Unconstrained mutation predicts everything and is useless

In his 2001 book Science and Selection, David Hull proposed that prefixing variation with "blind" or "random" was unnecessary:

In sum, statements about the sorts of variation that function in selection processes need not include any reference to their being "blind," "random," or what have you. All of the terms that have been used to modify "variation" are extremely misleading. Hence, we see no reason to put any adjective before "variation" in our definition of selection.

While "blind" and "random" are unfortunate word choices, one can't allow completely unconstrained mutations in evolutionary models - or else they lose all predictive value. If mutation is a completely unconstrained process literally: anything goes.

As far as I can see, Donald Campbell got this issue right long ago - but used the unfortunate term "blind" to describe it. Campbell basically said that variation was generated on the basis of existing knowledge. Since then he has been much misunderstood on this topic, due to a poor choice of terminology.

Alas, while technically correct, the idea of generating variation on the basis of existing knowledge doesn't help modellers very much. Most modellers of evolutionary dynamics typically stick with undirected mutations, which often work pretty well, and are simple to model.

Replicator rot

It does appear that some people have become confused about the role of high-fidelity copying in evolutionary processes - much as Henrich and Boyd (2002) claim.

Hull (1988) apparently based his theory of evolution on replicators, which he defined as follows:

replicator: an entity that passes on its structure largely intact in successive generations

The "largely intact" is problematical - cumulative adaptive evolution doesn't depend on high-fidelity transmission of structure - since low-fidelity transmission can be compensated for by error correction.

Let's call this mistake "replicator rot".

Aunger says this on the topic:

Any evolutionary process, including the cultural kind, needs only to exhibit features that correlate from one generation to the next. This quality is what biologists call heredity. Replication is a more precise claim about how evolution works — it suggests that a special kind of agent causes the recurrence of cultural features: a replicator. Some evolutionary approaches — competitors to memetics, such as sociobiology and evolutionary psychology — invoke only genetic heredity in their explanation of culture. I disagree. Socially transmitted information is central to the nature of culture. But when it is transmitted, is it replicated? That’s the crucial question.

He goes on to conclude that memes are replicators.

Dennett explains the basic evolutionary algorithm as being based on:

heredity or replication: the elements have the capability to create copies or replicas of themselves.

I think Dennett saves himself from replicator rot by including the terms "heredity" and "copies" - but it's a close call.

Blackmore and Dawkins mostly avoid the worst of the replicator rot as well - but they do so at the expense of giving the word "replicator" a counter-intuitive technical definition which avoids any mention of high-fidelity copying - with Dawkins (1982) saying:

I define a replicator as anything in the universe of which copies are made.

...and Blackmore (1999, p.5) saying:

a replicator is anything of which copies are made

Academic critics - such as Henrich and McElreath (2003) - typically finger Dawkins, Dennett and Blackmore - but their supporting evidence is usually inaccurate or vague.

Sue has said: "by definition, the information people copy is a replicator" and Dawkins has said: "Anything that is imitated is a form of replication". It seems clear that they don't intend any "high-fidelity copying" implications of the term "replicator" - and their explicit definitions of the term "replicator" confirm that this is indeed the case.

Dawkins did publicly succumb to replicator rot later on. In 2005's The God Delusion, page 191:

In its most general form, natural selection must choose between alternative replicators. A replicator is a piece of coded information that makes exact copies of itself, along with occasional inexact copies or 'mutations'.

This is simply wrong. Natural selection can choose between any items, whether they are frequently copied exactly or not.

In 2013, Dawkins offered:

Anything transmitted with high fidelity from brain to brain by imitation is a meme.

This is true - but it suggests that computer-based transmission doesn't count, and that high-fidelity copying is required. These both seem like highly dubious ideas.

