Home > The role of mate choice in biocomputation:

The role of mate choice in biocomputation:


The role of mate choice in biocomputation:

Sexual selection as a process of search, optimization, and diversification 

Geoffrey F. Miller and Peter M. Todd 

Published in:  W. Banzof & F. H. Eeckman (Eds.) (1995). Evolution and Biocomputation: Computational Models of Evolution. Lecture Notes in Computer Science 899. (pp. 169-204). Springer-Verlag. 


The most successful, complex, and numerous species on earth are composed of sexually-reproducing animals and flowering plants.  Both groups typically undergo a form of sexual selection through mate choice: animals are selected by conspecifics and flowering plants are selected by heterospecific pollinators.  This suggests that the evolution of phenotypic complexity and diversity may be driven not simply by natural-selective adaptation to econiches, but by subtle interactions between natural selection and sexual selection.  This paper reviews several theoretical arguments and simulation results in support of this view.  Biological interest in sexual selection has exploded in the last 15 years (see Andersson & Bradbury, 1987; Cronin, 1991), but has not yet been integrated with the biocomputational perspective on evolution as a process of search and optimization (Holland, 1975; Goldberg, 1989).  In the terminology of sexual selection theory, mate preferences for "viability indicators" (e.g.  Hamilton & Zuk, 1982) may enhance evolutionary optimization, and mate preferences for "arbitrary traits" (e.g.  Fisher, 1930) may enhance evolutionary search and diversification.  Specifically, as a short-term optimization process, sexual selection can: (1) speed evolution by increasing the accuracy of the mapping from phenotype to fitness and thereby decreasing the "noise" or "sampling error" characteristic of many forms of natural selection, and (2) speed evolution by increasing the effective reproductive variance in a population even when survival-relevant differences are minimal, thereby imposing an automatic, emergent form of "fitness scaling", as used in genetic algorithm optimization methods (see Goldberg, 1989).  As a longer-term search process, sexual selection can: (3) help populations escape from local ecological optima, essentially by replacing genetic drift in Wright's (1932) "shifting balance" model with a much more powerful and directional stochastic process, and (4) facilitate the emergence of complex innovations, some of which may eventually show some ecological utility.  Finally, as a process of diversification, sexual selection can (5) promote spontaneous sympatric speciation through assortative mating, increasing biodiversity and thereby increasing the number of reproductively isolated lineages performing parallel evolutionary searches (Todd & Miller, 1991) through an adaptive landscape.  The net result of these last three effects is that sexual selection may be to macroevolution what genetic mutation is to microevolution: the prime source of potentially adaptive heritable variation, at both the individual and species levels.  Thus, if evolution is understood as a biocomputational process of search, optimization, and diversification, sexual selection can play an important role complementary to that of natural selection.  In that role, sexual selection may help explain precisely those phenomena that natural selection finds troubling, such as the success of sexually-reproducing lineages, the speed and robustness of evolutionary adaptation, and the origin of otherwise puzzling evolutionary innovations, such as the human brain (Miller, 1993).  Implications of this view will be discussed for biology, psychology, and evolutionary approaches to artificial intelligence and robotics. 

Keywords: sexual selection, mate choice, optimization, speciation, evolutionary innovation, genetic algorithms, biodiversity 


Sexual selection through mate choice (Darwin, 1871) has traditionally been considered a minor, peripheral, even pathological process, tangential to the main work of natural selection and largely irrelevant to such central issues in biology as speciation, the origin of evolutionary innovations, and the optimization of complex adaptations (see Cronin, 1991).  But this traditional view is at odds with the fact that the most complex, diversified, and elaborated taxa on earth are those in which mate choice operates: animals with nervous systems, and flowering plants.  The dominance of these life-forms, and the maintenance of sexual reproduction itself, has often been attributed to the advantages of genetic recombination.  But recombination alone is not diagnostic of animals and flowering plants: bacteria and non-flowering plants both do sexual recombination.  Rather, the interesting common feature of animals and flowering plants is that both undergo a form of sexual selection through mate choice.  Animals are sexually selected by opposite-sex conspecifics (Darwin, 1871; see Cronin, 1991), and flowering plants are sexually selected by heterospecific pollinators such as insects and hummingbirds (Sprengel, 1793; Darwin, 1862; see Barth, 1991).  Indeed, Darwin's dual fascination with animal courtship (Darwin, 1871) and with the contrivances of flowers to attract pollinators (Darwin, 1862) may reflect his understanding that these two phenomena shared some deep similarities. 

The importance of mate choice in evolution can be appreciated by considering the special properties of neural systems as generators of selection forces.  The brains and sensory-motor systems of organisms make choices that affect the survival and reproduction of other organisms in ways that are quite different from the effects of inanimate selection forces (as first emphasized by Morgan, 1888).  This sort of psychological selection (Miller, 1993; Miller & Freyd, 1993) by animate agents can have much more direct, accurate, focused, and striking results than simple biological selection by ecological challenges such as unicellular parasites or physical selection by habitat conditions such as temperature or humidity.  Recently, several biologists have considered the evolutionary implications of "sensory selection", perhaps the simplest form of psychological selection (see Endler, 1992; Guilford & Dawkins, 1991; Ryan, 1990; Ryan & Keddy-Hector, 1992).  This paper emphasizes the evolutionary effects of mate choice because mate choice is probably the strongest, most common, and best-analyzed type of psychological selection.  But there are many other forms of psychological selection both within and between species.  For example, the effects of psychological selection on prey by predators results in mimicry, camouflage, warning coloration, and protean (unpredictable) escape behavior.  Artificial selection on other species by humans, whether for economic or aesthetic purposes, is simply the most self-conscious and systematic form of psychological selection.  Thus, we can view sexual selection by animals choosing mates as mid-way between brute natural selection by the inanimate environment, and purposive artificial selection by humans. 

But the big questions remain: What distinctive evolutionary effects arise from psychological selection, and in particular from sexual selection through mate choice? And how does sexual selection interact with other selective forces arising from the ecological and physical environment? The traditional answer has been that sexual selection either copies natural selection pressures already present (e.g.  when animals choose high-viability mates) making it redundant and impotent, or introduces new selection pressures irrelevant to the real work of adapting to the econiche (e.g.  when animals choose highly ornamented mates), making it distracting and maladaptive.  In this paper we take a more positive view of sexual selection.  By viewing evolution as a "biocomputational" process of search, optimization, and diversification in an adaptive landscape of possible phenotypic designs, we can better appreciate the complementary roles played by sexual selection and natural selection.  We suggest that the success of animals and flowering plants is no accident, but is due to the complex interplay between the dynamics of sexually-selective mate choice and the dynamics of naturally-selective ecological factors.  Both processes together are capable of generating complex adaptations and biodiversity much more efficiently than either process alone.  Mate choice can therefore play a critical role in biocomputation, facilitating not only short-term optimization within populations, but also the longer-term search for new adaptive zones and new evolutionary innovations, and even speciation and the macroevolution of biodiversity. 

This paper begins with a discussion of the historical origins of the idea of mate choice (section 2) and the evolutionary origins of mate choice mechanisms (section 3).  We then explore how mate choice can improve biocomputation construed as adaptive population movements on fitness landscapes, by allowing faster optimization to fitness peaks (section 4), easier escape from local optima (section 5), and the generation of evolutionary innovations (section 6).  Moving from serial to parallel search, we then consider how sexual selection can lead to sympatric speciation and thus to evolutionary search by multiple independent lineages (section 7).  Finally, section 8 discusses some implications of these ideas for science (particularly biology and evolutionary psychology) and some applications in engineering (particularly genetic algorithms research and evolutionary optimization techniques).  This theoretical paper complements our earlier work on genetic algorithm simulations of sexual selection (Miller, accepted, a; Miller & Todd, 1993; Todd & Miller, 1991, 1993); in further work we will test these ideas with more extensive simulations (Todd & Miller, in preparation) and comparative biology research (Miller, accepted, b; Miller, 1993). 

The evolution of economic traits through natural selection versus the evolution of reproductive traits through sexual selection 

Darwin (1859, 1871) clearly distinguished between natural selection and sexual selection as different kinds of processes operating on different kinds of traits according to different kinds of evolutionary dynamics.  For him, natural selection improved organisms' abilities to survive in an environment that is often hostile and always competitive, while sexual selection honed abilities to attract and select mates and to produce viable and attractive offspring.  But this critical distinction between natural and sexual selection was lost with the Modern Synthesis (Dobzhansky, 1937; Huxley, 1942; Mayr, 1942; Simpson, 1944), when natural selection was redefined as any change in gene frequencies due to the fitness effects of heritable traits, whether through differential survival or differential reproduction.  The theory of sexual selection through mate choice had been widely dismissed after Darwin, and this brute-force redefinition of natural selection to encompass virtually all non-random evolutionary processes did nothing to revive interest in mate choice. 

Fisher (1915, 1930) was one of the few biologists of his era to worry about the origins and effects of mate choice.  He developed a theory of "runaway sexual selection," in which an evolutionary positive-feedback loop is established (via genetic linkage) between female preferences for certain male traits, and the male traits themselves.  As a result, both the elaborateness of the traits and the extremity of the preferences could increase at an exponential rate.  Fisher's model could account for the wildly exaggerated male traits seen in many species, such as the peacock's plumage, but it did not explain the evolutionary origins of female preferences themselves, and was not stated in formal genetic terms.  Huxley (1938) criticized Fisher's model in a hostile and confused review of sexual selection theory, which kept Darwin's theory of mate choice in limbo for decades to come. 

In the last 15 years, however, there has been an explosion of work on sexual selection through mate choice.  The new population genetics models of O'Donald (1980), Lande (1981), and Kirkpatrick (1982) supported the mathematical feasibility of Fisher's runaway sexual selection process.  Behavioral experiments on animals showed that females of many species do exhibit strong preferences for certain male traits (e.g.  Andersson, 1982; Catchpole, 1980; Ryan, 1985).  New comparative morphology has supported Darwin's (1871) claim that capricious elaboration is the hallmark of sexual selection: for instance, Eberhard (1985) argued that the only feasible explanation for the wildly complex and diverse male genitalia of many species is evolution through female preference for certain kinds of genital stimulation.  Evolutionary computer simulation models such as those of Collins and Jefferson (1992) and Miller and Todd (1993) have confirmed the plausibility, robustness, and power of runaway sexual selection.  Once biologists started taking the possibility of female choice seriously, evidence for its existence and significance came quickly and ubiquitously.  Cronin (1991) provides a readable, comprehensive, and much more detailed account of this history. 

Largely independently of this revival of sexual selection theory, Eldredge (1985, 1986, 1989) has developed a general model of evolution based on the interaction of a "geneological hierarchy" composed of genes, organisms, species, and monophyletic taxa, and an "ecological hierarchy" composed of organisms, "avatars" (sets of organisms that each occupy the same ecological niche), and ecosystems.  Phenotypes in this view are composed of two kinds of traits: "economic traits" that arise through natural selection to deal with the ecological hierarchy, and "reproductive traits" that arise through sexual selection to deal with other entities (e.g.  potential mates) in the geneological hierarchy.  Eldredge (1989) emphasizes that the relationship between economic success and reproductive success can be quite weak, and that reproductive traits are legitimate biological adaptations — as shown by recent research on mate choice and courtship displays (see Cronin, 1991).  Eldredge also grants geneological units their own hierarchy separate from the ecological one, but does not emphasize the possibility of evolutionary dynamics occurring entirely within the geneological hierarchy, without any ecological relevance.  The one exception is Eldredge's discussion of how "specific mate recognition systems" (SMRSs) might be disrupted through stochastic effects, resulting in spontaneous speciation.  But other processes occurring purely within the geneological hierarchy, such as Fisher's (1930) runaway process, are not mentioned.  Thus, even in his authoritative review of macroevolutionary theory (Eldredge, 1989), which consistently views evolutionary change in terms of movements through adaptive landscapes, Eldredge overlooks the adaptive autonomy of sexual selection, and the adaptive interplay between sexual selection and natural selection. 

