Climbing Mount Improbable

by Richard Dawkins

  The ladder web, then, is one solution to the problem of how to impede escaping prey, one that is especially effective against moths. Another technique used by some species of spider is the sprung trap. Hyptiotes's web is not a full orb, but is reduced to a triangle with only four spokes. There is an additional line attached at the point of the triangle which keeps the whole web taut. But this main guy rope, instead of being attached directly to a firm surface, is held by the spider. Indeed, the spider herself forms a living link in the attachment to a firm surface. She pulls the line taut with her front legs, and uses her third pair of legs to hold a loop of slack. She sits dangerously still in this suspended position, and waits. When an insect blunders into the web the spider reacts instantly. She releases the trap, which collapses over the insect and at the same time jerks her on towards it. She may spring the trap in two or three further stages, gathering up the slack and releasing more silk behind. The insect is now hopelessly entangled in the collapsed net. The spider wraps the victim in yet more silk and carries it away as a thickly swathed parcel. Only then does she finally bite the poor creature, inject digestive juices and slowly suck its liquefied remains out through the wall of the silken parcel. The triangular web is now not in a fit state to be used again, and it has to be rebuilt from scratch.

  Presumably Hyptiotes is solving the problem that a web under tension, although good for catching an insect in the first place, is vulnerable to powerful struggling. If you are an insect caught by sticky threads, it is easier to pull yourself free if those threads are under tension than if they are slack. If the threads are slack, there is nothing for you to pull against and you become ever more comprehensively mired in sticky silk. Like a supersonic plane whose optimal wing shape for taking off is different from the optimum for fast flight, a spider webs optimal tension for initially catching prey is different from the optimum tension for keeping them entangled. Some planes solve their dual optimum problem by compromise: being not too bad {54} at either task. Others — swing-wing fighter planes — get the best of both worlds by varying the geometry of their wings, albeit at a cost in complicated mechanism. Hyptiotes builds a variable-tension web.

  Ordinary orb web spiders seem to go for the high tension that is best for initial capture, and rely on their own speed about the web to grapple the prey into submission before it can escape. Other spiders seem to go for the opposite solution, and build webs with threads that are loose in the first place (Figure 2.8). Pasilobus builds a triangular web with a single thread bisecting the main angle. The sticky capture threads are reduced to a few loosely hanging loops. The clever thing about these loose loops — this is another elegant discovery by Michael and Barbara Robinson in New Guinea — is that they come away specially easily at one end. An insect such as a moth, having blundered into a thread and stuck to it, quickly breaks the thread at the special low-shear junction, but remains tethered at the other end of the thread. The victim now flies round and round like a toy plane on a string. It is an easy task for the spider to haul in the line and dispatch the prey. The advantage of this arrangement may again be {55}

  Figure 2.8 Pasilobus triangular web with quick-shear threads.

  partly that the insect can't struggle free because everything is so loose that it can't get a solid purchase. Or it may be that the main benefit of the quick-release threads is one that harks back to an earlier member of our list of problems: how to absorb the impact of a fast-flying insect without bouncing it back like a trampoline. As in the other triangular web, it seems probable that Pasilobus's triangle is the reduced descendant of a full orb web. At any rate, there is another genus, Poecilopachys, which uses the same quick-shear principle in a full orb web. In this case, unlike most orbs, the web is horizontal, not vertical. If we think of Pasilobus's triangle as a reduced version of Poecilopachys's orb, the ultimate reduction in the same direction is the single thread of the bolas spider Mastophora (Figure 2.9). A bolas (or bola) is a weapon originally invented by native South Americans and still used by gauchos for hunting (for example) rheas, the large flightless birds of the pampas. It consists of a weight, such as a pair of balls or stones, on the end of a rope. It is slung towards the prey with the purpose of entangling its legs and bringing it down. The young Charles Darwin experimented with the bolas while on horseback and managed to catch his own horse — to the amusement of the gauchos though not, presumably, of the horse. The bolas spider's prey are always male moths of the family Noctuidae, and for a reason. Noctuid female moths lure their mates from a distance by releasing a unique {56}

