Chinese junk and Chinese Whispers
I wrote the foreword to Sue Blackmore’s The Meme Machine, and used the occasion to offer an answer to one of the main criticisms of meme theory. The criticism was that memes, unlike genes, don’t have high-fidelity replication. As the generations go by, the critic complains, the information will degenerate, a condition fatal to evolution. In DNA, the sequence ATGCGATTC will be precisely copied (or, if miscopied, will contain a definite, discretely identifiable error). But when a meme, such as a nursery rhyme, is copied – say from father to child – the replication is imprecise. The child’s voice is higher, her vowels are not pronounced in exactly the same way, and so on. Therefore memes are not like genes and can’t be the basis for evolution because the replication fidelity is not high enough.
The criticism is superficially plausible but demonstrably wrong. I answered it with a series of thought experiments, versions of the childhood game of ‘Chinese Whispers’ (‘Telephone’ to American children). Imagine twenty children in a line. I whisper a sentence to the first child. It might be ‘Down in a deep dark dell sat an old cow munching a beanstalk.’ The first child whispers what she has heard to the second child, and so on down the line. The twentieth child then sings out the ‘evolved’ version of the sentence. It might have become distorted, provoking hilarity. But, if the phrase is short, and especially if it means something in the children’s own language, there’s a good chance that it will survive, intact, to the end of the line. Now, if we take a particular instance of the game where the phrase emerges correctly at the far end of the line, I make the following point. It doesn’t matter that each child’s vocalization of the phrase is not a precise copy of the preceding child’s. One child might have an Irish accent, the next a Scottish accent, the next a Yorkshire accent and so on. Because the phrase is meaningful to them in their own shared language, each child will ‘normalize’ it. Scottish vowels are different from Yorkshire vowels, but the difference doesn’t detract from the content. An Australian child hears the words, recognizes them as drawn from the shared lexicon of the English language, and correctly renders them in such a way that the next child, with an American accent, can understand the phrase and pass it on.
There might be a ‘mutation’ somewhere along the way. For example, say child number 14 changes ‘dark’ to ‘dank’, and the mutant form, ‘dank’, is then copied to the end of the line. That would be interesting in itself. But let’s take the case where no such mutation occurs. Now suppose that an experimenter with a tape recorder eavesdrops on each whispering child down the line. She separates the nineteen recordings as separate little tapes and scrambles them in a hat. Independent observers are then given the tapes and asked to rank them in order of resemblance to the original message as spoken to the first child. You know what the result would be. Assuming no mutations of the dark/dank kind, there will be no tendency for tapes that come from early in the sequence to be better than those that come late. There is no tendency for degeneration as the copying process proceeds. That alone should be enough to confound the criticism I mentioned.
But let’s take the thought experiment one stage further. Suppose the message is in a language the children do not know. Suppose it is Arma virumque cano, Troiae qui primus ab oris. Again, we know what will happen, without needing to do the experiment. The children, knowing no Latin, can only imitate phonetically. The final sounds emerging from the twentieth child will have lost almost all resemblance to the original sound spoken to the first child. Moreover, if we do the experiment of shuffling the tapes in a hat, once more we know what the result will be. Independent observers will certainly be able to rank the tapes in order of resemblance to the original message: there will be a steady deterioration in resemblance to the original, from the first tape to the last.
You could do parallel experiments not with words but with skills: for example, a skill in carpentry (simulating the transmission from master carpenter to apprentice and so on through twenty ‘generations’). Here, the equivalent ‘normalization’, playing a role equivalent to the shared language, will, I suspect, be an appreciation of what the skill is designed to achieve. If, for instance, the master is teaching the apprentice how to hammer in a nail, the apprentice will probably not imitate the precise number and strength of hammer blows. Rather, he will imitate the goal the master is trying to achieve, namely ‘head of nail flush with wood’ and he’ll go on hammering until the goal is achieved. The goal is what he will imitate, and that is what will be passed on to the next ‘apprentice’.
