The Memetic Mind, Not: Where Dennett Goes Wrong

On the face of it, Dennett and I have very different views about cultural evolution. To be sure, we both believe that Dawkins’s initial insight is valid: that culture is an evolutionary regime unto itself in which the benefits of cultural success accrue to cultural entities, not human individuals or populations. Where Dennett talks only of memes, I make an explicit distinction between memes and a cultural correlate of the phenotype (for which I have yet to adopt a term of art).

While Dennett allows memes to exist both in the external world and in the mind, most of his discussion is about memes in the mind moving from one mind to another. Indeed, I’d be curious to know what Dennett thinks exists in the mind apart from memes; of what, for example, does the neonate’s mind consist of? By contrast, I insist that memes exist in the external world, as observable (and memorable) properties of objects, events, and processes. The cultural correlates of the biological phenotype emerge as mental processes in brains as those brains engage with memes.

We thus have rather, if not utterly, different views about cultural evolution. As I have been thinking these things through, however, I have begun to suspect that our difference is more in how we assign roles in the process of cultural evolution to the mechanisms of human thought and action than in our conception of those mechanisms (though we no doubt have our differences there as well). And that’s the line I wish to investigate in this post. I will concentrate that investigation on a single essay:

From Typo to Thinko: When Evolution Graduated to Semantic Norms [PDF], in Evolution and Culture. Stephen C. Levinson and Pierre Jaisson, eds. The MIT Press: 2006.

All quotations are from that paper.

* * * * *

After some preliminary this and that, Dennett gets down to business, taking up an argument where Dan Sperber (2000) (PDF) distinguishes between true copying and triggered “re-productions.” Sperber’s conditions for true copying are quite strict, too strict Dennett believes. And I’m with Dennett on this. Sperber, in Dennett’s retelling, asks us to consider the difference between copying “a nonsense scribble and a five-pointed star.” We are not good at copying nonsense scribbles. The copies would be poor, and copies of copies would be poorer still. Copies of five-pointed stars would fare much better. In fact, if the original had visible irregularities, some of the copies might be more accurate than the original (p. 138):

However, Sperber maintains, the succession of stars would not really be copies of their predecessors, since the “copyists” would normalize to the recipe for the drawing procedure, ignoring the details of the individual productions. Is Sperber looking at the right level of fineness? Dawkins’s point is that a finite repertoire of such triggered productions is not just a good trick for human beings who want to heighten their transmission fidelity, it is a Good Trick discovered several billion years ago by natural selection. Sperber distinguishes copying from merely triggering the production of a similar effect, but a repertoire of such triggers, called an alphabet, is what makes high-fidelity copying possible, both in cells and in human culture.

I agree with this.

Having introduced the notion of an alphabet into the discussion, Dennett continues with a succession of examples involving language (p. 139):

Suppose Tommy writes the letters “SePERaTE” on the blackboard, and Billy “copies” it by writing “seperate.” Is this copying or triggered reproduction? The normalization to all lower case letters shows that Billy is not slavishly copying Tommy’s chalk marks but rather is being triggered to execute a series of canonical, normalized acts: make an “s”; make an “e,” etc. It is thanks to these letter norms that Billy can “copy” Tommy’s word at all.

Billy, of course, has already internalized the norms for using lowercase and uppercase letters. If this weren’t the case, if Billy were unfamiliar with the conventions of English orthography–perhaps because is unfamiliar with English or perhaps he only speaks it, then he would copy the upper and lowercase forms as Tommy has written them.

This, no doubt, is obvious to you, and having me underline it in this way is tedious and perhaps even insulting. I do this because that’s what I want to concentrate on, what knowledge is in one’s head as one reads a signal (or some other cultural artifact or practice).

Dennett continues on, this time bringing other norms into play, other bits of internalized knowledge, as well us other actors to utilize them:

And he [Billy] does copy Tommy’s spelling error, unlike Molly, who “copies” Tommy by writing “separate,” responding to a higher norm, at the level of word spelling. Sally then goes a step higher, “copying” the phrase “separate butt equal”— all words in good standing in the dictionary—as “separate but equal,” responding to a recognized norm at the phrase level. Can we go higher? Of course. Anybody who, when “copying” the line in a recipe “Separate three eggs and beat the yolks until they form stiff white cones,” would replace “yolks” with “whites,” knows enough about cooking to recognize the error and correct it. Above spelling and syntactic norms are a host of semantic norms as well.

This ‘assent’ to ever more comprehensive norms is the point of Dennett’s essay. He is arguing that this is typical of culture, of memes, but not of biology, of genes. While there are norms in the genetic realm, they are considerably more restricted.

