In principle, it would seem a valuable heuristic procedure to assume that an animal is optimizing something under a given set of constraints, and to try to work out what those constraints are. This is a restricted version of what McFarland and his colleagues call the ‘reverse optimality’ approach (e.g. McCleery 1978). As a case study I shall take some work with which I happen to be familiar.
Dawkins and Brockmann (1980) found that the digger wasps (Sphex ichneumoneus) studied by Brockmann behaved in a way that a naive human economist might have criticized as maladaptive. Individual wasps appeared to commit the ‘Concorde Fallacy’ of valuing a resource according to how much they had already spent on it, rather than according to how much they could get out of it in the future. Very briefly, the evidence is as follows. Solitary females provision burrows with stung and paralysed katydids which are to serve as food for their larvae (see Chapter 7). Occasionally two females find themselves provisioning the same burrow, and they usually end up fighting over it. Each fight goes on until one wasp, thereby defined as the loser, flees from the area, leaving the winner in control of the burrow and all the katydids caught by both wasps. We measured the ‘real value’ of a burrow as the number of katydids which it contained. The ‘prior investment’ by each wasp in the burrow was measured as the number of katydids which she, as an individual, had put into it. The evidence suggested that each wasp fought for a time proportional to her own investment, rather than proportional to the ‘true value’ of the burrow.
Such a policy has great human psychological appeal. We too tend to fight tenaciously for property which we have put great effort into acquiring. The fallacy gets its name from the fact that, at a time when sober economic judgement of future prospects counselled abandoning the developing of the Concorde airliner, one of the arguments in favour of continuing with the half completed project was retrospective: ‘We have already spent so much on it that we cannot back out now.’ A popular argument for prolonging wars gave rise to the other name for the fallacy, the ‘Our boys shall not have died in vain’ fallacy.
When Dr Brockmann and I first realized that digger wasps behaved in like manner, I was, it has to be confessed, a little disconcerted, possibly because of my own past investment of effort (Dawkins & Carlisle 1976; Dawkins 1976a) in persuading my colleagues that the psychologically appealing Concorde Fallacy was, indeed, a fallacy! But then we started to think more seriously about cost constraints. Could it be that what appeared to be maladaptive was better interpreted as an optimum, given certain constraints? The question then became: Is there a constraint such that the wasps’ Concordian behaviour is the best they can achieve under it?
In fact the question was more complicated than that, because it was necessary to substitute Maynard Smith’s (1974) concept of evolutionary stability (‘ESS’—see Chapter 7) for that of simple optimality, but the principle remains that a reverse optimality approach might be heuristically valuable. If we can show that an animal’s behaviour is what would be produced by an optimizing system working under constraint X, maybe we can use the approach to learn something of the constraints under which animals actually do work.
In the present case it seemed that the relevant constraint might be one of sensory capacity. If the wasps, for some reason, cannot count katydids in the burrow, but can metre some aspect of their own hunting efforts, there is an asymmetry of information possessed by the two combatants. Each one ‘knows’ that the burrow contains at least b katydids, where b is the number she herself has caught. She may ‘estimate’ that the true number in the burrow is larger than b, but she does not know how much larger. Under such conditions Grafen (in preparation) has shown that the expected ESS is approximately the one originally calculated by Bishop and Cannings (1978) for the so-called ‘generalized war of attrition’. The mathematical details can be left aside; for present purposes what matters is that the behaviour expected by the extended war of attrition model would look very like the Concordian behaviour actually shown by the wasps.
If we were interested in testing the general hypothesis that animals optimize, this kind of post hoc rationalization would be suspect. By post hoc modification of the details of the hypothesis, one is bound to find a version which fits the facts. Maynard Smith’s (1978b) reply to this kind of criticism is very relevant: ‘… in testing a model we are not testing the general proposition that nature optimizes, but the specific hypotheses about constraints, optimization criteria, and heredity’. In the present case we are making a general assumption that nature does optimize within constraints, and testing particular models of what those constraints might be.
The particular constraint suggested—inability of the wasp’s sensory system to assess the contents of a burrow—is in accordance with independent evidence from the same population of wasps (Brockmann, Grafen & Dawkins 1979; Brockmann & Dawkins 1979). There is no reason to regard it as an irrevocably binding limitation for all time. Probably the wasps could evolve the capacity to assess nest contents, but only at a cost. Digger wasps of the related species Ammophila campestris have long been known to make an assessment of the contents of each of their nests every day (Baerends 1941). Unlike Sphex, which provisions one burrow at a time, lays an egg, then fills the burrow in with soil and leaves the larva to eat the provision on its own, Ammophila campestris is a progressive provisioner of several burrows concurrently. A female tends two or three growing larvae, each in a separate burrow, at the same time. The ages of her various larvae are staggered, and their food needs are different. Every morning she assesses the current contents of each burrow on a special early morning ‘inspection round’. By experimentally changing the contents of burrows, Baerends showed that the female adjusts her whole day’s provisioning of each burrow according to what it contained at the time of her morning inspection. The contents of the burrow at any other time of day have no effect on her behaviour, even though she is provisioning it all day. She appears, therefore, to use her assessment faculty sparingly, switching it off for the rest of the day after the morning inspection, almost as though it was a costly, power-consuming instrument. Fanciful as that analogy may be, it surely suggests that the assessment faculty, whatever it is, may have overhead running costs, even if (G. P. Baerends, personal communication) these consist only in the time consumed.
