Replication and Reconstruction. A quick note on Scott-Phillips 2017

An interesting article from Thom Scott-Phillips has been recently published in the Journal of Cognition and Culture: A (Simple) Experimental Demonstration that Cultural Evolution is not Replicative but Reconstructive – and an Explanations of Why this Difference Matters.

The article describes an experiment that nicely illustrates (“make flesh” in the words of Scott-Phillips) a thought experiment proposed by Dan Sperber. Shortly, imagine a Chinese whispers game, in which chains of individuals have to reproduce two drawings. One is a familiar configuration (in the specific case, the first three letters of the latin alphabet), while the other one is a meaningless scribble (see the image below, from Scott-Phillips’ paper).


Throughout the transmission chains, the former shape is maintained, whereas the latter is not, and it tends to be transformed in something more simple and symmetrical (see the image below, from Scott-Phillips’ paper – “Reconstruction conditions”).


While these results are quite intuitive, I agree with Scott-Phillips (and Sperber) that they exemplify an important point: cultural stability does not need to be supported by high-fidelity social learning. Perhaps the stimulus chosen as “attractor” (i.e. “ABC”) may be misleading as, in fact, it is an attractor only in virtue of the existence of previous high-fidelity social learning: “ABC” would probably not remain stable in a chain composed by individuals that are not familiar with the latin alphabet. But the argument is more general: “attractor” is a statistical notion, and it is a description of what needs to be explained. In this case, previous knowledge of the latin alphabet is one of the “factors of attraction” (I think) that explains the stability of the chain.

From here there are – I agree – far reaching consequences. First, it does not make much sense to concentrate our explicative efforts on a “special” human capacity for imitation or similar. Second, models of cultural evolution that consider the process of transmission as a “replication plus random error” of cultural traits are not modelling what is interesting to model.

Scott-Phillips underlines an important point when he writes:

if […] propagation is reconstructive, then stability arises from the fact that a subclass of cultural types are easily re-producible, while others are not, and hence a casual explanation of stability comes from an explanation of  some types are easily re-producible (and why they are re-produced), while others are not. What differentiates the reproducible from the unreproducible?

Scott-Phillips discusses all the above in a clear and concise way so, if you do not know what I am talking about, please have a look at the paper!

Said so (of course this was coming at some point!), I think that the experiment does not show that “cultural propagation is reconstructive”. In the “Replications Conditions” individuals do not have any problem to reproduce almost exactly the same scribble throughout the chain (see the figure above), showing that cultural transmission can be also strongly preservative, independently from the stimulus presented. The change of the condition is as simple as explicitly asking to the participants to trace the image.

Don’t get me wrong: I believe that the point stressed by Scott-Phillips is extremely important and overlooked by “standard” cultural evolution. I am myself working on very similar matters, and I hope that all these works will converge in a more cognitively-oriented cultural evolution theory. In the same time I feel that to convince cultural evolutionists of the importance of reconstruction in cultural transmission one needs to show that (i) “real” propagation – cultural transmission in the wild – is more similar to the “Reconstruction” than to the “Replication” condition of the experiment, and (ii) this creates interesting differences in the resulting population-level dynamics (probably through modelling). Otherwise, cultural evolutionists will have good arguments to claim that black-boxing cognition is a reasonable strategy when explaining culture. So, let’s all get back to work!

9 thoughts on “Replication and Reconstruction. A quick note on Scott-Phillips 2017

    1. Yes, I mentioned above that ‘ABC’ may be an attractor only for individuals familiar with the latin alphabet. I would expect Chinese characters or Arabic scripts to have similar effect of the scribble.

  1. Hi. I replied in a recent blog post titled: “Attraction in conventional evolutionary theory”. To paraphrase the most relevant bit:

    Reconstruction is not confined to cultural evolution. It happens in the organic realm as well. Stability of DNA-based creatures is not explained simply by invoking high-fidelity copying. Living fossils illustrate stability comes from other sources. Nobody in their right mind would argue that Alligators or Ginkgo Biloba trees resemble their ancestors from millions of years ago only because of high fidelity copying. That would be failing to give longevity and fecundity their due. It’s not that mutations affecting leaf shapes and leg lengths never arise due to high copying fidelity. Rather these stable forms represent adaptive peaks: sweet spots in the fitness landscape that are hard to improve on. In dynamical systems theory such spots are sometimes known as ‘attractors’ in state space.

    In both organic and cultural evolution, stability is explained by a mixture of high copying fidelity, longevity and fecundity. Characterizing stability in organic evolution as only the result of copying fidelity is a mistake. In both organic and cultural realms, some entities are also better at reproducing themselves than other ones, and are more long-lived than other ones. The fitness landscapes they evolve on have stable adaptive peaks that result in stable forms that can last for hundreds of millions of years. We can use much the same theory of stability on adaptive peaks (or ‘attraction’ if you prefer) in both organic and cultural evolution.

  2. Hi Alberto. Thanks again for bringing attention to my paper, and for the thoughtful commentary.

    An astute reader has brought to attention that there may be a slight misreading of the paper. You write that “The change of the condition is as simple as explicitly asking to the participants to trace the image”. This is not correct. The difference between the two conditions, replication and reconstruction, was not in the instructions. On the contrary, the instructions were the essentially same in the two conditions (participants were just asked to draw the image). The key difference was the means by which they did this. In the reconstruction condition the participants were shown the image for two seconds, and asked to draw what they had seen. In the replication condition they placed the new piece of paper on top of the image, and traced it. The replication condition is effectively a copy-and-paste situation, the purpose of which is to provide a baseline comparison for what happens in the reconstruction condition. (I have learned, incidentally, that ‘trace’ does not translate well into Italian and some other romance languages. It means, in this context, to follow on the top piece of paper the still-visible lines of the bottom piece of paper. It is how I was taught, as young boy, how to write the letters of the alphabet.)

