Neural networks in the brain consist of large agglomerations of individual cells with a layer of cells taking input (e.g. from the eyes, or from another network), another layer providing output (e.g. to the muscles, the voice, or another network), and many layers in between. Each neuron has connections to many others and the strength of these connections varies according to their history. At any given state of the network a certain kind of input will produce a certain kind of output but this relationship is not fixed. The network can be trained, for example by consistently pairing certain sorts of input, and this experience changes its responses to new inputs. In other words, it can remember.
This kind of memory is nothing like the memory of a digital computer, with its fixed locations, nor like a tape recorder with its more or less faithful duplication of everything fed into it. In a brain, every input builds on what went before. In a lifetime of complex experiences we do not store each one away in a black box to be retrieved later when we need it -rather, every experience comes into a complex brain and has a greater or lesser effect on what it finds there. Some things have virtually no effect and are completely unmemorable (we could not function otherwise). Some have enough of an effect to stay briefly in short–term memory but are then lost, while some lead to dramatic changes so that precise events can easily be reconstructed, whole poems recited or that special face never forgotten.
Effective memes will be those that cause high fidelity, long–lasting memory. Memes may be successful at spreading largely because they are memorable rather than because they are important or useful. Wrong theories in science may spread simply because they are comprehensible and fit easily with existing theories, and bad books may sell more copies because you can remember the title when you get to the bookshop -though, of course, we do have strategies for overcoming these biases. An important task of memetics will be to integrate the psychology of memory with an understanding of memetic selection.
Some people argue that memes are not digital (Maynard Smith 1996)
and that only digital systems can support evolution. Certainly genes are digital and certainly digital storage is far preferable to analogue. We all know that digital video- and audio–recordings look and sound better than their analogue predecessors; a digital system allows information to be stored and transmitted with far less loss of information even over noisy channels. However, there is no law that says that evolution has to be digitally based – the issue is really one of the quality of replication.
What, then, makes for a good quality replicator? Dawkins (1976) sums it up in three words – fidelity, fecundity, and longevity. This means that a replicator has to be copied accurately, many copies must be made, and the copies must last a long time – although there may be trade–offs between the three. Genes do well on all three counts, and being digital gives them high fidelity copying. So what about brains?
Our memory is obviously good enough for us to learn several languages, recognise thousands of photographs from one showing, and recall the major events of our lives over periods of decades. Is this good enough to support memetic evolution? I think this is an empirical question that could be put to the test. In the future, memeticists might be able to develop mathematical models to determine just how high the fidelity of memory must be to support memetic evolution, and compare that with the known performance of human memory. My guess is that our memories will be found to be quite good enough, whether they ultimately turn out to be digital or not.
Second, memes depend on being transmitted from one person to another and, by definition, this is done by imitation. We have already seen how poorly understood imitation is but we may at least make a simple prediction. Actions that are easy to imitate will make for successful memes and ones that are difficult to imitate will not.
Apart from that, the effective transmission of memes depends critically on human preferences, attention, emotions and desires – in other words, the stuff of evolutionary psychology. For genetic reasons we are driven by the desire for sex, for sex of different kinds, for food, for better food, for avoiding danger and for excitement and power. Evolutionary psychology already provides us with lots of information that explains why some memes are picked up again and again while others make no impact. We need to use that information and build on it.
To conclude – it is true that we do not understand in detail how memes are stored and transmitted. But we have plenty of clues and we certainly know enough to get started.
Memetic evolution is ‘Lamarckian’
Biological evolution is not Lamarckian and cultural evolution is – or so I have heard. This apparent difference has been frequently noted, and many treat it as a problem (Boyd and Richerson 1985; Dennett 1991; Gould 1979, 1991; Hull 1982; Wispé and Thompson 1976). In a recent discussion of artificial life, the British biologist John Maynard Smith asked what features are necessary for any evolving system – natural or artificial – and suggested ‘digital coding and non–Lamarckian inheritance’ (Maynard Smith 1996, p. 177). So is memetic evolution really Lamarckian? And what would be the significance for memetics if it were?
First, the term ‘Lamarckian’ has come to refer to just one aspect of the evolutionary theory of Jean–Baptiste de Lamarck. Lamarck believed in all sorts of things that have now been rejected, including the inevitability of progress in evolution and the importance of organisms striving towards their own improvement. However, what is now referred to as ‘Lamarckism’ is the principle of the inheritance of acquired characteristics. That is, if you learn something or undergo some change during your lifetime, you can pass it on to your offspring.
Lamarckism (in this sense) is not true of biological evolution, at least in sexually reproducing species. The way inheritance works (which was not understood in Darwin’s or Lamarck’s time) makes it impossible. This is sometimes known as ‘Weismann’s barrier’ after August Weismann who, at the end of the nineteenth century, pointed out what he called the ‘continuity of the germ–plasm’. In more modern terms we can see it like this – using the example of sexual reproduction and human beings.
The genes are coded in DNA and stored in pairs of chromosomes in every cell of your body. At any location on a chromosome different people may have different alleles (versions) of the same gene and the total make–up of genes in each individual is known as their genotype. Correspondingly, the various characteristics of the final person is known as the phenotype. The genes are not a blue print or a map of the future phenotype; they are instructions for building proteins. These instructions control the development of the embryo as it grows, and of the adult as it develops in its own unique environment. The result is a phenotype which is highly dependent on the genotype it started from but is in no sense a copy of that genotype or completely determined by it.
