Brief Candle in the Dark

by Richard Dawkins

  All my biomorph-style programs employed artificial selection, not natural selection. I could only dream about the much more difficult problem of how to simulate natural selection in an interesting way. The fact that it’s difficult is itself instructive. One could imagine building into my biomorph program a selection criterion such as ‘spikiness’ or ‘roundness’. And indeed I did exactly that as an experiment. This bypassed the human eye as selection agent, and it did work. But it wasn’t biologically very interesting. In order to simulate survival in a ‘world’, it would be necessary to construct that world, with its own ‘physics’, its own (ideally three-dimensional) geography, its own rules for how biomorphs should interact with other objects and other biomorphs in that world, rules for how not to occupy the same physical space as other objects, and so on. In the years since The Blind Watchmaker was published, clever programmers have developed such artificial worlds with their own ‘physics’, for example Steve Grand with his Creatures, Torsten Reil with his Natural Motion, and the various fantasy environments of the Second Life type. Out of my league, and anyway I’ve shaken off the addiction to programming.

  Arthromorphs

  The evolution of evolvability is all about floodgates opening to new creative improvement. The Los Alamos conference where I introduced the idea became a kind of metaphor for the idea itself, because that conference really did release something a bit like a creative surge in my own mind (and probably in other participants’ minds as well). For me that surge was to reach its climax in Climbing Mount Improbable, which I regard as my most underrated book (it is the least read of them all, although it is probably the most innovative after The Extended Phenotype).

  And now here’s another gate that conference opened. It was there that I met Ted Kaehler. One of Apple’s star programmers, Ted has the sort of creatively original mind we have come to associate with that artistically innovative corporation. He was there partly to assist with computer demonstrations (including mine), but his expertise and interests extend way beyond such technicalities, and I had many discussions with him on evolutionary ideas. I saw more of him later, when he was working with Alan Kay’s Apple-sponsored education project in Los Angeles, the project whose high-pressure think-tank I was briefly privileged to join when I stayed with the lovely Gwen Roberts and did most of my work on the colour biomorphs (see page 288). Ted and I brainstormed with increasing enthusiasm – it’s a wonderful feeling, as I described in the wasp chapters, when joint thinking goes fast and well. We obsessed about the evolution of evolvability, especially segmentation; and together we hatched a plan to write a new biomorph-style artificial selection program concentrating on segmented arthropod-like artificial creatures, and incorporating other overtly biological principles of embryology. We called our new artificial creatures ‘arthromorphs’.

  The original Blind Watchmaker biomorphs had nine genes. The ‘Los Alamos’ version had sixteen. The colour version had thirty-six. Each enlargement of the genome opened what I have been calling floodgates, unleashing an expansion of evolutionary ‘creativity’, albeit constrained in ‘constructive’ ways, for example by segmentation or ‘kaleidoscopic mirrors’. But each of these enhancements relied on a major intervention by the programmer. I had to go back to the drawing board and write a whole lot of new code. And, in a way, that is a suitable metaphor for the evolution of evolvability, for I do think that in real biology the radical watershed events we are talking about – like the origin of segmentation, the origin of multicellularity, the origin of sex, or the origin of five-way symmetry in echinoderms – are rare and rather catastrophic upheavals, a little bit analogous to a major rewrite of a computer program. Indeed, the analogy even extends to ‘debugging’, for we can be sure that when a revolutionary mutation is incorporated into the gene pool by selection it will have knock-on effects, which need to be ironed out in its wake: ironed out by subsequent selection in favour of a retinue of minor mutations that smooth away the adverse side-effects of a generally beneficial major mutation.

  But real biology knows an intermediate tier of mutation, less revolutionary than the origin of multicellularity, sex, segmentation or novel ‘mirrors’ of symmetry, but more radical than the ordinary point mutations whereby a Watson–Crick nucleotide changes to another of the gang of four – C, T, G or A. This intermediate category includes duplications (or the inverse, deletions) of whole stretches of chromosome. Gene duplication is the main way in which genomes get bigger. In The Ancestor’s Tale (specifically in The Lamprey’s Tale) I described the process for the particular case of haemoglobin. To recap briefly: we have five different ‘globin’ chains, coded by different genes in different parts of the genome. And the point is that all five are descended from a single ancestral globin coded by a single ancestral gene. The ancestral gene (which is still the only one possessed by our remote and primitive cousins the lampreys) was successively duplicated in evolution to make the multiple ‘globin genes’ we have today. Normally when we speak of evolutionary divergence we mean the splitting of an ancestral species into two. Two populations of walking, breathing animals split and go their separate ways. Here, on the other hand, while we are still talking about evolutionary divergences, we are talking about splits that each occurred within a single individual, so that the two descendant molecular lineages persisted, side by side, within the bodies of future individuals through all future generations.

  Incidentally, I am often asked whether our improved understanding of genomics has changed what I would say if I were to rewrite The Selfish Gene. The answer is no – a reluctant no in some ways, for scientists take pride in changing their minds when new evidence comes in. My ‘gene’s-eye view’ of 1976 is, if anything, strengthened by new considerations such as the gene duplication discussed in The Lamprey’s Tale. This is because we now see evolutionary divergence at the gene level within individuals, which downplays the importance of the individual (as opposed to the gene) as the level at which selection acts.