Here Dennett sets out the parallel (p. 139):

DNA has an alphabet—the famous ACGT—and words, the three-letter codons that “spell” the twenty amino acids. In fact, the high fidelity of genetic transmission depends on the subcellular machinery being triggered to “recognize” and “re-produce” a small repertoire of types, whose idiosyncracies, if any, are ignored, not slavishly copied: “make a cytosine,” “make a guanine,” etc. There are error-correcting enzymes as well, but they don’t ascend (as far as we know!) above the level of a spell checker, correcting “typos” by brute template matching against the original.

So far so good. A bit later Dennett tells us:

Put otherwise, DNA error-correcting enzymes have always responded to semantic norms, but just local or proximal semantic norms—make a G—as contrasted to more distal semantic norms—make a codon for asparagine or make some lysozyme or make a protein that blocks serotonin uptake, or even make something that will fight off infection.

And now things get interesting. Once we start talking about making lysozyme we’ve left the realm of DNA replication and are talking about something else, either organismic maintenance or ontogenesis. We’re in the phenotypic realm rather than the genetic realm.

That’s where I have a problem with Dennett. In the linguistic examples, from orthography, to spelling, syntax and semantics, Dennett places the whole bunch in the memetics realm. When he then offers a set of biological examples from, from “make a cytosine” through “make something that will fight off infection”, he sets them in parallel with his linguistic series, but fails to acknowledge that his biological examples span two spheres, that of genes and that of phenotypes. I want to make a similar distinction in the linguistic bunch, placing the most localized cases in the memetic realm (upper vs. lower case, perhaps spelling) and the rest in a different realm, whatever you want to call the cultural correlate of biological phenotypes.

Dennett sees the argument he developed in the linguistic example as being completely general (pp. 141-142):

When Sperber notes that in cultural transmission “the information provided by the stimulus is complemented with information already in the system” (2000: 171) he is right, but the same is true of DNA replication. The main difference, so far as I can see, is that unlike DNA replication, human cultural replication is accomplished by processes of highly variable semantic depth, responding to perceived (and misperceived) “copying” errors relative to norms at many levels. The alphabets of written languages provide us with the most vivid and best-understood system of such norms of replication, but the phenomenon of semantic norms is not directly tied to language. Musical notation relies on the staff to digitalize the roughly inked spots, so that a musician can see at a glance that a chord is A-C#-E-G even though the A is written almost twice as far beneath the staff as it is “supposed” to be. A sketch of a new sort of axle for a wagon need not make the wheels exactly round; the user of the sketch will recognize those irregular closed curves as representations of wheels, which are to be round, of course. As we move through our various apprenticeships in life, we learn to perceive new families of categories—new alphabets, in an extended sense—from which to construct high-fidelity copies. Only a skilled potter can see at a glance what another potter is doing and copy it or teach it to others.

While I believe he is correct in insisting upon the generality, I also think that, in every case, he needs to distinguish between memetic ‘triggers’ and the complex cascade of mental activity that swirls around them. In my view the memetic triggers are out there in the physical world, they are the properties of things, events, and processes. The mental swirls and eddies, they are the cultural phenotypes, the cultural interactors. These are the entities that on which cultural selection operates.

And THAT’s why I am insisting, that in reasoning about cultural evolution, we must distinguish between memetic realm and some other realm, the realm of cultural phenotypes. Frankly, I’m sorely tempted simply to call this other realm by a common name, the mind or the mental realm. But won’t, not just yet.

THAT was just a sudden impulse, one whose implications I’ve not had time to consider. But, given Dennett’s propensity to view the mind as a stew of memes, that’s a terminological matter that would but our difference quite clearly.

I also realize that, at this point, I’ve not really argued my case, not in this post, though I’ve certainly argued it elsewhere. And I’ll argue it some more. But not here and not now.

More later.

* * * * *

I drafted the body of this post yesterday, July 1, but wrote the introductory paragraphs this morning (July 2). You may notice that in the second paragraph I suggest that the cultural correlates of biological phenotypes are mental entities. I’m still liking that usage. But such things require more than a day to simmer.

  • Tim Tyler

    Dennett on meme phenotypes:

    "Memes are supposed to be analogous to genes, the replicating entities of the cultural media, but they also have vehicles, or phenotypes; they are like not-so-naked genes. They are like viruses (Dawkins, 1993). As with viruses, there is a phenotype/genotype distinction, but just barely."