In my Blackmore foreword, I used the example of another manual skill: folding paper to make an origami Chinese junk. Its plausibility as a meme is illustrated by the fact that, when I introduced it to my boarding school, it spread like a measles epidemic. Even more interesting, I learned the skill from my father who had, in his turn, learned it when it spread as an epidemic through the very same school a quarter-century earlier.
When imitating an origami skill, each child imitates, not the exact hand movements of the preceding child but a ‘normalized’ version consisting of a perception of what the preceding child is trying to do. For example, the ‘apprentice’ child will infer that the ‘teacher’ child is trying to fold exactly into the middle. If the ‘teacher’ is clumsy and his fold actually overshoots the midline slightly, the ‘apprentice’ will ignore the error and try to fold exactly into the midline. And now, the equivalent of the ‘tapes in the hat’ is to ask independent observers to rank the nineteen Chinese junks. Assuming no major mutations (which again would be interesting in their own right), there will be no tendency for junks later in the sequence to show any ‘degeneration’ compared to junks early in the sequence. There’ll be good ones and bad ones dotted around the line because, I strongly suspect, skilled children will not attempt to imitate obvious incompetence like folding a little way past the midline, but will ‘normalize’.
There may be good objections to the meme/gene analogy but ‘degeneration’ because of low-fidelity replication is not one of them.
If we wanted to perform an experiment in memetics, what might we do? Perhaps take a word with a conventional pronunciation, invent a ‘mutant’ mispronunciation and broadcast it, daily, to tens of thousands of people; then, later, investigate whether the mutant pronunciation takes over the meme pool and becomes the norm. An expensive protocol, unlikely to attract funds from a grant-giving agency. Fortunately, however, by sheer happenstance, the expensive part of the experiment is sometimes done for us. Trains of the London underground system have an on-board tannoy, which daily broadcasts, to tens of thousands of passengers, the names of the stations. The pre-mutant pronunciation of Marylebone station is (something like) ‘marry-le-bön’. The mutant pronunciation on the Bakerloo Line, spoken by a young woman’s recorded voice, is ‘marley-bone’. All that remains for the experiment is to ask a random sample of commuters on the Bakerloo Line how they pronounce the name, and repeat the sample at yearly intervals; then investigate the dispersive spread of the meme by sampling the British population as a whole. My guess is that the mutant form has already started to propagate quite widely. I facetiously suggest that the last bastion to fall will be the Marylebone Cricket Club, the famous MCC.
Models of the world
Under Horace Barlow’s influence, I came to see an animal’s sensory systems, especially the set of tuned recognition neurones in the brain, as a kind of model of the world in which the animal lives. In the same vein, I presented the genes of the animal as a digital description of past worlds – a kind of statistical average of the living conditions of the animal’s ancestors, environments in which the ancestors had survived. I saw the gene pool of a species as an averaging computer, averaging the properties of ancestral worlds. Similarly, brains, as they learn, average the statistical properties of the world that the individual animal has experienced during its own lifetime. As the sculptor which is natural selection chips away at a gene pool, turning it into a descriptive model of averaged anc
estral worlds, so individual experience sculpts brain models of the current world. In both cases the models are updated by data from the world, but on the timescale of generations for the gene pool model, the timescale of individual development for the brain model. I’ve long liked this poem by Julian Huxley (with whom, for some reason, I identified as an undergraduate), and I quoted it in A Devil’s Chaplain:
The world of things entered your infant mind
To populate that crystal cabinet.
Within its walls the strangest partners met,
And things turned thoughts did propagate their kind.
For, once within, corporeal fact could find
A spirit. Fact and you in mutual debt
Built there your little microcosm – which yet
Had hugest tasks to its small self assigned.
Dead men can live there, and converse with stars:
Equator speaks with pole, and night with day;
Spirit dissolves the world’s material bars –
A million isolations burn away.
