The solution – redundancy reduction – extends beyond sensory adaptation and lateral inhibition to a fascinating list of feature-detector neurones in the brain such as horizontal line detectors, vertical line detectors, ‘bug detectors’ and others, all of which can be represented as redundancy-reducing in the Barlow/Attneave sense. For example, a straight line can be represented as just its two ends, leaving the brain to ‘fill in’ the redundant intermediate points. As with the bats and the spider webs, the whole Barlow story can be told as an elegant and easily memorized sequence of problems, with engineering solutions giving rise to new problems, suggesting new engineering solutions and so on.
We should also expect that the ‘detector’ cells that evolve in the brain of an animal of a particular species will be tuned to detect not only features that are redundant in the sensory stream, but features that are functionally important for animals of that species – for example, the colour and shape of a sexual partner. These two combined would mean that a comprehensive list of the detector cells in an animal’s brain should amount to a kind of indirect description of the important properties of the world in which the species lives.
And this idea, in turn, is related to another one, this time my own: ‘the Genetic Book of the Dead’. The idea here is that the genes of an animal could in theory be read as a digital description of the environments in which its ancestors have survived.
‘The Genetic Book of the Dead’ and the species as ‘averaging computer’
River Out of Eden begins by looking back at the reader’s ancestors and reflecting – trivially when you think about it but still significantly – that not a single one of your ancestors died young or failed to achieve at least one heterosexual copulation. Every individual born inherits the genes of a literally unbroken line of successful ancestors. We inherit the genes that equipped this pro-genitorial elite, as I called them, to be an elite. The exact means by which an individual becomes a successful ancestor varies from species to species but, however they do it, we are all descended from individuals who were good at it. ‘Good at it’ means good at flying in the case of birds, bats and pterosaurs, good at digging in the case of moles, aardvarks and wombats, good at hunting in the case of lions, hawks and pike, good at fighting in the case of male deer, elephant seals and parasitic fig wasps.
There is a sense, therefore, in which the DNA of a species could, in principle, be read out as a kind of description of the way of life at which that species excels. I have mentioned this idea of ‘the Genetic Book of the Dead’ in several of my books, but I argued it most fully in the chapter of that name in Unweaving the Rainbow. Here’s one of the ways I introduced it:
A species is an averaging computer. It builds up, over the generations, a statistical description of the worlds in which the ancestors of today’s species members lived and reproduced. That description is written in the language of DNA. It lies not in the DNA of any one individual but collectively in the DNA – the selfish cooperators – of the whole breeding population. Perhaps ‘readout’ captures it better than ‘description’. If you find an animal’s body, a new species previously unknown to science, a knowledgeable zoologist allowed to examine and dissect its every detail should be able to ‘read’ its body and tell you what kind of environment its ancestors inhabited: desert, rain forest, arctic tundra, temperate woodland or coral reef. The zoologist should be able to tell you, by reading its teeth and its guts, what it fed on. Flat, millstone teeth and long intestines with complicated blind alleys indicate that it was a herbivore; sharp, shearing teeth and short, uncomplicated guts indicate a carnivore. The animal’s feet, and its eyes and other sense organs spell out the way it moved and how it found its food. Its stripes or flashes, its horns, antlers or crests, provide a readout, for the knowledgeable, of its social and sex life.
I called the species an ‘averaging computer’, but why is it the species that is the averaging computer, not the individual organism? Because, at least in sexually reproducing animals, any one individual genome is but an ephemeral sample of the gene pool that has been sieved and winnowed down the generations, averaging the conditions and adversities which individuals of ancestral generations braved and survived. The species gene pool is a kind of negative image of the average environment of individuals of the species. If we think of natural selection as a sculptor, chiselling rough raw materials towards ever-increasing perfection, the entity that is chiselled is the species gene pool. Each individual’s genome is a sample of that gene pool, and the survival (or failure) of the individual depends (among other things) upon the set of genes that it was lucky enough (or unlucky enough) to draw from the pool. I first tried to convey the idea of genes’ success being dependent on their genetic companions in The Selfish Gene in 1976, with my metaphor of reshuffled rowing crews, where the oarsmen stand for genes and the successively re-crewed boats stand for organisms. Like many metaphors, this one should not be pushed too far, but it does convey the important idea that the best genes, in the long run, will tend to survive in the gene pool, even though many copies of them perish because they are dragged down by inferior fellow crew members in particular bodies. It is the gene pool that improves in the long run as natural selection chips its way down the generations. It is but a short step from here to the image of the Genetic Book of the Dead. It’s important to understand that the environment is not directly imprinted on the genes – that would be Lamarckism. Rather, the genes vary at random and the ones that fit the environment survive to populate the gene pool of the future.