    I think he acknowledges meme phenotypes - though it seems to be true that he understates them. Meme products include skyscrapers and aeroplanes - they are pretty significant, and have gone far beyond the 'virus' stage.

  • Yeah, he acknowledges them, back in 1998 (a draft), and then simply ignores them. The paper I'm working from, which is more recent (drafted in 2000, published in 2006), is as close as Dennett gets to a detailed discussion of mechanisms. That's where one would expect some discussion of cultural phenotypes, he yet says nothing about them. He doesn't mention them in his 2009 Cold Spring Harbor paper, which would have been for a technical audience), nor in his 2010 encyclopedia article. I can only conclude that, as far as he's concerned, they are of no account. That's a major slip-up on his part.

  • Tim Tyler

    Another reference (1990) where he gives examples:

    "Memes are also invisible, and are carried by meme-vehicles - pictures, books, sayings (in particular languages, oral or written, on paper or magnetically encoded, etc."

    Dennett does recognize meme products.

  • Hone Pākehā

    I think forcing the concepts of identity and replication to culture is misleading. Cognition is
    way too complex and as Sperber and Claidière (Biological Theory 1, 2006, 20-22) point out, macrostability doesn’t imply microheritability.

    Dennett’s illusion of cultural self-reproduction is probably rooted in a belief in Fodor’s Language of Thought (Harvard University Press, 1975). If there’s a Mentalese vocabulary the reproduction of an idea is the translation of some ‘text’ to a natural language and then its retranslation to Mentalese by another individual. However, verbal comprehension isn’t simply a deterministic translation. Introducing inference is still not good news for Dennett because ostensive-inferential communication relies on a non-demonstrative logic, too. Somewhere on the early pages of his Intentional Stance (MIT Press, 1987), he says that more or less 90% of human communication is successful. Again, if we define success by the rate of identical Mentalese words, this sounds plausible. But if you adopt a complex, stochastic view of human comprehension, say the only plausible framework in contemporary pragmatics: Relevance Theory (Sperber and Wilson, Relevance (2nd ed.), Blackwell, 1995), communicative success turns out to be merely apparent. The catch is that failure becomes obvious only when audiences do something wildly different from what speakers expect as the result of communication.

    As far as I understand it, the real take-home lesson of Sperber (2000) was the following:
    On one hand, biological selection happens on the phenotypic level (related to the genotype in a non-deterministic way) but replication takes place in the genes via a less than perfect but deterministic way. (Testers check out a new computer game and based on their feedback the code of the most entertaining version will be copied and developed further.)

    On the other hand, a supposed cultural replication would take place at the same level as selection, that is at the ideotype(=cultural phenotype), which in turn is “derived” non-deterministically from the memotype. Replication itself would also happen via non-demonstrative logic and thus the new “meme” would be already two non-determinisms far from its parent “meme”. (Testers check the game and then,
    after a few failed attempts to break the code, they decide to write a similarly functioning one, starting its code from a blank screen.)

    As a consequence of top-down comprehension procedures, the constituents (whatever they are) of ‘A’ are typically vastly different from those of ‘B’ and no correspondence (and
    inheritance) can be posited between constituents. Neither can these concepts be applied on higher levels because of the promiscuity of cultural causality, and because partition is so contingent on actual interpretations. Dennett's finite repertoire (Mentalese words?) doesn't exist in culture.

    Shortly, identity and correspondence between cultural items are very problematic; therefore heritability, copying fidelity, selective pressure, reproductive fitness, etc. are not very good ideas in culture. Which means that cultural artifacts are newly created rather than replicated and Sperber is right and Dennett is wrong.

  • “As a consequence of top-down comprehension procedures, the constituents (whatever they are) of ‘A’ are typically vastly different from those of ‘B’ and no correspondence (and inheritance) can be posited between constituents.”

    A and B? Are these the cultural productions of two individuals interacting where one is ‘copying’ the other?

    But, yes, Dennett gets language wrong. He talks as though, when Jack says something to Jill, he’s just sending tokens of mentalese through a tube from which Jill extracts them. Hence what Jill gets out of the tube is exactly what Jack put into it. Language doesn’t work like that at all.

  • Hone Pākehā

    Yes, I meant A and B as in Sperber's (2000) three conditions for true replication.

  • Tim Tyler

    Dan Sperber and Nicolas Claidie seem simply confused to me. They say:

    "we believe that a proper understanding of the mechanisms of cultural propagation drawing on the work of cognitive and social scientists [snip refs] contradicts the idea that culture exhibits inheritance in the strict sense needed for the theory of evolution by natural selection to apply straightforwardly to it."

