mechanism (Baker & Bellis 1993; 1995). The "upsuck"
hypothesis originated with the work of Fox and his colleagues showing that the
normally higher pressure in the uterus relative to that in the vagina reverses
direction immediately following orgasm (Fox et al. 1970). On the basis of
sexual selection theory, Baker and Bellis (1993) proposed that female orgasm
selectively biases the retention of sperm of one male over another when a
female has multiple mates. The selectivity aspect of the hypothesis predicts
that female orgasm should not always occur during intercourse and will be
associated with the characteristics of male partners. Provisional data suggest
that female orgasm can, in fact, lead to sperm retention (Baker & Bellis
1993). If so, data also suggest that female patterns of orgasm would favor the
sperm of men who are extra-pair partners (Baker & Bellis 1993) and who
possess high developmental stability — a characteristic that may be associated
with increased genetic fitness of offspring in ancestral environments
(Thornhill et al. 1995). If this and other predictions that follow from this
hypothesis are shown to be robust, female orgasm may exhibit special design for
choosing sires who produce viable offspring.
However, other aspects of female
orgasm may exhibit special design for pair-bonding with good social partners.
During female orgasm the neurohormone oxytocin is released (Blaicher et al.
1999). As this hormone plays a role in pair-bond formation in non-human mammals
(Young et al. 1998), it may play a pair-bonding role in women as well (Turner
et al. 1999). However, the characteristics that make for a good pair-bonding
partner (e.g., a highly investing male) may make for a suboptimal sire, and
vice versa (Gangestad & Thornhill 1997a). Female orgasm may exhibit mixed
design because it may sometimes cause women to be pair-bonded to men who make
good sires but low-investment social partners. Similarly, it may also sometimes
cause women to retain the sperm of men who make good social partners but poor
sires. We stress that much empirical work would need to be done to demonstrate
these points. Moreover, it may be the case that selection has reduced the
incidence of mistakes by neurophysiologically decoupling the two possible
functions of orgasm (e.g., Thornhill & Furlow 1998).
5. The exaptationist program
Gould and Lewontin have also
argued that the focus on adaptationist hypotheses (even if rigorously put to
the test so that erroneous inferences of adaptation are minimized) is not harmless.
It leads scientists to ignore more prevalent, more important, and more
interesting hypotheses for trait design. For instance, of the alternative
explanations for trait design listed in the prior section, Gould and Lewontin
have been particularly insistent that adaptationists consider hypotheses of
constraint, spandrel, and exaptation (Gould 1991a; Gould & Lewontin 1979;
Gould & Vrba 1982; Lewontin 1979; 1983).
[The
constraint argument] holds ... that the basic body plans of organisms are so
integrated and so replete with constraints upon adaptation ... that
conventional styles of selective arguments can explain little of interest about
them. It does not deny that change, when it occurs, may be mediated by natural
selection, but it holds that constraints restrict possible paths and modes of
change so strongly that the constraints themselves become much the most
interesting aspect of evolution. [Gould & Lewontin 1979, p. 594]
Similarly, Gould has also argued
that exaptations are both more numerous and more important than adaptations
(Gould 1991a; Gould & Vrba 1982). Moreover, he argues that this is
particularly true with respect to the evolutionary study of human behavior and
psychology where the list of exaptive uses to which the human brain is put
"is a mountain to the adaptive molehill" (Gould 1991a, p. 59).
The problem with the pluralistic
approach is that Gould and Lewontin have not provided any evidentiary standards
for testing these alternatives (Daly 1991). Indeed, Lewontin has acknowledged
this problem (Lewontin 1978, p. 228): "In a sense … biologists are forced
to the extreme adaptationist program because the alternatives, although they
are undoubtedly operative in many cases are untestable in particular
cases." The lack of evidentiary criteria for testing alternatives has
sometimes led participants in the debate to accept non-adaptationist
explanations for traits very uncritically, and subsequent research has often
vindicated adaptationist explanations for them (Alcock 1998). In the absence of
rigorous evidentiary criteria for accepting alternatives, even exaptationist
story telling is just-so story telling.
5.1 Testing adaptationist
hypotheses is a necessary part of the pluralistic program advocated by Gould
and Lewontin.