    This point should/might address the concerns you raise. You write that “individuals do not have any problem to reproduce almost exactly the same scribble throughout the chain”, but this is not true. The reconstitution condition shows that, absent the means to effectively copy-and-paste, individuals do struggle to accurately reproduce stimuli if those stimuli are not recognised as tokens of a type. If this was not the case, then the two seeds would have evolved in a similar manner, but they did not.

    In most cases of cultural transmission, the means to copy-and-paste are not available, and reconstruction is what happens. An exception is online – which is one reason, incidentally, why your general interest in the effects of digital technology on cultural transmission is apposite.

    1. Hi Thom,
      Thank you for your comment! Let me first reiterate that I am completely on the same page in respect of the main point of your experiment.
      Regarding your specific comment: you say the difference was not in the instruction, however in the paper you write:
      “In the Replication conditions, they were also given a A6 piece of blank paper and a pen, but in this case, they were asked to trace the image.”
      To me, this can be considered “instructions”, or, you would admit, it’s open to ambiguity as it is written.

      In any case, I do not think this matters too much. I think the main argument I was trying to elaborate in the post is – to paraphrase the last sentence of your comment – whether the means of copy-and-paste are not available in most cases of cultural transmission. You think this is the case and I agree (really!), but I also think that this needs to be validated empirically. Your experiment (I keep my opinion) shows that there are two options (copy-and-paste and reconstruction) and that they have different consequences (which is great!). However, you mention yourself that you learnt the alphabet through copy-and-paste, and that digital technologies have this effects. Another example I am thinking about: quite a few things in school must be learned by heart, from ways to solve divisions to more complicated maths stuff or similar, where would you put those? As I was implying in the post, finding a way to assess the “degree of copy-and-pasteness” in different domains and then map domains according to that would make for a nice paper.

      ps: you are right, in Italian “trace” would be intuitively translated as “tracciare” which is not what you meant, but I imagined you were indeed referring to “ricalcare” so I think I’ve got it from the beginning.

  3. To my eyes tmtyler is offering the beginning of a potentially more radical criticism of the distinction advanced by TomScottPhillips. Phillips proposes that cultural stability is the product not of replication, as in genetics or biology, but of reconstruction, which involves various kinds of variation, typing, and mechanisms of correction. Tmtyler objects that organic transmission – as in genetic inheritance – is not actually replication either, but also involves mechanisms of reconstruction. I would suggest that this too stops a step (or more) short: the same could be said in the inorganic realm as well, as in the reproduction of crystal structures or even, say, of the neutron, proton, and electron composition of atoms (i.e. elements). What Phillps refers to as “reconstruction” would not be a specifically “cognitive” model but a general ontological postulate (applying across domains of being), and “replication” would not be an opposed model but rather a subspecies of reproduction differentiated from other kinds of reconstruction by a high degree of fidelity. This would denature the opposition between replication and reconstruction, since reconstruction would be the general characteristic of transmission, reproduction, inheritance, identity etc. as such, while replication would just be a special case of reconstruction.

    1. Thank you for your comment! Thom Scott-Phillips – and in fact Dan Sperber, to whom ultimately all these ideas are due…- claim indeed that in cultural evolution replication is just a special case of reconstruction. You propose that this is true in all realms (not only in culture, and not even only in biology). This is an interesting point, and I am inclined to think that there is some true to it. However, I think the issue is that, at least for biology, it is convenient to think to transmission as analogous to replication in many cases. This (again this is my personal opinion) allowed to build parsimonious models that ended up to be very useful for the development of evolutionary biology. The question is: is this useful for cultural evolution too? Sperber would think it is not, while many others working in cultural evolution think it is. I am personally convinced that, while the criticism of Sperber & co. is legitimate, it is a way to frame interesting questions for which we do not have – yet – clear answers.

  4. I think some clarity in the debate can be obtained by tabooing the term “transmission” and using the term “copying” instead. Information is copied in both organic and cultural evolution (with some degree of fidelity). The alleged “reconstruction” process isn’t part of copying. Copying is relatively simple – either information is copied or it is not. Reconstruction would then be classified as part of the other components of the evolutionary algorithm: selection, mutation and merging/recombining.

    The idea that transmission includes reconstruction can then be identified as arising as a result of classifying multiple rounds of copying, selection and merging/recombining that take place inside individual minds as a single operation: “cultural transmission”. This kind of classification scheme can evidently obscure what is going on. It apparently leads to exceptionalism in cultural evolution – the idea that cultural evolution plays by its own special rules.

    The main problem here seems to be a substantial missing concept: mental or psychological evolution. There’s a tradition of expanding the application domain of Darwinism into psychology and neuroscience – starting with Skinner, Calvin and Edelman, Plotkin and Szathmary. The brain can optimize and generate adaptations too, and it does so in a manner consistent with known evolutionary principles operating on ideas, branching dendritic structures and branching electrical signals. See my article “On Memetics: Keeping Darwin in mind” for more about this.

    We do NOT need to tack “reconstruction” on to existing theories of cultural evolution! Evolutionary theory is already big enough to include organic, cultural and psychological evolution – and rather obviously, the latter includes reconstruction – and many other things. Scientists should try and establish simple and general theories with wide applicability – not have all kinds of special theoretical apparatus in different domains. Boyd, Richerson and Sperber all have a long history of not trying hard enough to fulfill this ideal.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s