Science in the Soul

by Richard Dawkins

  Although he was not talking about Lamarckism, Konrad Lorenz emphasized a related point for the case of learned behaviour, which is perhaps the most important kind of acquired adaptation. An animal learns to be a better animal during its own lifetime. It learns to eat sweet foods, say, thereby increasing its survival chances.*4 But there is nothing inherently nutritious about a sweet taste. Something, presumably natural selection, has to have built into the nervous system the arbitrary rule: ‘treat sweet taste as reward’, and this works because saccharine does not occur in nature whereas sugar does.

  The principle holds, too, for morphological characters. Feet that are subjected to wear and tear grow tougher and more thick-skinned. The thickening of the skin is an acquired adaptation, but it is not obvious why the change went in this direction. In man-made machines, parts that are subjected to wear get thinner not thicker, for obvious reasons. Why does the skin on the feet do the opposite? Because natural selection has worked in the past to ensure an adaptive rather than a maladaptive response to wear and tear.

  The relevance of this for would-be Lamarckian evolution is that there has to be a deep Darwinian underpinning even if there is a Lamarckian surface structure: a Darwinian choice of which potentially acquirable characters shall in fact be acquired and inherited. Lamarckian mechanisms cannot be fundamentally responsible for adaptive evolution. Even if acquired characters are inherited on some planet, evolution there will still rely on a Darwinian guide for its adaptive direction.

  Theory 3. Direct induction by the environment

  Adaptation, as we have seen, is a fit between organism and environment. The set of conceivable organisms is wider than the actual set. And there is a set of conceivable environments wider than the actual set. These two subsets match each other to some extent, and the matching is adaptation. We can re-express the point by saying that information from the environment is present in the organism. In a few cases this is vividly literal – a frog carries a picture of its environment around on its back. Such information is usually carried by an animal in the less literal sense that a trained observer, dissecting a new animal, can reconstruct many details of its natural environment.*5

  Now, how could the information get from the environment into the animal? Lorenz argues that there are two ways, natural selection and reinforcement learning, but that these are both selective processes in the broad sense.*6 There is, in theory, an alternative method for the environment to imprint its information on the organism, and that is by direct ‘instruction’. Some theories of how the immune system works are ‘instructive’: antibody molecules are thought to be shaped directly by moulding themselves around antigen molecules. The currently favoured theory is, by contrast, selective. I take ‘instruction’ to be synonymous with the ‘direct induction by the environment’ of Mayr’s Theory 3. It is not always clearly distinct from Theory 2.

  Instruction is the process whereby information flows directly from its environment into an animal. A case could be made for treating imitation learning, latent learning and imprinting as instructive, but for clarity it is safer to use a hypothetical example. Think of an animal on some planet, deriving camouflage from its tiger-like stripes. It lives in long dry ‘grass’, and its stripes closely match the typical thickness and spacing of local grass blades. On our own planet such adaptation would come about through the selection of random genetic variation, but on the imaginary planet it comes about through direct instruction. The animals go brown except where their skin is shaded from the ‘sun’ by blades of grass. Their stripes are therefore adapted with great precision, not just to any old habitat, but to the precise habitat in which they have sunbathed, and it is this same habitat in which they are going to have to survive. Local populations are automatically camouflaged against local grasses. Information about the habitat, in this case about the spacing patterns of the grass blades, has flowed into the animals, and is embodied in the spacing pattern of their skin pigment.

  Instructive adaptation demands the inheritance of acquired characters if it is to give rise to permanent or progressive evolutionary change. ‘Instruction’ received in one generation must be ‘remembered’ in the genetic (or equivalent) information. This process is in principle cumulative and progressive. However, if the genetic store is not to become overloaded by the accumulations of generations, some mechanism must exist for discarding unwanted ‘instructions’, and retaining desirable ones. I suspect that this must lead us, once again, to the need for some kind of selective process.