References

• Blackmore, Susan, J. (1999) The Meme Machine.
• Dawkins, Richard (1982) The Extended Phenotype.
• Dawkins, Richard (2005) The God Delusion.
• Dennett, Daniel C. (1996a) Darwin's Dangerous Idea.
• Henrich, Joseph and McElreath, Richard (2003) The evolution of cultural evolution..
• Henrich, Joseph and Boyd, Robert (2002) On modeling cognition and culture: Why replicators are not necessary for cultural evolution.
• Hull, David L. (1988) Science as a Process: An Evolutionary Account of the Social and Conceptual Development of Science.
• Tyler, Tim (2009) Against Replicator Terminology.
• Tyler, Tim (2010) The role of high-fidelity transmission in memetics.
• Tyler, Tim (2011) The claim that evolution doesn't require replicators.

Understanding of cultural symbionts in academia

Many modern academic students of cultural evolution seem to share a common problem with understanding how cultural evolution operates. Though some pay lip service to the idea, they don't seem to fully appreciate that culture's relationship with human hosts is a symbiosis.

Some quotes (some of which I have discussed before) illustrate the syndrome:

Kevin Laland and Gillian Brown, Sense and Nonsense (2004, p.253):

Social transmission can occur vertically (that is, from parents to offspring), obliquely (from the parental to the offspring generation; for instance, learning from teachers or religious elders) or horizontally (that is, within-generation transmission such as learning from friends or siblings). Of course, genetic inheritance is exclusively vertical and hence, as social transmission frequently occurs through some combination of these modes of information transmission, cultural evolution may commonly exhibit commonly exhibit quite different properties from biological evolution.

Paul Erlich The Evolution of Norms (2005):

Among humans, genes can only pass unidirectionally from one generation to the next (vertically), normally through intimate contact. But ideas (or “memes”) now regularly pass between individuals distant from each other in space and time, within generations, and even backwards through generations. Through mass media or the Internet, a single individual can influence millions of others within a very short period of time.

William Durham (1991, p.193) says:

genes usually cannot be transmitted independently of the reproduction of their carriers. This constraint obviously does not apply to memes.

Peter Richerson, in The Evidence for Culture Led Gene-Culture Coevolution: The Naturalization of Culture or the Culturalization of Human “Nature”? (2011) wrote:

We do know that culture is most ungene-like in many respects. Culture has the principle of inheritance of acquired variation (what one person invents another can imitate). We are not necessarily blind victims of chance imitation, but can pick and choose among any cultural variants that come to our attention and creatively put our own twist on them. we don’t have to imitate our parents or any other specific individuals but can always be open to a better idea, or own invention or someone else’s.

Alex Mesoudi, Cultural Evolution (2011) has a similar passage:

One of the more obvious differences between cultural and biological evolution involves the potential transmission pathways each involves. Genetic inheritance is often thought of as being exclusively vertical and biparental, with genetic information transmitted in equal amounts from two parents to a single offspring. In culture, on the other hand, one can learn beliefs, ideas, skills, and so forth, not just from one's biological parents (termed "vertical cultural transmission"), but also from other members of the parental generation ("oblique cultural transmission") and from members of one's own generation ("horizontal cultural transmission").

...though Mesoudi continues by acknowledging:

In fact many of these pathways of cultural transmission have parallels in biological evolution.

...although he fails to mention any of the key phenomena of mutualism, partasitism or symbiosis.

Most of the material above is completely wrong. Symbionts (parasites and mutualists) commonly pass "horizontally" between humans. Parasite genes are shared horizontally by kisses, sex, holding hands and sneezing. Mutualist symbionts and their genes are shared between humans at gardening shops, farms, seed shops and fruit shops. Oblique transmission and transmission "backwards" - down the host generations - work in a similar manner. It should be a matter of acute embarrassment among theorists of social evolution to have missed this.

Such symbiont exchange is by no means confined to humans or other creatures with culture - it occurs ubiquitously in the animal kingdom.

In my experience, many of the misunderstandings of memetics actually turn out to be misunderstandings of how biological evolution works. This example is a case in point.

