An Appetite for Wonder Page 16
But back to the controversy over nurture or nature. Male sedge warblers (to take just one example) have a complex and elaborate song, and they can perform it even when reared in isolation, never having heard another sedge warbler. The Lorenz–Tinbergen school would therefore have said it must be ‘innate’. But Lehrman emphasized the complexity of developmental processes and always wondered whether learning was involved in some less obvious way. For Lehrman, it wasn’t good enough to say that the young animal had been reared under deprived conditions. For him, the question was: ‘Deprived of what?’
Since Lehrman’s critique was published, ethologists have indeed discovered that many young songbirds, including sedge warblers, even when reared in isolation, learn to sing their correct species song by listening to their own fumbling efforts, repeating the good fumbles and discarding the bad. So that looks like nurture after all. But in that case, Lorenz and Tinbergen might reply, how do the young birds know which of their fumbles are good and which bad? Surely that ‘knowledge’ – a template for what their species song ought to sound like – has to be innate? All learning does is transfer the song pattern from the sensory part of the brain (the built-in template) to the motor side (the actual skill of singing the song).
Other species, by the way, such as the American white-crowned sparrow, also teach themselves to sing in this ‘fumbling’ way, but do need to have heard the species song earlier in life. It is as though the young bird takes a ‘tape recording’ before it can sing, and uses it as a template for teaching itself how to sing. And there are intermediates between the ‘learned tape recording’ and the ‘innate tape recording’ as templates for later learning.
This was the philosophical minefield into which Niko Tinbergen released me in 1962. I think he wanted to back away from his perceived association with Lorenz and saw me as a bridge towards the Lehrman camp. My experimental subject was to be not singing birds but baby chicks pecking. I did a series of experiments of which I’ll mention only one here.
Baby chicks straight out of the egg start pecking at small objects, presumably looking for food. But how do they know what to peck at? How do they know what’s good for them? One extreme would be for nature to endow them, before they have any experience at all, with a template picture of a grain of wheat in the brain. That’s unrealistic, especially in an omnivore. Do wheat grains and mealworms and barleycorns and millet seeds and beetle larvae have anything in common, as opposed to boring and inedible marks and stains? Yes, they do. For one thing, they are solid.
How do you recognize something as solid? One way is by surface shading. Look at the photographs of moon craters. They are the same photograph, but one is rotated through 180° relative to the other. My guess is that on the left you will see hollow craters and on the right solid flat-topped hills – and the other way around if you swivel the book upside down. The illusion has been known for a long time. It depends upon a preconception about where the light is coming from: in effect, a preconception about the location of the sun. Solid objects tend to be brighter on the side nearest the sun, which will usually be approximately above. A photograph of a solid object can therefore look hollow if you turn it upside-down, and vice versa.
The sun is seldom directly overhead, but the general direction of its light is more likely to be down than up. Therefore any predator seeking solid objects as possible prey can use surface shading cues based upon that assumption. And on the other side of the predator–prey arms race, natural selection might well favour prey animals that manage to disguise their solidity by ‘countershading’. Many species of fish are darker on top, lighter below, which tends to neutralize the natural tendency for sunlight to come from above, and thereby makes the fish look flatter. One fish, the ‘upside-down catfish’, is a genuine ‘exception that proves the rule’. It habitually swims upside-down and, sure enough, it is reverse countershaded: darker on its belly than its back.
A Dutch student of Tinbergen called Leen De Ruiter did some neat experiments on reverse countershaded caterpillars, who habitually rest upside-down. The upper picture shows Cerura vinula in its normal position. It looks flat and inconspicuous. The lower picture shows what it looked like when De Ruiter turned its twig upside-down: much more conspicuous to my eyes and – more significantly – to the eyes of jays, when De Ruiter used them as experimental predators.
But none of this says anything about whether – in jays or humans – the knowledge that the sun is normally overhead is innate or learned. The solid shading illusion seemed to me to provide a good opportunity to test the question, using baby chicks in deprivation experiments.
First, did chicks see the illusion? Apparently, yes. I photographed a half ping-pong ball lit asymmetrically, and printed the image to be about the size of a tempting grain or seed. When I viewed the photograph with the illuminated side at the top, the hemisphere looked solid. When I inverted the photograph, it didn’t. When chicks were offered a choice between the two orientations, they strongly chose to peck at the apparently solid picture, the one lit from above. This suggested that chicks possess the same ‘preconception’ we do, that the sun is normally overhead.
So far, so good; but these chicks, though young, were not completely naive. They were three days old and had been feeding in normal overhead light during this time. They might have had time to learn the appearance of solid objects illuminated from above.
To test this I did a crucial experiment. I reared chicks with light coming from below and tested them under the same conditions. So, at the time of testing, they had never had any experience of overhead light. As far as they were concerned, the world into which they had hatched was a world with a sun underneath them. Every solid object they had ever seen, whether food objects or parts of other chicks, was lighter underneath than on top. I expected that, when tested with the two ping-pong ball photographs, they would prefer to peck at the one illuminated from below.
