Home » Blog

Speculations: Abstracting Evolution

3 January 2017 One Comment

This post is aiming to get down some thoughts around how the superset of evolutionary models can be framed. It’s almost certainly work that has been done somewhere before but I’m struggling to find it so it seemed useful to lay out what I’m looking for. 

Evolutionary models are extraordinarily powerful, in part because they are extremely flexible. At their worst they are tautologies – the criticism of “survival of the fittest” as an idea is well founded even if it’s description of Darwinism is not – but at their best they provide an understanding that stretches from our capacity to engineer molecules, to an understanding of our origins and possibilities, to the design and development of various forms of artificial intelligence, the optimisation algorithms that is based on, and our understanding of the development of culture.

Within that enormous scale there seems to be quite a lot of sloppiness. Much of the popular literature on biological evolution focusses on refining down to a strict definition of what evolution is, and what is evolving, largely as a response to the political challenge of creationism and intelligent design advocates. In turn these strict definitions fail to cover the emerging complexities of how different parts of biological systems, genes, genomes, cells, organisms, communities, species, ecologies are subject to selection and differential persistence. The simplifying assumption that biological evolution is equivalent to DNA sequence evolution is both enormously powerful and obviously incomplete. The easy parallels made with genetic optimization algorithms and the apparently “biological” look of items that sometimes arise from them makes for easy analogies but common analysis is harder.

One of the first scientific insights that captured my imagination was an intuition that biological systems are set up in such a way as to be evolvable. The layering of just enough chemical diversity in the form of proteins onto a neat sequential instruction set. The way a simple linear molecule is set up to enable huge structural diversity. Most fascinating of all these systems must necessarily have evolved to be evolvable. And yet at the same time they can be quite recalcitrant to our naive attempts to apply what we think of as the same evolutionary process – randomise gene, express protein, select for function – in the lab. Like most persistent systems and many biological systems, the molecular evolvability of proteins is both resistant to small scale perturbation, but clearly based on the reconstruction of evolution over the long term, immensely flexible in the face of existential challenges.

Twenty years later I’m working within a model of culture and groups that has as one of its central claims, that it is an evolutionary model. Attempts to transplant evolutionary framings into those domains traditionally seen as belonging to the  academic humanities have generally been unsuccessful. The claim of the Cultural Science program is that this has been due to a misdiagnosis of what it is that is under selection.

In cultural science it is meaningfulness that evolves, ‘demically’. However, it is misleading to represent this as referring to ‘units of meaningfulness’, as if merely seeking to re-label ‘basic cultural unit’ with ‘basic unit of meaningfulness’. Instead, our claim is ontological and seeks to develop the idea that cultural evolution is the emergence of meaningfulness from webs of associations and relations and also by negotiation and use within a deme and between demes. Importantly, this is not a thing, or even information per se, but a structure of associations in action. It is these dynamic demic associations that evolve.

Hartley and Potts (2015), Cultural Science p126

Ironically this is not a paragraph that will very meaningful to many evolutionary biologists. One important point is that identifying the “unit of selection” is a challenge for any evolutionary theory. Here it is claimed that the unit of selection for culture is inherently complex, a dynamic network of narratives and meanings. These can be probed using traditional humanistic and social sciences techniques, discourse analysis, close and distant reading, critique of framing, ethnography and sociology. My own internal analogy is that the results of these studies are to the true “genes of culture” as the image of bands on a gel (showing my age there!) are to the real operation of DNA in a complex organism – simplifications built on techniques that frame the underlying complexity in a way that both makes it comprehensible but also reinforces the framing of the technique.

John Welch in a recent paper makes a similar point in response to calls for reform of (biological) evolutionary theory.

It is argued that a few inescapable properties of the field make it prone to criticisms of predictable kinds, whether or not the criticisms have any merit. For example, the variety of living things and the complexity of evolution make it easy to generate data that seem revolutionary (e.g. exceptions to well-established generalizations, or neglected factors in evolution), and lead to disappointment with existing explanatory frameworks (with their high levels of abstraction, and limited predictive power).

Welch (2016), http://doi.org/10.1007/s10539-016-9557-8

Biological systems are damn good at co-opting effects, systems to their own end. It is in the nature of evolved and evolvable systems to be capable of opportunistically taking advantage of whatever they can get their hands on. If the anthropomorphic language bothers you; those systems/collectives more capable of extracting benefiting opportunistically from the environment will, all other things being equal, persist more consistently than those that do not. Eugene Koonin in another very recent paper cautions against focussing too much on selection, in comparison to the role of neutral sequence change and diversification. That attention to the mechanisms of diversification is as important as to those of selection is obvious. The assumption that diversification, or rather change, and selection, are coupled in some form to the idea of replication, and particularly that there is a sequential pattern to be followed is a less obvious error, but one common particularly in cases of modelling or algorithmic design. See for instance slide 12 in this slidedeck from Knowles and Watson (2017).

