The Extended Phenotype

by Richard Dawkins

  Symons (1979) makes the computer myth explicit:

  I wish to point out that Dawkins’s implication—through the use of words like ‘robot’ and ‘blindly’—that evolutionary theory favors determinism is utterly without foundation … A robot is a mindless automaton. Perhaps some animals are robots (we have no way of knowing); however, Dawkins is not referring to some animals, but to all animals and in this case specifically to human beings. Now, to paraphrase Stebbing, ‘robot’ can be opposed to ‘thinking being’ or it can be used figuratively to indicate a person who seems to act mechanically, but there is no common usage of language that provides a meaning for the word ‘robot’ in which it would make sense to say that all living things are robots [p. 41].

  The point of the passage from Stebbing which Symons paraphrased is the reasonable one that X is a useless word unless there are some things that are not X. If everything is a robot, then the word robot doesn’t mean anything useful. But the word robot has other associations, and rigid inflexibility was not the association I was thinking of. A robot is a programmed machine, and an important thing about programming is that it is distinct from, and done in advance of, performance of the behaviour itself. A computer is programmed to perform the behaviour of calculating square roots, or playing chess. The relationship between a chess-playing computer and the person who programmed it is not obvious, and is open to misunderstanding. It might be thought that the programmer watches the progress of the game and gives instructions to the computer move by move. In fact, however, the programming is finished before the game begins. The programmer tries to anticipate contingencies, and builds in conditional instructions of great complexity, but once the game begins he has to keep his hands off. He is not allowed to give the computer any new hints during the course of the game. If he did he would not be programming but performing, and his entry would be disqualified from the tournament. In the work criticized by Symons, I made extensive use of the analogy of computer chess in order to explain the point that genes do not control behaviour directly in the sense of interfering in its performance. They only control behaviour in the sense of programming the machine in advance of performance. It was this association with the word robot that I wanted to invoke, not the association with mindless inflexibility.

  As for the mindless inflexibility association itself, it could have been justified in the days when the acme of automation was the rod and cam control system of a marine engine, and Kipling wrote ‘McAndrew’s Hymn’:

  From coupler-flange to spindle-guide I see Thy Hand, O God—Predestination in the stride o’ yon connectin’-rod.

  John Calvin might ha’ forged the same—

  But that was 1893 and the heyday of steam. We are now well embarked on the golden age of electronics. If machines ever had associations with rigid inflexibility—and I accept that they had—it is high time they lived them down. Computer programs have now been written that play chess to International Master standard (Levy 1978), that converse and reason in correct and indefinitely complex grammatical English (Winograd 1972), that create elegant and aesthetically satisfying new proofs of mathematical theorems (Hofstadter 1979), that compose music and diagnose illness; and the pace of progress in the field shows no sign of slowing down (Evans 1979). The advanced programming field known as artificial intelligence is in a buoyant, confident state (Boden 1977). Few who have studied it would now bet against computer programs beating the strongest Grand Masters at chess within the next 10 years. From being synonymous in the popular mind with a moronically undeviating, jerky-limbed zombie, ‘robot’ will one day become a byword for flexibility and rapid intelligence.

  Unfortunately I jumped the gun a little in the passage quoted. When I wrote it I had just returned from an eye-opening and mind-boggling conference on the state of the art of artificial intelligence programming, and I genuinely and innocently in my enthusiasm forgot that robots are popularly supposed to be inflexible idiots. I also have to apologize for the fact that, without my knowledge, the cover of the German edition of The Selfish Gene was given a picture of a human puppet jerking on the end of strings descending from the word gene, and the French edition a picture of little bowler-hatted men with clockwork wind-up keys sticking out of their backs. I have had slides of both covers made up as illustrations of what I was not trying to say.

  So, the answer to Symons is that of course he was right to criticize what he thought I was saying, but of course I wasn’t actually saying it (Ridley 1980a). No doubt I was partly to blame for the original misunderstanding, but I can only urge now that we put aside the preconceptions derived from common usage (‘… most men don’t understand computers to even the slightest degree’—Weizenbaum 1976, p. 9), and actually go and read some of the fascinating modern literature on robotics and computer intelligence (e.g. Boden 1977; Evans 1979; Hofstadter 1979).

  Once again, of course, philosophers may debate the ultimate determinacy of computers programmed to behave in artificially intelligent ways, but if we are going to get into that level of philosophy many would apply the same arguments to human intelligence (Turing 1950). What is a brain, they would ask, but a computer, and what is education but a form of programming? It is very hard to give a non-supernatural account of the human brain and human emotions, feelings and apparent free will, without regarding the brain as, in some sense, the equivalent of a programmed, cybernetic machine. The astronomer Sir Fred Hoyle (1964) expresses very vividly what, it seems to me, any evolutionist must think about nervous systems:

  Looking back [at evolution] I am overwhelmingly impressed by the way in which chemistry has gradually given way to electronics. It is not unreasonable to describe the first living creatures as entirely chemical in character. Although electrochemical processes are important in plants, organized electronics, in the sense of data processing, does not enter or operate in the plant world. But primitive electronics begins to assume importance as soon as we have a creature that moves around … The first electronic systems possessed by primitive animals were essentially guidance systems, analogous logically to sonar or radar. As we pass to more developed animals we find electronic systems being used not merely for guidance but for directing the animal toward food …

  The situation is analogous to a guided missile, the job of which is to intercept and destroy another missile. Just as in our modern world attack and defense become more and more subtle in their methods, so it was the case with animals. And with increasing subtlety, better and better systems of electronics become necessary. What happened in nature has a close parallel with the development of electronics in modern military applications … I find it a sobering thought that but for the tooth-and-claw existence of the jungle we should not possess our intellectual capabilities, we should not be able to inquire into the structure of the Universe, or be able to appreciate a symphony of Beethoven … Viewed in this light, the question that is sometimes asked—can computers think?—is somewhat ironic. Here of course I mean the computers that we ourselves make out of inorganic materials. What on earth do those who ask such a question think they themselves are? Simply computers, but vastly more complicated ones than anything we have yet learned to make. Remember that our man-made computer industry is a mere two or three decades old, whereas we ourselves are the products of an evolution that has operated over hundreds of millions of years [pp. 24–26].

  Others may disagree with this conclusion, although I suspect that the only alternatives to it are religious ones. Whatever the outcome of that debate, to return to genes and the main point of this chapter, the issue of determinism versus free will is just not affected one way or the other by whether or not you happen to be considering genes as causal agents rather than environmental determinants.

  But, it will pardonably be said, there is no smoke without fire. Functional ethologists and ‘sociobiologists’ must have said something to deserve being tarred with the brush of genetic determinism. Or if it is all a misunderstanding there must be some good explanation, because misunderstan
dings that are so widespread do not come about for no reason, even if abetted by cultural myths as powerful as the gene myth and the computer myth in unholy alliance. Speaking for myself, I think I know the reason. It is an interesting one, and it will occupy the rest of this chapter. The misunderstanding arises from the way we talk about a quite different subject, namely natural selection. Gene selectionism, which is a way of talking about evolution, is mistaken for genetic determinism, which is a point of view about development. People like me are continually postulating genes ‘for’ this and genes for that. We give the impression of being obsessed with genes and with ‘genetically programmed’ behaviour. Take this in conjunction with the popular myths of the Calvinistic determinacy of genes, and of ‘programmed’ behaviour as the hallmark of jactitating Disneyland puppets, and is it any wonder that we are accused of being genetic determinists?

  Why, then, do functional ethologists talk about genes so much? Because we are interested in natural selection, and natural selection is differential survival of genes. If we are to so much as discuss the possibility of a behaviour pattern’s evolving by natural selection, we have to postulate genetic variation with respect to the tendency or capacity to perform that behaviour pattern. This is not to say that there necessarily is such genetic variation for any particular behaviour pattern, only that there must have been genetic variation in the past if we are to treat the behaviour pattern as a Darwinian adaptation. Of course the behaviour pattern may not be a Darwinian adaptation, in which case the argument will not apply.

  Incidentally, I should defend my usage of ‘Darwinian adaptation’ as synonymous with ‘adaptation produced by natural selection’, for Gould and Lewontin (1979) have recently emphasized, with approval, the ‘pluralistic’ character of Darwin’s own thought. It is indeed true that, especially towards the end of his life, Darwin was driven by criticisms, which we can now see to be erroneous, to make some concessions to ‘pluralism’: he did not regard natural selection as the only important driving force in evolution. As the historian R. M. Young (1971) has sardonically put it, ‘… by the sixth edition the book was mistitled and should have read On the Origin of Species by Means of Natural Selection and All Sorts of Other Things’. It is, therefore, arguably incorrect to use ‘Darwinian evolution’ as synonymous with ‘evolution by natural selection’. But Darwinian adaptation is another matter. Adaptation cannot be produced by random drift, or by any other realistic evolutionary force that we know of save natural selection. It is true that Darwin’s pluralism did fleetingly allow for one other driving force that might, in principle, lead to adaptation, but that driving force is inseparably linked with the name of Lamarck, not of Darwin. ‘Darwinian adaptation’ could not sensibly mean anything other than adaptation produced by natural selection, and I shall use it in this sense. In several other places in this book (e.g. in Chapters 3 and 6), we shall resolve apparent disputes by drawing a distinction between evolution in general, and adaptive evolution in particular. The fixation of neutral mutations, for instance, can be regarded as evolution, but it is not adaptive evolution. If a molecular geneticist interested in gene substitutions, or a palaeontologist interested in major trends, argues with an ecologist interested in adaptation, they are likely to find themselves at cross-purposes simply because each of them emphasizes a different aspect of what evolution means.