But the time is now right to take sexual selection seriously in both roles: (1) as a potentially autonomous evolutionary process that can operate entirely within Eldredge's "geneological hierarchy", and (2) as a potentially important complement to natural selection that can facilitate adaptation to Eldredge's "ecological hierarchy" in various ways.  The remainder of this paper focuses on this second role.  But to understand the dynamic interplay between natural and sexual selection, we must first understand their different characteristic dynamics. 

Natural selection typically results in convergent evolution onto a few (locally) optimal solutions given pre-established problems posed by the econiche.  In natural selection by the ecological niche or the physical habitat, organisms adapt to environments, but not vice-versa (except in relatively rare cases of tight co-evolution — see Futuyama & Slatkin, 1983).  This causal flow of selection from environment to organism makes natural selection fairly easy to study empirically and formally, because one can often identify a relatively stable set of external conditions (i.e.  a "fitness function") to which a species adapts.  Moreover, natural selection itself is primarily a hill-climbing process, good at exploiting adaptive peaks, but somewhat weak at discovering them. 

By contrast, sexual selection often results in an unpredictable, divergent pattern of evolution, with lineages speciating spontaneously and exploring the space of phenotypic possibilities according to their capriciously evolved mate preferences.  In sexual selection, the mate choice mechanisms that constitute the selective "environment" can themselves evolve under various forces, including the current distribution of available phenotypes.  Thus, the environment and the adaptations — the traits and preferences — can co-evolve under sexual selection, as Fisher (1930) realized.  The causal flow of sexual selection forces is bi-directional, and thus more complex and chaotic.  The resulting unpredictable dynamics may look entirely anarchic, without structure and due entirely to chance, but are in fact "autarchic", in that a species evolving through strong selective mate choice is a self-governing system that in a sense determines its own evolutionary trajectory.  Indeed, sexual selection could be considered the strongest form of biological self-organization that operates apart from natural selection — but it is a form almost entirely overlooked by those who study self-organization from a biocomputational perspective (see e.g.  Langton et al., 1993; Kauffman, 1993). 

If one visualizes sexual selection dynamics as branching, divergent patterns that explore phenotype space capriciously and autonomously, and natural selection dynamics as convergent, hill-climbing patterns that seek out adaptive peaks, then their potential complementarity can be understood.  The overall evolutionary trajectory of a sexually-reproducing lineage results from the combined effects of sexual selection dynamics and natural selection dynamics (plus the stochastic effects of genetic drift and neutral drift) — an interplay of capriciously directed divergence and ecologically directed convergence.  This interplay might help explain evolutionary patterns that have proven difficult to explain under natural selection alone, particularly the abilities of lineages to optimize complex adaptations, to escape from local evolutionary optima, to generate evolutionary innovations, and to split apart into sympatric species. 

This interplay between capricious, divergent sexual selection and natural selection is analogous to the interplay between genetic mutation and natural selection.  The major difference is that the high-level variation in phenotypic design produced by sexual selection is much richer, more complex, and typically less deleterious than the low-level variation in protein structure produced by random genetic mutation.  Thus, many of the phenomena that seem difficult to account for through the interaction of low-level genetic mutation and natural selection, might be better accounted for through the interaction of higher-level sexual-selective effects and natural selection.  But we should consider the evolutionary origins of mate choice before we consider its evolutionary effects.  

Why mate choice mechanisms evolve 

Darwin (1871) analyzed the evolutionary effects but not the evolutionary origins of mate preferences.  Fisher (1915, 1930) went further in discussing how mate preferences might co-evolve with the traits they prefer, by becoming genetically linked to them, but he too did not directly consider the selection pressures on mate choice itself.  Recently, the question of how selective mate choice can evolve has occupied an increasingly important position in sexual selection theory (e.g.  Bateson, 1983; Kirkpatrick, 1982, 1987; Pomiankowski, 1988; Sullivan, 1989); the issue becomes particularly acute when mate choice is costly in terms of energy, time, or risk (Iwasa et al., 1991; Pomiankowski, 1987, 1990; Pomiankowski et al., 1991). 

The mysterious origins of mate choice can be made clearer if the adaptive utility of choice in general is appreciated.  Little sleep is lost over the issues of how habitat choice, food choice, or nesting place choice could ever evolve given their costs; the same acceptance ought to apply to mate choice.  Animal nervous systems have two basic functions: (1) generating adaptive survival behavior that registers, and exploits or avoids, important objects and situations in the ecological environment, such as food, water, prey, and predators (which we collectively call "ecological affordances"), and (2) generating adaptive reproductive behavior that registers and exploits important objects in the sexual environment, such as viable, fertile, and attractive mates (which we collectively call "reproductive affordances").  Current theories of how animals make adaptive choices among ecological affordances are substantially more sophisticated than theories of how animals make adaptive choices among reproductive affordances.  However, by seeing both ecological affordances and reproductive affordances as examples of "fitness affordances" in general (Miller & Freyd, 1993; Todd & Wilson, 1993), we can see the underlying similarity between both sorts of adaptive choice behavior.  The key to choosing food adaptively is to have evolved a food-choice mechanism that has internalized the likely survival effects of eating different kinds of foods: from an evolutionary perspective, the internally represented utility of a food item should reflect its objectively likely prospective fitness effects on the animal, given its energy requirements, biochemistry, gut morphology, etc.  By analogy, the key to choosing mates adaptively is to evolve a mate choice mechanism that has internalized the likely long-term fitness consequences of reproducing with different kinds of potential mates, given a certain recurring set of natural and sexual selection pressures.  The adaptive benefit of choice in each case is that negative fitness affordances that threatened survival or fertility in the past can be avoided, and positive fitness affordances that enhanced survival or fertility in the past can be exploited.  Thus, choice is a way of internalizing ancestral selection pressures into current psychological mechanisms. 

This view of the evolution of choice suggests that mate choice mechanisms can be analyzed according to normative criteria of adaptiveness.  The internally represented sexual attractiveness of a potential mate should reflect its objectively likely prospective fitness value as a mate, in terms of the likely viability and sexual attractiveness of any offspring that one might have with it.  Thus, the efficiency and normativity of a mate choice mechanism could in principle be assessed with the same theoretical rigor as a mechanism for any other kind of adaptive choice.  Mate choice is well-calibrated if the perceived sexual attractiveness of potential mates is highly correlated with the actual viability, fertility, and attractiveness of the offspring they would produce.  The observable traits of potential mates that correlate primarily with offspring survival prospects can be termed "viability indicators" (Zahavi, 1975), and the observable traits that correlate primarily with offspring reproductive prospects can be called "aesthetic courtship displays" of the sort analyzed by Darwin (1871) and Fisher (1930).  In fact, most sexually-elaborated traits such as the peacock's tail will probably play both roles to some extent, with their large costs making them useful viability indicators (e.g.  Petrie, 1992) but the details of their design making them attractive aesthetic displays (e.g.  Petrie et al., 1991). 

Now we can ask, what actually gets "evolutionarily internalized" from the environment (Shepard, 1984, 1987) in the case of mate preferences? Mate choice mechanisms may in some cases evolve to "represent" the recent history of a population's evolutionary trajectory through phenotype space, that is, the recent history of natural selection and sexual selection patterns that have been operating in the population.  Sustained, directional movement through phenotype space typically implies that directional selection is operating, or that a fitness gradient is being climbed in a certain direction.  Mate preferences which are in agreement with this directional movement, internalizing the species' recent history, will then be more successful, assuming the movement continues.  In this case, mate preferences can be described as "anticipatory" assessments of past selection pressures that will probably continue to be applied in the future, in particular to one's offspring. 

This picture of how mate preferences evolve has clear implications for sexual-selection dynamics.  If a population has not been moving through phenotypic space, e.g.  it is perched atop an adaptive peak due to stabilizing selection, as most populations are most of the time, then mate preferences will probably evolve to favor potential mates near the current peak, and they will tend to reinforce the stabilizing natural selection that is currently in force.  (If biased mutation tends to displace individuals from the peak more often in one direction than in another, then mate preferences may evolve to counteract that recurrent deleterious mutation by having a directional component — see Pomiankowski et al., 1991.) But if a population has been evolving and moving through phenotype space, then mate preferences can evolve to "point" in the direction of movement, conferring more evolutionary "momentum" on the population that it would have under natural selection alone.  These sorts of directional mate preferences (Kirkpatrick, 1987; Miller & Todd, 1993) can be visualized as momentum vectors in phenotype space that can keep populations moving along a certain trajectory, in some cases even after natural-selective forces have shifted. 

Another effect could be seen when a population has been splitting apart due to sympatric or allopatric divergence.  In this case, mate preferences in each sub-population can evolve to favor breeding within the sub-population, and not between sub-populations, thereby reinforcing the speciation.  The divergent mate preferences of two populations splitting apart can be visualized as vectors pointing in different directions.  These sexual-selective vectors will reinforce and amplify the initial effects of divergence by imposing disruptive (sexual) selection against individuals positioned phenotypically in between the parting populations.  Thus, directional mate preferences will often evolve to be congruent with whatever directional natural selection (if any) is operating on a population, whether it applies to a unified population or one splitting apart into subspecies.  Sexual selection may thereby smooth out and reinforce the effects of natural selection. 

But sexual selection vectors can often point in different directions from natural-selection vectors, resulting in a complex evolutionary interplay between these forces.  The evolution of mate preferences can be influenced by a number of factors other than natural selection for mate preferences in favor of high-viability traits.  For example, stochastic genetic drift can act on mate preferences as it can act on any phenotypic trait; this effect is important in facilitating spontaneous speciation and in the capriciousness of runaway sexual selection.  Intrinsic sensory biases in favor of certain kinds of courtship displays, such as louder calls or brighter colors, may affect the direction of sexual selection (Endler, 1992; Guilford & Dawkins, 1991; Ryan, 1990; Ryan & Keddy-Hector, 1992).  Also, an intrinsic psychological preference for novelty, as noted by Darwin (1871) and in work on the "Coolidge effect" (Dewsbury, 1981), may favor low-frequency traits and exert "apostatic selection" (Clarke, 1962), a kind of centrifugal selection that can maintain stable polymorphisms, facilitate speciation, and hasten the evolution of biodiversity.  Thus, a number of effects may lead mate choice mechanisms to diverge from preferring the objectively highest-viability mate as the sexiest mate.  These effects will often make sexual-selective vectors diverge from natural-selective gradients in phenotype space, and give sexual selection its capricious, divergent, unpredictable nature.  Now that we have considered the evolutionary origins of mate preferences, we can consider their evolutionary effects. 

Ecological optimization can be facilitated by selective mate choice 

Natural selection is often analyzed theoretically, and implemented computationally, as a more or less simple "fitness function" that maps from phenotypic traits to reproductive success scores (Goldberg, 1989).  But natural selection as it actually operates in the wild is often a horribly noisy, irregular, and inaccurate process.  Predators might often eat the prey animal that has the better vision, larger brain, and longer legs, simply because that animal happened to be closer at dinner time than the duller, blinder, slower animal over the hill.  A lethal virus may attack and eliminate the animal with the better immune system simply because that animal happened to drink from the wrong pond.  Anyone who doubts the noisiness and inaccuracy of natural selection should consider the relative speed at which animals evolve in the wild versus under artificial selection by human breeders, who cull undesirable traits with much more accuracy and thoroughness.  Maynard Smith (1978, p.  12) observed that evolution can happen up to five orders of magnitude (100,000 times) faster under artificial selection than under typical natural selection, at least over the short term. 

The fundamental reason for this disparity is that Nature (i.e.  the physical habitat or biological econiche) has no incentive to maximize the selective efficiency or accuracy of natural selection, whereas human breeders do have incentives to maximize the efficiency and accuracy of artificial selection.  Likewise, animals choosing mates have very heavy incentives to maximize the efficiency and accuracy of their mate choice, and thereby the efficiency and accuracy of the sexual selection that they impose.  Thus, it would be extremely surprising if the selective efficiency and accuracy of natural selection were typically as high as that of sexual selection through mate choice.  Habitats and econiches are not well-adapted to impose natural selection, whereas animals are well-adapted to choose mates and thereby to impose sexual selection.  (This difference is often obscured in genetic algorithms research, where fitness functions are specifically designed by humans to be efficient and accurate selectors.) 