Several authors have suggested
that testing adaptationist hypotheses might be a way to show that a trait has
been constrained in its design (Dennett 1995; Mayr 1983; Sober & Wilson
1998). As noted above, adaptationists sometimes make predictions about trait
design based on optimization models or special design arguments that generally
assume no developmental or genetic constraints. If these predictions fail, one
reason may be because trait design is influenced by genetic constraints that
are difficult to identify and include in optimization models. If efforts to
build in new assumptions about selection fail to produce a model that yields
correct predictions, genetic constraints that limit the potential
genotype-phenotype mappings and thereby force selection to make design tradeoffs
between traits become a more likely reason for lack of model fit. Sober and
Adaptationism
is sometimes understood as a claim about nature — that organisms are well
adapted (or even perfectly adapted) to their environments. At other times,
however, adaptationism is understood as a method for investigating nature. This
is the idea that a useful procedure for studying an organism is to ask,
"What would the organism be like if it were well adapted to its environment?"
Posing this question does not commit one to the position that the organism
actually is well adapted. Perhaps the population inhabits a novel
environment and has not had time to adapt. Perhaps the most adaptive behaviors
never arose by mutation. Perhaps maladaptive behaviors spread by the process of
random genetic drift.... Even so, this kind of failure can be highly
instructive because it allows deviations from the optimal phenotype to be
discovered and interpreted. [pp. 11-12, citations omitted, emphasis in
original]
Dennett (1995) uses an analogy
of chess-playing to make this point. Sometimes, to even up a game with a weaker
player, a strong player takes on a handicap, such as no King’s bishop. Suppose
the handicap was to constrain movement of the pieces in some way (e.g., by not
moving a piece twice in a row, by not moving a queen as a bishop, by not
castling). The player writes down the limitation at the beginning of the game
on a piece of paper, but does not tell the opponent. How should the opponent discover
the limitation? By playing the game as if there were no
limitation, with an eye toward seeing some limitation. Until the player
makes some non-optimal move, there is no evidence of any specific limitation.
So it is with figuring out the ways of Mother Nature. The adaptationist assumes
no specific limitation until seeing clear evidence of limitation because that
is a good way of detecting limitations. Another way of knowing the limitation,
of course, would simply be to peek at the piece of paper. But here the analogy
clearly breaks down. Mother Nature, Dennett notes, doesn’t write down her
limitations. We cannot discover her limitations except by observing her ways.
This is, in fact, a general
principle for testing alternative hypotheses to adaptation. Adaptationism is
not only useful for discovering constraints on trait design, but empirical
demonstration of constraint, exaptation and spandrel requires an
adaptationist approach. Commitment to the scientific enterprise requires that
we not accept claims about constraint, exaptation or spandrel in the absence of
evidence. It is certainly possible that genetic constraints are so prevalent
that optimization theory will be impotent to explain evolutionary outcomes and
that exaptation is so common that traits will rarely exhibit special design for
any particular effect. However, in the absence of evidence one way or another,
we should be agnostic as to whether a given trait is optimally designed or
constrained; exapted to a particular effect or specially designed for it.
Building an empirical case that certain features of a trait are best explained
by exaptation, spandrel, or constraint requires a demonstration that the
trait’s features cannot be better accounted for by adaptationist hypotheses.
Confidence in alternative hypotheses for trait design only increases after
consideration of all plausible adaptationist hypotheses and their failure to
live up to special design scrutiny. We discuss several examples to make this
point.
5.2 Example of spandrel.
Spandrels differ from
adaptations in that they were not reproductively beneficial in ancestral
environments. Rather, spandrels must have been reproductively neutral or even
costly over the period of their evolution. Spandrels evolved because they were
genetically linked to other traits that were favored by selection. Many
psychological phenomena currently thought of as pathologies are good candidates
as maladaptive spandrels (e.g., schizophrenia). Pathological spandrels of the
psyche presumably are the byproducts of developmental mechanisms that evolved
to produce particular psychological outcomes but produce pathologies when
subjected to stressful environmental or genetic perturbations. Of course,
adaptations can be costly (and they often are), but they evolved because they
also had compensating beneficial effects in the ancestral environment that
drove their evolution. Building an empirical case that a costly trait is a
maladaptive spandrel thus requires some demonstration: (1) that the trait is
not an adaptation; and (2) that the trait is linked (by pleiotropy or linkage
disequilibrium) to another trait that presumably was favored by selection. The
first requirement must involve rigorous testing of adaptationist hypotheses and
a systematic failure to find special design evidence for any of the
hypothesized functions.