These academic students of cultural evolution usually go on to say that - because of these differences, we need new models to deal with the situation - and then they go on to develop elaborate extended genotype models to deal with the situation. No! That is not how science is done. The existing models of organic symbiosis handle all these cases just fine. We do not need a raft of new models just to deal with the case of organisms whose genes happen not to be made out of DNA.

Mesoudi's defense of this practice reads:

Nevertheless, most quantitative models of genetic inheritance are indeed based on the assumption of vertical inheritance, making it necessary to construct models tailored specifically to the cultural case.

Not everyone in academia gets this wrong. David Hull, for example was pointing out this mistake back in 1988:

In this connection, commentators often state that biological evolution is always vertical whereas conceptual evolution is likely to be "horizontal". By this they mean that the transmission of characteristics in biological evolution is always from parent to offspring (ie, inheritance). Characteristics always follow genes. In point of fact, biological evolution is not always vertical, even when characteristics follow genes. For example, it is horizontal when bacteria, paramecia, etc. exchange genetic material. Horizontal transmission can even be cross-lineage, as when viruses pick up genes from an organism belonging to one species and transmit them to an organism belonging to a different species.

There are a few cases of recognition of symbiosis:

Most mathematical models of cultural evolution derive from epidemiology. The terminology of "horizontal transmission", "vertical transmission" and "oblique transmission" comes from epidemiology. Epidemiology itself is mostly - though not exclusively - concerned with symbiosis.

Boyd and Richerson (1985) have three sentences on symbiosis. They say:

Horizontal transmission is analogous in some ways to the transmission of a pathogen

...and...

The item of culture being spread horizontally acts like a microbe that reproduces and spreads rapidly because it is "infective" and has a short generation length compared to the biological generation length of the host. Fads and fashions and technical innovations are familiar examples.

Boyd and Richerson (2005, p.165) has a paragraph on symbiosis:

The nonparentally transmitted parts of culture are analogous to microbes. Our immune system evolved to kill microbial pathogens but it also allows us to acquire helpful symbionts. As we know all too well, microbial pathogens are common, despite the sophistication of the immune system. One reason is that we are not the only players in this game. Natural selection helps parasites trick our immune system. Since microbial populations have short generation times and large populations, parasite adaptation can be very rapid. The psychology of social learning is like an immune system in that it is adapted to absorb beneficial ideas but resist maladaptive ones. And, like the immune system it is not always able to keep up with rapidly evolving cultural “pathogens.”

This section is pure memetics. They also implicitly endorse symbiosis in their section on "selfish memes" (p.153-154).

In "The Role of Evolved Predispositions in Cultural Evolution" they say:

An empirical study of the spread of heroin addiction describes the close resemblance of its dynamics to the spread of disease that requires intimate contact (Hughes and Crawford 19721. Addiction is spread along chains of close friendship. Addicts remain infectious only in the early stapes of addiction, while the p1easurabte aspect of the drug still outweighs the manifest disability of advanced addiction. Only a limited population of susceptible individuals is at risk of acquiring the addiction even if exposed. Many simple epidemiological principles probably apply to pathological cultural traits - e.g., parents notice that the incidence of minor microbial infections and various obnoxious habits in children increase together when they first go to school. Crowded classrooms of young susceptibles are the ideal environment for the spread of pathogens of both types by horizontal transmission among the children!

There's a similar section in their paper: "Built For Speed, Not For Comfort".

...and there is a fairly specific endorsement of the idea from Peter Richerson here:

I think it is near to undeniable that cultural variants are sometimes selected to become selfish pathogens along the lines that Dawkins suggested. Since some cultural variants can spread rapidly among people, as in the case of fads, they rather resemble the life cycle of a viral or bacterial pathogen.

Rivals to gene-based terminology

In my memetics book, I propose that we use the term "gene" for the heritable elements in evolution - and ignore or discard the numerous dictionaries and textbooks that claim that genes are "molecular units of heredity" - or that they have anything to do with DNA or RNA.

