An Appetite for Wonder Read online

  We recognized eight distinct grooming acts, which we presumed would show up as FAPs if we had the time to do a frame-by-frame analysis like the one we did with the chicks drinking: FR (rub front feet together), TG (rub tongue between front feet), HD (wipe head with front feet), FM (rub one or other middle foot between front feet), BM (rub one or other middle foot between back feet), BF (rub back feet together), AB (wipe abdomen with back feet), WG (wipe wings with back feet). Using a Dawkins Organ, we recorded the sequences of these eight grooming acts, plus MV (move away) and NO (stand still, doing nothing).

  The graph shows the probability, given that the fly is now doing HD, that it will do FR next (‘lag’ = 1, probability very high), next but one (very low probability), next but two (high probability), next but three (low probability) etc. You can see that there is a pronounced tendency to alternate, and also that there is a general die-away (as you’d expect) in predictability as we look to the more distant future – longer and longer ‘lags’.

  That picture was for the particular case of FR following HD. We plotted the same kind of graph for all the possible transitions, and put the graphs into a table.

  You can see that many of the transitions follow the same zig-zag pattern, although some are exactly out of phase with each other. The bottom row (UNC) shows the uncertainty attached to predictions of the future, following each behaviour, calculated using the Shannon Information Index, in the same way as for the chick drinking study.

  We also tried the experiment of using the human ear to identify patterns in animal behaviour. For this, we used a Dawkins Organ rendition of fly grooming behaviour, but eliminated the true intervals between musical notes. I told the computer to reduce all the intervals to a single standard short interval and then we simply listened to the ‘music’. It sounded rather like ‘modern’ (as opposed to ‘traditional’) jazz. Also rather like the ‘singing’ Elliott computer of my juvenile insomniac dalliance – I suppose the comparison might be interesting. I thought the human ear might be a promising piece of apparatus to use in detecting patterns in animal behaviour, but didn’t follow up the method; I merely report it here as an interesting curiosity. If the World Wide Web had existed in those days, I would surely have uploaded the fly washing music and you could now dance to it. As things are, I’m afraid those Dipteran Melodies are gone for ever, like the Lost Chord.

  I cannot claim that our fly study, or the other studies of decision-making that preceded it, really tell us much about how animal brains work. I see them more as explorations of methods: not just methods of doing research on animal behaviour, but methods of thinking. Marian and I did a lot more work on the flies, but it’s all published and I won’t write any more about it here. It did, however, feed into my next big writing project: a long theoretical paper on ‘Hierarchical organisation as a candidate principle for ethology’. This is the subject of a later section.

  Meanwhile, in 1973, Niko Tinbergen won the Nobel Prize in Physiology or Medicine (jointly with Konrad Lorenz, his co-founder of ethology, and Karl von Frisch, the discoverer of the legendary bee dance). Just one year later, in 1974, Niko reached Oxford’s mandatory retirement age of sixty-seven, and the university agreed to appoint a successor as Reader in Animal Behaviour. ‘Reader’ was a rather prestigious rank at Oxford, now, I think, fallen into desuetude in a move to bring the title of ‘professor’ into line with American custom by sprinkling it about more liberally – the rather unkindly dubbed ‘Mickey Mouse professors’. I was very content where I was as lecturer, and had no ambition to apply for the job.

  Most people thought of Mike Cullen as Niko’s natural successor. Perhaps for that very reason, in order to make a clean break, the majority of the appointment committee went for David McFarland. As Hans Kruuk wrote in his biography of Tinbergen, ‘one could hardly have found anyone more unlike Niko’. Though controversial in many quarters, David’s appointment was in some ways an inspired one, at least if you take the view that a new appointment is an opportunity for a new departure. His science was highly theoretical, indeed mathematical. He brought to it the intuitions of a mathematician, and he surrounded himself with trained mathematicians and engineers who could do the algebra. The talk in the coffee room switched from gulls and sticklebacks in the field to feedback control systems and computer simulations.

  Perhaps it was a microcosm of the way biology was changing. I was young and not yet set in my ways. ‘If you can’t beat them, join them’ was my attitude. So I set to work to learn control theory from the engineers and mathematicians who now surrounded me. And what better way to learn it than hands-on? I again indulged my passion – or vice – for computer programming, and wrote a program for a digital computer (‘my’ PDP-8), enabling it to behave like an analogue computer. To this end, I invented yet another computer language, which I called SysGen.

  Unlike the propositions in a conventional computer language like Fortran, which are executed sequentially, SysGen statements were executed ‘simultaneously’ – not really simultaneously, of course, because a digital computer does everything sequentially at bottom; but they could be written in any order. My task in writing the SysGen Interpreter program was to persuade the digital computer to behave as if the operations were simultaneous: a virtual analogue computer. As with an analogue computer, results were displayed as a set of graphs on an oscilloscope screen.