The Universe can live and work and plan,
At last made God within the mind of man.
I would now add another verse, in imitation of the Huxley style:1
Ancestral worlds invade your species’ genes,
Encoding long forgotten deaths and lives.
Digital texts enshrining what survives,
Distilled from genomes now in smithereens.
What went before? What happened? Who can say?
Yet all is written in your DNA.
Reverting to Julian Huxley’s poem and the brain models constructed on the timescale of individual development, many of my public lectures during the 1990s were devoted to the theme. I was especially inspired by the virtual reality software to which I was introduced during my Christmas Lectures year, and which I demonstrated to the children at the Royal Institution. I brought it all together in the chapter of Unweaving the Rainbow called ‘Reweaving the world’.
When we think we are looking ‘out’ at the real world, there is a strong sense in which we are looking at a simulation, constructed in the brain but constrained by information flowing in from the real world. It is as though the brain contains cupboards full of models waiting to be pulled out at the behest of information flowing in from the sense organs. In a sort of ‘exception that proves the rule’ sense, visual illusions persuade us of this, as Richard Gregory has shown us in his books (and in his Simonyi Lecture in Oxford). The Necker Cube is a famous illusion, which I also used in The Extended Phenotype (see page 318) as an analogy for my two ways of looking at natural selection: the gene’s-eye view and the ‘vehicle’ view. It is a two-dimensional pattern which is equally compatible with two alternative three-dimensional models in the cupboard. The brain could have been designed to plump for one of the two models and stick to it. Actually what it does is take first one model out of the cupboard, ‘see’ that for a few seconds, then put it back and take the other model out. So what we see is first one cube, then the other, then the first again, and so on.
Other famous illusions such as the Devil’s Tuning Fork (see below), the Impossible Triangle (which I demonstrated during my Royal Institution Christmas Lectures1) and the Hollow Mask Illusion all make the same point more dramatically.
We move through a constructed world, a world of virtual reality. If we are sane, undrugged and awake, the constructed virtual reality through which we walk is constrained by sense data in ways that conduce to our survival: it is the real world, not a world of dreams or hallucinations, in which we must survive. Computer software allows us to walk through imaginary worlds, fantasy worlds, Greek temples or Fairylands, science-fiction landscapes of alien planets. When we turn our head, accelerometers in the helmet register the movement, and the images the computer presents to our eyes shift in register. We seem to be turning inside the Greek temple and now see a statue that had been ‘behind us’. And when we dream at night, the brain’s own virtual reality software emancipates itself from reality and we walk through splendid mansions of the mind, or run from constructed monsters in panic-ridden nightmares.
In Unweaving the Rainbow and in my lectures of the 1990s, I fantasized about a surgeon of the future, in a separate room from her patient, walking through his intestine, which is realistically simulated using data from an endoscope inside him. When she turns her head to the side, the tip of the endoscope swings in sympathy. She forges on through the virtual (yet constrained by endoscopic reality) intestine until she locates the tumour ahead of her. She wields the virtual chain saw in her toolkit, and the appropriate microsurgical blade at the tip of the endoscope mirrors her large arm movements in miniature and delicately cuts out the tumour. A parallel fantasy had a plumber of the future doing the same kind of thing, walking – or even swimming – down a virtual drain while his movements are mirrored in the real drain by a small robot deployed to clear a blockage. The key is constraint: the virtual worlds in which we move are not purely fantastic but guided along tramlines constrained to be usefully close to reality.
Our human mental cupboards are especially well furnished with face models, and we eagerly retrieve them given even the smallest encouragement through our optic nerves. This accounts for the numerous stories of Jesus or the Virgin Mary appearing in a slice of toast or a damp wall. The Hollow Mask Illusion (which also featured in my Christmas Lecture1) is the most spectacular manifestation of our face-happy model-retrieving facility. It’s also significant that there is a named brain deficiency, prosopagnosia, sufferers from which can see normally except that they can’t recognize faces, even faces of people they know well and love.