Published arguments over whether
attention deficit/hyperactivity disorder (ADHD) is an adaptation or a
maladaptive spandrel illustrate our point. According to the Diagnositic and
Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV, American
Psychiatric Association 1994), ADHD is diagnosed by showing that an individual
shows "a persistent pattern of inattention and/or
hyperactivity-impulsivity that is more frequent and severe than is typically
observed in individuals at a comparable level of development," (DSM-IV
1994, p. 78). In addition to symptoms of inattention, hyperactivity and
impulsivity, diagnosis requires that there be "clear evidence of
clinically significant impairment in social, academic or occupational
functioning."
Shelley-Tremblay and Rosen
(1996) and Jensen et al. (1997) ask how this suite of costly traits could be
present in 5% of the population. They argue that distractible, risk-taking
individuals might have had a competitive advantage in two ancestral settings —
intraspecific fighting and gathering as coastal waders. In these dangerous
settings, survival would depend on being "response-ready."
Response-ready individuals would be hypervigilant, rapid scanning the visual
and auditory environment, quick to pounce or flee and motorically hyperactive. Conversely, "the excessively contemplative, more phlegmatic
individual would have been ‘environmentally challenged’" (Jensen et al.
1997, p. 1674) in these settings.
Special design evidence in favor
of this hypothesis would indicate that individuals with ADHD respond better to
cues of danger (perhaps particularly those encountered in ancestral
environments; e.g., large animal movement or aggressive human movement). Do
ADHD individuals respond more quickly and effectively to these cues? At
present, the evidence suggests not. In a critical review, Goldstein and Barkley
(1998) argue that, in fact, individuals diagnosed with ADHD appear to be
deficient in response readiness (though, it should be noted that crucial tests
of response readiness to ancestral cues of danger have not been
performed). Because ADHD appears to lack special design for response readiness,
the evidence seems to be more consistent with the hypothesis that ADHD is a
maladaptive spandrel that persists despite
selection, not because of it (e.g., Gangestad & Yeo 1997). Of course, a
slower response to ancestral cues of danger, and demonstrating the precise
adaptation to which it was linked, would enhance the empirical case that ADHD
was a maladaptive spandrel.
Goldstein and Barkley (1998)
dismiss Shelley-Tremblay and Rosen’s (1996) adaptationist hypothesis as a
"just-so story" (sensu Gould & Lewontin 1979). The fault with Shelley-Tremblay and Rosen’s argument, however, lies
not in the fact that they adopted an adaptationist approach but rather that
their approach was insufficiently adaptationist. Shelley-Tremblay
and Rosen did not lay out evidence for special design of ADHD or even critical
tests of special design. Ironically, it was the authors who doubted this
adaptationist hypothesis, Goldstein and Barkley (1998), who rolled out the
evidence pertinent to special design — illustrating that an adaptationist
approach is critical to tests of hypotheses about adaptations as well as hypotheses
about alternative evolutionary scenarios.
5.3 Example of genetic or
developmental constraint.
Gould and Lewontin (1979) argue
that there are some constraints acting on organismal design that cannot be
included in optimization models because they are unpredictable and
unquantifiable. Genetic constraints are often due to historical events that
cannot be predicted beforehand from a general theory and so the genetic
constraint hypothesis does not readily lend itself to making positive
predictions about trait design. The genetic constraint hypothesis does predict,
however, that even after adaptationists have included all the constraints that
they can, their models will still fail to accurately explain trait design. The
hypothesis that genetic constraints have forced the organism to make design
tradeoffs between traits is mutually exclusive with the hypothesis that the
trait is optimally designed for its function. Before one can conclude that a
trait is suboptimally designed for a function because of a genetic constraint,
one must show a systematic failure of optimization models to explain trait
design.
Gould has argued that allometric
relationships sometimes exist because of constraints (Gould & Lewontin
1979). Adaptation that produces changes in the size of one trait may cause
changes in the size of other traits (in a non-optimal fashion) simply because a
developmental plan deeply ingrained in the genome does not allow unlinked
growth of the traits. Of course, the existence of an allometric relationship by
itself is not evidence of suboptimal trait design due to genetic constraint.