Sphex ichneumoneus, not being a progressive provisioner, and tending only one burrow at a time, presumably has less need than Ammophila for a burrow-assessment faculty. By not attempting to count prey in the burrow, it can save itself not only the running expenditure that Ammophila seems so careful to ration; it can also save itself the initial manufacturing costs of the necessary neural and sensory apparatus. Probably it could benefit slightly from having an ability to assess burrow contents, but only on the comparatively rare occasions when it finds itself competing for a burrow with another wasp. It is easy to believe that the costs outweigh the benefits, and that selection has therefore never favoured the evolution of assessment apparatus. I think this is a more constructive and interesting hypothesis than the alternative hypothesis that the necessary mutational variation has never arisen. Of course we have to admit that the latter might be right, but I would prefer to keep it as a hypothesis of last resort.
Imperfections at one level due to selection at another level
One of the main topics to be tackled in this book is that of the level at which natural selection acts. The kind of adaptations we should see if selection acted at the level of the group would be quite different from the adaptations we should expect if selection acts at the level of the individual. It follows that a group selectionist might well see as imperfections, features which an individual selectionist would see as adaptations. This is the main reason why I regard as unfair Gould and Lewontin’s (1979) equating of modern adaptationism with the naive perfectionism that Haldane named after Voltaire’s Dr Pangloss. With reservations due to the various constraints on perfection, an adaptationist may believe that all aspects of organisms are ‘adaptive optimal so
lutions to problems’, or that ‘it is virtually impossible to do a better job than an organism is doing in its given environment’. Yet the same adaptationist may be extremely fussy about the kind of meaning he allows to words like ‘optimal’ and ‘better’. There are many kinds of adaptive, indeed Panglossian, explanations, for example most group-selectionist ones, which would be utterly ruled out by the modern adaptationist.
For the Panglossian the demonstration that something is ‘beneficial’ (to whom or to what is often not specified) is a sufficient explanation for its existence. The neo-Darwinian adaptationist, on the other hand, insists upon knowing the exact nature of the selective process that has led to the evolution of the putative adaptation. In particular, he insists on precise language about the level at which natural selection is supposed to have acted. The Panglossian looks at a one-to-one sex ratio and sees that it is good: does it not minimize the wastage of the population’s resources? The neo-Darwinian adaptationist considers in detail the fates of genes acting on parents to bias the sex ratio of their offspring, and calculates the evolutionarily stable state of the population (Fisher 1930a). The Panglossian is disconcerted by 1:1 sex ratios in polygynous species, in which a minority of males hold harems and the rest sit about in bachelor herds consuming almost half the population’s food resources yet contributing not at all to the population’s reproduction. The neo-Darwinian adaptationist takes this in his stride. The system may be hideously uneconomical from the population’s point of view, but, from the point of view of the genes influencing the trait concerned, there is no mutant that could do better. My point is that neo-Darwinian adaptationism is not a catch-all, blanket faith in all being for the best. It rules out of court most of the adaptive explanations that readily occur to the Panglossian.
Some years ago, a colleague received an application from a prospective graduate student wishing to work on adaptation, who was brought up a religious fundamentalist and did not believe in evolution. He believed in adaptations, but thought they were designed by God, designed for the benefit of … ah, but that is just the problem! It might be thought that it did not matter whether the student believed adaptations were produced by natural selection or by God. Adaptations are ‘beneficial’ whether because of natural selection or because of beneficient design, and could not a fundamentalist student be usefully employed in uncovering the detailed ways in which they were beneficial? My point is that this argument will not do, because what is beneficial to one entity in the hierarchy of life is harmful to another, and creationism gives us no grounds for supposing that one entity’s welfare will be preferred to another’s. In passing, the fundamentalist student might pause to wonder at a God who goes to great trouble to provide predators with beautiful adaptations to catch prey, while with the other hand giving prey beautiful adaptations to thwart them. Perhaps He enjoys the spectator sport. Returning to the main point, if adaptations were designed by God, He might have designed them to benefit the individual animal (its survival or—not the same thing—its inclusive fitness), the species, some other species such as mankind (the usual view of religious fundamentalists), the ‘balance of nature’, or some other inscrutable purpose known only to Him. These are frequently incompatible alternatives. It really matters for whose benefit adaptations are designed. Facts such as the sex ratio in harem-forming mammals are inexplicable on certain hypotheses and easily explicable on others. The adaptationist working within the framework of a proper understanding of the genetical theory of natural selection countenances only a very restricted set of the possible functional hypotheses which the Panglossian might admit.
