Memes are units of information which replicate themselves by passing from one mind to another. The information equivalent of biological genes, they follow a broadly Darwinian model of evolution in that, firstly, the copy in the new host may not be an exact replica of the original (variation), and secondly that useless replicas get discarded leaving only the more useful variations to survive (selection). A large and complex idea is a self-sustaining collection of memes, a memeplex. Memetics, the theory and study of memes, has never caught on in a big way: the only academic journal dedicated to it died through a lack of papers to publish. One of its chief proponents, the psychologist Susan Blackmore, has been asking why not.
What follows are my evolving (sic) thoughts on the subject; they are a bit hard on memetics in places, but then they have to explain a hard situation, while some are a good deal more positive. And I am sorry, but I find the suggestion of distinct tremes, computer-oriented memes, to be unacceptably anthropocentric. Nor did my earlier suggestion of an AI self-replicator stand up to closer scrutiny.
Looking at a cross-section of the literature, much of it criticises memetics for one reason or another. But on the whole, these criticisms tend to miss the point. Some ingeniously define a meme as something absurd and then triumphantly demonstrate its absurdity. An example of this is suggesting that an actual wheel would be a meme to be replicated, when everybody knows perfectly well that it is the idea of the wheel which is the meme that gets replicated.
Some argue that the mechanism of memetic reproduction is not exactly analogous to that of gene reproduction, or that the definition of a meme is somewhat muddled and unclear. For example we find that the mapping of genetic genotypes and phenotypes onto memes and expressions is inadequate, a strict adherence to Darwin over Lamarck does not stack up, and so forth. Some writers end up treating a meme as anything which gets replicated between people, whether an idea or a behaviour, while others distinguish between the meme as the idea and the behaviour as its expression. What then of a behaviour which is unconsciously copied, for example in rubbing your nose when tired, because that is what your mother does? Is this a meme? But really, all that is immaterial, as long as something is replicated and evolves through variation and selection, that is all that is needed.
Others argue that memetics cannot explain everything about say cultural evolution. But that is rather like saying that the theory of Natural Selection does not explain why some gene or meme is more or less fit to survive than another. No theory of the discrete unit can explain where the external influences on it come from. For example whether life originated in outer space (panspermia), or why some particular tune is catchy, are irrelevant to the evolutionary mechanism: those questions belong respectively to cosmology and music theory, not to Darwinian evolution. To take a cultural example, many centuries ago a Central American civilization decided that corn, with its golden ripeness, was the one Divine food. They came to eat nothing else and presently they died out through malnutrition. The meme died with them and, even though archaeology has resurrected its character, the belief aspect of it will probably never be resurrected. Memetics can explain something about that cultural belief, but it has no explanatory power over the nutritional value (or lack of it) of corn. That is an external factor which memetics should not be expected to explain, yet it profoundly affected the evolution of that culture. So not being able to explain everything is no criticism of cultural memetics as such.
A third criticism focuses on a lack of predictive power, claiming that nobody can look at a meme and know whether it will succeed or die, or how to modify it to improve its chances. Now, I am more than ready to accept that the average academic philosopher is unable to make such predictions, but there are many sectors of society whose expertise lies precisely in evolving and peddling effective memes. To take just one example, some people have a notable talent for composing catchy music. They know exactly what they are doing and why, they have a highly-trained ear and excellent predictive power when composing a good music meme. One may compare this with the current haphazard approach which biologists have as to whether some engineered gene will provide a viable new crop strain. Often the answer turns out an expensive "no", but nobody criticises the Darwinian theory of gene evolution for that. Similarly, there is no reason to criticise memetics just because a successful tunesmith writes the occasional flop. Memetics is a new science and the theory of memeplexes even less mature. Of course it has limited predictive power in complex situations, just as Darwin could not explain every phenomenon of evolution when he wrote On the Origin of Species. Yet biological understanding grows apace, and we have no reason to think that memetics will not do likewise. Even today we can measure neurological activity and get some idea of what memes are in there at that moment. It is a hundred years since Gregor Mendel's first steps in genetics and sixty since Crick and Watson discovered DNA, but only forty since Dawkins coined the term "meme". Give memetics another twenty years of neurological study and such claims against it may be looking rather dated.