The only way to demonstrate that the growth patterns of body parts could be
genetically constrained is to show that the pattern of covariations between
parts is inconsistent with a priori notions of what would be optimal
growth patterns.
Consider the finding that the
sizes of mammalian brain components are predictable with a high degree of
accuracy from absolute brain size by a non-linear function (Finlay &
Darlington 1995; Finlay et al., 2001). The authors suggest that these linked
regularities are attributable to constraints on brain development. In testing
their constraint hypothesis, the authors took an adaptationist approach by
making an argument based on optimality. Because each species will be under
selection for different cognitive abilities, Finlay and Darlington argued that
the optimal response is for different species to invest in different brain
components, which should result in a non-allometric relationship with overall
brain size. This is a valid adaptationist hypothesis and its subsequent
rejection with the finding of an allometric relationship increases confidence
in the constraint hypothesis. By using an adaptationist approach, Finlay and
her colleagues have made a prima facie case for constraint on mammalian
brain development. (But see Barton [1999; 2001] on departures
from perfect allometric relationships in primate brain size and for a critique
of the analyses of Finlay et al., 2001.)
5.4 Example of exaptation.
Exaptation is not necessarily
mutually exclusive to adaptation. A trait can first be exapted to a new
beneficial effect and then subsequently be modified by selection for that
effect. As we have noted, Gould and Vrba (1982) refer to the initial trait as a
primary exaptation and the subsequent modifications as secondary
adaptations. If some aspects of the initially exapted trait exhibit special
design for the new effect, some reconstruction of the selective history of the
trait is possible. For example, while feathers may have initially served a
thermoregulatory function and later exapted to flight, flight feathers (found
on the bird’s wings and tail) are longer (Gill 1990) and stiffer (Corning &
Biewener 1998) than necessary to serve a thermoregulatory function. Moreover,
unlike body feathers, the vanes of wing feathers are asymmetrical with the
thinner vane being oriented toward the wind during flight (Gill 1990). Flight
feathers clearly exhibit at least secondary modification for flight. We may
then infer that flight exerted a selective force on the construction of flight
feathers.
It is only when adaptationists
fail to find any evidence of phenotypic modification for a particular effect
that they will be unable to make inferences about the selective history of a
trait. As such, all plausible adaptationist hypotheses must be considered,
subjected to special design scrutiny, and rejected before a conclusion of
exaptation without secondary adaptation (i.e., pure exaptation) may be drawn.
Consider again men’s preference
for a small WHR in women. A claim that the preference for small WHR is an
exaptation for any specific effect requires demonstrating that the preference
fails to exhibit special design for that effect. This requires considering what
the trait would be like if it did exhibit special design for that effect. For
instance, if the preference evolved because a small WHR ancestrally promised
greater reproductive value, one might expect that men would specifically prefer
it when choosing a long-term mate in whose offspring the man invests. A failure
to support this prediction would strengthen a claim that the preference evolved
for some other reason and was exapted to choosing a mate of greater
reproductive value. Similarly, if the preference evolved because a small WHR
suggested lack of pregnancy, one might expect that men would prefer women of
low WHR as both long-term and short-term mates. If it has historically been a
cue of infectious disease, one might expect men to actively avoid sex with
women of high WHR. Moreover, one might expect that its importance would vary
with the prevalence of disease (Gangestad & Buss 1993). Other predictions
could be made about the special design implications of these adaptationist
hypotheses. A systematic failure to support these predictions would strengthen
claims that the preference is not the product of phenotypic modification by
selection for these effects, but was instead exapted to these effects.
Moreover, such a failure would make it impossible to make inferences about the
selective forces that shaped men’s preference for small WHR.
Parenthetically, if men’s
preference for small WHR does not exhibit special design for any of these
possible functions, then it should be more evolvable than if it is specially
designed for one of them. Traits that lack specialized design for a particular function are more likely to be versatile and therefore more
evolvable (West-Eberhard 1998). In other words, once the preference becomes
specially designed for a particular function, it may be more difficult to
modify the preference for one of the other functions. Thus, inferences about
the future evolvability of a trait could possibly be made by testing
adaptationist hypotheses and subjecting them to special design scrutiny.
5.5 Testing alternative
hypotheses for traits that have undergone adaptation.