One of the main messages of this book is that, for many purposes, it is better to regard the level at which selection acts as neither the organism, nor the group or any larger unit, but the gene or small genetic fragment. This difficult topic will be debated in later chapters. For the present, it is sufficient to note that selection at the level of the gene can give rise to apparent imperfections at the level of the individual. I shall discuss ‘meiotic drive’ and related phenomena in Chapter 8, but the classic example is the case of heterozygous advantage. A gene may be positively selected because of its beneficial effects when heterozygous, even though it has harmful effects when homozygous. As a consequence of this, a predictable proportion of the individual organisms in the population will have defects. The general point is this. The genome of an individual organism in a sexual population is the product of a more or less random shuffling of the genes in the population. Genes are selected over their alleles because of their phenotypic effects, averaged over all the individual bodies in which they are distributed, over the whole population, and through many generations. The effects that a given gene has will usually depend upon the other genes with which it shares a body: heterozygous advantage is just a special case of this. A certain proportion of bad bodies seems an almost inevitable consequence of selection for good genes, where good refers to the average effects of a gene on a statistical sample of bodies in which it finds itself permuted with other genes.
Inevitable, that is, as long as we accept the Mendelian shuffle as given and inescapable. Williams (1979), disappointed at finding no evidence for adaptive fine-adjustment of the sex ratio, makes the perceptive point that
Sex is only one of many offspring characters that would seem adaptive for a parent to control. For instance, in human populations affected by sickle-cell anaemia, it would be advantageous for a heterozygous woman to have her A eggs fertilized only by a-bearing sperm, and vice versa, or even to abort all homozygous embryos. Yet if mated to another heterozygote she will reliably submit to the Mendelian lottery, even though this means markedly lowered fitness for half her children … The really fundamental questions in evolution may be answerable only by regarding each gene as ultimately in conflict with every other gene, even those at other loci in the same cell. A really valid theory of natural selection must be based ultimately on selfish replicators, genes and all other entities capable of the biased accumulation of different variant forms.
Amen!
Mistakes due to environmental unpredictability or ‘malevolence’
However well adapted an animal may be to environmental conditions, those conditions must be regarded as a statistical average. It will usually be impossible to cater for every conceivable contingency of detail, and any given animal will therefore frequently be observed to make ‘mistakes’, mistakes which can easily be fatal. This is not the same point as the time-lag problem already mentioned. The time-lag problem arises because of non-stationarities in the statistical properties of the environment: average conditions now are different from the average conditions experienced by the animal’s ancestors. The present point is more inescapable. The modern animal may be living in identical average conditions to those of an ancestor, yet the detailed moment to moment occurrences facing either of them are not the same from day to day, and are too complex for precise prediction to be possible.
It is particularly in behaviour that such mistakes are seen. The more static attributes of an animal, its anatomical structure for instance, are obviously adapted only to long-term average conditions. An individual is either big or small, it cannot change size from minute to minute as the need arises. Behaviour, rapid muscular movement, is that part of an animal’s adaptive repertoire which is specifically concerned with high speed adjustment. The animal can be now here, now there, now up a tree, now underground, rapidly accommodating to environmental contingencies. The number of such possible contingencies, when defined in all their detail, is like the number of possible chess positions, virtually infinite. Just as chess-playing computers (and chess-playing people) learn to classify chess positions into a manageable number of generalized classes, so the best that an adaptationist can hope for is that an animal will have been programmed to behave in ways appropriate to a manageable number of general contingency classes. Actual contingencies will fit these general classes only approximately, and apparent mistakes are therefore bound to be made.
/>
The animal that we see up a tree may come from a long line of tree-dwelling ancestors. The trees in which the ancestors underwent natural selection were, in general, much the same as the trees of today. General rules of behaviour which worked then, such as ‘Never go out on a limb that is too thin’, still work. But the details of any one tree are inevitably different from the details of another. The leaves are in slightly different places, the breaking strain of the branches is only approximately predictable from their diameter, and so on. However strongly adaptationist our beliefs may be, we can only expect animals to be average statistical optimizers, never perfect anticipators of every detail.
So far we have considered the environment as statistically complex and therefore hard to predict. We have not reckoned on its being actively malevolent from our animal’s point of view. Tree boughs surely do not deliberately snap out of spite when monkeys venture on to them. But a ‘tree bough’ may turn out to be a camouflaged python, and our monkey’s last mistake is then no accident but is, in a sense, deliberately engineered. Part of a monkey’s environment is non-living or at least indifferent to the monkey’s existence, and the monkey’s mistakes can be put down to statistical unpredictability. But other parts of the monkey’s environment consist of living things that are themselves adapted to profit at the expense of monkeys. This portion of the monkey’s environment may be called malevolent.