Zoology students have an easier life than medics, although it wasn’t always so. Peter Medawar, in 1965, quoted one of eight examination questions set in 1860 for students of comparative anatomy at University College London:
By what special structures are bats enabled to fly through the air? and how do the galeopitheci, the pteromys, the petaurus, and petauristae support themselves in that light element? Compare the structure of the wing of the bat with that of the bird, and with that of the extinct pterodactyl: and explain the structures by which the cobra expands its neck, and the saurian dragon flies through the atmosphere. By what structures do serpents spring from the ground, and fishes and cephalopods leap on deck from the waters? and how do flying-fishes support themselves in the air? Explain the origin, the nature, the mode of construction, and the uses of the fibrous parachutes of arachnidans and larvae, and the cocoons which envelop the young; and describe the skeletal elements which support, and the muscles which move the meoptera and the metaptera of insects. Describe the structure, the attachments, and the principal varieties of form of the legs of insects; and compare them with the hollow articulated limbs of nereides, and the tubular feet of lumbrici. How are the muscles disposed which move the solid setae of stylaria, the cutaneous investment of ascaris, the tubular peduncle of pentalasmis, the wheels of rotifera, the feet of asterias, the mantle of medusae, and the tubular tentacles of acinae? How do entozoa effect the migrations necessary to their development and metamorphoses? how do the fishes polypifera and porifera distribute their progeny over the ocean? and lastly, how do the microscopic indestructible protozoa spread from lake to lake over the globe? 1
Medawar quoted this preposterous exam question in evidence against the widespread view that science, as it advances, becomes less and less easy to master because there is more and more to learn. His characteristically provocative reply was that we actually have to learn less than our Victorian predecessors, because multitudinous raw facts have become subsumed under relatively few general principles, the greatest of which was bequeathed by Darwin.
Medawar had a point; but, not for the first time, this laughing cavalier of the mind was exaggerating. He would have to admit that most papers in Nature and Science today can be read only by specialists in their respective fields. Nevertheless, weaving facts together into a functional story is a powerful aid to memory, and it’s one that I started using early on in my lecturing career at Oxford and Berkeley, and especially as a tutor at Oxford. This is what I meant when I said that adaptationism can have pedagogical advantages. The particular way I use it as a teacher is to take a problem faced by an animal and pose it as an engineer might. Then I list various solutions that might occur to the engineer, naming the pros and cons of each. Finally I come on to the solution actually adopted by natural selection. This provides a narrative flow which grips and becomes an effortless guide to memory.
I put the technique through its paces in the second chapters of both The Blind Watchmaker (the example of bat sonar) and Climbing Mount Improbable (the example of spider webs) and I’ll reprise the examples here to illustrate it. First, the bats. The problem facing a bat is how to find its way around at night. Given that birds have got daytime aerial hunting sewn up, bats were driven to hunt by night. And that posed a problem. It’s dark. An engineer might think of various solutions, each with its own knock-on problems: emit your own light like some deep-sea fish; feel your way with long antennae like whip-scorpions; perfect an extreme sense of hearing like owls, so that the tiniest rustle betrays prey, an extreme sense of smell like moles or extreme sense of touch like star-nosed moles; or, finally, sonar: emit loud sounds and exploit the echoes. Of these engineering solutions, the one actually adopted by bats is sonar. In various ways bats time the echoes of their own ultrasonic shrieks, and calculate the position, and rate of change of relative position, of obstacles and prey.
But this raises further problems. Precise timing of the interval between a sound and its echo is improved if the sound is short. But the shorter and more staccato the sound, the harder it is to make it really loud; and it needs to be very loud because echoes are so faint. Could an engineer find a way to get the best of both worlds? One way is to make the sound not staccato. Have a longer cry but modulate its pitch: swoop down (or up) through an octave or so during the course of each shriek. Now the shriek is not short and it can therefore be loud. What is short is the time it spends at each pitch. When the echo bounces back, the brain ‘knows’ that high-pitched echoes are from early parts of the shriek, low-pitched echoes from late parts of the shriek. The Warden of my Oxford College at the time I was writing The Blind Watchmaker, the physicist Arthur Cooke, had worked on the top secret British radar project (then called RDF) in the Second World War, and he told me one evening at dinner that the same technique was used by radar engineers, under the name ‘chirp radar’. Another engineering solution is to exploit the Doppler shift (the reason the ambulance siren drops in pitch as it rushes past you). Some bats make good use of this when tracking a moving target such as an insect prey.
Move on to the next engineering problem. Echoes, to repeat the point, are necessarily much fainter than the originating sounds: in danger of being too faint to hear. Possible engineering solutions: make the cries exceedingly loud, and/or make the ears exceedingly sensitive. But those two solutions tread on each other’s toes. Extremely sensitive ears are in danger of being deafened by extremely loud cries. Early Second World War radar faced an analogous problem, and again Arthur Cooke told me at dinner that the engineers solved it by designing what they called ‘send–receive’ radar. And – would you believe? – the exactly equivalent solution is adopted by some bats. Temporarily switch your ears off just before you shriek, by tugging with a special-purpose muscle on the bones that transmit sound from the eardrum. Relax the muscle immediately after shrieking, so your ear is restored to peak sensitivity in time to hear the echo. This cycle: tug, shriek, relax muscle, listen for echoes, tug again . . . has to be repeated for every shriek; and amazingly the repeat rate can climb as high as 50 per second, faster than a machine gun, as the bat comes in for the kill, the final approach to an insect prey.