The problem with the approach we
have advocated is that it assumes that one may make inferences about
exaptation, spandrel, or constraint only after special design analyses or
optimization arguments fail to explain the features of a trait. However, given
that the relation between the trait’s features and its effects may have been
influenced by a combination of selection, genetic or developmental constraint,
and exaptation, how is the scientist to make inferences about the explanatory
power of other hypotheses when special design or optimization analyses succeed
in demonstrating adaptation for a particular function? First, the special
design approach implies that a trait performs its function reasonably well, but
it does not imply that the trait must perform it optimally. Even after
demonstrating adaptation by special design, one may still build a case for
constraint by showing that the trait is not optimally designed to perform its
function, as in the case of the vertebrate eye.
However, we can only offer
suggestions for demonstrating that a trait was initially exapted to an effect
when it also exhibits special design for it. In some instances, it may be
possible to make an inference of exaptation by examining the phylogenetic
history of the trait, as in the case of birds’ feathers evolving from reptiles’
scales for a thermal insulation effect and subsequently becoming exapted to
flight. If the trait is complex, the scientist may be able to do separate
design analyses of its components in the hope that selection has only modified
some of its components, but not others. If components fail to exhibit special
design for the effect, they may have been exapted to it instead.
For instance, after further
demonstrating allometric relationships between brain volumes across mammalian
species (see above), Finlay, Darlington and Nicastro (2001) argued that
mammalian brains develop from a fairly simple developmental plan with few
parametric variations, a plan that is conserved because it promotes
evolvability but necessarily also entails constraint. According to this plan,
larger brains necessarily develop larger neo-cortical areas. One implication is
that the large expansion of the human brain need not have been due to selection
for existing neo-cortical functions, despite the fact that the neo-cortex
accounts for most of the expansion. Rather, selection could have favored some
larger subcortical feature, resulting in a large neo-cortex as a byproduct.
Possibly, then, the large neo-cortex was exapted to its distinctly human
functions (e.g., language, advanced tool-making, certain theory of mind tasks,
etc.). Testing this proposal requires adaptationist methodology. For one, it
implies that increased size of a noncortical structure was favored, which means
that some noncortical feature should be larger than expected on the basis of
allometric relationships.
Second, it implies lack of
special design of cortical structures for these distinctly human functions —
unless secondary adaptation has also occurred. Demonstrating that the enlarged
human neo-cortex has been exapted for its many beneficial effects, and an
adaptation for none, is admittedly a burdensome task that requires considering,
testing, and systematically rejecting many adaptationist hypotheses for
neo-cortical design. But then again, the claim is quite broad. For instance, it
is possible that the neo-cortex has been constrained in its size relative to
the rest of the brain and that selection has been free to play with how it is
structured, a point raised by Finlay and Darlington (1995) in their initial
paper. Components of the human neo-cortex seem to exhibit structural design for
different tasks, such as language, planning, attention, and theory of mind tasks
(Adolphs 2001; Damasio 1994; Pinker 1994). Many of these components could be
adaptations that evolved sometime prior to the evolution of human beings such
that the increase in neo-cortex size over evolutionary time enhanced its
performance at its pre-existing functions rather than leading to exaptation to
new tasks. Testing this possibility will require identifying the functions that
selection initially designed the neocortex for, which could be done by using
comparative studies in conjunction with design analyses.
On the other hand, such tasks
could be unique human functions: (1) for which selection designed the
neo-cortex; (2) to which the neo-cortex was exapted; or (3) that arose by some
combination of exaptation and secondary adaptation. Under the second scenario,
the components of the neo-cortex presumably came to exhibit complex structural
design for different tasks from developmental processes that evolved for other
purposes (i.e., ELM’s). We have suggested several lines of evidence that could be
useful in building a case against ELM hypotheses (sections 3.1.6.1-3.1.6.3).
For instance, demonstrating that the neo-cortex exhibits developmental
specificity for certain theory of mind tasks would strongly suggest that
exaptation is not a complete explanation for neo-cortical design. Neonates and
children do seem to exhibit a cognitive predisposition towards developing many
theory of mind tasks (Flavell 1999), and presumably this reflects a structural
predisposition in the neo-cortex (Adolphs 2001). The evidence of developmental
specificity for some aspects of language learning is even stronger (Pinker
1994), even if it has not been demonstrated for all aspects (Gomez & Gerken
2000), and language is, of course, strongly rooted in the neo-cortex (Pinker 1994).