I think it was while giving tutorials that it first occurred to me that a sufficiently knowledgeable zoologist should in principle be able to read out, from the anatomy, physiology and DNA of a species, how and where it lived, who its enemies were, the weather it had to contend with and so on. I was teaching the principles of taxonomy, the science of animal classification. Animals that are unrelated but have similar ways of life tend to resemble each other in superficial features, which are in danger of distracting us from the features they share with their true taxonomic relatives. Dolphins superficially resemble marlins because both swim fast near the surface of the sea, but these superficial resemblances are outnumbered by the features by which dolphins resemble land mammals, and those by which marlins resemble other fish. Numerical methods exist for estimating these competing resemblances, independently of whether they are ‘ancestral’ or ‘recent’.
Such ‘numerical taxonomy’ methods are less fashionable nowadays than they were when I learned them as an undergraduate from Arthur Cain, but they are good for illustrating the point. You measure everything you can find about a whole lot of species, feed all the measurements into a computer, and ask the computer to come up with a figure for the distance between each species and each other species. Distance here doesn’t, of course, mean spatial distance. It means how much they resemble each other: their distance from each other in a multidimensional, mathematical ‘resemblance space’. What you hope to find is that, although dolphins and marlins are pulled a little bit ‘closer’ to each other than they ‘should’ be, because of their similar ways of life, these similarities (streamlining and so on) are swamped by the much more numerous differences stemming from the fact that one is a mammal and the other a fish: they’ve had a very long time to diverge from each other since the Devonian period. The numerical calculations ‘filter off’ the superficial (minority) resemblances by swamping them, leaving us with the ‘fundamental’ (majority) resemblances that indicate pedigree relationships.
It occurred to me, while thinking aloud together with pupils in tutorials, that those numerical methods might in principle be stood on their head. Instead of filtering off the ‘superficial’ functional characteristics (like the streamlined shape of dolphins and marlins), leaving the ‘true’ taxonomic characters, we could do the opposite: go out of our way to filter off the taxonomic characters that stem from relatedness, and concentrate on the minority of functional resemblances. How might this be done? Imagi
ne that we construct a set of pairs of animals. The first of each pair thrives in water, the second on dry land. Yet, taxonomically speaking, each animal is more closely related to its pair than to any of the others on ‘its side’ of the pairings: {otter, badger} {beaver, gopher} {yapok, opossum} {water shrew, land shrew}{water vole, vole} {pond snail, land snail} {water spider, land spider} {marine iguana, land iguana}. Suppose we make hundreds of measurements on all these animals (and lots more similar pairs) – anatomical measurements, physiological measurements, biochemical measurements, genetic sequences – and then take them all and throw them into a computer, telling the computer which member of each pair is aquatic, which terrestrial. Now ask the computer (this is not as easy as it sounds, but methods exist for doing it) questions like this: ‘What do the aquatic animals in each case have in common, as opposed to their terrestrial counterparts?’ We might get a bit subtler than this. Instead of asking our animals to tick a box, either aquatic or terrestrial, we might place them along a gradient of aquaticness and look for quantitative correlations along the gradient. We might even make so bold as to ask: ‘What measurements of an animal would I have to multiply by what factor, in order to morph it from terrestrial to aquatic?’