  ‘Genes for conformity, xenophobia, and aggressiveness are simply postulated for humans because they are needed for the theory, not because any evidence for them exists’ (Lewontin 1979b). This is a fair criticism of E. O. Wilson, but not a very damning one. Apart from possible political repercussions which might be unfortunate, there is nothing wrong with cautiously speculating about a possible Darwinian survival value of xenophobia or any other trait. And you cannot begin to speculate, however cautiously, about the survival value of anything unless you postulate a genetic basis for variation in that thing. Of course xenophobia may not vary genetically, and of course xenophobia may not be a Darwinian adaptation, but we can’t even discuss the possibility of its being a Darwinian adaptation unless we postulate a genetic basis for it. Lewontin himself has expressed the point as well as anybody: ‘In order for a trait to evolve by natural selection it is necessary that there be genetic variation in the population for such a trait’ (Lewontin 1979b). And ‘genetic variation in the population for’ a trait X is exactly what we mean when we talk, for brevity, of ‘a gene for’ X.

  Xenophobia is controversial, so consider a behaviour pattern that nobody would fear to regard as a Darwinian adaptation. Pit-digging in antlions is obviously an adaptation to catch prey. Antlions are insects, neuropteran larvae with the general appearance and demeanour of monsters from outer space. They are ‘sit and wait’ predators who dig pits in soft sand which trap ants and other small walking insects. The pit is a nearly perfect cone, whose sides slope so steeply that prey cannot climb out once they have fallen in. The antlion sits just under the sand at the bottom of the pit, where it lunges with its horror-film jaws at anything that falls in.

  Pit-digging is a complex behaviour pattern. It costs time and energy, and satisfies the most exacting criteria for recognition as an adaptation (Williams 1966; Curio 1973). It must, then, have evolved by natural selection. How might this have happened? The details don’t matter for the moral I want to draw. Probably an ancestral antlion existed which did not dig a pit but simply lurked just beneath the sand surface waiting for prey to blunder over it. Indeed some species still do this. Later, behaviour leading to the creation of a shallow depression in the sand probably was favoured by selection because the depression marginally impeded escaping prey. By gradual degrees over many generations the behaviour changed so that what was a shallow depression became deeper and wider. This not only hindered escaping prey but also increased the catchment area over which prey might stumble in the first place. Later still the digging behaviour changed again so that the resulting pit became a steep-sided cone, lined with fine, sliding sand so that prey were unable to climb out.

  Nothing in the previous paragraph is contentious or controversial. It will be regarded as legitimate speculation about historical events that we cannot see directly, and it will probably be thought plausible. One reason why it will be accepted as uncontroversial historical speculation is that it makes no mention of genes. But my point is that none of that history, nor any comparable history, could possibly have been true unless there was genetic variation in the behaviour at every step of the evolutionary way. Pit-digging in antlions is only one of the thousands of examples that I could have chosen. Unless natural selection has genetic variation to act upon, it cannot give rise to evolutionary change. It follows that where you find Darwinian adaptation there must have been genetic variation in the character concerned.

  Nobody has ever done a genetic study of pit-digging behaviour in antlions (J. Lucas, personal communication). There is no need to do one, if all we want to do is satisfy ourselves of the sometime existence of genetic variation in the behaviour pattern. It is sufficient that we are convinced that it is a Darwinian adaptation (if you are not convinced that pit-digging is such an adaptation, simply substitute any example of which you are convinced).

  I spoke of the sometime existence of genetic variation. This was because it is quite likely that, were a genetic study to be mounted of antlions today, no genetic variation would be found. It is in general to be expected that, where there is strong selection in favour of some trait, the original variation on which selection acted to guide the evolution of the trait will have become used up. This is the familiar ‘paradox’ (it is not really very paradoxical when we think about it carefully) that traits under strong selection tend to have low heritability (Falconer 1960); ‘…evolution by natural selection destroys the genetic variance on which it feeds’ (Lewontin 1979b). Functional hypotheses frequently concern phenotypic traits, like possession of eyes, which are all but universal in the population, and therefore without contemporary genetic variation. When we speculate about
, or make models of, the evolutionary production of an adaptation, we are necessarily talking about a time when there was appropriate genetic variation. We are bound, in such discussions, to postulate, implicitly or explicitly, genes ‘for’ proposed adaptations.

  Some may balk at treating ‘a genetic contribution to variation in X’ as equivalent to ‘a gene or genes for X’. But this is a routine genetic practice, and one which close examination shows to be almost inevitable. Other than at the molecular level, where one gene is seen directly to produce one protein chain, geneticists never deal with units of phenotype as such. Rather, they always deal with differences. When a geneticist speaks of a gene ‘for’ red eyes in Drosophila, he is not speaking of the cistron which acts as template for the synthesis of the red pigment molecule. He is implicitly saying: there is variation in eye colour in the population; other things being equal, a fly with this gene is more likely to have red eyes than a fly without the gene. That is all that we ever mean by a gene ‘for’ red eyes. This happens to be a morphological rather than a behavioural example, but exactly the same applies to behaviour. A gene ‘for’ behaviour X is a gene ‘for’ whatever morphological and physiological states tend to produce that behaviour.