In the case of demic cultural evolution sketched above it is not clear we even expect replication. Cultures may divide and split but that is not a necessary part of the model1, rather that they compete and differentially survive. We can talk about the persistence and continuity of human microbiome (and its successful or unsuccessful transmission to a new-born child) in ways that sound Lamarckian at one level. That a microbiome evolves is clear. That it has an effect on the adaptive fitness of the host is clear, and that its characteristics likely have a long term effect on the population genetics of the host species(s) is at the very least plausible. To understand the whole system we need a highly sophisticated notion of what it is that is evolving. In practice we focus on the persistence of one identifiable object (genetic markers in the host, the presence of a specific sequence – or set of sequences – within the microbiome) for any given study. This is the point Welch makes, that there are principled reasons for focussing on the evolution of a specific identifiable object.

The reductionist agenda gave us the view of genes as digital DNA sequences and from that the full power of the evolutionary synthesis. One logical endpoint of Welch’s argument is that we have to accept this is not the full story that other elements of the system are evolving. Koonin cautions us to be more open to what forms of selection and diversification matter, but to my mind falls into the same trap, the assumption that all genes are DNA sequences, and therefore that all evolution can be examined through statistical analysis of DNA sequences. Evolution may “only make sense in the light of population genetics” as Lynch tells us, but the DNA sequence is, at least in my analogy, on the trace, of a trace of what the real “gene” is.

To generalise

It seems to me that question of identification must lie at the centre of any abstract model of evolutionary framing. An evolutionary model or framing is one that helps us analyse why a specific recognizable object persists. This is necessarily circular as we likely recognise an object precisely because it is in some sense persistent. The choice to focus on a particular object is contextual, and semantic, ideally supplemented by an analysis of the extent to which other framings can be neglected. The great success of the evolutionary synthesis rests on the fact that seeing DNA sequences as genes has enormous explanatory power. It is not complete but it gets us a long way.

It’s not clear to me that replication is a necessary component of an evolutionary model. Differential persistence amongst a collection of objects seems enough to me. Change or variation within objects is not strictly necessary, but systems without it would appear to be rather boring. Both starting a model with diversity, as population genetics generally does, or starting with consistency and then incorporating change, as is the case for evolutionary optimizations, are both feasible. In those cases where there is a Platonic conception that provides the means of recognising relevant objects, they will be formally equivalent through a time transformation. The assumption of a universal common ancestor as a way of identifying cognate genes is one example. A more trivial example is whether to assign time point zero in an computational optimisation prior to or after the first introduction of variation.

If replication is part of the model it can take many forms and is not necessarily linked to variation. Variation can take many forms. Differential persistence – selection – may take many forms. Differential persistence is linked in a complex way to those things outside of the objects of interest, broadly the environment, but the environment is also a set of evolving systems. All of these may be sequential or continuous or some combination of those. Some combinations of replication, variation, differential persistence and environment will be stable, some presumably will not. The interesting metaquestion is the characteristics of such systems that lead to different forms of behaviour and interesting, or useful, dynamics of the system as a whole. How can recognizable attributes of the object be linked to survival and can lineages be constructed? If recognition and identification is at the centre of these models then an inability to reconstruct lineage will appear the same as the loss of the object in question.

At the highest level, this kind of model does become tautologous. That which persists, persists, so long as we can recognise it. It is through the characterisation of objects and the dynamics of the system that we can return to scientific models of specific systems. These are all models, as soon as we focus our attention on one set of objects we recognise we are neglecting the full complexity of the system. But it does seem productive to consider how those different models can be compared to each other, and how we can understand the dynamics of differing systems. That in turn offers the opportunity to turn that back around and ask what are we missing, if we seek to focus on a different element (the persistence of a demic narrative, or of a particular marker in a microbiome), how does that change our view?

This may in turn allow us to understand how multiple evolving systems interact. E.O. Wilson’s Consilience remains probably the largest scale attempt to frame everything in evolutionary terms. Where Wilson fails to my mind is in finding a way to tackle the complexity of interaction between the layers he talks about, as though chemistry can neatly be separated from biology, biology from psychology, psychology from sociology and on and on. Nonetheless he makes a strong point about the success of these approaches in tackling specific classes of problem, those that respond well to a reductionist approach. What is missing is a means of putting those back together that works with the complexity of interactions between what were never really layers in the first place, merely convenient categorisations for our identification of objects.

I’m sure this work has been done somewhere, but I’m slightly at a loss as to which discipline it would be in.

1. Strictly speaking some form of splitting or replication is required for the model to continue in the long term if some cultures are extinguished but that’s a consequence not a central point.


  • Paweł Szczęsny

    I have a feeling I’ve been to the place you are currently visiting. And I came back disappointed. One of the major issues of making models of cultural evolution is that you cannot model outside of the system. This is actually way more problematic that it seems. Genes do not care what you think of them. When you are studying culture, your thoughts are meant to interact with the subject you are studying (talk to any cultural studies scientist), emancipate it, influence it. This is actually way more explicit than I could believe at first glance.

    In other words, scientific community will actively reject or adapt your model. Think how impact factor was born and what’s its role at the moment.

    Maybe the question isn’t how to reasonably build an evolutionary model of science, but how to build a model that is reasonable?