But I was delighted to be proved wrong. The chicks overwhelmingly pecked at the photograph illuminated from above. If you accept my interpretation, this means that the chicks are genetically equipped by ancestral natural selection with something equivalent to ‘advance information’: in the world in which they are to live, the sun will normally shine from above. My experiment had pinpointed a true example of innate information which is not reversed by a positive attempt to teach the contrary.
I can’t think of any group of humans who habitually live with underfloor lighting. If they exist, it would be interesting to test them in the same way I tested my chicks. I thought about offering an intuitive guess as to what the result would be, but I honestly prefer not to place a bet. Wouldn’t it be fascinating if we too saw the illusion innately? Having been surprised by the chicks, I’d be only slightly more surprised if humans did the same. We may never know, but there could be ways to do the experiment on very young babies. They don’t peck, but they do fixate their eyes on objects that interest them, and you can measure that. Could a developmental psychologist offer babies a version of my ping-pong ball experiment and measure the time they spend staring at each of the two photographs? Would it be considered unethical to use underfloor lighting for a baby’s room for the first few days of life? I can’t see why, but who knows what the verdict of a modern ‘ethical committee’ might be?
In the end, my work on ‘nature or nurture’ constituted only a small part of my doctoral research,45 and it was relegated to an appendix in my thesis. The main part of my thesis had little in common with it, except that it also involved pecking in chicks. And it was also an attempt to illustrate a point of philosophical interest – although taken from a different part of philosophy. It became possible through an improved technique for recording pecks.
Bevington Road, and especially its satellite research stations in the great gull colonies of the north, ran a system of ‘slaves’ – young unpaid volunteers who wanted a brief taste of the Tinbergen experience before going to university. Among them were Fritz Vollrath (who later returned to O
xford to head a flourishing group working on spider behaviour, and remains a close friend) and (also from Germany) Jan Adam. Jan and I found an immediate affinity, and we worked together. He had remarkable workshop skills – combining the very different virtues of my father and Major Campbell – and, fortunately, these were the days before health and safety regulations interfered to protect us from ourselves and sap our initiative. Jan and I had the freedom of the departmental workshops: lathes, milling machines, bandsaws and all. We (that is to say Jan, with me as willing apprentice – the younger brother syndrome again, I suppose) built an apparatus to automate the counting of chick pecks, using delicately hinged little pecking keys, elegantly made from scratch by Jan, with sensitive micro-switches. Previously, when working on the surface shading illusion, I had counted pecks by hand. Suddenly, I was in a position to collect huge quantities of data automatically. And this opened the door to a completely different kind of research, motivated by a different philosophy, Karl Popper’s philosophy of science, which I learned from Peter Medawar.
As I have already explained, I had come to know of Medawar early on through my father, who was a schoolfriend of his. As British biology’s star intellectual, Medawar came to give a visiting lecture at his old Oxford department when I was an undergraduate there, and I remember the excited buzz in the standing-room-only audience waiting for this tall, handsome, gracious figure to arrive (‘This lecturer has never been thought ungracious in his life,’ as a later critic said of him). The lecture prompted me to read Medawar’s essays, later anthologized in The Art of the Soluble and Pluto’s Republic,46 and it was from them that I learned about Karl Popper.
I became intrigued by Popper’s vision of science as a two-stage process: first the creative – almost artistic – dreaming up of a hypothesis or ‘model’, followed by attempts to falsify predictions deduced from it. I wanted to do a textbook Popperian study: dream up a hypothesis that might or might not be true, deduce precise mathematical predictions from it, and then try to falsify those predictions in the lab. It was important to me that the predictions should be mathematically precise. It was not enough to predict that a measurement X should be larger than Y. I wanted a model that would predict the exact value of X. And this kind of exact prediction demanded large quantities of data. Jan’s apparatus for counting massive numbers of pecks gave me the opportunity. Instead of pecking at photographs of ping-pong balls, my birds pecked at little coloured hemispheres mounted on Jan’s hinged windows, which triggered micro-switches. They preferred blue over red over green, but that wasn’t what interested me. I wanted to know what governed each individual pecking decision, whichever colour it was directed towards. And this, of course, was only a specimen of a more general question about how decisions are made at any time by any animal.
Medawar elsewhere made the point that scientific research doesn’t develop in the same orderly sequence as the final published ‘story’. Real life is messier than that. In my own case it was so messy that I can’t remember what gave me the idea for my ‘Popperian’ experiments. I remember only the finished story which, as Medawar would have expected, gives an implausibly tidy impression.
The finished story is that I dreamed up an imaginary ‘model’ of what might be going on inside a chick’s head when it makes a decision between alternative targets, did some algebra to deduce precise, quantitative predictions from the model, then tested them in the lab. The model itself was a ‘drive/threshold’ model. I postulated that there was a variable (‘drive’ to peck) in the bird’s head, whose graph was continuously wiggling up and down as the drive strengthened or weakened (perhaps at random; it didn’t matter). Every time the drive happened to rise above the threshold for a colour, the bird was capable of pecking at that colour (something else, for which I developed and tested another model which I’ll mention later, determined the timing of pecks). Blue, being a preferred colour, had a lower threshold than green. But if the drive rose above green’s threshold, it automatically had to be above blue’s threshold as well. What would the bird do then? I postulated that it would be indifferent between the two colours, since both thresholds were exceeded: it would ‘toss a coin’ to decide between them. So the model predicted that a bird’s choices over a long period would consist of periods of pecking at only the preferred colour, interspersed with periods of choosing at random between the two. There would be no periods of consistently positive choice of the less preferred colour.