The pedagogic advantage of the ‘Darwinian engineer’ approach is that the facts become strung together in a memorable narrative rather than having to be learned piecemeal. Indeed, there’s a good chance that students will anticipate facts even before they are told them, which is good training for how to dream up fruitful hypotheses that might be worth testing in research.
For example, bats often fly about in the company of hundreds of other bats. How might they solve the problem of their echoes being inadvertently jammed by all the other bat cries and echoes? Here’s an idea that might occur to a student, thinking like a ‘Darwinian engineer’. Imagine taking a movie film, cutting it up into its separate frames, shuffling the separate frames in a hat and splicing them together again at random. Now the story would no longer make sense: indeed, there would be no ‘story’, no consecutive narrative. In the same way, to an individual bat, all the other bats’ echoes would sound like the equivalent of my movie film made of random frames: easily ignored because of their random unpredictability in relation to any ‘story so far’. Only the individual bat’s own echoes would form a coherent narrative, making sense when ‘spliced’ to their predecessors in the sequence of echoes. Experimental psychologists use the same kind of argument to solve the ‘cocktail party problem’: how do we manage to understand one conversation at a cocktail party when our ears are assailed by dozens of other conversations all around us?
I used the same ‘Darwinian engineer’ technique in chapter 2 of Climbing Mount Improbable, but this time using the example of spider webs instead of bat sonar. Once again, start with a problem: how might a spider extend the effective length of its prey-catching limbs? Again offer up various hypothetical solutions, culminating in the elegantly economical solution natural selection actually adopted: the silken web. And repeat the process for sub-problem
s and sub-sub-problems that subsequently open up. In a later chapter of the same book, called ‘The fortyfold path to enlightenment’, I followed the same formula when talking about the design of eyes. Here, I carried the ‘design engineer’ approach to what some might consider an absurd extreme, but I hope an instructive one. A lens is a simple device, but the computational problem it solves is actually surprisingly sophisticated. I chose to dramatize this by imagining a computer picking up rays of light and bending them through a precisely calculated angle in order to focus an image on a screen. That would be ridiculously complicated, yet the task is easily solved by a lens, a device so simple that – as I demonstrated in my Royal Institution Christmas Lectures – it can be approximated by a sagging, transparent plastic bag full of water, whose poorly focused image can be improved step by step up a smooth gradient on ‘Mount Improbable’. It’s a metaphor for how easy it can be in practice to evolve something apparently complicated in theory. Long touted since Darwin’s own time as his nemesis, the eye is easily evolved and indeed has evolved several dozen times independently, all around the byways of the animal kingdom.
The value of the ‘Darwinian engineer’ approach for explaining things hit me much earlier, when I was inspired, separately, by two Cambridge eye physiologists, W. A. H. Rushton and H. B. Barlow. I had met Rushton when I was still a schoolboy, for he had two sons at Oundle, one of whom was my exact contemporary. We played the clarinet together in the school orchestra and Google tells me (no surprise) that he later made his career as an academic musicologist. Presumably because the eminent Professor Rushton had two sons at the school, he agreed to give a talk to the sixth form biology group.
Rushton made an interesting distinction between analogue and digital signalling systems. In an analogue telephone the continuously varying pressure wave of the spoken sound is transduced into a parallel voltage wave in a wire, which is transduced back again into sound in the earpiece at the other end. The problem is that, if the wire is long, the electrical signal attenuates and has to be boosted by an amplifier. Boosting inevitably introduces random noise. That doesn’t matter if there are only a few boosting stations along the line. But given a sufficiently large number of boosting stations, the accumulated noise overwhelms the signal and the conversation becomes an unintelligible hiss. This is why nerves, at least long ones, cannot work like (analogue) telephone wires.
Nerves aren’t wires carrying electric currents, they are even less hi-fi, more like fizzing trails of gunpowder acting as a fuse, with the added complication of the ‘nodes of Ranvier’, which can be seen as discrete boosting stations. The upshot is that a nerve has the noisy equivalent of hundreds of boosting stations strung out along its length. How might an engineer solve the noise problem? By abandoning all hope of conveying information via the height (voltage) of the wave. Instead, turn the wave into a spike, whose height is fixed or anyway irrelevant. Convey information not by the height of the spike but by the chattering pattern of a sequence of variable spikes. For example, signal a loud sound with a rapid burst of spikes in quick succession; a quiet sound by few spikes, well separated in time.