Given the relative noisiness and inefficiency of natural selection itself, how did the "organs of extreme perfection and complication" that Darwin (1859) so admired ever manage to evolve? We believe they do so with substantial assistance from selective mate choice, at least in animals and flowering plants.  As we saw in the previous section, sexually reproducing animals have strong incentives to internalize whatever natural-selection pressures are being applied to their population in the form of selective mate preferences.  For example, these preferences can inhibit mating with individuals that probably survived by luck rather than by genetic merit, whatever genetic merit means given current natural-selective and sexual-selective pressures.  By avoiding mates that have low apparent viability but happen to still be alive anyway, parents can keep from having offspring that would probably not be so lucky.  Conversely, by mating with individuals who clearly show high viability and sexual attractiveness, parents may give their offspring a genetic boost with respect to natural and sexual selection for generations to come.  For example, an average individual who mates with someone with twice their viability or attractiveness may increase their long-term reproductive success (e.g.  number of surviving grand-children) by roughly 50% compared to random mating, by having their genes "hitch-hike" in bodies with the better genes of their mate.  This inheritance of genetic and economic advantage through mate choice can have several important effects on the optimization of complex adaptations, because the brains and sensory systems involved in mate choice can act as highly efficient "lenses" for reflecting, refracting, recombining, amplifying, and focusing natural selection pressures. 

First, the noisiness of natural selection can be substantially reduced by mate choice, leading to smoother, faster evolutionary optimization.  It might take a while for mate preferences to accurately internalize the current regime of natural selection, but once in place, such preferences can exert much more accurate, less noisy selection than natural selection itself can.  For example, natural selection by viruses alone (a biological selector) might yield a low correlation between heritable immune system quality and reproductive success, because the infected animals might be too sick to have a full-sized litter, but still manage to have several offspring despite their illness.  But mate choice based on observed health and immune capacity may boost this correlation much higher, if conspecifics refuse to mate at all with an individual who bears the viral infection, and thereby lower the sick individual's reproductive success to nil.  The higher the correlation between heritable phenotypic traits and reproductive success, the faster the evolution (Fisher, 1930).  Mate choice can therefore heavily penalize individuals who show a tendency to get sick, whereas natural selection heavily penalizes only those individuals who actually have fewer offspring or die.  Here, the brains and sensory systems involved in mate choice act to focus the noisy, diffuse, unreliable forces of natural selection into smoother, steeper gradients of sexual selection.  Thus, much of the work of constructing and optimizing complex adaptations may be performed by mate choice mechanisms tuned to reflect natural-selection pressures, rather than by the natural-selection pressures themselves. 

Of course, most animals that fail to reproduce — especially in r-selected species that produce large numbers of offspring with little parental care — will do so because they were spontaneously aborted, failed to hatch, or died before reproductive maturity due to illness, starvation, or predation.  Out of the countless eggs and sperm that adult salmon release during mating, only a very few zygotes will survive the rigors of childhood and up-river migration to successfully choose mates and spawn themselves.  Natural selection may eliminate almost all of the individuals in a particular generation in this way.  As Darwin (1859) noted in his discussion of the inevitability of competition, the manifest capability of organisms to reproduce far outstrips the carrying capacity of their environment, so natural selection will eliminate the vast majority of individuals.  In contrast, even the most intensive mate choice in highly polygynous species will not cull the remaining reproductively-mature individuals from the mating game with anything like this kind of ferocious efficiency.  A large number of bachelor males may not leave behind any offspring, but most of the females and a significant number of males will, making sexual selection look like a much weaker force in terms of the percentages of individuals affected.  But the efficiency of a selective process depends most heavily on the correlation between heritable phenotypic features and selective outcomes.  In natural selection, this correlation may often be quite low, because, as stressed earlier, Nature typically has no incentive to increase its selective efficiency.  By contrast, this correlation may be quite high in sexual selection, because animals have large incentives to increase their mate-choice efficiency.  Thus, although sexual selection typically affects fewer individuals per generation than natural selection, sexual selection may account for most of the nonrandom change in heritable phenotypic traits — i.e.  most of the evolution. 

Second, mate choice can magnify relative fitness differences, thereby increasing the speed and robustness of optimization.  In genetic algorithms research, populations often converge to have nearly similar performance on the objective fitness function after a few dozen generations, and further optimization becomes difficult because the relatively small fitness differences are insufficient to result in much evolution.  Methods for "fitness scaling" such as linear rescaling or rank-based selection can overcome this problem by mapping from small differences in objective fitness (corresponding to ecological success) onto large differences in reproductive success (Goldberg, 1989).  We believe that in nature, sexual selection can provide an automatic form of fitness scaling that helps populations avoid this sort of evolutionary stagnation.  Again, sexually reproducing animals have incentives to register slight differences in the observed viability of potential mates and to mate selectively with higher-viability individuals.  The result of this choosiness will be automatic fitness scaling that maintains substantial variance in reproductive success and thereby keeps evolution humming along even when every individual is similar in fitness (e.g.  when near some optimum).  Here, brains and sensory systems act through the categorizing power of mate choice so as to magnify small fitness differences, effectively separating individuals who would otherwise have indistinguishable fitnesses (and have the same number of offspring) into different distinguishable fitnesses — and thereby greatly increasing the variance in the number of offspring. 

Third, mate choice mechanisms can pick out phenotypic traits that are different from those on which natural selection itself acts, but that are highly correlated with natural-selective fitness.  For example, bilateral symmetry may be an important correlate of ecological success for many vertebrates.  But natural selection might increase the degree of symmetry in a particular lineage only very indirectly through its effects on several different correlates of symmetry, such as locomotive efficiency (individuals with asymmetric legs won't be able to get around as well and so will be selected against on the grounds of their locomotive inefficiency, rather than being selected against for asymmetry per se).  By contrast, mate preferences for perceivable facial and bodily form can directly select for symmetry in a way that natural selection cannot.  1.  In general, mate choice can complement natural selection by operating on perceivable phenotypic attributes that underlay a wide array of economic traits, but which would typically be shaped only indirectly by a number of different, weak, indirect natural selection pressures.  To continue our analogy between brains and optical devices, mate choice mechanisms can act as panoramic lenses, bringing into view a wider array of phenotypic features than natural selection alone would tend to focus on. 

Natural selection is extremely efficient at eliminating major genetic blunders, such as highly deleterious mutations or disruptive chromosome duplications — it simply prevents the afflicted individual from reaching reproductive maturity.  But the more subtle task of shaping and optimizing complex adaptations may be more difficult for direct ecological selection pressures to manage.  Natural selection alone can of course accomplish wonderful things, given enough time: 3.5 billion years of prokaryote evolution (amounting to many trillions of generations) has produced some quite intricate biochemical adaptations in these single-celled organisms.  But for larger-bodied animals with slower generation times, we believe that selective mate choice plays a major role in the optimization of complex adaptations.  For such species, the efficacy of natural selection may depend strongly on shaping the mate choice mechanisms that "take over" via sexual selection and do much of the difficult evolutionary work. 

There is suggestive data that support this claim.  Bateson (1988) replotted data from Wyles, Kunkel, and Wilson (1983), and found a strong positive correlation across several taxa between rate of evolution (assessed by a measure of morphological variability across eight traits) and relative brain size (see Figure 1).  For example, song birds have larger brains than non-song birds, and apparently evolve faster; humans have the largest brains of all primates, and apparently evolve the fastest.  Bateson (1988) interpreted this correlation in terms of larger brains allowing better habitat choice, a stronger "Baldwin effect" (in which the ability to learn actually speeds up the evolution of unlearned traits — see Hinton and Nowlan, 1987), and various forms of "behaviorally induced environmental change" — but he overlooked the potential effects of brain size on sexual selection patterns.  We believe it is more important that larger brains allow more powerful and subtle forms of selective mate choice.  Indeed, the vastly enlarged human brain has allowed us not only to (unconsciously) impose strong sexual selection on members of our own species (Darwin, 1871; Miller, 1993), but also to impose very strong artificial selection on members of other species (Darwin, 1859).  The correlation between brain size and rate of evolution provides a suggestive start for studies of the relationship between the capacity for selective mate choice and the rate and course of evolution, but clearly much more data is needed on this issue 

Escaping evolutionary local optima through sexual selection .  

Escaping local optima: The relative power of "sexual-selective drift", genetic drift, and neutral drift 

Populations can become perched on some adaptive peak in the fitness landscape through the optimizing effect of sexual and natural selection acting together.  But many such peaks are only local evolutionary optima, and better peaks may exist elsewhere.  Once a population has converged on such a locally optimal peak then, how can it move off that peak, incurring a temporary ecological fitness cost, to explore the surrounding adaptive landscape and perhaps find a higher-fitness peak elsewhere? Wright's (1932, 1982) "shifting balance" theory was designed to address this problem of escaping from local evolutionary optima.  He suggested that genetic drift operating in quasi-isolated populations can sometimes allow one population to move far enough away from its current fitness peak that it enters a new adaptive zone at the base of a new and higher fitness peak.  Once that population starts to climb the new fitness peak, its genes can spread to other populations, so that the evolutionary innovations involved in climbing this peak can eventually reach fixation throughout the species.  Thus, the species as a whole can climb from a lower peak to a higher one. 

Wright's (1932) model anticipated some of the recent concerns about how to take "adaptive walks" that escape from local optima in rugged fitness landscapes (Kaufmann, 1993).  In very rugged landscapes, short steps (defined relative to the landscape's ruggedness) of the sort generated by genetic point mutations are unlikely to allow individuals or populations to escape a local optimum.  This is similar to Darwin's (1883) problem of how minor mutations can accumulate into useful adaptations if they have no utility in their initial form.  But jumping further across the landscape does not guarantee success, either: longer steps of the sort generated by macromutations (as favored by Goldschmidt, 1940) are unlikely to end up anywhere very reasonable; most mutations are deleterious, and major mutations even more so.  The central problem is to make the "foray length" of population movements away from local optima able to exploit the "correlation length" of the adaptive landscape, and allow directional excursions away from the current adaptive peak to explore the surrounding fitness landscape.  Wright's shifting balance model suggests that genetic drift might provide enough random jiggling around the local optimum to sometimes knock the population over into another adaptive zone, but the analysis of adaptive walks in rugged fitness landscapes (Kaufmann, 1993) indicates that this is unlikely to be a common occurrence. 

Our model of population movement in phenotype space via mate choice is similar to Wright's shifting balance theory, but it provides a mechanism for exploring the local adaptive landscape that can be much more powerful and directional than random genetic drift: sexual selection.  Here, we are relying on a kind of "sexual-selective drift" resulting from the stochastic dynamics of mate choice and runaway sexual selection, to displace populations from local optima.  We suspect that with mate choice, the effects of sexual-selective drift will almost always be stronger and more directional than simple genetic drift for a given population size, and will be more likely to take a population down from a local optimum and over into a new adaptive zone.  Genetic drift relies on passive sampling error to move populations down from economic adaptive peaks, whereas sexual selection relies on active mate choice, which can overwhelm even quite strong ecological selection pressures.  Our simulations have shown that with directional mate preferences in particular, populations move around through phenotype space much more quickly than they would under genetic drift alone (Miller & Todd, 1993).  Thus, sexual selection can be seen as a way of making Wright's shifting balance model much more powerful, by allowing active mate choice dynamics to replace passive genetic drift as the main source of evolutionary innovation. 