There have been some other proposals for terms that convey this meaning:

• Mneme. Richard Semon (1904) wrote:

  Instead of speaking of a factor of memory, a factor of habit, or a factor of heredity and attempting to identify one with another, I have preferred to consider these as manifestations of a common principal, which I shall call the mnemic principal.

• Meme. In his book The Mocking Memes - A Basis for Automated Intelligence, Evan Louis Sheehan writes:

  I define memes to include every sort of pattern that serves as a template for its own replication.

• Replicator. David Hull (1988b) proposed replicators fill the role of the carriers of heredity in evolving systems.
  

Deploying Richard Semon's term "mneme" in the modern era seems rather impractical.

Evan Louis Sheehan's "meme" tries to hijack an existing term. "Meme" has an established meaning which does not obviously need to change. I think the attempt fails.

The proposals of David Hull and Evan Louis Sheehan also suffer from a technical problem - since they only include copyable heredity information, and not all heritable information is capable of being copied.

How do genes differ from ordinary information? In other words, what is an example of information that is not inherited? Conventionally, there is no inheritance without some living thing being involved. Also, information that is destroyed is not inherited. Other forms of information could potentially be inherited by some living thing or another.

So: "gene" still seems to be better overall. Of course, this raises the issuse of what name should we give to small chunks of nucleic acid. Im my book, I wrote:

Those are "genes" too, of course, and can normally simply be referred to as such - but if a term is really wanted to refer specifically to nucleic acid chunks while excluding other forms of inheritance - they could be called "organic genes", "cellular genes", "nuclear genes" or "DNA genes" - depending on exactly what you actually meant.

What about snappy abbreviations?

"Denes" is my pick for "DNA genes" - with "denetics" referring to their study.

"Nenes" is my pick for "nuclear genes" - with "nenetics" referring to their study.

Update 2013-05-19: Dawkins said:

Completely unknown to me when I coined "meme" in 1976, the German biologist Richard Semon wrote a book called Die Mneme (English translation The Mneme (London, Allen & Unwln, 1921)) in which he adopted the "mneme" coined in 1870 by the Austrian physiologist Ewald Hering.

Alex Mesoudi's take on memes

I tracked down Alex Mesoudi's take on memes. He describes memetics as being a "fad". In Cultural Evolution, page 42, he says:

Memetics makes the neo-Darwinian assumption that culture can be divided into discrete units that are inherited in a particulate fashion, like genes. It also assumes that memes are transmitted with high fidelity, this being one of the defining characteristics of a replicator according to Dawkins.

For me, this is disappointing material. Mesoudi fails to provide references. As far as I can tell, Dawkins did not, in fact, claim that high fidelity transmission was one of the defining characteristics of a replicator.

What Richard Dawkins actually says in The Selfish Gene (page 17) is:

A third characteristic of replicator molecules which would have been positively selected is accuracy of replication. If molecules of type X and type F last the same length of time and replicate at the same rate, but X makes a mistake on average every tenth replication while Y makes a mistake only every hundredth replication, Y will obviously become more numerous.

Dawkins clearly says that accuracy of replication is a trait of replicator molecules which is subject to positive selection. That means that the replicator molecules varied in their replication accuracy. This is completely the opposite of high fidelity transmission "being one of the defining characteristics of a replicator" - since it is saying that replicator molecules varied in their transmission fidelity and were subject to selection based on that variation.

Mesoudi is attributing to Dawkins the idea that high fidelity transmission is a defining characteristic of replicators - which is the opposite of what Dawkins said in The Selfish Gene. Dawkins went on to offer an explicit definition of the term "replicator" in The Extended Phenotype - and that doesn't include the concept of high-fidelity transmission either.

This material about memetics assuming high fidelity transmission appears to be F.U.D. As far as I can tell, the origin of it is a 1985 book on the topic - "Culture and the Evolutionary Process" by Robert Boyd and Peter Richerson. I figure they probably needed selling points for their variation of cultural evolution - and the ability to handle analog transmission was one of the ones they explicitly went for - on page 37. For that to be true, previous authors needed to have assumed discreteness. They took the conventional meaning of the word "replicator", and either failed to notice or ignored the fact that Dawkins (1982) had explicitly defined it as:

I define a replicator as anything in the universe of which copies are made.