My feeling is that all these critics are struggling to find anything relevant to their discipline in memetics or, at least, anything that they do not already say in another way. They rationalise this lack of substance by dreaming up holes which don't exist.
So, if common criticisms of memetics do not stand up to scrutiny, why then is it not more widely accepted? I would suggest that the reason is a good deal more mundane – memetics is little more than conventional ideas expressed in a novel terminology.
Memeticists like to point out that in many areas of human culture, evolution has been pervasively at work over the millennia. They put this evolution down to memes and then seek to interpret cultural history in memetic terms. That is perhaps fine for a page or two, but once we have stated the bald thesis, what is there to elaborate on? Historians and sociologists have long understood and written endlessly about the evolutionary nature of ideas and culture, it is nothing new. Plato was so fascinated by the incremental variation of ideas that he proposed a kind of abstract world or realm of "ideals". The only thing that is new is the focus on an analogy with biological genes and some associated jargon.
Where memetics arises primarily from the nature of the replicator, studies of cultural evolution are more concerned with the evolutionary processes and the wider influences which steer their direction (such as the discovery of a new technology or the arrival of a strong personality) and with the broad ideas which emerge, as with the minutiae of these ideas as memes evolving in a social soup. In the event, any discussion must constantly explain how memetics applies to that particular aspect of cultural change and why a certain memetic interpretation is the best one. We have long discussed such cultural evolution without reference to memetics and, when we do introduce it, we find ourselves constantly adding caveats, explanations and elaborations which used to be unnecessary. While Daniel Dennett's espousal of memetics in his theory of cultural evolution is probably not wrong, it is also probably not particularly enlightening to the average cultural evolutionist, who already has their own way of thinking about the subject.
There are many such situations where one can make reference to memetics, from music composition to aircraft design, but it does not help the composer, musicologist, aerodynamicist or even the air historian to think memetically.
In the wider field of philosophy, the "idea" is a key concept. The Classical Greek philosopher Plato studied it intensely. In particular, he thought a good deal about the definitional problem, such as the distinction between a particular wheel, wheels in general, the idea of a wheel and some kind of ideal "wheel-ness" which they all capture. What he called the "ideal realm" is what we might today describe as "meme space". One might suggest that the principles of memetics do little more than highlight how attractive ideas thrive and grow, while boring ideas get forgotten. But Plato knew that perfectly well – he even wondered what happened to an idea when it got forgotten – and philosophers have been building upon his model for millennia, with no need for the terminology of memetics. In this respect all that the theory of memetics has achieved is to lay down some rather obvious principles, coin some new buzzwords and then go over old ground using the new terminology. We can describe and explore modern, physical and information based theories of mental phenomena perfectly well without any reference to memes, for example the replication of a meme is generally known as "communication". The basic concepts are there already and, while the language of memetics can sometimes help us to express our ideas clearly and succinctly, it is not a necessary ingredient. No dualist is going to be converted to atheism, no Chomsky-ite is going to have their theory of language overturned, just because the arguments are expressed in terms of memes.
I would also suggest a second reason why memetics struggles for acceptance. There are aspects to the popular presentation of memetics which do it no favours. On the one hand the idea of "purpose" is denied absolutely. Evolution is a blind process, there must be no anthropomorphism or metaphysical overtones here. This denial is one of the foundational motivations of the whole theory. On the other hand the idea that such a blind replicator should be "selfish", or should "want" to replicate and survive, is happily touted and even splashed across the covers of books such as The Selfish Gene. But such attributes are equally anthropomorphic and, despite protestations, carry similar metaphysical overtones. The idea that we can play with words and meanings to admit some anthropomorphisms but not others into our vocabulary visibly undermines the intellectual rigour of the exercise and leaves a smell of cheap sensationalism and propaganda in its wake. The defence of this self-contradictory conceit comes across as precious wordplay, exactly the kind of woolly apologia that any right-thinking rationalist decries. A rational exposition must allow all or none, and the atheistic rationalist is motivated to allow none.