So, that’s an interesting biological solution to an engineering problem. But, as with the bats and spiders, one solution leads on to the next problem, which needs another engineering solution. And this brings me to the second of my Cambridge influences, Horace Barlow. My first wife Marian and I met Horace (named after his grandfather Sir Horace Darwin, Charles’s son) when we were at Berkeley, California, and attended his lectures as a visiting professor on sensory physiology. These lectures were notable for the fact that Horace usually turned up at least half an hour late. It was worth the wait. An immensely clever man, he was also a fount of idiosyncratic amusement. You could tell when a joke was on its way some seconds before it arrived, just by watching his face. The Barlow paper that inspired us dated from about ten years before we heard his lectures (it was the reason we went out of our way to attend them), and it completely changed my approach to teaching about sensory systems. Indeed, we both became obsessed with the Barlow paper, and for a period it dominated many of our science conversations with each other. The very name ‘Horace Barlow’ became a kind of shorthand between us to refer to a whole thread of thought that we were sharing at the time. My lectures on behavioural physiology to Berkeley students at the time became dominated by the ‘Darwinian engineer’ approach.
You remember I said just now that nerves signal loud sounds not by the height of the spike but by the frequency or timing of the spikes (and the same goes for high temperature, bright light etc.). That’s true, but it raises a further engineering problem. If the frequency of nerve spikes is simply proportional to the intensity of the signal, the necessary information is indeed conveyed, but the process is wasteful – and wasteful in a profoundly interesting way. The wastage is remediable – by removing ‘redundancy’. What is redundancy?
The state of the world at any one moment is pretty much the same as it was in the previous moment: the world doesn’t change at random, capriciously. Like journalists reporting news, nerves reporting on the state of the world need to send a signal only when there is a change. Don’t say: ‘It’s loud it’s loud it’s loud it’s loud it’s loud it’s loud . . .’ Instead, say: ‘A loud sound has started. Assume no change until further notice.’ This is where ‘redundancy’ comes in as a technical term in information theory. Once you know the current state of the world, further reports of the same state are redundant. Redundancy is the inverse of information. Information is a mathematically precise measure of ‘surprise’. In the time domain, information means changes in the state of the world from one moment to the next, because only changes have surprise value. Redundancy in this context means ‘sameness’. The receiver of multiple messages doesn’t have to monitor all channels all the time: only those that signal a change. This could only fail to be helpful if the world changed randomly and capriciously all the time. Which, fortunately – well, obviously – it doesn’t.
Redundancy filtering was Barlow’s engineering solution to the problem of economical signalling in the time domain, and – sure enough – it is implemented by nervous systems in the form of sensory adaptation. Most sensory systems send a rapid burst of spikes every time they detect a change, after which the spike rate settles down to a low or even zero rate until there is another change.
There’s an analogous engineering problem in the spatial domain. If you think of an eye (or a digital camera) looking at a scene, most cells in the retina (or pixels in the camera) will be seeing the same thing as their neighbours in the retina (or camera). This is because the world’s scenes are not capriciously random, pepper and salt, but are typically made up of large patches of uniform colour like the sky or a whitewashed wall. Away from an edge, every pixel sees the same as its neighbours, and to report it is a waste of pixels. The economical way to convey the information is for the sender to report on edges and for the receiver (the brain in this case) to ‘fill in’ the swathes of uniform colour between the edges.
Barlow pointed out that this engineering problem, too, has its neat, redundancy-reducing solution in biology. It’s called lateral inhibition. Lateral inhibition is the equivalent of sensory adaptation, but in the spatial, not the temporal domain. Each cell in the array of ‘pixels’, in addition to sending nerve spikes to the brain, inhibits its immediate neighbours. Cells that are sitting in the middle of a patch of uniform colour are inhibited from all sides, and therefore send only few, if any, spikes to the brain. Cells that are sitting on the edge of a patch of colour receive inhibition from their neighbours on only one side. So the brain gets the majority of its spikes from edges: the redundancy problem is solved, or at least mitigated.
Barlow introduced his article – and it was this that especially grabbed Marian’s and my imagination – with a mind-stretching thought experiment. Imagine that for every pattern the brain might ever want to recognize – every tree, every predator, every prey, every face, every letter of t
he alphabet, of the Greek alphabet – there was one nerve cell, hooked up to the retina in such a way that it fired when its ‘own’ shape fell on the retina. Each of these brain cells is wired up to a ‘keyhole’ combination of pixels so that it fires only when the correct ‘keyhole’ shape is seen. It also has to be wired up negatively to the ‘anti-keyhole’ (all the pixels other than the keyhole) otherwise it would fire when it saw a blank field of light covering the whole keyhole. That sounds fine, but on second thoughts it can’t be true. Remember that all the shapes needing to be recognized by these overlapping keyholes may be presented in thousands of different orientations and from any distance. The number of overlapping keyholes (with the rest of the retina in each case being an anti-keyhole) would be so prodigiously large that their corresponding brain cells would have to be more numerous than all the atoms in the world. Fred Attneave, an American psychologist who independently thought up the same idea as Barlow, estimated that the volume of the brain would have to be measured in cubic light years!
Brief Candle in the Dark Page 34