Now imagine that you acquire some new characteristics by, say learning a language, practising playing the piano, or building up your thigh muscles – that is, your phenotype changes. There is no way that this
change in your body can affect the genes that you pass on – although it can affect
whether
you pass some on or not. The genes your children will inherit are derived directly from the genes you inherited and it is this continuous line that is known as the ‘germ line’. Conceivably, if genes worked as some kind of stored blueprint or map then changes in the phenotype could be fed back to change the map but this is not the case. Conceivably, the process of meiosis, in which cells divide to make ova and sperms, could be affected by changes in the phenotype, but this does not happen and in any case, the ova a woman carries in her ovaries are already there at her birth. We must imagine the germ line going on continuously, with the genes being shuffled and recombined in each generation. These genes instruct the phenotypes which then set off on their own and are either successful or not, but the phenotypes do not instruct the genes.
Even though Lamarckian inheritance cannot happen in such a system, there have been many experiments looking for it. Weismann himself cut off the tails of mice for many generations with no obvious effect on the length of their offspring’s tails. However, this is not strictly a test of the theory because Lamarck argued that organisms had to strive towards improvements, as when giraffes stretch their necks, or birds practise flying, and Weismann’s mice presumably did not strive to have their tails cut off. In Russia, the official science of Lysenko was based on Lamarck–ism but produced no progress in biology and was disastrous for Soviet agriculture because their plant breeding programmes failed.
Lamarck’s idea is still popular and appears in many guises, including memories of past lives being attributed to ‘genetic memory’, and psychic powers being explained by ‘spiritual evolution’. Perhaps it is popular because it implies that there is some point in all our hard work or some benefit to our children if we struggle to improve ourselves. But from the purely genetic point of view there is no such benefit. Popular it may be, but it is simply not true.
At least, Lamarckism is not true for sexual species. For other kinds of organism the idea is simply inapplicable. The most common creatures on this planet are unicellular organisms, such as bacteria, which reproduce by cell division. For these ubiquitous creatures there is no clear distinction between genotype and phenotype, genetic information is exchanged in various ways, and there is no clear germ line. So the whole idea of Lamarckian inheritance is irrelevant.
What then of cultural evolution? The answer depends critically on how you draw the analogy between genes and memes and, as I have stressed before, we must be very careful whenever we use this analogy.
One way to draw the analogy is to stick to the notion of the human genotype, phenotype and generation. In this case, acquired characteristics are certainly passed on, as when religions are transmitted from parents to children generation after generation. But memes do not stick to biological generations and can jump about all over the place. If I invent a brilliant new recipe for pumpkin soup, I can pass it on to you and you can pass it on to your granny and she can pass it on to her best friend. Also, this is not inheritance in the biological sense and the genes are not affected. So it is not Lamarckian.
A more interesting way to use the analogy is to forget about phenotypes and biological generations and look at memes and memetic generations. In the case of the soup there were three generations between me and your granny’s best friend. In each generation, the recipe went from brain to behaviour in the kitchen and on to the next brain (that is if you watched me make the soup). Was there inheritance of acquired characteristics? Let us say that the meme in my brain is the equivalent of the genotype and my behaviour in the kitchen is the equivalent of the phenotype. Then, yes, inheritance is Lamarckian because if I put in too much salt on this occasion, or you forget one of my special herbs, or fail to copy the way I shred the garlic, then you will pass on this new version when your granny watches you, and so the new phenotype will have acquired the characteristics accordingly.
But what if you did not watch me make the soup? What if I sent you the recipe in the post and you passed it on to your granny and she made a photocopy for her friend? Now the situation is quite different. We may draw the analogy with biology this way. The written recipe is like the genotype, it contains the instructions for making the soup. The soup is like the phenotype. The delicious taste of the soup is the reason why the recipe is copied – your granny only asked for a copy of the recipe because she liked the soup. In this case, if she fails to follow the recipe correctly her alterations may affect the chance of someone else wanting the recipe, but the alterations will not be passed on because they are only in the soup itself (phenotype) not in the written recipe (genotype). In this case the process is perfectly analogous to the biological situation and is
not
Lamarckian.
I shall call these different modes of transmission ‘copy–the-product’ and ‘copy–the-instructions’. Music provides a slightly different example. Let us suppose that my daughter plays a beautiful piece of music for her friends and one of them wants to learn to play it too. Emily could either play it many times until her friend can copy it accurately (copy–the-product), or simply hand her the written music in a book (copy–the-instructions). In
the first case, any changes Emily makes will be passed on, and if there follows a series of pianists copying each other, the composition may gradually change, incorporating the errors or embellishments of each player. In the second case, the individual playing styles of each pianist will not have any effect because copies of the (unembellished) written music are passed on. In the first case the process appears Lamarckian but in the second case it does not.
In the biological world, sexual species work by copying–the-instructions. The genes are the instructions that are copied, the phenotype is the result and is not copied. In the world of memes, in which both processes are used, you could argue for calling ‘copy–the-instructions’ Darwinian, and ‘copy–the-product’ Lamarckian, but I suggest this would only lead to more confusion. I deliberately described the soup and the music in ways that allowed the two modes of replication to be easily separated, but in the real world they may be inextricably mixed. From me to your granny’s friend the instructions on making the soup might go from brain to piece of paper, to behaviour, to another brain, to a computer disk and another piece of paper and to another brain – with lots of different flavoured soups being made along the way. Which is the genotype and which the phenotype in each case? Are we to count memes as only the instructions in the brains or the ones on paper too? Are the behaviours memes or meme–phenotypes? If the behaviour is the phenotype, what then is the soup? There are lots of possibilities in memetic evolution because memes are not confined by the rigid structure of DNA. The ways they spread are legion. But we can only decide whether memetic evolution is
really
Lamarckian if we can answer these questions. We seem to be at an impasse.