  Ted and I, when drawing up the specification for our Arthromorphs program, didn’t attempt to simulate haemoglobin-style gene duplication per se. However, our new program did incorporate a form of gene duplication (and deletion), which proved to be highly instructive. Whereas all my previous biomorph programs had a fixed repertoire of genes (nine, sixteen or thirty-six for the three versions), arthromorphs had a variable number of genes, the number of genes itself being subject to mutation. Can you see that we were moving in the direction of letting evolution do its own rewriting of the software, whereas previously, for each macro-mutated advance in the evolvability of biomorphs, I had had to sit down and write a whole lot more code?

  Segmentation was deeply woven into the fabric of arthromorph embryology, although it was genetically allowable to have only a single segment. Left/right mirroring was a default constraint: all arthromorphs were left/right symmetrical. Each segment consisted of an oval body part (its shape and size under genetic control) with the capacity to grow a pair of symmetrical limbs, each limb with the capacity to bifurcate in a claw. So far, so arthropodan. The number of joints in each limb was under genetic control, as were the size of each and the angle of each joint; and the same went for the size and angle of the terminal claws.

  Where it starts to get embryologically more interesting is that groups of neighbouring segments (in series) share fields of influence. For example, the first three (say) segments might be nearly the same as each other but more different from the next two segments, and different again from the next four segments – a structure redolent of head, thorax, abdomen (see Arthromorph 1 on the following page). Each of these (of course it didn’t have to be three: that number was itself subject to genetic variation) groups of segments we called a tagma (plural tagmata), which is the correct name in arthropod biology. But the segments within a tagma didn’t have to be literally identical. Each segment was influenced by its own tagma-specific genes, which were free to mutate independently of the other segments. The comparative uniformity within a
tagma was achieved by multiplying the genetic quantities of each segment by a number (a ‘gene’) specific to that tagma. Arthromorph 2 pictured here is similar to Arthromorph 1, except that segment 3, though recognizably a member of tagma 1, has longer legs than the other two segments of tagma 1. Tagma 3 as a whole has different legs, too.

  At a higher level, there were other genes which multiplied the values of all the genes of the whole organism, across all tagmata. Finally, we added ‘gradient’ genes, which multiplied the other genetic effects by an increasing (or decreasing) number as you proceeded posteriorly along the organism (or along a tagma). Increases (and decreases) in numbers of tagmata, and of segments within each tagma, were achieved by gene duplication (or deletion).

  Such was arthromorph embryology, and you’ll notice that it was more complicated than biomorph embryology in biologically interesting ways. It pushed me to the limit of my programming skill, so that I had to lean on Ted’s superior experience. I did the coding myself (in Pascal, which was not Ted’s preferred language and has now, as I’ve already mentioned, been largely superseded), but Ted guided me with emailed suggestions written in a kind of pseudo computer language pretty much akin to a formal subset of English. At times, I suspect he must have got a little impatient with my slowness – not up to the standards of a professional Apple software engineer – but he was always very kind and we got it finished in the end. Once the difficult embryology routine was written, it was a simple matter to embed it in a version of the original biomorph program to handle the choosing for ‘breeding’ on the screen. Below is a ‘zoo’ (perhaps flea circus is a better term) of a selection of the countless arthromorphs that I was able to breed by artificial selection, once the program was finished.

  Christopher Langton’s series of Artificial Life conferences continued, as did a related series called Digital Biota. Chris himself was present at the second of these, held at Magdalene College, Cambridge in 1996, and I was invited to give the keynote address, on the title ‘The view from real life’: obviously an attempt to ground the geeks in real biology as they explored the enchantments of their virtual worlds. The conference, for me, was chiefly memorable because of a wonderful impromptu speech by Douglas Adams (reprinted in The Salmon of Doubt) and because it was there that I met Steve Grand, author of Creation: Life and How to Make It, a tour de force to match the virtuosity of his artificial life program, Creatures. It was also there that I was introduced to the amazing possibilities of virtual worlds in which ‘avatars’, owned by players from all over the world, could roam amid fantastic castles and palaces, casinos and streets, all built and even policed as a communal project. Although I am intrigued by the feats of programming that go into such virtual worlds, I find something a little creepy about the extremes reached by people who, through their avatars, inhabit them. The phrase ‘get a life’ has become a cliché, but you can’t help feeling it ought to be hung up in prominent town squares in Second Life, where the inhabitants have even been known to ‘marry’ people whom they have never met in the flesh, and subsequently get ‘divorced’ over ‘unfaithfulness’ in cyberspace. Ah well, perhaps it is the way of the future and one day I shall virtually eat my own real words.1

  The cooperative gene

  My emphasis on the species gene pool (rather than the individual genome) as the text of a Genetic Book of the Dead also serves to shore up another central plank of my world-view: the cooperative gene. This is the obvious but important idea that, from a functional point of view, the gene is impotent outside the context of the other genes in the gene pool – which is to say the other genes with which it has to share a large number of individual bodies, distributed in space and time. There is a whole chapter, ‘The selfish cooperator’ devoted to it in Unweaving the Rainbow, but it is an idea that was foreshadowed (despite the book’s title) in The Selfish Gene:

  A gene that cooperates well with most of the other genes that it is likely to meet in successive bodies, i.e. the genes in the whole of the rest of the gene pool, will tend to have an advantage.