In this definition, high fidelity transmission is conspicuous by its absence.

As far as I can tell, that is how the idea of high fidelity transmission is a defining characteristic of memes most likely arose. It is a rather sad story - but it is not clear why it is that this misunderstanding still being perpetuated some 26 years later.

Memetics doesn't require high fidelity transmission. The most successful memes do indeed exhibit high-fidelity transmission - just as Dawkins said. They have migrated onto the internet, where transmission is digital - resulting in hundreds of millions of bit-identical copies of the latest Lady Gaga song - but that doesn't mean that every single meme is transmitted with high fidelity. That would be ridiculous. The idea that high fidelity transmission is required by memetics is just a bunch of nonsense.

The researchers involved have been advised about this before. For example here:

In a section entitled “cultural variants are not replicators”, they repeat the false claim that copying must be perfect for a replicator to count as such

...and here:

Boyd and Richerson first introduce replicators as “material objects that are faithfully copied”. It certainly isn't Richard Dawkin's view who described replicators as “any entity in the universe of which copies are made”.

Also, I went on about this in my video review of Not By Genes Alone.

I will repeat the message again here: memetics does not require high fidelity data copying. The idea that it does is just a basic misunderstanding - originating - as far as I can tell with Boyd and Richerson, 1985, p.37. Their claim seems patronising to me. It is basic information theory that you can produce high fidelity information transmission (which is much more important for adaptive evolution) from low fidelity data transmission.

Boyd and Richerson were not attacking a straw man - since I actually have seen two people mess this point up:

• John Maynard-Smith - In "Evolution-Natural and Artificial" (1999), John claims that adaptive evolution requires digital inheritance - and it makes some other rather silly claims for good measure. That is somewhat embarassing, but it has nothing to do with memetics - because Maynard-Smith was not a supporter.

• David Hull - In "Science as a Process" (1988) David defined a replicator as being: "an entity which passes on its structure largely intact in successive replications."

Hull goes on to talk about entire populations passing on their structure "largely intact". What is "structure"? Hull says to see "Social Being" by Harre (1979) for more about that. However, I think this leaves only a little room for a sympathetic interpretation involving information since "information" and "structure" seem to be two rather different concepts. I think this puts David Hull's attempt to form a basis for evolutionary theory into the category of being wrong - or, at best, misleading.

Hull makes one passing mention of "memes" in "Science as a Process", on page 406. It is plain from his other writings that he endorsed the term. However, I do not endorse Hull's formulation. I think he made a mistake - or at least was being misleading. However, please don't attribute to memetics Hull's mistake. Lumsden and Wilson made some mistakes too, but I don't go around saying that gene-culture evoultion has got it all wrong.

The idea of high fidelity transmission is genuinely important for evolutionary theory - because information needs to be transmitted with high fidelity in large-scale culmulative adaptive evolution (with some caveats about incredibly strong selection pressures and mysterious sources of directed mutations). However it is possible to have high-fidelity information copying with low fidelity data transmission, as Shannon and Von Neumann previously explained long ago. We actually see such things in the modern world in the form of self-encrypting computer viruses. The data is wildly different in each generation, but the information is transmitted with high fidelity. So, it is high fidelity transmission of information - not data - from place to place that is the correct idea which associates the concept of "high fidelity" with the fundamentals of evolutionary theory.Also, not all of the transmitted information has to be conveyed with high fidelity, just some of it. You need some kind of mutual information between ancestors and descendants, or evolution is not going to be very "cumulative". There's a page with more details about this point here.

This is not a completely trivial point, and I can understand how people could get into a muddle about it. However, now that we have the internet, the time for getting into pointless muddles should be over.