  Imagine, for instance, a form of mammal-like life in which a stout ‘umbilical nerve’ enabled a mother to ‘dump’ the entire contents of her memory in the brain of her fetus. The technology is available even to our nervous systems: the corpus callosum can shunt large quantities of information from right hemisphere to left. An umbilical nerve could make the experience and wisdom of each generation automatically available to the next, and this might seem very desirable. But without a selective filter, it would take few generations for the load of information to become unmanageably large. Once again we come up against the need for a selective underpinning. I will leave this now, and make one more point about instructive adaptation (which applies equally to all Lamarckian types of theory).

  The point is that there is a logical link-up between the two major theories of adaptive evolution – selection and instruction – and the two major theories of embryonic development – epigenesis and preformationism.*7 Instructive evolution can work only if embryology is preformationistic. If embryology is epigenetic, as it is on our planet, instructive evolution cannot work.

  Briefly, if acquired characters are to be inherited, embryonic processes must be reversible: phenotypic change has to be read back into the genes (or equivalent). If embryology is preformationistic – the genes are a true blueprint – then it may indeed be reversible. You can translate a house back into its blueprint. But if embryonic development is epigenetic; if, as on this planet, the genetic information is more like a recipe for a cake than a blueprint for a house, it is irreversible. There is no one-to-one mapping between bits of genome and bits of phenotype, any more than there is mapping between crumbs of cake and words of recipe. The recipe is not a blueprint that can be reconstructed from the cake. The transformation of recipe into cake cannot be put into reverse, and nor can the process of making a body. Therefore acquired adaptations cannot be read back into the ‘genes’ on any planet where embryology is epigenetic.

  This is not to say that there could not, on some planet, be a form of life whose embryology was preformationistic. That is a separate question. How likely is it? The form of life would have to be very different from ours, so much so that it is hard to visualize how it might work. As for reversible embryology itself, it is even harder to visualize. Some mechanism would have to scan the detailed form of the adult body, carefully noting down, for instance, the exact location of brown pigment in a sun-striped skin, perhaps turning it into a linear stream of code numbers, as in a television camera. Embryonic development would read the scan out again, like a television receiver. I have an intuitive hunch that there is an objection in principle to this kind of embryology, but I cannot at present formulate it clearly.*8 All I am saying here is that, if planets are divided into those where embryology is preformationistic and those, like Earth, where embryology is epigenetic, Darwinian evolution could be supported on both kinds of planet, but Lamarckian evolution, even if there were not other reasons for doubting its existence, could be supported only on the preformationistic planets – if there are any.

  Theory 4. Saltationism

  The great virtue of the idea of evolution is that it explains, in terms of blind physical forces, the existence of undisputed adaptations whose functionally directed statistical improbability is enormous, without recourse to the supernatural or mystical. Since we define an undisputed adaptation as an adaptation that is too complex to have come about by chance, how is it possible for a theory to invoke only blind physical forces in explanation? The
answer – Darwin’s answer – is astonishingly simple when we consider how self-evident Paley’s Divine Watchmaker must have seemed to his contemporaries. The key is that the coadapted parts do not have to be assembled all at once. They can be put together in small stages. But they really do have to be small stages. Otherwise we are back again with the problem we started with: the creation by chance of complexity that is too great to have been created by chance!

  Take the eye again, as an example of an organ that contains a large number of independent coadapted parts, say N. The a priori probability of any one of these N features coming into existence by chance is low, but not incredibly low. It is comparable to the chance of a crystal pebble being washed by the sea so that it acts as a rounded lens. Any one adaptation on its own could, plausibly, have come into existence through blind physical forces. If each of the N coadapted features confers some slight advantage on its own, then the whole many-parted organ can be put together over a long period of time. This is particularly plausible for the eye – ironically in view of that organ’s niche of honour in the creationist pantheon. The eye is, par excellence, a case where a fraction of an organ is better than no organ at all; an eye without a lens or even a pupil, for instance, could still detect the looming shadow of a predator.