  For example, a number of attributes are desirable in an efficient carnivore’s body, among them sharp cutting teeth, the right kind of intestine for digesting meat, and many other things. An efficient herbivore, on the other hand, needs flat grinding teeth, and a much longer intestine with a different kind of digestive chemistry. In a herbivore gene pool, any new gene that conferred on its possessors sharp meat-eating teeth would not be very successful. This is not because meat-eating is universally a bad idea, but because you cannot efficiently eat meat unless you also have the right sort of intestine, and all the other attributes of a meat-eating way of life. Genes for sharp, meat-eating teeth are not inherently bad genes. They are only bad genes in a gene-pool that is dominated by genes for herbivorous qualities.

  This is a subtle, complicated idea. It is complicated because the ‘environment’ of a gene consists largely of other genes, each of which is itself being selected for its ability to cooperate with its environment of other genes.

  I would happily bring out a new book called The Cooperative Gene, but the book itself would be identical, word for word, with The Selfish Gene.1 There is no paradox here. The selfish genes that survive do so in their environment. That environment of course includes the external environment that we can see: the climate, the predators and parasites, the food supply and so on. But an even more important part of the environment of any gene is the other genes in the gene pool of the species – which means the set of genes with which it is statistically likely to share a body. A gene in isolation can have no phenotypic effects, and the phenotypic effects that it does have depend on the other genes that are present in the body in the short term, the gene pool in the long term. Natural selection, at each locus independently, favours whichever allele cooperates with the other genes with whom it shares a succession of bodies: and that means it cooperates with the alleles at those other loci, which cooperate in their turn. Cooperation is the name of the game. The effect is that cartels of mutually cooperating genes build up in gene pools. If a member of one cartel were plucked out and plonked down in another cartel, the result would not be a success. My understanding of this important point was heavily influenced by research of the E. B. Ford school at Oxford. Ford and his colleagues showed by hybridization experiments that complex characteristics of moths break down when genes are exposed to a foreign ‘genetic climate’, the foreign genetic climate of another species. This work made a big impression on me in my undergraduate tutorials with Ford’s junior colleague Robert Creed. Here’s how I put it in Unweaving the Rainbow. I apologize for such long quotations, but I have no better words to express a point which has been many times misunderstood.

  It is tempting to speak of the ‘whole cheetah’ or the ‘whole antelope’ as being selected, ‘as a unit’. Tempting, but superficial. Also lazy. It requires some extra thinking work to see what is really going on. Genes that program the development of carnivorous guts flourish in a genetic climate that is already dominated by genes programming carnivorous brains. And vice versa. Genes that program defensive camouflage flourish in a genetic climate that is already dominated by genes programming herbivorous teeth. And vice versa. There are lots and lots of ways of making a living. To mention only a few mammal examples, there is the cheetah way, the impala way, the mole way, the baboon way, the koala way. There is no need to say that one way is better than any other. All of them work. What is bad is to be caught with half your adaptations aimed at one way of life, half aimed at another.

  This kind of argument is best expressed at the level of the separate genes. At each genetic locus, the gene most likely to be favoured is the one that is compatible with the genetic climate afforded by the others, the one that survives in that climate through repeated generations. Since this applies to each one of the genes that constitute the climate – since every gene is potentially part of the climate of every other – the result is that a species gene pool tends to coalesce into a gang of mutually compatibl
e partners.

  And that is the very important sense in which genes are simultaneously ‘selfish’ and ‘cooperative’: a cornerstone of what, in my more presumptuous moments, I might call my ‘world-view’. It is a much more coherent and penetrating way to think about the evolution of cooperation than limp-wristed hand-waving about selection of the organism ‘as a unit’.

  Universal Darwinism

  In 1982, the centenary of Darwin’s death was marked by commemorations around the world, perhaps most prominently at Cambridge, where the young Charles read for his undergraduate degree in theology, and where he ‘walked with Henslow’ and collected beetles. I felt honoured to be invited to speak, and I chose as my title ‘Universal Darwinism’. My idea was that natural selection is more than just the driving force of evolution in the life forms that we know, on this planet; there is, as far as we know, no other force capable of filling the role of ultimate responsibility for adaptive evolution. ‘Adaptive’ is a necessary word in that sentence. Random genetic drift is responsible for much, if not most, of evolutionary change at the molecular level. But it cannot be responsible for functional, adaptive evolution. Only natural selection, as far as we know and as far as anybody has so far imagined, produces organs that work as if an engineer had designed them: wings that fly, eyes that see, ears that hear, stings that paralyse. Or so I asserted. My stated implication was that, if we ever discover life elsewhere in the universe, it will turn out to be Darwinian life: it will have evolved along some local equivalent of Darwinian principles.