Aside from classical genetic drift (sampling error in small populations), "neutral drift" through adaptively neutral mutations (Kimura, 1983) might conceivably play an important role in allowing populations to explore high-dimensional adaptive landscapes.  The idea is this: the more dimensions there are to an adaptive landscape, the less problematic local optima will be, because the more equal-fitness "ridges" there will be from one optimum to another in the space.  A local optimum may be a peak with respect to each of two phenotypic dimensions, but it is unlikely to be a peak with respect to each of a thousand dimensions, so there will be plenty of room for adaptively neutral exploration of phenotype space (see Eigen, 1992; Schuster, 1988). Under this model, populations can drift around through adaptive landscapes without incurring fitness costs for doing the exploration. 

The neutral drift theory is usually applied to molecular evolution (DNA base pair substitutions typically do not change expressed protein functionality), but it could in principle extend to morphology and behavior.  For example, if quadrupedalism and bipedalism happen to have equal locomotive efficiency in a certain environment (such as the Pleistocene savanna of Africa), a population might drift from the former to the latter without incurring much fitness cost in between, and without natural selection in favor of bipedalism per se. Although both ways of moving may be equal in locomotive efficiency, they have very different implications with respect to other potential activities such as tool use.  Once the population drifts into bipedalism, it will happen to enter a new adaptive zone wherein natural selection can favor new adaptations for tool use, resulting in an evolutionary innovation with respect to tool use.  Thus, if the problem of local optima in high-dimensional adaptive landscapes really is over-stated, then neutral drift from one adaptive zone to another might facilitate the discovery of evolutionary innovations associated with different adaptive peaks. 

However, we believe that for complex phenotypic adaptations at the level of morphology and behavior, the problems of local optima are not so easily overcome.  The evolutionary conservatism characteristic of many morphological and behavioral traits in many taxa suggests that neutral drift has trouble operating on such traits.  Still, so little is known about neutral drift above the level of molecules that such arguments are not convincing.  We can however ask, if neutral drift theory does apply to complex phenotypic traits, is neutral drift through phenotype space likely to be faster with or without the capricious dynamics of sexual selection? Here again, we believe that populations capable of mate choice are more likely to move along fitness ridges and exploit the possibilities of neutral drift, because mate choice can confer more mobility and momentum on evolving populations 

The role of sexual dimorphism in escaping local optima through sexual selection 

As Darwin (1871) noted, females are often choosier than males about their mates, so sexual selection often acts more strongly on males.  Sexually dimorphic selection pressures will often result in sexually dimorphic traits, although dimorphism in a trait tends to evolve much more slowly than the trait itself (Lande, 1980, 1987).  Thus, Darwin was able to use sexual dimorphism as a diagnostic feature for a trait having evolved through sexual selection.  But the effects of sexual dimorphism on longer-term evolutionary processes have rarely been considered. 

Highly elaborated male courtship displays, whether behavioral or morphological, are often costly in terms of the male's "economic" success with respect to the surrounding econiche.  Indeed, according to Zahavi's (1975) handicap theory, this cost is indirectly the reason why elaborated displays can evolve under sexual selection.  If we view a dimorphic population as situated in an adaptive landscape that represents purely ecological (economic) fitness, then the females will be situated close to the fitness peak, while the males will be situated some distance from the peak, and thus lower on the fitness landscape.  As the male displays become more elaborated and more costly, the males will travel further away from the fitness peak that represents economic optimality. 

Thus, sexual dimorphism in courtship traits leads to a kind of sexual division of labor with respect to the job of exploring adaptive landscapes.  Males get pushed off economic fitness peaks by the pressure of female choice in favor of highly elaborated, costly courtship displays.  Due to the typical lack of male choosiness, the females can stay more comfortably situated near the economic fitness peak.  Thus, males become the explorers of the adaptive landscape, compelled to wander through the space of possible phenotypic designs by the demands of female choice to "bring home" a sexy, interesting, and expensive courtship display.  The economic costs of wandering through phenotype space are compensated for by the reproductive benefits of attracting mates with a costly, elaborated courtship display.  In most species most of the time, the males will reach some equilibrium distance (Fisher, 1930; Kirkpatrick, 1982), close enough to the economic fitness peak to survive, but far enough away to demonstrate their viability and to incur the costs of an elaborate display, and the species will be recognized as having some sexually dimorphic traits. 

But sometimes, in some species, the males might stumble upon a new adaptive zone in the course of their wanderings.  That is, a sexually-elaborated trait, or some phenotypic side-effect of it, could prove economically useful, and would be subject to favorable natural selection.  The males would then start to climb the new economic fitness peak; and once the males reach a level of economic benefit on this new peak that exceeds the benefit obtainable on the old fitness peak, then there can be selection for females as well to move from their position on the old peak to the new, higher, peak.  This selection on females would act to eliminate the sexual dimorphism that maintains the useful new traits in the males alone, so that the females too could inherit the new trait (from their fathers initially).  Thus, once the males enter a new adaptive zone and start to climb a higher fitness peak, a combination of natural selection and reduced sexual dimorphism may move the entire population, males and females, to the top of the new fitness peak.  Populations that successfully shift from one adaptive peak to another will show little sexual dimorphism for the original courtship traits that brought them into the region of the new peak, since selection on the females will have worked to remove it; instead, they will be recognized as beneficiaries of an evolutionary innovation characteristic of both males and females.  So it may be difficult to recognize modern species that have undergone this peak-jumping process except through careful analysis of the fossil record; computer simulation may be more useful in determining whether this peak-jumping mechanism is plausible (see Todd & Miller, in preparation). 

This possibly rapid shift between fitness peaks resembles what Simpson (1944) called "quantum evolution" or what Eldredge and Gould (1972) called a "punctuation".  The quantum evolution term is apt because our theory suggests that populations capable of sexual dimorphism can do a kind of "quantum tunneling" between adaptive peaks: the normal economic costs that slow movement across low-fitness valleys between peaks can be over-ridden by geneological (sexually selected) benefits to the males, allowing them to traverse the valleys much more quickly.  The females can then join the males once a new peak is actually discovered.  The result could be much more rapid movement between peaks than would be possible under natural selection alone. 

This rapid tunneling between peaks looks strange from the perspective of the purely economic adaptive landscape that represents only natural selection pressures.  But that landscape is not the whole picture: the effects of sexual selection establish a separate "reproductive landscape" with different dimensions and perhaps a different topography for males and females.  The economic and reproductive landscapes together combine to form a master adaptive landscape; what looks like paradoxical downhill movement or quantum tunneling in the purely economic landscape traversed by natural selection may actually be hillclimbing in the combined landscape that includes sexual selection pressures. 

But won't these initially economically-unfeasible excursions by the males threaten their survival, and hence that of the species as a whole? Sexual selection is often maligned for just this reason, as "a fascinating example of how selection may proceed without adaptation" (Futuyma, 1986, p.  278), on the principle that the economic costs of highly elaborated male courtship displays might predispose a species to extinction — e.g.  as argued by Haldane (1932), Huxley (1938), and Kirkpatrick (1982).  But as Pomiankowski (1988) has emphasized, the relationship between male economic success and population viability is quite complex and unclear.  Reproductive output in sexually-reproducing species is typically limited by the number of females, not by the number of males.  The population's rate of replacement will not necessarily be decreased by the loss of male viability due to elaborated courtship displays.  On the contrary: "a population denuded of males will have more resources available for females and so may support an absolutely larger reproductive output for a given resource base" (Pomiankowski, 1988).  Thus, the population-level costs of sexually-elaborated traits may be minimal, and the individual-level benefits may be large, due to sexual selection.  This makes quantum tunneling between adaptive peaks through sexual selection a plausible mechanism for generating evolutionary innovations and escaping local ecological optima. 

At first glance, our proposal bears an uncomfortable resemblance to traditional sexist images of males going out to hunt and sometimes returning with meat for the benefit of their families.  But females may also do some important exploration of the adaptive landscape, with respect to different phenotypic dimensions.  Under Fisher's (1930) runaway selection model for example, female preferences and male traits both become elaborated through sexual selection.  Females become ever-choosier and more discriminating.  The benefits of selective mate choice can favor the evolution of new sensory, perceptual, and decision-making adaptations in females, despite their economic costs.  Thus, while males are exploring the space of possible secondary sexual characteristics and behavioral courtship displays under sexual selection, females may be exploring the space of possible sensory, perceptual, and cognitive traits.  If the females happen upon a mate choice mechanism such as a new form of color vision or better timbre perception that also happens to have economic benefits in their econiche, then we would expect such mechanisms to be further modified and elaborated through natural selection, and inherited by males also, eventually showing low dimorphism.  Thus, females can also tunnel between peaks in the space of possible perceptual systems, deriving the reproductive benefits of selective mate choice even when a perceptual system shows little ecological benefit. 

In summary, sexual selection provides the easiest, fastest, and most efficient way for populations to escape local ecological optima.  Sexual dimorphism with respect to courtship traits and mate preferences allows a sexual division of labor in searching the adaptive landscape.  Many morphological and behavioral innovations that currently show economic utility and low sexual dimorphism may have originated as parts of male courtship displays.  Likewise, many sensory, perceptual, and decision-making innovations could have originated as components of female choice mechanisms, and later have been modified for ecological applications.  Those innovations that did not happen to show any ecological utility remained in their sexually dimorphic form, and are typically not recognized as innovations at all.   

Sexual selection and evolutionary innovations  

The mystery of evolutionary innovations 

Evolutionary innovations are important because natural selection crafts adaptations out of innovations: "Innovation is the mainspring of evolution" (Jablonski & Bottjer, 1990, p.  253).  Classic examples of major evolutionary innovations include the bony skeleton of vertebrates, the jaws of gnathostomes, the amniote egg, feathers, continuously growing incisors, large brains in hominids, the insect wing, and insect pollination of angiosperms (Cracraft, 1990).  But the complete list of major evolutionary innovations is almost endless, being virtually synonymous with the diagnostic characters of all successful higher taxa, and the complete list of minor innovations would include essentially all diagnostic characters of all species. 

But, for all their biological importance and large number, the causal origins of evolutionary innovations have been long contended and remain poorly understood.  Virtually every major evolutionary theorist has tackled the problem of evolutionary innovations, e.g.  Darwin (1859, 1871, 1883), Romanes (1897), Weismann (1917), Wright (1932, 1982), Simpson (1953), Mayr (1954, 1960, 1963), and Gould (1977).  But the major questions remain unresolved (see Nitecki, 1990, for a recent review).  This section reviews the history of evolutionary thinking about innovations; section 6.2 examines the most baffling features of innovations; section 6.3 suggests that sexual selection through mate choice can help explain the strange pattern of innovations in animals and flowering plants; section 6.4 outlines some limits to our hypothesis; and section 6.5 concludes the discussion of innovations. 

Darwin, particularly in the sixth edition of the Origin of species (Darwin, 1883), worried about the early evolutionary stages of "organs of extreme perfection" such as the human eye and the bird's wing.  How could these innovations be preserved and elaborated before they could possibly assume their later survival function (such as vision or flight)? The problem for Darwin was to account for the origin of phenotypic innovation that was more complex and well-integrated than what random mutation could produce, but that was not yet useful enough in the struggle for existence to have been favored by natural selection.  Mutations seemed able to generate only trivial or disastrous phenotypic changes, so could not account for the origins of useful innovations, whereas natural selection could only optimize innovations already in place.  Nor could Darwin convince skeptics that some mysterious interplay between mutation and selection could account for evolutionary innovations. 

Darwin's difficulty in accounting for evolutionary innovations was one of the weakest and most often-attacked aspects of his theory of natural selection.  Even his most ardent followers were anxious about this problem.  Romanes (1897) was very concerned to show how "adaptive characters", or evolutionary novelties, originate.  For him, this was the central question of evolutionary theory, much more important than the question of how species originate, but one that he was  never able to answer to his own satisfaction.  Simpson (1953) later proposed that "key mutations" can cause a lineage to enter a new "adaptive zone" such that the lineage undergoes an adaptive radiation, splitting apart into a large number of species to exploit all the ecological opportunities in that new adaptive zone.  Similarly, Mayr (1963) defined an evolutionary innovation as "any newly acquired structure or property that permits the performance of a new function, which, in turn, will open a new adaptive zone" (Mayr, 1963, p.  602).  However, both Simpson and Mayr were better able to describe innovation's effects than to explain its causes.  Their notion that major innovations are closely associated with adaptive radiations has been a persistent theme in innovation theory, appearing more recently under the guise of "key evolutionary innovations" in Liem (1973, 1990), and "key characters" in Van Valen (1971). 