  To repeat, the key to the Darwinian explanation of adaptive complexity is the replacement of instantaneous, coincidental, multi-dimensional luck, by gradual, inch by inch, smeared-out luck. Luck is involved, to be sure. But a theory that bunches the luck up into major steps is more incredible than a theory that spreads the luck out in small stages. This leads to the following general principle of universal biology. Wherever in the universe adaptive complexity shall be found, it will have come into being gradually through a series of small alterations, never through large and sudden increments in adaptive complexity.*9 We must reject Mayr’s fourth theory, saltationism, as a candidate for explanation of the evolution of complexity.

  It is almost impossible to dispute this rejection. It is implicit in the definition of adaptive complexity that the only alternative to gradualistic evolution is supernatural magic. This is not to say that the argument in favour of gradualism is a worthless tautology, an unfalsifiable dogma of the sort that creationists and philosophers are so fond of jumping about on. It is not logically impossible for a full-fashioned eye to spring de novo from virgin bare skin. It is just that the possibility is statistically negligible.

  Now, it has recently been widely and repeatedly publicized that some modern evolutionists reject ‘gradualism’, and espouse what John Turner has rather irreverently called theories of evolution by jerks. Since these are reasonable people without mystical leanings, they must be gradualists in the sense in which I am here using the term: the ‘gradualism’ that they oppose must be defined differently. There are actually two confusions of language here, and I intend to clear them up in turn. The first is the common confusion between ‘punctuated equilibrium’ and true saltationism.*10 The second is a confusion between two theoretically distinct kinds of saltation.

  Punctuated equilibrium is not macro-mutation, not saltation at all in the traditional sense of the term. It is, however, necessary to discuss it here, because it is popularly regarded as a theory of saltation, and its partisans quote, with approval, Huxley’s criticism of Darwin for upholding the principle of Natura non facit saltum.*11 The punctuationist theory is portrayed as radical and revolutionary and at variance with the ‘gradualistic’ assumptions of both Darwin and the neo-Darwinian synthesis. Punctuated equilibrium, however, was originally conceived as what the orthodox neo-Darwinian synthetic theory should truly predict, on a paleontological timescale, if we take its embedded ideas of allopatric speciation seriously. It derives its ‘jerks’ by taking the ‘stately unfolding’ of the neo-Darwinian synthesis, and inserting long periods of stasis separating brief bursts of gradual, albeit rapid, evolution.

  The plausibility of such ‘rapid gradualism’ is dramatized by a thought experiment of Ledyard Stebbins.*12 He imagines a species of mouse, evolving larger body size at such an imperceptibly slow rate that the differences between the means of successive generations would be utterly swamped by sampling error. Yet even at this slow rate Stebbins’ mouse lineage would attain the body size of a large elephant in about sixty thousand years, a timespan so short that it would be regarded as instantaneous by paleontologists. Evolutionary change too slow to be detected by micro-evolutionists can nevertheless be too fast to be detected by macro-evolutionists.*13 What a paleontologist sees as a ‘saltation’ can in fact be a smooth and gradual change so slow as to be undetectable to the micro-evolutionist. This kind of paleontological ‘saltation’ has nothing to do with the one-generation macro-mutations that, I suspect, Huxley and Darwin had in mind when they debated Natura non facit saltum. Confusion has arisen here, possibly because some individual champions of punctuated equilibrium have also, incidentally, championed macro-mutation. Other ‘punctuationists’ have either confused their theory with macro-mutationism, or have explicitly invoked macro-mutation as one of the mechanisms of punctuation.