  Figure 2.9 Bolas spider.

  perfume. The bolas spider lures males to their deaths by synthesizing a closely similar perfume.The ‘bolas’ is a weighty bob on the end of a single thread of silk which the spider holds in one ‘hand’. It waves the bolas around until it entangles a moth, then hauls it in. It is altogether a more high-tech affair than the gaucho's simple bag of stones. It is in fact a tightly coiled rope of silk embedded in a drop of water, like one of the sticky beadlets on an orb web. When the spider slings its bolas, the silk unreels automatically just as an anglers line unreels when he casts. If the moth is hit, it sticks, and flies round and round. The rest of the story is much the same as for the spiders with the easy-shear threads. The moth is reeled in and fanged. The bolas spider lives in South America and it is a wonderful thought that the Indians may have got the idea for their bolas by observing it in action.

  We've been looking at variants on and reductions of the standard orb web. It is time to return to the orb web itself. At the end of the previous chapter we raised the question of how to take a computer model of artificial selection like the biomorph program and turn it into a model of natural selection, with blind nature instead of a human eye doing the choosing. We agreed that the snag with biomorphs was that they had nothing corresponding to a real, physical world in which to survive and be successful or unsuccessful. We could imagine some biomorphs behaving like predators; perhaps imagine them chasing other biomorphs behaving like prey. But there seems to be no natural, uncontrived way to decide which features of biomorphs will make them good, or not so good, at catching prey or at escaping predators. The human eye may see a pair of slavering, rapacious fangs mounted at one end of a biomorph (Figure 1.16, p. 33). But these gaping jaws, however fearsome they seem to our imaginations, cannot prove themselves in practice because they don't move, don't inhabit a world of real physics in which their sharpness can penetrate real shell or hide. The fangs and skin are only patterns of pixels on a two-dimensional fluorescent screen. Sharpness and toughness, brittleness and venomousness, these quantities have no meaning on the computer screen beyond contrived meanings defined as arbitrary numbers by the programmer. You can lash up a computer game in which numbers battle against other numbers, but the graphic clothing of the numbers {57} is cosmetic and superfluous. ‘Arbitrary’ and ‘contrived’ strike the player as understatements. It was at this point at the end of the previous chapter that, with relief, we fell back on the spider web. Here was a piece of nature that could be simulated non-arbitrarily.

  Orb webs in real life do their business largely in two dimensions. If the mesh is too coarse, flies pass straight through. If the mesh is too fine, rival spiders will achieve nearly the same result at less cost in silk, and will therefore leave behind more progeny to carry on their economically more prudent genes. Natural selection finds the efficient compromise. A web drawn on a computer screen has properties that interact, in ways that are hardly arbitrary at all, with flies drawn on the same screen. Size of mesh is a quantity that really means something on the computer screen, in relation to size of computer ‘fly’. Total quantity of line (‘cost of silk’) is another such quantity. The ratio between the two that defines efficiency can be measured with an acceptably small allowance of contrived artificiality. It is even possible to import some more sophisticated physics into the computer model, and Fritz Vollrath (from whom I learned much of what I have written in this chapter), with his physicist
colleagues Lorraine Lin and Donald Edmonds, have made a good start. It is easier to simulate ‘elasticity’ and ‘breaking strain’ in computer ‘silk’ than it is to simulate, say, ‘nimbleness’ in ‘dodging’ a computer ‘predator’, or ‘alertness’ in ‘spotting’ one. But in this chapter we shall be more concerned with models of web-building behaviour itself.