Over this long history, several kinds of explanations have been offered to explain the emergence of evolutionary innovations.  Goldschmidt (1940) suggested that macromutations could produce fully functioning novelties in the form of "hopeful monsters".  The problem is that random macromutations are overwhelmingly unlikely to generate the sort of structural complexity and integration characteristic of innovations even in their early stages.  Complex innovations cannot be explained by undirected random mutation.  On the other hand, Fisher (1930) took the Darwinian hard line and maintained that innovations could indeed be produced purely through natural-selective hill-climbing.  The difficulty with this view is that it ignores the problem of local optima, as discussed in section 5.  Significant innovation corresponds to fairly substantial movement through a multi-dimensional adaptive landscape.  But because many adaptive landscapes have complex structures (Eigen, 1992; Kauffman, 1993), with many peaks, ridges, valleys, and local optima, long movements through such landscapes may often require escaping from local optima.  As section 5.1 emphasized, this problem of escaping local optima is probably more serious at the level of complex phenotypic design than at the level of genetic sequences or protein shapes (cf.  Eigen, 1992) — and most evolutionary innovations of interest to biologists are innovations in complex phenotypic design.  Thus, the evolution of a new phenotypic innovation may often reflect escape from a local adaptive optimum and the discovery of a better solution elsewhere in the space of possible phenotypes (Wright, 1932; Patterson, 1988).  Finally, other theorists have stressed the role of phenotypic structure in allowing for innovations, through phenotypic by-products of other adaptive change (Mayr, 1963), through various mechanisms of phenotypic self-organization (e.g.  Eigen, 1992; Kauffman, 1993), and through changes in developmental mechanisms, particularly "heterochronies" that affect the relative timing of the development of different traits (Bonner, 1982; Goodwin et al., 1983; Gould, 1977; Muller, 1990; Raff, 1990; Raff & Raff, 1987).  These sorts of phenotypic constraints and correlations are probably important, but as we will see, they cannot explain the most striking features of the distribution of evolutionary innovation.  There are three major problems for these traditional theories about evolutionary innovation; these will be examined in turn.   

Three puzzling aspects of evolutionary innovation 

First, there is a disparity between the huge number of minor varietal innovations and the small number of ecologically useful innovations.  Darwin (1883, p.  156) stressed this problem when he quoted Milne Edwards: "Nature is prodigal in variety but niggardly in innovation.  Why ...  should there be so much variety and so little real novelty?".  The vast majority of characteristic innovations are "inconsequential" (Liem, 1990); they are what Francis Bacon called "the mere Sport of Nature" when he disparaged the apparently pointless variety of animals, plants, and fossils (quoted in Cook, 1991.) Only very few of the initially inconsequential minor innovations may lead to major innovative evolutionary shifts in form or function that allow the invasion of major new habitats and adaptive zones.  But if evolutionary innovations spread through populations under the influence of traditional natural selection for their ecological utility, why do so few innovations show the sort of ecological utility that characterizes key innovations? 

Second, there is often a disparity in time between the causal origin of an innovation and the ultimate ecological and evolutionary effect of an innovation.  The causes of evolutionary innovations must be clearly separated from their possible effects on diversification, niche exploitation, or adaptive radiation (Cracraft, 1990).  "Key innovations" that allow a monophyletic taxon to radiate outwards into a number of new niches can only be identified post-hoc, after their success has been demonstrated evolutionarily.  Immediately after they originate, evolutionary innovations are just innovations pure and simple.  Their prospective future ecological utility as fully elaborated traits cannot bring them into being initially.  If we wish to understand the actual causal origins of evolutionary innovations, we must look within the species where the innovation originated, not at the ultimate macroevolutionary consequences of the innovation.  Liem has stressed this point, observing that "An evolutionary novelty may remain in a stasis for extended times when it does not convey an improvement in the matter/energy transfer" (Liem, 1990, p.  161), and "historical tests also show that there is often a great delay between the emergence of a KEI [key evolutionary innovation] and the onset of the diversification it is assumed to cause" (Liem, 1990, p.  165), due to its newfound ecological utility.  Earlier, he also noted that "adaptive radiations will not occur until after an evolutionary novelty has reached a certain degree of development" (Liem, 1973, p.  426).  Jablonski (1986, 1990) has also observed that many innovations fail to persist, let alone trigger a diversification indicative of ecological utility.  Thus, to account for innovations, we must explain the origin and elaboration of many integrated morphological and behavioral systems that only rarely manifest much survival utility.  We seem to need a form of iterative Darwinian selection other than natural selection for ecologically useful survival traits. 

Third, the distribution of innovations in animals and flowering plants is not random with respect to phenotypic features, but is highly concentrated in features subject to sexual selection.  Traditional theories of innovation through natural selection or through phenotypic constraints and correlations have trouble accounting for this distribution, which is seen most clearly when we consider the methods of biological taxonomy.  The most common features used by taxonomists to distinguish one species from another should logically be the sorts of features most characteristic of (at least minor) evolutionary innovations.  This is an almost tautological result of the fact that taxa, including species, are in some sense made up of their innovations (Weismann, 1917): their innovations are their critical defining features.  The most commonly used defining features for species appear to be primary and secondary sexual traits, and behavioral courtship displays, which Mayr (1960) would have misinterpreted as "species recognition signals".  And a great many of these traits, used in the identification of species of animals and flowering plants and discussed in speciation research, are just the sort of characteristics most likely to have arisen by sexual selection through mate choice.  Studies of evolutionary innovation that rely on reconstructing explicit phylogenies often rely on such features.  For example, in Cracraft's (1990, pp.  31-35) analysis of evolutionary innovations in the Pionopsitta genus of South American parrots, every single one of the 30 innovations discussed was a distinctive plumage color pattern or plumage growth pattern that could have been elaborated through mate choice, such as "bright orange-red shoulder patch", "crown bright red in male, not female", "yellow collar around head", or "crown and back of neck black".  Moreover, it is often easier in taxonomy to identify the species of a male than of a female animal, because secondary sexual characters are typically more elaborated in males, whereas females more often retain camouflaged and ancestral forms (Eberhard, 1985). 

So, in Eldredge's (1989) terminology, reproductive rather than economic traits are often used to distinguish between species.  In section 7.1, we claim that speciation can result from a stochastic divergence of mate choice criteria in a sympatric population leading to a disruption of the mate recognition system within a given species.  If so, most traits distinguishing one species from another — that is, most minor evolutionary innovations — are sexual characters or courtship displays that arose through mate choice.  Moreover, the biological species concept, which views species as reproductively isolated populations, virtually demands that the innovations that distinguish one species from another must function as reproductive isolators — that is, as traits subject to selective or assortative mate choice.  Thus, both the empirical methods of taxonomists and the theoretical presuppositions of the biological species concept, suggest that most evolutionary innovations in animals and flowering plants arose through sexual selection acting on traits capable of creating reproductive isolation between populations, particularly primary and secondary sexual traits, and courtship behaviors. 

To explain evolutionary innovations then, we need to account for the following facts.  (1) Most innovations are too complex and well-integrated to have resulted simply from random mutation or genetic drift, and are too structurally and functionally novel (i.e.  functionally non-neutral) to have resulted simply from neutral drift.  (2) Many innovations require escape from an evolutionary local optimum, which natural-selective hill-climbing tends to oppose.  (3) Most innovations (i.e.  most traits taxonomically useful in distinguishing species) show very little ecological utility and do not result in adaptive radiations.  (4) Those innovations that do eventually show ecological utility often show a long delay between their origin and their proliferation.  (5) Finally, most innovations in animals and flowering plants are heavily concentrated in phenotypic traits subject to mate choice, and this distribution cannot be explained by models of innovation through general phenotypic correlations and constraints.  In general then, the origins of evolutionary innovations must be explained in terms of some kind of selection between individuals that has little effect on ecological success and that only rarely leads to macroevolutionary success.  "Irrespective of whether innovations are perceived as `large' or `small', they all must arise and become established at the level of individuals and populations, not higher taxa" (Cracraft, 1990, p.  28).  Thus, innovations that characterize an entire population or species must be explained at some level above that of simple mutation or developmental constraints, but below that of macroevolutionary `sifting' between species (Vrba & Gould, 1986), and aside from that of natural selection for ecological utility.   

The role of mate choice in generating evolutionary innovations 

Sexual selection through mate choice can account for all of these features of evolutionary innovation in animals and flowering plants.  Thus, Darwin's "prodigal variety", may arise from a long-overlooked wellspring of innovation — the effects and side-effects of mate choice.  These sexually-selected varietal novelties could be called "courtship innovations." From these humble origins, a few incipient courtship innovations may continue to be elaborated into more and more complex morphological and behavioral characteristics.  At various points in this evolutionary course of elaboration, a tiny minority of courtship innovations and their phenotypic by-products will happen to show some ecological utility, and may be modified to form new "economic innovations" that have some ecological utility. And a tiny minority of these economic innovations will prove important enough that they allow adaptive radiations and later come to be recognized as "key innovations." Thus, the causal origins of key innovations may often be the same as the causal origins of courtship innovations: elaboration of a trait by sexual selection through mate choice.  The net result of sexual selection's innovativeness may be that sexual selection is to macroevolution what genetic mutation is to microevolution: the prime source of potentially adaptive heritable variation, at both the individual and species levels 

What kinds of evolutionary innovations can be generated through sexual selection? 

Our theory that many evolutionary innovations arise at first through the effects of selective mate choice, or as side-effects of sexually-selected traits, must be clarified and given some caveats.  First, and most obviously, the theory applies only to biological systems where mate choice operates in some fashion.  We have lumped together flowering plants and animals because they both undergo a form of sexual selection by animals with nervous systems, either heterospecific pollinators or conspecifics.  Evolutionary innovations in asexual lineages, and in sexually reproducing organisms that are too simple to exercise heritable patterns of nonrandom mate choice, must be explained in some other way.  But since innovations seem to emerge much more slowly and sparsely in lineages without mate choice, there is less that needs explaining.  Thus, we would expect the frequency distribution of evolutionary innovations to be highly skewed across lineages, clustered in species subject to high levels of selective mate choice.  As sections 6.1 and 7.2 argue, this is just what we see. 

Second, selective mate choice can directly affect only those phenotypic traits that are perceivable to the animal doing the selecting, given its sensory and perceptual capabilities.  Thus, mate choice typically applies to macroscopic morphology and manifest behavior.  But it also applies indirectly to any microscopic morphology, physiology, neural circuitry, or biochemistry that affects the appearance of the perceivable traits or behaviors, e.g.  the iridescence of bird feathers carried by microscopic diffractive structures on feathers, the complex courtship behavior generated by hidden neural circuits, or the persistent bird song allowed by an efficient energy metabolism.  Furthermore, elaboration of these sexually-selected traits may often have phenotypic side-effects on many other traits, and ecologically useful innovations may sometimes emerge from these side-effects.  So we would expect the frequency distribution of evolutionary innovations across phenotypic traits to be highly skewed, clustered around traits that are directly subject to mate choice (such as genitals, secondary sexual morphology, and courtship behaviors), and spreading outwards from these traits to others that are structurally, behaviorally, or developmentally correlated. 