  I’m not sure how useful SysGen was in practice, but inventing the language, and writing the Interpreter program for it, certainly helped me to understand not just control theory but also the integral calculus. It gave me a much better idea of what it means to integrate. I was mindful of my maternal grandfather’s recommendation of Calculus Made Easy, by his old mentor Silvanus Thompson (who, as quoted earlier, was fond of saying, ‘What one fool can do, another can’). Thompson introduces his explanation of integration with another phrase that has stuck in my brain: ‘So we had best lose no time in learning how to integrate.’ I had only half understood integration in Ernie Dow’s lessons, and SysGen gave me the sort of hands-on experience that assists comprehension.

  Similar in intention, but much easier and less time-consuming, was my attempt to understand Chomsky-style linguistics by the hands-on method. I wrote a computer program to generate random sentences, which might not have been very meaningful but were always scrupulously grammatical. This is easy – and that very fact is instructive – given that your programming language allows procedures (subroutines) to call themselves recursively. This was true of Algol-60, the programming language that I favoured at the time under the influence of Roger Abbott, who had brilliantly succeeded in writing an Algol compiler for the PDP-8. Algol subroutines could call themselves, unlike the contemporary version of that old workhorse of scientific programmers, IBM’s Fortran language. Mention of Fortran reminds me of a nice in-joke told by Terry Winograd, pioneer of artificial intelligence. Some time in the 1970s I attended a fascinating conference in Cambridge on the state of the art of artificial intelligence programming, and Winograd was the star lecturer. At one point he gave vent to a wonderful piece of sarcasm: ‘Now, you may be one of those who says, “Fortran was good enough for my grandfather, it’s good enough for me.”‘

  Given that your programming language allows procedures to call themselves recursively, writing programs to deliver correct grammar is remarkably – elegantly – easy. I wrote a program that had procedures with names like NounPhrase, AdjectivalPhrase, PrepositionalClause, RelativeClause etc., all of which could call any other procedure, including themselves, and it generated random sentences like this one:

  (The adjective noun (of the adjective noun (which adverbly adverbly verbed (in noun (of the noun (which verbed))))) adverbly verbed)

  Parse it carefully (as I have done here using brackets, although the computer didn’t generate them but left them implicit) and you’ll see that it is grammatically correct although not exactly dripping with information. It makes syntactic but not semantic sense. The computer could easily inject sema
ntics (if not sense) by replacing ‘noun’, ‘adjective’ etc. with particular, randomly chosen instances of nouns and adjectives. Thus you could inject a vocabulary from a chosen domain, such as pornography or ornithology. Or you could inject the vocabulary of francophoney metatwaddle – as Andrew Bulhak was later to do when he wrote his hilarious ‘Postmodernism Generator’, which I quoted in A Devil’s Chaplain:

  If one examines capitalist theory, one is faced with a choice: either reject neotextual materialism or conclude that society has objective value. If dialectic desituationism holds, we have to choose between Habermasian discourse and the subtextual paradigm of context. It could be said that the subject is contextualised into a textual nationalism that includes truth as a reality. In a sense, the premise of the subtextual paradigm of context states that reality comes from the collective unconscious.

  This randomly generated garbage makes about as much sense as many a journal devoted to the metatwaddle of ‘literary theory’, and Bulhak’s program is capable of generating a literally indefinite quantity of it.

  Two more programming projects date from around this time in my life, both of which also, as it turned out, served to hone my skills for the future rather than deliver results of more immediately practical usefulness. The first of these was a program to translate from one computer language to another: specifically, from BASIC to Algol-60. It worked well for those two languages, and would have worked, with minor detailed changes, to translate from any computer language of that general algorithmic type to any other. My second project of this time was STRIDUL-8: a program to make the PDP-8 computer sing like a cricket.

  I had been inspired to work on crickets by my Berkeley friend, the neurobiologist David Bentley; and my entomologically inclined graduate student Ted Burk (now a professor in Nebraska) was keen to do his doctoral thesis on them. David kindly sent me some eggs of the Pacific field cricket Teleogryllus oceanicus. They hatched in Oxford and soon we had a thriving colony, which Ted looked after, feeding them on lettuce. While Ted productively pursued his own research on the crickets’ behaviour, I conceived a parallel project using computer-generated courtship song. That research project was never completed, but I did complete the writing of STRIDUL-8, and it worked pretty well.

  My testing apparatus was a seesaw, made out of balsa wood so it was very light – as it had to be for crickets. It was really just a long balsa-wood passage, closed with netting at each end and over its top, and resting on a hinged fulcrum in the middle. Only one cricket was placed in the passage at a time, and it was free to walk from one end to the other as often as it liked. Whichever end it approached tipped down, as a seesaw should, and this fact was recorded by a micro-switch, which importantly also reversed the location of the sound. There were two little loudspeakers, one at each end of the seesaw. Cricket song was played through whichever of the two loudspeakers was at the opposite end of the seesaw from the cricket. So, imagine you are a female cricket, sitting somewhere towards the west end of the corridor. Song is playing from the east. You like what you hear, so you start to walk east. When you near the east end, your weight tilts the seesaw to the east, tripping the micro-switch and thereby informing the computer, which now switches the song to the west end loudspeaker. So you turn and walk west, and the whole process happens in reverse. Preferred songs therefore generate a large number of seesaw reversals and these were automatically counted by the computer. Whether the cricket thought she was pursuing an ever-receding coy male, or whether she thought the male was jumping capriciously over her head, or whether she thought at all, is impossible to say. Unpreferred songs would generate only a small number of rockings of the seesaw. Indeed, if a song was positively aversive, the cricket would stay at the opposite end of the passage and not generate any seesaw tips at all.