  In writing the simulation rules for a computer spider, the programmer has the benefit of lots of detailed research on the rules actually followed by real spiders, and the decision-points that punctuate the stream of spider behaviour. Professor Vollrath and the members of his international spider research group are in the forefront of this research, and they are therefore well placed to embody the knowledge in a computer program. In fact writing a computer program is a pretty good way to summarize knowledge about any set of rules. Sam Zschokke is the member of the group who has taken on the task of summarizing, in computer form, the descriptive information about the observed movements of web-building spiders. His program is called ‘MoveWatch’. Peter Fuchs and Thiemo Krink, building on {58} work by Nick Gotts and Alun ap Rhisiart, have concentrated on the inverse task of programming ‘computer spiders’ catching ‘computer flies’. Their program is called NetSpinner.

  Figure 2.10 is Move Watch's picture of the movements of an individual Araneus diadematus as she built one particular web. Note that these are not pictures of webs, although they look superficially like it. What we have here is a telescoping in time of the movements in time of a spider. It was made by videotaping the spider as she built her {59}

  Figure 2.10 Computer-tracing of a particular (Araneus diadematus) spider's positions as it spins a web. Move Watch program written by Sam Zschokke: (a), (b) preliminaries; (c) radii; (d) auxiliary spiral; (e) sticky spiral; (f) all movements superimposed.

  web. Her position at successive instants was fed into the computer in the form of a pair of grid coordinates. Then the computer drew lines between the successive positions. The ‘sticky spiral’ lines (Figure 2.10e), for instance, represent the trajectory of the spider while she was building the sticky spiral. They do not represent the exact positions of any silk threads. If they did, they'd be more evenly spread. As it is, they are concentrated in ‘waves’, reflecting the fact that the spider used the temporary, auxiliary spiral as a support while she built the sticky spiral (Figure 2.10d).

  These diagrams do not represent models of the behaviour of computer spiders. Instead, they are computer descriptions of real spider behaviour. We now turn to NetSpinner, the complementary program which behaves like a kind of idealized, theoretical spider. It can be made to behave like any of a great variety of theoretical spiders. NetSpinner simulates artificial spider behaviour, in the same kind of way as the biomorph program simulated the anatomy of insect-like creatures. It builds ‘webs’ on the computer screen, using behavioural rules whose details vary under the influence of ‘genes’. As in biomorphs, the genes are just numbers in the computer's memory, and they are transmitted forwards from generation to generation. Within each generation, the genes influence the ‘behaviour’ of the artificial spider and hence the shape of the ‘web’. For instance, one gene might control the angle between radial spokes: mutation in this gene would change the number of spokes, by making a numerical adjustment to a behavioural rule in the computer spider. As in the biomorph program, the genes are allowed to alter their values slightly, at random, as the generations go by. These mutations show themselves as changes in web shape, and are hence subject to selection.

  You can think of the six webs of Figure 2.11 as though they were biomorphs (ignore the spots for the moment). The web at top left is the parent. The other five are mutant offspring. Of course in real life webs don't give birth to webs; spiders (who build webs) give birth to spiders (who build webs). But actually there is an important sense in which what I have just said about webs here could also be said of bodies. Genes (which build human parents) give rise to genes (which build human children). In the computer model, the genes that built {60}

  Figure 2.11 Computer-generated webs, bombarded with computer-flies. NetSpinner program written by Peter Fuchs and Thiemo Krink.

  the parent web at top left (via their influence on the behaviour of a notional spider which we don't actually see on the screen) are the genes that were mutated to give rise to the genes that built the daughter webs in the other five slots.

  Of course we could, as if we were choosing a biomorph for breeding, choose by eye one of the six webs for breeding. What this would mean is that its genes would be the ones chosen to go forward (subject to mutation) to the next generation. But that would be artificial selection. The whole point of switching from biomorphs to spider webs was that we saw a possible opportunity to simulate natural selection: selection by measured efficiency at catching ‘flies’ rather than selection by human aesthetic whim.