Third, as a corollary of the previous point about phenotypic side-effects, our theory may have fairly limited application to evolutionary innovation in the traits of flowering plants, apart from flowers themselves.  Pollinators can directly select for flower traits such as shape, color, smell, and size, but it is unclear how easy it would be for floral innovations to become modified into ecologically useful new kinds of seeds, fruits, or chemical defenses, much less new kinds of twigs, leaves, or roots.  (On the other hand, the evolution of insectivorous plants such as the Venus Fly-Trap was probably facilitated by the ease with which attractive flowers can be modified from pollination to predation functions.) Moreover, despite the fact that the complexity of plant behavior has often been underestimated (see Darwin, 1876; Simon, 1992), plants cannot use shifts in behavior and habit to smooth the way for changes of morphological function as easily as animals do (Darwin, 1883; Bateson, 1988).  As a result, the modification of courtship innovations into economic innovations in plants may be more difficult than in animals.  However, polymorphism and sympatric speciation could almost certainly be facilitated through flower selection by pollinators, as the data from Eriksson and Bremer (1992) suggest. So the effects of pollinator choice might at least explain the higher speciation rates and high rates of floral innovation in flowering plants.   

Summary: An overview of evolutionary innovation through sexual selection 

Species perched on adaptive peaks will generally have mate choice mechanisms complementary to the natural-selective pressures keeping them there, so long periods of stasis will ensue for most species, most of the time.  But occasionally, directional preferences, or intrinsic perceptual biases in preferences, or genetic drift acting on preferences, can lead to runaway dynamics that take a population (or at least the males) away from the ecological fitness peak.  So the effects of mate choice can be visualized as vectors that pull populations away from adaptive peaks out on long forays into the unknown, where they may or may not encounter new ecological opportunities and evolve economically useful traits.  If they do not encounter new opportunities, little is lost: the males will have sexually dimorphic courtship innovations, and the females will have mate choice mechanisms, both of which have some economic costs but substantial reproductive benefits.  But if they do encounter new opportunities, much is gained: if the male courtship innovation or the female mate choice mechanism happens to be modifiable into a useful economic innovation, then it will be elaborated through natural selection and its degree of sexual dimorphism will decrease.  The lucky population will enter a new adaptive zone, rapidly climb the new peak, and may often become reproductively isolated from other populations.  The result could look like a period of rapid evolution concentrated around a speciation event, just as described by punctuated equilibrium theory (Eldredge & Gould, 1972).  Moreover, if the new adaptive zone happens to be particularly large and fruitful, and the economic innovation proves particularly advantageous, then the event will look like the establishment of a key evolutionary innovation, and may lead to the formation of new higher taxa.   


Sympatric speciation through sexual selection 

Parallel computation can be faster than serial computation.  This principle also applies to evolutionary processes of "biocomputation".  At one level, the adaptive power on natural selection exploits parallelism across the genes, gene complexes, and individuals within a population.  But at another level, a single population exploring an adaptive landscape is not as efficient as a set of populations exploring in parallel.  As section 5.2 discussed, sexual dimorphism between males and females allows one sub-population to stay perched on an old adaptive peak while another explores the surrounding phenotype space for other adaptive peaks.  Are there any more powerful methods of parallel search in biocomputation, that would allow many "search parties" to branch out across the adaptive landscape? 

Speciation does exactly that.  When a biological lineage splits apart into reproductively isolated subpopulations, one search party is replaced by two independent parties.  Here again, we can ask whether mate choice and sexual selection can help biocomputation, this time through facilitating speciation.  Though vitally interested in both speciation and mate choice, Darwin did not seem to perceive this connection, and the Origin of species (1859) in fact offered no clear mechanism of any sort whereby speciation could happen.  The biologists of the Modern Synthesis (e.g.  Dobzhansky, 1937; Huxley, 1942; Mayr, 1942) saw species as self-defined reproductive communities, and yet often argued against the idea that sexual selection, the obvious agent of reproductive self-definition, could induce speciation, because their attitude towards Darwin's theory of selective mate choice was so hostile. 

Instead, two major theories of speciation developed during the Modern Synthesis, and both suggested that speciating populations are split apart by some divisive force or "cleaver" external to the population itself.  The cleaver separates the population in twain genetically and phenotypically, and then reproductive barriers arise afterwards through genetic drift or through selection against hybridization.  In Mayr's (1942) model of allopatric speciation, the cleaver is a new geographic barrier arising to separate previously interbreeding populations.  For example, a river may shift course to isolate one population from another.  Some combination of genetic drift, the "founder effect", and disruptive selection then causes the two newly isolated groups to diverge phenotypically.  Once enough phenotypic divergence accumulates, the populations can no longer interbreed even when the physical barrier disappears, and so are recognized as separate species.  Speciation for Mayr was a side-effect of geographical separation. 

In Dobzhansky's (1937) model of sympatric speciation, the cleaver is more abstract: it is a low-fitness valley in an adaptive landscape, rather than a barrier in geographic space.  For example, an adaptive landscape might develop two high-fitness peaks (econiches) separated by a low-fitness valley.  This valley could establish disruptive selection against interbreeding between the peaks, thereby driving an original population to split and diverge towards the separate peaks in two polymorphic subpopulations.  Dobzhansky further suggested that after divergence, reproductive isolation evolves through selection against hybridization: since hybrids will usually fall in the lower-fitness valley, mechanisms to prevent cross-breeding between the separate populations will tend to evolve.  Thus the evolution of reproductive isolation (speciation itself) is viewed as a conservative process of consolidating adaptive change rather than a radical process of differentiation.  Vrba (1985) and Futuyma (1986) concur that speciation serves a conservative function, acting like a `ratchet' in macroevolution: only reproductive isolation allows a newly diverged population to effectively consolidate its adaptive differentiation; otherwise, the parent species will tend to genetically re-absorb it. 

A recent development in sympatric models is Paterson's (1985) concept of specific mate recognition systems (SMRSs).  SMRSs are phenotypic mechanisms each species uses to maintain itself as a self-defining reproductive community — in our terms, a set of mate choice mechanisms for assortative mating.  A species is thus considered the largest collection of organisms with a shared SMRS.  In Paterson's view, sympatric disruption and divergence of these SMRSs themselves (through some unspecified processes) can lead to speciation.  Eldredge (1989, p.  120) emphasizes the potential macroevolutionary significance of SMRSs: "significant adaptive change in sexually reproducing lineages accumulates only in conjunction with occasional disruptions of the SMRSs." 

Historically, the acceptability of sympatric models has depended on the perceived ability of disruptive selection to generate stable polymorphisms and eventually reproductive isolation.  A large number of experiments reviewed by Thoday (1972) show that disruptive selection is sufficient to generate phenotypic divergence even in the face of maximal gene flow between populations (which Mayr, 1963, p.  472, saw as the Achilles' heel of sympatric speciation models), and that mechanisms of reproductive isolation would evolve to avoid hybrids and consolidate that divergence.  Computer models by Crosby (1970) showed that sympatric speciation could occur when populations choose different micro-habitats, evolve stable polymorphisms through disruptive selection, and then evolve reproductive barriers to avoid hybridization.  But the speciation debate has continued to grind down to a question of whose cleaver is bigger: Mayr's (1942) geographic barriers or Dobzhansky's (1937) fitness valleys. 

To address this issue, we (Todd & Miller, 1991) developed a computer simulation of sexual selection that allowed for the possibility of "spontaneous" sympatric speciation through the interaction of assortative mating and genetic drift acting in a finite population.  We found that spontaneous speciation could indeed happen, even in the absence of any geographic isolation and even without any natural selection.  The rate of speciation increased with mutation rate and depended on the exact type of mate preference implemented.  Preferences for individuals similar to one's own phenotype yielded the highest speciation rate, while inherited preferences for individuals with particular specific phenotypes yielded lower rates of speciation.  In further investigations we found that spontaneous speciation also happens robustly with directional mate preferences, when the directional preference vectors happen to diverge and split the population into two subpopulations heading off on different trajectories through phenotype space (Miller & Todd, 1993); and that speciation can happen robustly as well when an individual's mate preferences are learned from the phenotypes of their parents through the process of "sexual imprinting learning" (Todd & Miller, 1993 

Sexual selection and the origins of biodiversity 

There is some biological evidence that speciation rates are indeed higher when selective mate choice plays a more important role.  Ryan (1986) found a correlation between cladal diversity in frogs and complexity of their inner ear organs (amphibian papilla), which are responsible for the operation of female choice on male calls.  He reasoned that "since mating call divergence is an important component in the speciation process, differences in the number of species in each lineage should be influenced by structural variation of the inner ear [and hence the operation of mate choice]" (p.  1379).  Immelmann (1972, p.  167) has argued that mate preferences derived from imprinting on the phenotypes of one's parents may speed speciation in ducks, geese, and the like: "imprinting may be of special advantage in any rapidly evolving group, as well as wherever several closely related and similar species occur in the same region [i.e.  sympatric situations].  Interestingly enough, both statements really do seem to apply to all groups of birds in which imprinting has been found to be a widespread phenomenon...".  The enormous diversity of insects (at least 750,000 documented species, maybe as many as 10 million in the wild) might seem at first sight to contradict the notion that mate choice facilitates speciation, since few (except Darwin) seem willing to attribute much mate choice to insects.  But Eberhard (1985, 1991, 1992) has shown that male insect genitalia evolve largely through the effects of cryptic female choice, in such as way that speciation could be promoted. 

Further evidence for speciation through mate choice comes from a consideration of biodiversity and the numbers of species across different kingdoms and phyla.  There seems to be a striking correlation between a taxon's species diversity and the taxon's evolutionary potential for sexual selection through mate choice, resulting in highly skewed richness of species across the five kingdoms.  Recent estimates of biodiversity suggest there may be somewhere between 10 and 80 million species on earth (May, 1990, 1992).  But of the 1.5 million or so species that have actually been identified and documented so far by taxonomists, the animal kingdom contains about 1,110,000, the plant kingdom contains about 290,000, the fungi contain about 90,000, the protists contain about 40,000, and the monera contain only about 5000 (Cook, 1991).  (It should be noted that sampling biases might account for a small amount of the skewness here: many animals and plants are larger and easier to notice and to classify than fungi, protists, or monera.) Although the vast majority of species in each kingdom can undergo some form of genetic recombination through sexual reproduction, only in the animals and the flowering plants is selective mate choice of central importance.  Of the 290,000 documented species of plants, about 250,000 are angiosperms (flowering plants) fertilized by animal pollinators.  And of the 1,110,000 documented species of animals, those with sufficient neural complexity to allow for some degree of mate choice (particularly the arthropods, molluscs, and chordates) are much more numerous than those without.  Thus, species diversity is vastly greater among taxa wherein a more or less complex nervous system mediates mate choice, either a conspecific's nervous system in the case of animals or in a heterospecific pollinator's nervous system in the case of flowering plants. 

This pattern is the opposite of what we might expect if allopatric speciation were the primary cause of biodiversity.  The effects of geographic separation (allopatry) should obviously be weaker for species whose reproduction is mediated by a mobile animal.  Animals can search over wide areas for mates and pollinators can fly long distances.  So allopatric speciation would predict lower species diversity among taxa whose reproduction is mediated by mobile animals with reasonably complex nervous systems — exactly the opposite of what we observe.  To further explore the role of selective mate choice in creating species biodiversity, we need to analyze the degree of mate choice in the various taxa more accurately, adjust the speciation rates between taxa for number of generations of evolution (and thus organism size), and if possible take into account the amount of geographic spread and migratory range of the species involved.  In this way, we hope to gain more evidence to show that sympatric speciation through mate choice, particularly through assortative mating, is the best explanation available for the extreme biodiversity of animals and flowering plants, and is thus the most powerful mechanism for dividing up and spreading out evolution's exploratory search of the adaptive landscape 

Implications and applications 

Implications for biology and psychology 

Biologists have been exploring the nuances of natural selection almost continuously since Darwin's time, and much has been learned.  By contrast, Darwin's (1871) theory of sexual selection through mate choice was virtually ignored until about 15 years ago, so the implications of sexual selection are only beginning to be realized.  This paper has made some strong claims about how natural selection and sexual selection might interact to explain long-standing mysteries in biology, such as how complex adaptations get optimized, how species split apart, and how evolutionary innovations get constructed before they show any ecological utility.  From the perspective of traditional natural selection research and the Modern Synthesis, these claims may look strange and implausible.  But Darwin may not have found them so.  Taking mate choice seriously does not mean abandoning Darwinism, adaptationism, optimality theory, game theory, or anything else of proven value in biology.  It simply means recognizing a broader class of selection pressures and a richer set of evolutionary dynamics than have been analyzed so far. 