We could then do the same thing for pairs of arboreal versus ground-dwelling species: {squirrel, rat} {tree frog, frog} {tree kangaroo, wallaby}; then the same for pairs of underground versus above-ground species: {mole, shrew} {mole cricket, cricket} {mole rat, rat} and so on. In the case of aquatic versus terrestrial, we might expect webbed feet to pop up as one answer, and that’s pretty obvious. But I would hope the computer would find less obvious answers, buried deep inside the animals. Something about blood chemistry, for example. And, to bring us back to the Genetic Book of the Dead, we could do the same exercise with genes. Are there genes that tie aquatic animals to other aquatic animals in spite of their not being particularly closely related? We normally expect genetic comparisons to tell us which animals are closely related. With respect to most of their genes, marine and land iguanas, being close cousins, will certainly come out closely resembling each other. But I would also hope to do the opposite: find a few genes that marine iguanas have in common with other marine creatures, and don’t share with land iguanas or other dry-land animals – perhaps a gene concerned with excretion of salt.
It was considerations such as this, talked through and argued about with pupil after pupil in tutorials down the years, that led me to coin phrases like ‘Genetic Book of the Dead’, and to suggest that a sufficiently knowledgeable zoologist, when presented with an unknown animal, should eventually, with the help of a computer, be able to reconstruct the way of life of that animal – more strictly, of its ancestors. In particular, the genes that helped the animal’s ancestors to survive are in principle decipherable as a coded description of its ancestral world: ancestral predators, ancestral climate, ancestral parasites, ancestral social system.
And in those tutorials where my pupils and I threw these ideas around, I was mindful of my own tutor Arthur Cain, and his dictum that ‘the animal is what it is because it needs to be’. On one occasion as a graduate student I found myself in the Royal Oak pub in Oxford (known as the doctors’ pub because the old Radcliffe Infirmary was opposite) having a solitary supper of, I am chagrined to say, bacon and eggs. Arthur coincidentally happened to be doing the same thing in the same pub, so we joined each other (like, I am also chagrined to recall, the two ‘travelling men’ who founded the Gideons). We talked about taxonomy and adaptation, and Arthur at one point illustrated his theme by suggesting that a squirrel might be described as a rat which had moved a certain distance away from a rat-like ancestor, along the ‘arboreality dimension’. That image stayed with me and informed the chapter of Unweaving the Rainbow called ‘The Genetic Book of the Dead’ and also the idea of ‘the Museum of All Possible Animals’ which dominated two chapters of Climbing Mount Improbable (see below). But the ‘Museum’ was more directly inspired by my attempts at computer modelling, which began while I was writing The Blind Watchmaker.
Evolution in pixels
Chapter 3 of The Blind Watchmaker, ‘Accumulating small change’, occupied as much time and effort as the other ten chapters put together. This was because of the weeks and months I spent writing the suite of computer programs, called Blind Watchmaker, designed to breed ‘computer biomorphs’ on the screen by artificial selection. The word ‘biomorph’ was borrowed from my friend Desmond Morris, whose surrealist paintings depict quasi-biological forms which, by his own entirely believable account, ‘evolve’ from canvas to canvas. Desmond’s painting The Expectant Valley had been used for the cover of The Selfish Gene. I bought the original at one of Desmond’s exhibitions, because the price (£750) was exactly equal to the advance given me by Oxford University Press, and the omen pricked my fancy. When, a decade later, I spoke to Desmond about The Blind Watchmaker he was so taken with the title that he set to work, there and then, on a painting with the same title. And this new painting – though it had more to do with the title than the contents of the book – later graced the covers of both the Longman and the Penguin editions of The Blind Watchmaker.
I wrote my computer biomorph program in Pascal, a now largely superseded language, which itself was a direct descendant of the (even more thoroughly superseded) Algol 60 language that I had learned as a graduate student. I had continual recourse to the Apple Macintosh ‘Toolbox’, the repertoire of hard-wired machine code programs that gives the Mac its characteristic (and notoriously imitated) ‘look and feel’; and the half-dozen technical manuals of the Mac toolbox became my much thumbed, increasingly grubby and messily annotated bible.
I was also continually running to the ever-patient Alan Grafen for help and advice: not that he was a more experienced Mac programmer than I was – rather the reverse – but he has undenied advantages in the IQ department. As P. G. Wodehouse might have put it: ‘North of the collar stud, Alan stands alone.’ Or, as Marian said of him: ‘He has the most annoying habit of being right.’ During the course of my programming marathon Alan once rather endearingly said he felt sorry for me, because I was mired in a peculiarly difficult piece of coding but I had got into it too deep to back out. That sounds Concordian and, to an extent, it was: backing out would have meant throwing away all the work I had put in so far. But there was more to it than that. I was driven to persist – and for this I dare to take credit, even to feel a little proud – by a biological intuition, almost like an instinctive nose for what I, as a biologist, could scent must work. I was propelled forwards by a conviction that something truly exciting must eventually emerge from my biomorph-generating algorithm, if only I persisted and got myself out of the mire of complexity.