  Now look at the spots on the picture. These are ‘flies’ that the computer has shot at random at the webs. If you look carefully, you'll notice that it's the same set of randomly positioned flies that has been shot at all six webs. This is the sort of thing that a computer, as {61} opposed to real life, does all the time unless you go out of your way to tell it not to. It is not important in this case, and it even eases comparison between the webs. Comparison means partly that the computer counts up the number of flies ‘caught’ by each of the six webs. If that were all, the web at bottom right would win the contest, for its sticky spiral embraces the largest number of flies. But sheer number of flies is not the only important variable. There is also the cost of the silk. The web at top middle uses the least amount of silk, so if that were the sole criterion it would win the competition. The true winner is the web that catches the most flies minus a cost function computed from the length of silk. By this more sophisticated calculation, the winner is the web at bottom middle. This is the one chosen to breed and pass the genes that built it on to the next generation. As in the biomorph program, this process of breeding from winners over many generations fosters a gradual evolutionary trend. But whereas with the biomorphs the direction of the trend was guided purely by human whim, in the case of NetSpinner the direction of evolution is automatically guided towards improved efficiency. It is what we hoped for: a computer model of natural selection rather than artificial selection. And what evolves under these conditions? It is really rather gratifying how lifelike are the webs that emerge in an overnight run of forty generations (Figure 2.12).

  The pictures I've shown so far were produced by NetSpinner II, which is mainly the work of Peter Fuchs (NetSpinner I was a preliminary version that I shall not discuss). Later versions of the NetSpinner program, rewritten by Thiemo Krink, steal a march on biomorphs in an additional important respect. NetSpinner III incorporates sexual reproduction. Biomorphs, and NetSpinner II, reproduced only asexually. What can it mean to say that computer spiders reproduce sexually? You don't literally see spiders copulating on the screen, though no doubt that could be managed, complete with the occasional cannibalistic climax. What the program does is arrange the genetic liaisons of sexual reproduction, the recombining of half of one parent's genes with half of the other parent's genes.

  Here's how it works. In any one generation there is a population, or ‘deme’, of half a dozen spiders, each of whom builds a web. The shape {62}

  Figure 2.12 Overnight evolution of web by NetSpinner depicted every five generations.

  of the web is governed by a ‘chromosome’ or string of genes. Each gene works by influencing a specific web-building ‘rule’, as we saw above. The webs are then bombarded with ‘flies’. The ‘goodness’ of the web is calculated in the same way as before, as a function of the number of ‘flies’ caught minus a function of the ‘silk’ used. A fixed proportion of the population of spiders dies in every generation, and it is the ones with the least efficient webs that die. The remaining spiders mate with one another at random to produce a new generation of spiders. ‘Mating’ means that the chromosomes of the two spiders �
��line up’ and exchange a portion of their length. This sounds bizarre and contrived until you remember that it is exactly what real chromosomes, of ourselves as well as spiders, actually do in sexual reproduction. {63}

  The process continues and the population evolves, generation after generation, but with one further refinement. There is not just one deme of six spiders but (say) three semi-separate demes (Figure 2.13). Each of the three demes evolves in isolation except that, from time to time, an individual ‘migrates’ to another deme, carrying its genes with it. We'll return to the theory behind this in Chapter 4. For the moment we can briefly say that all three demes evolve towards improved webs: webs that are better at catching flies economically. Some demes may run up evolutionary blind alleys. Migrating spider genes can be thought of as an injection of fresh ‘ideas’ from another population. It is almost as though a successful sub-population sends out genes that ‘suggest’ to a less successful population a better way to solve the problem of building a web.

  In generation one, in all three demes, there is a wide variety of web shapes, most of them not particularly efficient. As in the asexual example of Figure 2.12, what we observe, as the generations go by, is a gradual narrowing down of the variation towards a better and more efficient shape. But now sexual reproduction sees to it that ‘ideas’ are shared within demes, so the different members of each deme are rather similar to each other. On the other hand they are genetically cut off from the other demes, so there are noticeable differences between demes. At one point in generation eleven, the genes of two webs migrate across from Deme 3 to Deme 2, thereby ‘infecting’ Deme 2 with ‘ideas’ from Deme 3. By generation fifty — actually long before this in some cases — the webs have evolved to become good, stable, efficient fly-catching devices.