Psychology has barely begun to recognize the role of natural selection in constructing mental and behavioral adaptations, much less the role of sexual selection in doing so.  One of our motivations for exploring the interaction of natural and sexual selection was our conviction that sexual selection may have played a critical role in the evolution of our unique human morphology (Szalay & Costello, 1991) and psychology (Miller, 1993).  The evolution of the human brain can be seen as a problem of escaping a local optimum: the ecologically efficient 500 cc.  brain of the Australopithecenes, who were perfectly good at bipedal walking, gathering, scavenging, and complex social life with their normal ape-sized brains.  During the rapid encephalization of our species in the last two million years, through the Homo habilis and Homo erectus stages up through archaic Homo sapiens, our ancestors showed very little ecological progress: tool making was at a virtual stand-still, the hunting of even small animals was still quite inefficient, and we persisted alongside unencephalized Australopithecene species for well over a million years.  These facts suggest that large brains did not give our lineage any significant ecological advantages until the last 100,000 years, when big-game hunting and complex tool-making started to develop quite rapidly — long after we had attained roughly our present brain size.  Instead, we propose that brain size probably evolved through runaway sexual selection operating on both males and females (Miller, 1993).  Human encephalization represents the most mysterious example of escape from a local ecological optimum, and we think the runaway dynamics of selective mate choice had everything to do with this escape 

Applications in genetic algorithms research and evolutionary design optimization 

If mate choice has been critical to the innovation, optimization, and diversification of life on our planet, we might expect that mate choice will also prove important in the design of complex artificial systems using genetic algorithms and other evolutionary optimization techniques.  Evolutionary engineering methods are often defended by claiming that we have a "sufficiency proof" that natural selection alone is capable of generating complex animals with complex behaviors.  But this is not strictly true: all we really know is that natural and sexual selection in concert can do this.  Indeed, the traditional assumption in genetic algorithms research that sexual recombination per se is the major advantage of sexual reproduction (Holland, 1975; Goldberg, 1989) may be misleading.  If instead the process of selective mate choice is what gives evolutionary power and subtlety to sexual reproduction, then current genetic algorithms work may be missing out on a major benefit of simulating sex.  For those interested in evolving robot control systems (e.g.  Cliff, Husbands, & Harvey, 1992; Harvey, Husbands, & Cliff, 1992, 1993) or other complex design structures (e.g.  Goldberg, 1989; Koza, 1993; see Forrest, 1993) through simulated natural selection, we suggest that incorporating processes of simulated sexual selection may help speed optimization, avoid local evolutionary optima, develop important new evolutionary innovations, and increase niche differentiation through speciation.  These effects may become particularly important as we move from pre-defined noise-free fitness functions to more complex, noisy, emergent fitness functions of the sort that arise when actually simulating ecosystems, coevolution, and other more naturalistic interactions.  Also, to the extent that the human brain evolved through runaway sexual selection (Miller, 1993), simulated sexual selection may help us cross the border between artificial life and artificial intelligence sometime in the future.  We will explore these issues in greater detail in upcoming papers (Miller, in press; Todd & Miller, in preparation 


Natural selection is fairly good at climbing fitness peaks in adaptive landscapes representing `economic' traits.  Sexual selection through mate choice has complementary strengths: it is good at making this natural-selective hill-climbing faster and more accurate, at allowing escape from local optima, at generating courtship innovations that may prove useful as economic innovations, and at generating biodiversity and niche differentiation through speciation.  The two processes together can yield a very powerful form of biocomputation that rapidly and efficiently explores the space of possible phenotypes, as shown by the diversity and complexity of animals and flowering plants on our planet.  We are the products not only of selection for survival, but also of selection for sexiness — dark-bright alloys forged in death and shaped by love. 


Geoffrey Miller has been supported by NSF Research Grant INT-9203229 and NSF-NATO Post-Doctoral Grant RCD-9255323.  For comments, support, advice, and/or inspiration relevant to this work, we are indebted to: Dave Cliff, Helena Cronin, Inman Harvey, Phil Husbands, Andrew Pomiankowski, Roger Shepard, and John Maynard Smith. 


    Barth, F.  G.  (1991).  Insects and flowers: The biology of a partnership. Princeton: Princeton Univ.  Press. 

    Bateson, P.  (Ed.).  (1983).  Mate choice.  Cambridge, UK: Cambridge U.  Press. 

    Bateson, P.  (1988).  The active role of behavior in evolution.  In M.-W.  Ho & S.  W.  Fox (Eds.), Evolutionary processes and metaphors, pp.  191-207.  New York: John Wiley. 

    Bonner, J.  T.  (Ed.).  (1982).  Evolution and development. Berlin: Springer-Verlag. 

    Clarke, B.  C.  (1962).  The evidence for apostatic selection.  Heredity (London), 24, 347-352. 

    Cliff, D., Husbands, P., & Harvey, I.  (1992).  Evolving visually guided robots.  In J.-A.  Meyer, H.  L.  Roitblat, & S.  W.  Wilson (Eds.), From Animals to Animats 2: Proceedings of the Second International Conference on Simulation of Adaptive Behavior, pp.  374-383.  Cambridge: MIT Press. 

    Cook, L.  M.  (1991).  Genetic and ecological diversity: The sport of nature. London: Chapman & Hall. 

    Cracraft, J.  (1990).  The origin of evolutionary novelties: Pattern and process at different hierarchical levels.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  21-44.  Chicago: U.  Chicago Press. 

    Cronin, H.  (1991).  The ant and the peacock: Altruism and sexual selection from Darwin to today. Cambridge: Cambridge Univ.  Press. 

    Crosby, J.  L.  (1970).  The evolution of genetic discontinuity: Computer models of the selection of barriers to interbreeding between subspecies.  Heredity, 25: 253-297. 

    Darwin, C.  (1859).  On the origin of species, 1st Ed.  London: John Murray. 

    Darwin, C.  (1862).  On the various contrivances by which orchids are fertilized by insects. London: John Murray. 

    Darwin, C.  (1871).  The descent of man, and selection in relation to sex. London: John Murray. 

    Darwin, C.  (1876).  The movements and habits of climbing plants, 2nd Ed.  New York: D.  Appleton & Co. 

    Darwin, C.  (1883).  On the origin of species, 6th Ed.  New York: D.  Appleton & Co. 

    Dewsbury, D.  A.  (1981).  Effects of novelty on copulatory behavior: The Coolidge Effect and related phenomena.  Psychological Review, 89(3), 464-482. 

    Dobzhansky, T.  (1937).  Genetics and the origin of species. (Reprint edition 1982).  New York: Columbia U.  Press. 

    Endler, J.  A.  (1992).  Signals, signal conditions, and the direction of evolution.  American Naturalist, 139, S125-S153. 

    Eberhard, W.  G.  (1985).  Sexual selection and animal genitalia. Cambridge: Harvard Univ.  Press. 

    Eberhard, W.  G.  (1991).  Copulatory courtship and cryptic female choice in insects.  Biol.  Rev., 66: 1-31. 

    Eberhard, W.  G.  (1992b).  Species isolation, genital mechanics, and the evolution of species-specific genitalia in three species of Macrodactylus beetles (Coleoptera, Scaraceidae, Melolonthinae).  Evolution, 46(6): 1774-1783. 

    Eigen, M.  (1992).  Steps towards life: A perspective on evolution. Oxford, UK: Oxford U.  Press. 

    Eldredge, N.  (1985).  Unfinished synthesis: Biological hierarchies and modern evolutionary thought. New York: Oxford U.  Press 

    Eldredge, N.  (1986).  Information, economics, and evolution.  Ann.  Review of Ecology and Systematics, 17, 351-369. 

    Eldredge, N.  (1989).  Macroevolutionary dynamics: Species, niches, and adaptive peaks. New York: McGraw-Hill. 

    Eldredge, N., & Gould, S.  J.  (1972).  Punctuated equilibria: An alternative to phyletic gradualism.  In T.  J.  M.  Schopf (Ed.), Models in paleobiology, pp.  82-115.  San Francisco: Freeman, Cooper. 

    Fisher, R.  A.  (1915).  The evolution of sexual preference.  Eugenics review, 7, 184-192. 

    Fisher, R.  A.  (1930).  The genetical theory of natural selection.  Oxford: Clarendon Press. 

    Forrest, D.  (Ed.) (1993).  Proceedings of the Fifth International Conference on Genetic Algorithms. San Francisco: Morgan Kaufmann. 

    Futuyma, D.  (1986).  Evolutionary biology. Sunderland, MA: Sinauer. 

    Futuyama, D., & Slatkin, M.  (Eds.).  (1983).  Coevolution. Sunderland, Massachusetts: Sinauer. 

    Goldschmidt, R.  B.  (1940).  The material basis of evolution. New Haven: Yale U.  Press. 

    Goodwin, B.  C., Holder, N., & Wylie, C.  C.  (Eds.).  (1983).  Development and evolution. Cambridge: Cambridge U.  Press. 

    Gould, S.  J.  (1977).  Ontogeny and phylogeny. Cambridge, MA: Harvard U.  Press. 

    Guilford, T., & Dawkins, M.  S.  (1991).  Receiver psychology and the evolution of animal signals.  Animal Behavior, 42, 1-14. 

    Haldane, J.  B.  S.  (1932).  The causes of evolution. London: Longman. 

    Harvey, I., Husbands, P., & Cliff, D.  (1992).  Issues in evolutionary robotics.  In J.-A.  Meyer, H.  L.  Roitblat, & S.  W.  Wilson (Eds.), From Animals to Animats 2: Proceedings of the Second International Conference on Simulation of Adaptive Behavior, pp.  364-373.  Cambridge: MIT Press. 

    Harvey, I., Husbands, P., Cliff, D.  (1993).  Genetic convergence in a species of evolved robot control architectures.  To appear in S.  Forrest (Ed.), Proceedings of Fifth International Conference on Genetic Algorithms. San Francisco: Morgan Kaufmann. 

    Hinton, G.  E., and Nowlan, S.  J.  (1987).  How learning guides evolution.  Complex Systems, 1, 495-502. 

    Huxley, J.  S.  (1938).  The present standing of the theory of sexual selection.  In G.  R.  de Beer (Ed.), Evolution: Essays on aspects of evolutionary biology, pp.  11-42.  Oxford: Clarendon Press. 

    Huxley, J.  S.  (1942).  Evolution: The modern synthesis. New York: Harper. 

    Iwasa, Y., Pomiankowski, A., & Nee, S.  (1991).  The evolution of costly mate preferences.  II.  The `handicap' principle.  Evolution, 45(6): 1431-1442. 

    Jablonski, D., & Bottjer, D.  J.  (1990).  The ecology of evolutionary innovation: The fossil record.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  253-288.  Chicago: U.  Chicago Press. 

    Jensen, J.  S.  (1990).  Plausibility and testability: Assessing the consequences of evolutionary innovation.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  171-190.  Chicago: U.  Chicago Press. 

    Kauffman, S.  A.  (1993).  Origins of order: Self-organization and selection in evolution. New York: Oxford U.  Press. 

    Kimura, M.  (1983).  The neutral theory of molecular evolution.  In M.  Nei & R.  K.  Koehn (Eds.), Evolution of genes and proteins, pp.  213-233.  Massachusetts: Sinauer. 

    Kirkpatrick, M.  (1982).  Sexual selection and the evolution of female choice.  Evolution, 36, 1-12. 

    Kirkpatrick, M.  (1987).  The evolutionary forces acting on female preferences in polygynous animals.  In J.  W.  Bradbury & M.  B.  Andersson (Eds.), Sexual selection: Testing the alternatives (pp.  67-82).  New York: Wiley. 

    Koza, J.  (1993).  Genetic programming. Cambridge, MA: MIT Press. 

    Lande, R.  (1980).  Sexual dimorphism, sexual selection and adaptation in polygenic characters.  Evolution, 34, 292-305. 