I think the people from academia who persist in spreading this kind of material around need to be explicit about where they are getting it from, quoting chapter and verse. So far, there hasn't been very much of that. Boyd and Richerson (1985) cited "The Selfish Gene" and "The Extended Phenotype". Page numbers please. Boyd and Richerson (1985, p.266) pointed to the forward to "The Meme Machine". There Dawkins defends against the charge that cultural evolution does not have good enough fidelity to exhibit adaptations - which is a perfectly reasonable point to make. Dawkins has plenty of material in those books which flatly contradicts the alledged business about high fidelity copying - so I figure Boyd and Richerson can't have been trying very hard to find a sympathetic interpretation. If people remain incapable of quoting the incriminating chapters and verses, I figure they are eventually going to have to shut up about it.

In the mean time, please try to remember: neither genetics nor memetics requires high-fidelity data copying. The idea that they do is just a fallacy.

Memetics is the cultural sub-branch of genetics, which is defined as being the science of heredity. How inheritance takes place is an implementation detail in genetics and, similarly, it is an implementation detail in memetics. This should not be that difficult to understand.

As for memes being "discrete" and "inherited in a particulate fashion", it is just a matter of fact that memeplexes can sometimes get sliced up during transmission, just as geneplexes can, and - after being sliced - some heritable information is on one side of the slice and some of it is on the other - producing two or more "discrete" pieces - or "particles" - with potentially different pathways to immortality or oblivion.

Do memeplexes have preferred "break" points - rather like the grooves on a bar of chocolate? Sometimes they do, and sometimes they don't - and DNA sequences are much the same - occasionally there are preferred points of division (e.g. see restriction enzymes), but usually there are not - and the divisions that take place during meiosis can take place practically anywhere.

Hull (2001, p.120) addresses the "particulate" criticism, as follows:

Some authors argue that no general analysis of selection process equally applicable to biological and conceptual evolution is possible because genes are "particulate" while the units in conceptual replication are highly variable and far from discrete. In point of fact neither biological nor conceptual replicators are all that particulate. In both cases, the relative size of the entities that function either as replicators or as interactors is highly variable and their boundaries sometimes quite fuzzy.

Contrast this with Mesoudi (2011, p.46):

However, whereas genetic inheritance is particulate, cultural inheritance in many cases appears to be non-particulate.

In what senses is genetic inheritance particulate, though? In sexual creatures, nucleotide sequences can - and do - divide at practically any point. Yes, there are start and stop codons - but that takes us out of the realm of genetics and into the world of development. From a genetics perspective, about the only sense in which genes are "particulate" is that they don't divide half-way through a base pair. In summary, the supposed "particulate" nature of genetic inheritance is just nonsense. Organic genetic expression is somewhat particulate - at least if you can cope with ideas like "particles within particles" and "overlapping particles" - but genetic inheritance, not so much.

Dawkins clearly allowed for "non-particulate" replicators in his 1983 "Universal Darwinism" essay - saying:

A full science of Universal Darwinism might consider aspects of replicators transcending their detailed nature and the time scale over which they are copied. For instance, the extent to which they are "particulate" as opposed to "blending" probably has a more important bearing on evolution than their detailed molecular or physical nature.

Dawkins is presumably talking about recombination there. There is a difference between genetic and cultural evolution in terms of what types of recombination are permitted - with most genetic recombination being in the form of splicing, while memetic recombination can involve averaging, extrapolation and interpolation. Memetics has permitted these kinds of recombination from the beginning. There is no version of memetics that does not allow for "blending" recombination - except in form of straw men in the minds of critics.

Population memetics no more requires particulate memes than population genetics requires particulate genes. In both cases, it is the researcher who selects the boundaries of the entities whose frequency they are measuring.

From my perspective, the alleged "discrete" and "particulate" problems supposedly associated with memes represent more daft storms in a teacup.

Laland and Brown said that cultural evolution had been characteried by Sense and Nonsense - and I think that the "discrete" and "particulate" criticisms take us well into the "nonsense" territory.

These are not real problems but rather show all the signs of being marketing material concocted by rival researchers keen to find some way of distinguishing their own ideas from those of others that came before them with much the same notions.