  Turning to macro-mutation, or true saltation, the second confusion I want to clear up is between two kinds of macro-mutation that we might conceive of. I could name them, unmemorably, saltation (1) and saltation (2), but instead I shall pursue an earlier fancy for airliners as metaphors, and label them ‘Boeing 747’ and ‘Stretched DC-8’ saltation. 747 saltation is the inconceivable kind. It gets its name from Sir Fred Hoyle’s much-quoted metaphor for his own cosmic misunderstanding of Darwinism. Hoyle compared Darwinian selection to a tornado, blowing through a junkyard and assembling a Boeing 747 (what he overlooked, of course, was the point about luck being ‘smeared out’ in small steps – see above). Stretched DC-8 saltation is quite different. It is not in principle hard to believe in at all. It refers to large and sudden changes in magnitude of some biological measure, without an accompanying large increase in adaptive information. It is named after an airliner that was made by elongating the fuselage of an existing design, not adding significant new complexity.*14 The change from DC-8 to Stretched DC-8 is a big change in magnitude – a saltation, not a gradualistic series of tiny changes. But, unlike the change from junk-heap to 747, it is not a big increase in information content or complexity, and that is the point I am emphasizing by the analogy.

  An example of DC-8 saltation would be the following. Suppose the giraffe’s neck shot out in one spectacular mutational step. Two parents had necks of normal antelope length. They had a freak child with a neck of modern giraffe length, and all giraffes are descended from this freak. This is unlikely to be true on Earth,*15 but something like it may happen elsewhere in the universe. There is no objection to it in principle, in the sense that there is a profound objection to the (747) idea that a complex organ like an eye could arise from bare skin by a single mutation. The crucial difference is one of complexity.

  I am assuming that the change from short antelope’s neck to long giraffe’s neck is not an increase in complexity. To be sure, both necks are exceedingly complex structures. You couldn’t go from no neck to either kind of neck in one step; that would be 747 saltation. But once the complex organization of the antelope’s neck already exists, the step to giraffe’s neck is just an elongation: various things have to grow faster at some stage in embryonic development; existing complexity is preserved. In practice, of course, such a drastic change in magnitude would be highly likely to have deleterious repercussions which would render the macro-mutant unlikely to survive. The existing antelope heart probably could not pump the blood up to the newly elevated giraffe head. Such practical objections to evolution by DC-8 saltation can only help my case in favour of gradualism, but I still want to make a separate, and more universal, case against 747 saltation.

  It may be argued that the distinction between 747 and DC-8 saltation is impossible to draw in practice. After all, DC-8 saltations, such as the proposed macro-mutational elongation of the giraffe’s neck, m
ay appear very complex: muscle blocks, vertebrae, nerves, blood vessels, all have to elongate together. Why does this not make it a 747 saltation, and therefore rule it out?

  We know that single mutations can orchestrate changes in growth rates of many diverse parts of organs, and, when we think about developmental processes, it is not in the least surprising that this should be so. When a single mutation causes a Drosophila to grow a leg where an antenna ought to be, the leg grows in all its formidable complexity. But this is not mysterious or surprising, not a 747 saltation, because the organization of a leg is already present in the body before the mutation. Wherever, as in embryogenesis, we have a hierarchically branching tree of causal relationships, a small alteration at a senior node of the tree can have large and complex ramified effects on the tips of the twigs. But although the change may be large in magnitude, there can be no large and sudden increments in adaptive information. If you think you have found a particular example of a large and sudden increment in adaptively complex information in practice, you can be certain the adaptive information was already there, even if it is an atavistic ‘throwback’ to an earlier ancestor.

  There is not, then, any objection in principle to theories of evolution by jerks, even the theory of hopeful monsters, provided that it is DC-8 saltation, not 747 saltation that is meant. No educated biologist actually believes in 747 saltation, but not all have been sufficiently explicit about the distinction between DC-8 and 747 saltation. An unfortunate consequence is that creationists and their journalistic fellow-travellers have been able to exploit saltationist-sounding statements of respected biologists. The biologist’s intended meaning may have been what I am calling DC-8 saltation, or even non-saltatory punctuation; but the creationist assumes saltation in the sense that I have dubbed 747, and 747 saltation would, indeed, be a blessed miracle.