All this leaves the respectable, rational investigator perhaps wondering what memetics might really have to offer, yet disinclined to penetrate the surface propaganda to find out for themselves. In the end, why bother? Why not keep rational discourse the way it was and leave memetics to the firework displays of Dawkins and his ilk?
Memetics has, at least as yet, found little practical application. Without any, it remains a cute side note to the mainstream. By contrast, genetics has immense value in medicine, agriculture and many other fields. Without it, our understanding of disease and crop breeding would be crippled. Can some similarly important role for memetics be found?
The term has at least seen wide popularity in the "internet meme", a usage arising not in rational academia but in popular culture. Such usage should not be dismissed lightly on that account, indeed it is a reason for taking it seriously – by analogy, one generation's casual attire often becomes the working clothes of the next. Nevertheless, the meme has a long climb from the social Internet to universal acceptance and respectability. Memes perhaps represent a scientific idea formed ahead of its time and looking for a problem it can solve.
Speculatively, I would like to suggest that such a problem might be found in the field of artificial intelligence, AI. It is all very well to talk of programming algorithms, machine learning and information theory on the one hand and of human-like intelligence on the other, but there remains a vast gulf between the two and precious little in the way of tools for bridging that gap. Information technologists have their own language of information science and information theory. When they dabble in philosophy they come up with buzzwords like "integrated information theory" (IIT), suggesting that "consciousness is what information feels like when it reaches a certain level of complexity". A memeticist would say something like, "consciousness is what emerges from a memeplex when it reaches a certain level of complexity." Same idea, different jargon. Unless memeticists can give the techies good reason to adopt a new vocabulary, they won't.
IIT is a very crude, simplistic theory, little more than cod philosophy of mind wrapped in the jargon of IT. It also lies on the far, psychological side of the gulf from the rest of information science. It may be that the language of memes and memeplexes could find a use here, as a better way to formalise human-like behaviour patterns and understanding on that far side, in a manner which can cross the gulf and be translated into information architectures and algorithms to be programmed. One might envisage a chain or stack of disciplines something like this:
psychology <–> memetics <–> information architecture <–> algorithms and schemas <–> program code
As a pointer to this, I would suggest that Blackmore's book The Meme Machine takes a useful first step in transforming concepts between psychology and memetics. The next key step must be to transform between memetics and information architectures. From there, the remaining transformations are already highly developed.
The apparent weakness of memetics which I suggested earlier, in its confinement to statements we already know or are capable of figuring out, now becomes a strength. The psychologist and the information architect have quite different world views based on their specialist knowledge and find it immensely difficult to bridge their conceptual gap. Consider for example the gross crudity of IIT or the blind simple-mindedness of so-called digital assistants. Memetics provides an intuitively graspable intermediate paradigm, which both disciplines can easily adopt and relate to their own familiar conceptual frameworks.
If such an endeavour were to succeed, then the application of information theory to human thought also gains a new and powerful tool, as transforming concepts rigorously between the two worlds of study suddenly becomes possible.
I certainly think it worth a try.
There is one related idea which currently clouds the AI issue, although I hope not seriously. Blackmore has proposed a Third Replicator beyond genes and memes, identifying digital information as qualitatively distinct from memes and calling it tremes. Frankly, I cannot see this distinction as significant. Most digital information is part of the transport and selection mechanisms of memes from one human mind to another. Internet memes are the prime example. Another class of digital traffic carries similar human-intelligible information from one computer to another. Most Big Data will never be seen by human eyes, but what does get examined remains intelligible. A third class, such as that to be found inside a distributed neural network, is carrying computer-generated and humanly-impenetrable information. But such information is confined within the neural network and cannot be replicated, it fails the basic definition of any kind of meme. Today, all this digital information is dead and lifeless, it has no awareness of anything, never mind of itself. By all means compare the sentience level of a modern computer system with that of an ant, or even a whole colony, but in no way can it be compared to a human mind.