I didn’t at first look directly at sequences of pecks. That was to come later, after I moved to California. I think the reason I didn’t test sequences at first was as unambitious as the fact that Jan’s apparatus could count pecks but not record the exact order in which they happened; and Jan himself had by now gone back to Germany, so wasn’t there to modify his apparatus. I think, too, that I was simply seduced by the Popperian elegance of deducing a mathematical formula which would predict some measured quantity from some other measured quantities.
The chicks happened to prefer blue over red over green. I imagined an experiment in which I would present Blue versus Green, Blue versus Red, and Red versus Green, counting the proportion P of pecks to the preferred colour in each case. This would give me three numbers (PBestWorst, PBestMedium, PMediumWorst). It’s only to be expected that PBestWorst would be larger than either of the other two. But could the model predict precisely how much bigger? Could I deduce from the model a formula to predict exactly what PBestWorst should be, if I fed in PBestMedium and PMediumWorst? Yes, that is exactly what I succeeded in doing. I defined algebraic symbols to stand for the time spent by the drive between various thresholds, did some school algebra (simultaneous equations as taught by Ernie Dow) to eliminate the unknown variables, and was pretty pleased when, at the end of pages of algebra, a simple, precise, quantitative prediction dropped out. The drive/threshold model predicts that
PBestWorst = 2(PBestMedium + PMediumWorst – PBestMedium • PMediumWorst) – 1.
I called this Prediction 1. The thing that interested me about Prediction 1 was that it is quantitatively precise.
So now to test it. Would the chicks obey the prediction? Yes: to my delight and amazement, in seven out of eight repeats of the experiment they did, very closely. The eighth experiment was way off, so much so that, to my acute embarrassment, when one of my papers was published in the journal Animal Behaviour,47 the printer removed the relevant point from the graph, thinking it must be a blemish on the block! Fortunately the offending datum was clearly present in the accompanying table, otherwise I might have been accused of dishonesty. I did another set of experiments on chicks, involving not pecking but walking into chambers illuminated by light of different colours. The graph shown here combines the two sets of experiments and plots the observed against predicted percentages for all 11 chick experiments.
If the model’s predictions were perfect, the points should all lie exactly along the diagonal line. With the exception of Experiment 8, as already mentioned, the Drive Threshold Model does a far better job than we ever dare to expect in animal behaviour experiments (physicists expect higher precision because there is usually less statistical error in their measurements).
I also used all the same data to test the predictions of an alternative model, one which simply assumed that each colour has a ‘value’ for the animal, and that the animal allocates its choices in proportion to the colour’s value. The two models gave similar predictions, so that if one is right the other one can’t help being nearly right. But the Drive Threshold Model was consistently more accurate in predicting the observed result. The ‘colour value’ model consistently overestimated PBestWorst. The ‘colour value’ model was falsified. The Drive Threshold Model triumphantly survived the attempt to falsify it, and indeed its predictions were (with the exception of the one experiment) remarkably accurate.
Does this good performance of the model really mean that there is something equivalent to a fluctuating ‘drive’ in the chick’s head, crossing ‘thresholds’, and that
something equivalent to tossing a coin happens when the drive is above more than one threshold? Well, Popper would say that the model survived a strong attempt to disprove it; but that says nothing about what the ‘drive’ and the ‘thresholds’ actually correspond to in the language of nerves and synapses. It is at least an interesting thought that you can make inferences about what is going on inside the head without cutting it open.
The same method of imagining a model and testing its predictions has proved enormously productive in many branches of science. In genetics, for example, you can infer the existence of chromosomes as one-dimensional linear sequences of genetic code without ever looking down a microscope, using only the data from breeding experiments. You can even work out the order in which the genes are arrayed along the chromosomes, and how far apart genes are from each other, entirely by imagining what might be the case and testing predictions in breeding experiments. As with my experiments on solidity and shading, I think of my Drive Threshold Model as an illustrative example of the kind of thing that can be done with a model, rather than as a conclusive discovery of what is really going on inside a chick’s head.
I elaborated the Drive Threshold Model in various directions (that’s also something that is supposed to happen according to Popperian philosophy) and tested nine predictions in all, with good success. One of these elaborations of the model, as I mentioned above, was an attempt to explain the exact timings of pecks (‘samplings’ of the position of the ‘drive’ relative to the ‘thresholds’). The predictions of this model stood up well against data on black-headed gull chicks from my colleague and close friend Dr Monica Impekoven, a visitor to Bevington Road from Switzerland. We published a joint paper on this work.48