I reprised the theme in The God Delusion, showing how wrong we are to be impressed by visions and apparitions, ghosts and djinns, angels and Blessed Virgins. Our brains are masters of the art of virtual reality. To knock up a vision of a haloed, glowing, robed figure is child’s play, and so is a still small voice in a storm. Many people are sincerely convinced that they have a personal experience of God: he speaks to them, appears to them in dreams and waking reveries. They need to be less impressed. Study Richard Gregory and the psychologists. Recognize the power of illusion. Understand how easily illusion morphs into delusion. God Delusion, for example.
The argument from personal incredulity
In The Blind Watchmaker I coined the phrase ‘argument from personal incredulity’ as a summation of the creationist’s principal ‘argument’. A less sarcastic rendering would be ‘argument from statistical improbability’ – or ‘argument from complexity’, for statistical improbability is the relevant measure of complexity and the relevant provoker of incredulity. Here’s how the argument always goes. A complex biological structure is extolled, having many parts arranged in a precise way. Any random rearrangement of the parts wouldn’t work. The number of possible rearrangements is calculated and found, of course, to be astronomically large. Therefore the complicated arrangement can’t have come about by chance. Therefore – here’s where the argument shoots itself in the foot – God must have done it.
Darwin himself devoted part of a chapter to what he called ‘organs of extreme perfection and complication’. He began with a famous sentence, much quoted by creationists:
To suppose that the eye with all its inimitable contrivances for adjusting the focus to different distances, for admitting different amounts of light, and for the correction of spherical and chromatic aberration, could have been formed by natural selection, seems, I freely confess, absurd in the highest degree.
You can tell, can’t you, from the tone of Darwin’s sentence, that he wasn’t going to leave it there. Doesn’t his tone send a clear signal that a ‘but’ or a ‘yet’ is about to follow? He might even have been leading his readers on, beckoning them towards him so that the punch, when it came, had the greater impact: ‘Yet reason tells me that . . .’ Google finds a mere 39,300 occurrences of the latter clause, compared to 130,000 for the immediately preceding phrase, ‘absurd in the highest
degree’. As Darwin himself said elsewhere, ‘Great is the power of steady misrepresentation . . .’
What’s wrong with the argument from statistical improbability, of course, is that natural selection is not a theory of chance. Natural selection is the non-random filtering of random variation, and the reason it works is that the improvement is cumulative and gradual. In The Blind Watchmaker I illustrated this with the metaphor of a combination lock, for example guarding the door of a bank vault. In the BBC Horizon film with the same name, I actually hammed up an attempt to open a real bank vault with a random number. The whole point of a combination lock is that you need stupendous luck to break in by random dial-twiddling. But, if the lock had a fault such that the vault progressively creaked open another chink every time the number on the dial moved in the right direction, any fool could break in. That’s the equivalent of gradualistic natural selection.
My later metaphor of Mount Improbable did the same explanatory duty. As briefly mentioned above, when I worked with Russell Barnes on the television film Root of All Evil? we simulated Mount Improbable in the ‘Garden of the Gods’ at Colorado Springs. I was filmed standing atop a sheer precipice to represent the ‘creationist’ or ‘huge stroke of luck’ side of the mountain, where achieving the improbable in a single step is equivalent to leaping from the foot to the summit in a single bound. Then the camera was re-sited and I had to tramp stolidly up the gentle, gradual slope on the ‘evolutionary’ side of the mountain: no problem evolving organs of limitless complexity, given only enough time and a gentle gradient of improvement with no sudden leaps. Television being what it is, of course, the gradient and the precipice were actually on different mountains (the ‘Inspector Morse Effect’, whereby the melancholy inspector is filmed entering one Oxford college and debouching into the quadrangle of another).
Brief Candle in the Dark Page 40