    Culture does exhibit inheritance - and evolution by natural selection applies pretty "straightforwardly" to it. Sperber and Claidie's counter-argument (invoking low fidelity transmission and reconstruction) is the same feeble one that Sperber's been presenting for years. Claiming that "that high fidelity is the exception rather than the rule" doesn't wash too well - once you get an internet connection and learn about the billions of identical copies out there. Lots of culture persists because of pretty ordinary and simple inheritance mechanisms involving direct copying - and that's been true since biblical times.

  • Tim Tyler

    Language actually can work quite a bit like that. It is designed to work that way. Several features of language contribute: the quantized, digital nature of words, the level of redundancy in spoken sentences and the existence of shared dictionaries. Digitization and redundancy combine to produce linguistic error correction. These language features mean that it's possible to play a game of Chinese whispers - and have serial chain transmission actually work.

  • Alas, not really. This observation applies ONLY to the physical form language takes, NOT to meaning. See my most recent post for more discussion of this:

  • And BEFORE biblical times? And just what portion of the population could read and could afford to own hand-copied written texts in biblical times? And forget about the internet. Those billions of copies, as you say, are utterly trivial. But coming up with an account of culture that works in the preliterate world is not at all trivial. That's the account we need.

  • Tim Tyler

    Some here may be interested in human evolution. Others might be interested in the present day and the future prospects. The past is all very well, but the main reason for studying it is to help us shape the future. So: don't let's write off the internet!

    It's true that most theories of cultural evolution should work equally well with current culture, or culture that is millions of years old. However, the significance of cultural transmission fades as you go back in time. Before ten thousand years ago, there was little division of labour, meaning that culture had to fit in an individual's head. Without good copying, memes would decay almost as fast as they could adapt, limiting cumulative adaptive evolution. Plus, ancient history is hard to investigate, and it's tricky perform relevant lab experiments.

    One of the major failings of cultural evolution in academia has been its obsession with the ancient past. It's been "lactose tolerance" this and "malaria resistance" that. Such a perspective makes very little sense. The modern world has a great abundance of riches for students of cultural evolution. I think that workers in the area should be fully exploiting these resources.

  • Hone Pākehā

    So you say that historical cultural evolution is not that important because memetics could not be accommodated to a little-division-of-labour society; so let's concentrate on internet-based, strictly copying communication because memetics can be applied to that?
    In other words, cultural evolution should be modelled in a simple, analogical framework because that fits a certain domain of interest; and that domain should be preferred because our chosen framework cannot be defended outside that domain?

    I can't see how the authors' insistence on proper (i.e. sufficiently broad) understanding is confused. I think confusion starts when you begin a circular argument and arbitrarily narrow down the explanandum.
    Darwinism couldn't be conceived without an "obsession with the ancient past," which you simply write off as a "major failing of cultural evolution in academia".

    If I'm misrepresenting your line or argumentation please excuse me. I'm still reliant on a pre-biblical technology of communication. If you could upload your brain-contents to the internet I could truly copy what you think instead of just, while guessing from the tiny pixels you've left on my screen, creating my own version.
    Your "future prospects" seem to need not only such a brain-copying system but also the total abolition of ostensive-inferential communication and a blockage of evolved cognitive mechanisms that feed such a communication before non-interpretative brain-to-brain copying can become a plausible model of cultural evolution.

  • I agree with you on this, of course. Tyler and Dennett (and others of their persuasion) are assuming that language is created and maintained by a committee of Designers who themselves stand outside the language system. I suppose this committee updates our minds when we are asleep; perhaps that's what dreams are.

    The point of Darwinian evolution is that it doesn't need such an external Designer.

  • Tim Tyler

    Fortunately, both Dennett and I are on record with opinions diametrically opposed to the positions you are attributing to us.

    Here's Dennett at a 2009 Harvard lecture:

    "If we think of words as virtual machines, then the question is 'who designed them?', and the answer is: 'nobody, evolution designed them'."

    I have an essay titled: "Tim Tyler - Are most words intelligently designed?" addressing exactly this question - and arguing against the position you attributing to me. You can check that out at:

  • Sure, that's what Dennett says. I wouldn't expect otherwise. But what he actually proposes about language contradicts that statements. What he actually proposes requires that language be designed from a POV outside language. He hasn't faced up to that contradiction in his thinking, and neither have you.

  • Tim Tyler

    Bill, you are not being clear. You haven't pointed out any contradiction - as far as I can see. There are just some vague accusations - which seem hardly worth bothering to respond to. Perhaps either spit out your alleged issue - or sit on it until you can put more resources into expressing it properly.