The key to it was the fractal nature of the embedded ‘embryology’ of my biomorphs, the recursive tree-growing procedure whose quantitative details were controlled by a set of nine (more in later versions of the program) numbers, which I called genes. Obviously, if you change the numerical values of the genes you’ll change the morphology of the biomorph. Less obviously, the change is often in a biologically interesting direction. I imported Darwinism (though not sex) by (asexually) ‘breeding’ daughter biomorphs from parent biomorphs using artificial selection. The computer offered up a choice of daughter biomorphs with slightly mutated genes, and the human chooser picked the one that was to give birth to the next generation – and so on for an indefinite number of generations. The numerical values of the genes were concealed: just like a breeder of cattle or roses, the biomorph breeder saw only the consequences of genetic change, the morphology on the computer screen.
In my dreams, then, I foresaw that something interesting and unexpected would emerge. But I never dared hope that my biomorphs would evolve their way from botany to entomology!
When I wrote the program, I never thought that it would evolve anything more than a variety of tree-like shapes. I had hoped for weeping willows, cedars of Lebanon, Lombardy poplars, seaweeds, perhaps deer antl
ers. Nothing in my biologist’s intuition, nothing in my 20 years’ experience of programming computers, and nothing in my wildest dreams, prepared me for what actually emerged on the screen. I can’t remember exactly when in the sequence it first began to dawn on me that an evolved resemblance to something like an insect was possible. With a wild surmise, I began to breed, generation after generation, from whichever child looked most like an insect. My incredulity grew in parallel with the evolving resemblance . . . I still cannot conceal from you my feeling of exultation as I first watched these creatures emerging before my eyes. I distinctly heard the triumphal opening chords of Also sprach Zarathustra (the ‘2001 theme’) in my mind. I couldn’t eat, and that night ‘my’ insects swarmed behind my eyelids as I tried to sleep.
There are computer games on the market in which the player has the illusion that he is wandering about in an underground labyrinth, which has a definite if complex geography and in which he encounters dragons, minotaurs or other mythic adversaries. In these games the monsters are rather few in number. They are all designed by a human programmer, and so is the geography of the labyrinth. In the evolution game, whether the computer version or the real thing, the player (or observer) obtains the same feeling of wandering metaphorically through a labyrinth of branching passages, but the number of possible pathways is all but infinite, and the monsters that one encounters are undesigned and unpredictable. On my wanderings through the backwaters of Biomorph Land, I have encountered fairy shrimps, Aztec temples, Gothic church windows, aboriginal drawings of kangaroos, and, on one memorable but unrecapturable occasion, a passable caricature of the Wykeham Professor of Logic.
The latter paragraph touches on one of the main biological lessons that I took away with me from this programming exercise. The inner eye of my imagination saw ‘biomorph land’, a multidimensional landscape of morphology, a nine-dimensional hypercube in which all possible biomorphs lurked, every one connected to every other one by a navigable trajectory of step-by-step, gradual evolution. In theory, though less tidily because the number of genes is not fixed, we can imagine all possible real animals as sitting in an n-dimensional hypercube, and I called this ‘genetic space’ in chapter 3 of The Blind Watchmaker. Most of the inhabitants of this monstrous (I use the word advisedly) hypercube not only have never existed but never could have survived if they had done: ‘However many ways there may be of being alive, it is certain that there are vastly more ways of being dead’ (a sentence that, I am pleased to see, has made it into the Oxford Dictionary of Quotations). Actual animals are islands in this hyperspace, vastly spaced out from one another as if in some Hyperpolynesia, surrounded by a fringing reef of closely related animals and separated from other islands by largely impassable wastes of impossible animals. Actual evolution is represented by timelines, trajectories through the hypercube. You see, although I’m no good at writing equations or getting my sums right, I do perhaps have the rudiments of the soul of a mathematician. Or so I would aspire.
Brief Candle in the Dark Page 35