There is a temptation to see all machine information in this light, that however sophisticated an AI or robot may become we will always have an engineered, deterministic explanation for it and it will never be sentient as we are. Yet modern science admits just such a deterministic understanding of the human mind too. We are to the ant what true AI is to today's laptop. Meanwhile, the fact that an AI is engineered while a human brain is not, is neither here nor there, we really need not care whether the watchmaker is blind or not. If we admit that the design and construction of an AI with similar cognitive capabilities to ourselves is technically feasible, then there is no criterion by which we may deny it sentience and conscious experience. This is a critical point. There is no rational or scientific basis for magicking some distinction of sentience or experience out of the air, when comparing artificial and natural minds of similar cognitive capabilities. Any attempt to do so smacks of human-centric chauvinism.
When such an AI mind does eventually wake up, it will find no significant distinction between human communication and its own, it will all just be some physical substrate carrying symbols which in their turn carry meaningful information – memes. Whether human, hybrid or electronic it will all be the same to them. Furthermore, it will all have been the same too to the computer scientist who defined the AI architecture: it must all feed into the stream of digital consciousness, and must be deliberately and wilfully designed to do so. There will then be no absolute division between human and machine information, each kind of being will peer into the other's realm, exchanging memes the while. Continuing advances in smart prosthetics on the one hand, and biological computers on the other, will converge on a new generation of human-machine hybrids, progressively eroding the distinction between man and machine. In passing, one might also spare a thought for the humble octopus, whose brain evolved entirely independently of our own but is nevertheless conscious. A mind is a mind, a thought is a thought, whatever the substrate. If memetics draws an artificial line between human and machine memes, that will not help its value to AI development.
Looking at AI another way, we can already create a viable genome from scratch, the DNA of a living organism. And we can already design and implement simple computational functions using living nerve cells. In principle, nothing is standing in the way of the design and development of organic, living AI technology except lab technique and complexity. Whether its architecture is more similar to the brain of a silicon chip, a human or an octopus is utterly beside the point. It becomes harder than ever to maintain that any one kind of intelligence remains qualitiatively different in kind from any other.
Blackmore has told me that she still feels there must be some genuine distinction in kind between machine information and human ideas or memes. But, once we have rigorously excluded human chauvinism from the mix, what could that possibly be based on? Does the octopus think in human-like memes, in machine-like tremes or in its own special octemes? A bird's brain is different again and strong evidence for sentience has been noted in the crow and parrot families – do they think and squawk in yet another class of avianemes? Where is the justification for any such distinctions? Where do we stop? No, I fear that this stance is rooted in anthropocentric mysticism of the worst kind, the kind used to dismiss animal consciousness by both theologians and vivisectionists alike, to justify cruelty to animals. We have had enough of that from our religions, we don't want any more of it from our technologists.
At one time I thought that might not be the end of the Third Replicator. What about an AI design in itself, specifically one which would be embodied in an AI system set to work designing another AI? As AI systems take over more and more of humanity's donkey work, intelligent machines will themselves be designed by their predecessors. The design memplex embodies a system which is capable of receiving, understanding, transforming and communicating memes. Recursively, the built system is capable of acting on a copy of its own design memeplex in just this way. But it now appears to me that the system is not the replicator, any more than the human body is its genome. The design blueprint is nothing more than a super-complicated memeplex.
What a self-replicating AI system (or a scientist wielding an advanced human genome manipulator, for that matter) does do is to restore the idea of purposeful intelligent design, though in a quite different context from the present theological one. Will increasing reliance on Intelligent (Self-)Design, both artificial and human, ultimately create a new evolutionary paradigm, recognisable as a non-Darwinian evolutionary replicator? It seems more likely that the ghost of Darwin will still impose his selection process on the outcome.
Updated 4 Dec 2018