  That, then, was my apparatus for measuring how much crickets like songs of different types. Play Song A for five minutes of the alternating seesawing regime, then do the same with Song B, and so on for many trials, properly randomized etc. Count seesaw tippings as a measure of how much the cricket liked each song. The point about computer-generated songs, as opposed to real ones, was to try to dissect out, in classic Tinbergen fashion, what it is about their own species song that crickets like. The computer would vary how it sang, in systematic ways. The initial plan was to start with a simulation of the species’ natural song and then change it – drop bits out, enhance other bits, vary the interval between chirps and so on. Later – it was a somewhat wild hope – I envisaged the computer being programmed instead to start with random song and ‘learn’ – or we could equivalently say ‘evolve’ – choosing ‘mutations’ step by step, until it progressively homed in on a synthetic preferred song. If the preferred song had turned out to be the natural song of Teleogryllus oceanicus, wouldn’t that have been sensational? And then if I had done the same thing with Teleogryllus commodus and the computer had homed in on its rather different song. What bliss that would have been for the researcher!

  In programming the computer to sing, I wanted to make it as versatile as possible. Versatility is what computers are good at. As with the analogue computer simulation, as with the language translation program, I wanted to program the general case. And this is where STRIDUL-8 came in: its language allowed you to specify any combination of pulses and intervals, and therefore any cricket song in the world. STRIDUL-8 had an intuitively reasonable bracket notation, which enabled the user to insert repeats, and repeats embedded within repeats, in a manner reminiscent of the grammar of language.

  STRIDUL-8 worked well. Its simulations of cricket song sounded like real crickets to human ears, and it was easy to program the computer to sing like any cricket species in the world. However, when I demonstrated the system to Dr Henry Bennet-Clark, world authority on the acoustics of insect sounds, newly arrived from Edinburgh to take up a position in Oxford, he made a face and said ‘Eeugh!’ STRIDUL-8 could only specify the pattern in time of pulses of sound, each pulse corresponding to one stroke of the wings against each other. I had made no attempt to simulate the actual wave form produced by each wing stroke, and this is what Henry objected to. He was right. STRIDUL-8, as it stood, could not have done justice to the European tree crickets of whose song Henry once wrote that if moonlight could be heard that is how it would sound. Temporarily discouraged, I put my whole cricket song project on the back burner, while I attended to other pressing tasks, notably a challenging invitation from Cambridge. And unfortunately I never returned to it: my cricketing days were over. I’ve often regretted it. I think most scientists have sad loose ends, projects started, never finished. If I ever had vague intentions to return to the crickets, they were thwarted by Moore’s Law: computers change so rapidly that if you leave a loose end of research untied for as long as I left mine, you find that the extant computers have all become newer, sexier models, and they have forgotten how to run your earlier programs. To find a computer that would run STRIDUL-8 today I’d have to go to a museum.

  Photographic Insert 3

  This picture of me with my parents, taken at a family wedding (my sister Sarah was a bridesmaid, so not with us at this point), unfortunately does not show the bright red of the cap I wore as a Chafyn Grove schoolboy. In my first term at Oundle I don’t think I was as happy as I looked for the camera. One of the best things about that school was Ioan Thomas, caught here encouraging an appetite for wonder at the living world.

  Life at Over Norton: the battered Land Rover with which we crashed through rough country; Wessex Saddlebacks landscaping the equally rough country that was then the garden of our cottage, c. 1951; my inventive father standing proudly by his patent pasteurizer; haymaking with the little grey Fergie.

  In the summer holidays I earned my keep bale-sledging. Bottom: Following in father’s footsteps: moving some family heirloom or other.

  Clockwise from bottom left: Peter Medawar before the stroke that changed his life; Niko in his element, at Ravenglass painting dummy eggs; ‘The deeply intelligen
t eyes, understanding what you meant even before the words came out . . . sceptical, quizzical tilt of the eyebrows, under the untidy hair.’ Mike Cullen, sadly missed mentor to so many; What to peck? Chicks that had never seen overhead light; George Barlow, my Berkeley friend and guide, came to Oxford later on sabbatical, and we went punting together on the Cherwell. (That’s not John Lennon, it’s Tim Halliday the newt expert.)

  Above: Hunting the Surrey puma; intrepid explorer scours the landscape for wild beasts. Bottom: Wild beasts or frightened boys? California National Guard raggedly confront the Peace People in Berkeley.