    Lande, R.  (1981).  Models of speciation by sexual selection on polygenic characters.  Proc.  Nat.  Acad.  Sci.  USA, 78, 3721-3725. 

    Lande, R.  (1987).  Genetic correlation between the sexes in the evolution of sexual dimorphism and mating preferences.  In Bradbury, J.  W., & Andersson, M.  B.  (Eds.), Sexual selection: Testing the alternatives, pp.  83-95.  New York: John Wiley. 

    Langton, C.  L., Farmer, J.  D., Rasmussen, S., & Taylor, C.  (Eds.).  (1992).  Artificial Life II. Addison-Wesley. 

    Liem, K.  F.  (1973).  Evolutionary strategies and morphological innovations: Cichlid pharyngeal jaws.  Systematic zoology, 22: 425-441. 

    Liem, K.  F.  (1990).  Key evolutionary innovations, differential diversity, and symecomorphosis.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  147-170.  Chicago: U.  Chicago Press. 

    Manning, J.  T.  (1993).  Fluctuating asymmetry, sexual selection and canine teeth in primates.  Proc.  Royal Soc.  London B, 251(1331), 83-87. 

    May, R.  M.  (1990).  How many species? Phil.  Trans.  Royal Soc.  London B, Biological Sciences, 330(1257), 293-304. 

    May, R.  M.  (1992).  How many species inhabit the earth? Scientific American, 267(4), 42-48. 

    Maynard Smith, J.  (1978).  The evolution of sex. Cambridge: Cambridge U.  Press. 

    Mayr, E.  (1942).  Systematics and the origin of species. (Reprint edition 1982).  New York: Columbia U.  Press. 

    Mayr, E.  (1954).  Change of genetic environment and evolution.  In J.  Huxley, A.  C.  Hardy, & E.  B.  Ford (Eds.), Evolution as a process, pp.  157-180.  London: George Allen & Unwin. 

    Mayr, E.  (1960).  The emergence of evolutionary novelties.  In S.  Tax (Ed.), Evolution after Darwin, Vol.  I., pp.  349-380.  Chicago: U.  Chicago Press. 

    Mayr, E.  (1963).  Animal species and evolution. Cambridge: Harvard U.  Press. 

    McKinney, F.  K.  (1988).  Multidisciplinary perspectives on evolutionary innovations.  Trends in ecology and evolution, 3, 220-222. 

    Miller, G.  F.  (1993).  Evolution of the human brain through runaway sexual selection.  Unpublished Ph.D.  thesis, Psychology Department, Stanford University. 

    Miller, G.  F.  (Accepted, a).  Artificial life through sexual selection.  For Cliff, D.  (Ed.), Artificial Intelligence and Simulation of Behavior Quarterly, Special issue on Artificial Life in the U.  K. 

    Miller, G.  F.  (Accepted, b).  Psychological selection in primates: The evolution of adaptive unpredictability in competition and courtship.  For A.  Whiten & R.  W.  Byrne (Eds.), Machiavellian Intelligence II. 

    Miller, G.  F.  & Freyd, J.  J.  (1993).  Dynamic mental representations of animate motion: The interplay among evolutionary, cognitive, and behavioral dynamics.  Cognitive Science Research Paper 290, University of Sussex.  Submitted as a target article for Behavioral and Brain Sciences. 

    Miller, G.  F., & Todd, P.  M.  (1993).  Evolutionary wanderlust: Sexual selection with directional mate preferences.  In J.-A.  Meyer, H.  L.  Roitblat, & S.  W.  Wilson (Eds.), From Animals to Animats 2: Proceedings of the Second International Conference on Simulation of Adaptive Behavior, pp.  21-30.  Cambridge, MA: MIT Press. 

    Moller, A.  P., & Hoglund, J.  (1991).  Patterns of fluctuating asymmetry in avian feather ornaments: Implications for models of sexual selection.  Proc.  Royal Soc.  London B, 245(1312), 1-6. 

    Morgan, C.  L.  (1888).  Natural selection and elimination.  Nature, Aug.  16, 370. 

    Muller, G.  B.  (1990).  Developmental mechanisms at the origin of morphological novelty: A side-effect hypothesis.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  99-130.  Chicago: U.  Chicago Press. 

    O'Donald, P.  (1980).  Genetic models of sexual selection.  Cambridge, UK: Cambridge University Press. 

    Nitecki, M.  (Ed.).  (1990). Evolutionary innovations. Chicago: U.  Chicago Press. 

    Paterson, H.  E.  H.  (1985).  The recognition concept of species.  In E.  S.  Vrba (Ed.), Species and speciation, Transvaal Mus.  Monogr.  4, 21-29. 

    Patterson, B.  D.  (1988).  Evolutionary innovations: Patterns and processes.  Evolutionary trends in plants 2: 86-87. 

    Petrie, M.  (1992).  Peacocks with low mating success are more likely to suffer predation.  Animal Behavior, 44, 585-586. 

    Petrie, M., Halliday, T., & Sanders, C.  (1991).  Peahens prefer peacocks with elaborate trains.  Animal Behavior, 41, 323-331. 

    Pimental, D., Smith, G.  J.  C., & Soans, J.  (1967) A population model of sympatric speciation.  American Naturalist,  101(922): 493-504. 

    Pomiankowski, A.  (1987).  The costs of choice in sexual selection.  J.  Theoretical Biology, 128, 195-218. 

    Pomiankowski, A.  (1988).  The evolution of female mate preferences for male genetic quality.  Oxford Surveys in Evolutionary Biology, 5, 136-184. 

    Pomiankowski, A.  (1990).  How to find the top male.  Nature, 347: 616-617. 

    Pomiankowski, A., Iwasa, Y., & Nee, S.  (1991).  The evolution of costly mate preferences.  I.  Fisher and biased mutation.  Evolution, 45(6): 1422-1430. 

    Raff, R.  A., Par, B., Parks, A., & Wray, G.  (1990).  Radical evolutionary change in early development.  In M.  Nitecki (Ed.), Evolutionary innovations, pp.  71-98.  Chicago: U.  Chicago Press. 

    Raff, R.  A., & Raff, E.  C.  (Eds.).  (1987).  Development as an evolutionary process. New York: Alan R Liss. 

    Romanes, G.  J.  (1897).  Darwin, and after Darwin.  II.  Post-Darwinian Questions.  Heredity and Utility, 2nd Ed.  Chicago: Open Court Publishing. 

    Ryan, M.  J.  (1986).  Neuroanatomy influences speciation rates among anurans.  Proc.  Nat.  Acad.  Sci.  USA, 83, 1379-1382. 

    Ryan, M.  J.  (1990).  Sexual selection, sensory systems, and sensory exploitation.  Oxford Surveys of Evol.  Biology, 7, 156-195. 

    Ryan, M.  J., & Keddy-Hector, A.  (1992).  Directional patterns of female mate choice and the role of sensory biases.  American Naturalist, 139, S4-S35. 

    Schuster, P.  (1988).  Stationary mutant distributions and evolutionary optimization.  Bull.  Mathematical Biology, 50(6), 635-660. 

    Simon, P.  (1992).  The action plant: Movement and nervous behaviour in plants. Cambridge, MA: Blackwell. 

    Simpson, G.  (1944).  Tempo and mode in evolution. New York: Columbia U.  Press. 

    Simpson, G.  (1953).  The major features of evolution. New York: Columbia U.  Press. 

    Sprengel, C.  K.  (1793).  Das entdeckte Geheimnis der Natur im Bau und in der Befruchtung der Blumen.  (The secret of nature revealed in the structure and pollination of flowers.) Berlin: F.  Vieweg.  (Reprinted 1972 by J.  Cramer, Lehre.) 

    Szalay, F.  S.  & Costello, R.  K.  (1991).  Evolution of permanent estrus displays in hominids.  J.  Human Evolution, 20, 439-464. 

    Sullivan, B.  K.  (1989).  Passive and active female choice: A comment.  Animal Behaviour, 37(4), 692-694. 

    Thoday, J.  M.  (1972).  Disruptive selection.  Proc.  of the Royal Soc.  of London B, 182: 109-143. 

    Thornhill, R.  (1992).  Fluctuating asymmetry and the mating system of the Japanese scorpionfly, Panorpa japonica.  Animal behavior, 44(5), 867-879. 

    Todd, P.  M., & Miller, G.  F.  (1991).  On the sympatric origin of species: Mercurial mating in the Quicksilver Model.  In R.  K.  Belew & L.  B.  Booker (Eds.), Proceedings of the Fourth International Conference on Genetic Algorithms, pp.  547-554.  San Mateo, CA: Morgan Kaufmann. 

    Todd, P.  M.  & Miller, G.  F.  (1993).  Parental guidance suggested: How parental imprinting evolves through sexual selection as an adaptive learning mechanism.  Adaptive Behavior, 2(1). 

    Todd, P.  M.  & Miller, G.  F.  (in preparation).  The role of mate choice in biocomputation II: Applications of sexual selection in search and optimization 

    Todd, P.  M., and Wilson, S.  W.  (1993).  Environment structure and adaptive behavior from the ground up.  In J.-A.  Meyer, H.  L.  Roitblat, and S.  W.  Wilson (Eds.), From Animals to Animats 2: Proceedings of the Second International Conference on Simulation of Adaptive Behavior (pp.  11-20).  Cambridge, MA: MIT Press/Bradford Books. 

    Weismann, A.  (1917).  The selection theory.  In Evolution in modern thought, by Haeckel, Thomson, Weismann, and Others, pp.  23-86.  New York: Boni and Liveright. 

    Williams, G.  C.  (1975).  Sex and evolution. Princeton: Princeton U.  Press. 

    Wright, S.  (1932).  The roles of mutation, inbreeding, crossbreeding, and selection in evolution.  Proc.  Sixth Int.  Congr.  Genetics, 1, 356-366. 

    Wright, S.  (1982).  Character change, speciation, and the higher taxa.  Evolution, 36, 427-443. 

    Wyles, J.  S., Kunkel, J.  G., & Wilson, A.  C.  (1983).  Birds, behavior, and anatomical evolution.  Proc.  Nat.  Acad.  Sci.  USA, 80, 4394-4397. 

    Van Valen, L.  M.  (1971).  Adaptive zones and the orders of mammals.  Evolution, 16, 125-142. 

    Vrba, E.  S.  (1983).  Macroevolutionary trends: New perspectives on the roles of adaptation and incidental effect.  Science, 221, 387-389. 

    Vrba, E.  S.   (1985).  Environment and evolution:  Alternative causes of the temporal distribution of evolutionary events.  South African Journal of Science, 81: 229-236. 

    Vrba, E.  S., & Gould, S.  J.  (1986).  The hierarchical expansion of sorting and selection: Sorting and selection cannot be equated.  Paleobiology, 12, 217-228. 

    Zahavi, A.  (1975).  Mate selection: A selection for a handicap.  Journal of Theoretical Biology, 53, 205-214.


The role of mate choice in biocomputation

1 Symmetry is a useful general-purpose cue of developmental competence (e.g.  Manning, 1993; Moller & Hoglund, 1991; Thornhill, 1992), because deleterious mutations, injuries, and diseases often disrupt symmetry, and because an animal choosing a mate need not know the detailed optimal form of a particular bilateral structure, it only needs the circuitry for comparing left to right.  And symmetrically-structured sensory surfaces and neural circuits (e.g.  eyes and brains) may make such symmetry judgments easy, because they facilitate the comparison of the corresponding left and right features of perceived objects.  The utility of symmetric body-plans as displays of developmental competence, and of symmetric brains and senses as mechanisms for choosing symmetric mates, could make developing a symmetric phenotype a common attractor state for many evolving lineages

Search more related documents:The role of mate choice in biocomputation:

Set Home | Add to Favorites

All Rights Reserved Powered by Free Document Search and Download

Copyright © 2011
This site does not host pdf,doc,ppt,xls,rtf,txt files all document are the property of their respective owners. complaint#downhi.com