However, other components of the neo-cortex may lack special design for their
effects. Finding that other neo-cortical components having uniquely human
effects generally lacked special design for those effects might actually
bolster the hypothesis that the neo-cortex was primarily exapted to theory of
mind and language capacities and only adapted to them secondarily.
In the end, it seems likely that
the design of the neo-cortex will probably be understood as a mixture of
adaptation, exaptation, and constraint precisely because it is a complex trait.
Determining which components are adaptations or exaptations for what effects
will undoubtedly take a great deal of work — work that must involve the testing
of adaptationist hypotheses.
6. Conclusion
There is a great need for a
consensus regarding the evidentiary criteria required to demonstrate that a
trait is an adaptation for a particular function. While people may differ with
respect to their expectations about whether a particular trait is an adaptation,
exaptation, or spandrel, many arguments can be resolved with good data that
bear on a commonly agreed upon standard of evidence. We have tried to
incorporate the best ideas from both sides in the debate in the attempt to help
forge a consensus.
Williams (1966) gave
adaptationists three guidelines for how to build a case that a trait is an
adaptation. First, the burden on the adaptationist is onerous. One must show
that plausible alternative explanations for the trait are unlikely to generate
the evidence about the trait. Second, one can acquire information about the
relative likelihood of alternatives by examining the features of the trait in
relation to what it does. Third, there is no fixed list of evidentiary criteria
that must be met before the scientist can make an inference of adaptation.
Williams did provide suggestions about the kinds of features of traits that
could give insight into the relative likelihood of alternatives and
adaptationists have since expanded on those (specificity, proficiency, precision,
efficiency, economy, complexity of design, reliable production, costliness,
etc.). However, the scientist must not lose sight of the ultimate burden of
proof. Different traits may require satisfaction of different evidentiary
criteria (e.g., cognitive and behavioral traits may require satisfaction of
more rigorous criteria than morphological traits). For instance, while we have
argued that it will be difficult to identify a trait’s function unless it
exhibits some specificity and proficiency for an effect, it is possible that
some cognitive and behavioral traits can come to exhibit specificity and
proficiency for performing a task without having been designed by selection for
that task (e.g., traits that develop from ELM’s). Demonstrating that a behavioral
or cognitive trait is unlikely to have developed from an ELM may require
additional evidence (e.g., developmental specificity, etc.).
At the same time, advocates of
an exaptationist program must live up to their own burden of proof. It is not
enough to assert that a trait has been exapted or that it is a spandrel without
meeting some rigorous standard of evidence. Good inference in any scientific
field requires a serious consideration of alternative hypotheses. Because
hypotheses about constraint, exaptation, and spandrel and hypotheses about
adaptation are often mutually exclusive to each other, we have argued that
confidence in these alternatives increases only when plausible adaptationist
hypotheses have been considered, subjected to special design scrutiny, and
systematically rejected. Consequently, adaptationism is not an ontological
commitment to the idea that traits or organisms are perfectly adapted to their
environment (Sober & Wilson 1998). Rather it is an epistemological approach
for discovering whether or not they actually are adapted to their
environment and for making testable inferences about the causes of trait
design.
References
Adolphs, R. (2001) The neurobiology of social cognition. Current Opinion in
Neurobiology 11: 231-239.
Alcock, J.
(1987) (Letter to editor) Natural History 96:4.
Alcock, J.
(1998) Unpunctuated equilibrium in the Natural History essays of Stephen
Jay Gould. Evolution
and Human Behavior 19:321-336.
Alexander, R. (1987) The biology of moral systems. Hawthorne, NY:
Aldine de Gruyter.
American Psychiatric Association
(1994) Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author.
Atran, S. (1998) Folk biology
and the anthropology of science: Cognitive universals and cultural particulars.
Behavioral and Brain Sciences 21: 547-569.
Baker, R. R., & Bellis, M.
A. (1993) Human sperm competition: Ejaculate manipulation by females and a
function for the female orgasm. Animal Behaviour 46:887-909.
Baker, R. R.,
& Bellis, M. A. (1995) Human sperm competition. London: Chapman & Hall.
Baillargeon, R. (1987) Young infants’ reasoning about the physical and spatial properties of a hidden object. Cognitive Developmen