Evolving ideas

(John Jacob Lyons, 18 September 2009)

The four dimensions of evolution identified in Jablonka and Lamb’s book ‘Evolution in Four Dimensions’ were described as; genetic, epigenetic, behavioural and symbolic. Although this is an extremely useful book, the dimensions proposed are not considered the best basis on which to carry forward research on evolutionary theory. The ‘symbolic’ dimension relates to one particular aspect of cultural evolution and, although there can be interaction between adaptive cultural change and genomic change, the use of the word ‘symbolic’ appears inappropriate and too narrow to denote this interaction. Genetic and epigenetic processes often operate in tandem and it may be misleading to nominate them as separate dimensions. An important deficiency is that the nominated dimensions are not from the same class of objects. They are not separate and different evolutionary processes and neither are they separate, independently generated drivers of evolution.

I want to suggest three independently generated, primary drivers of evolution at organism level that, I believe, would provide a better basis for future research; random genetic mutation, natural environmental change and behavioural change – including niche construction.

These categories can all be described as independent drivers of evolution. They are mutually exclusive but, of course, not exhaustive at this level of analysis. Sexual selection is a powerful driver and species would still evolve within the constraints of their existing gene-pool and without considering new mutations. Then there are the effects caused by artificial human selection; both within the human species (eg contraception) and those imposed on other species (eg selective breeding). However, I have highlighted three evolutionary drivers that, I believe, have been somewhat confused in previous work in evolutionary theory.

Each of these primary drivers is described below in a little more detail below.

Driver 1 – Random genetic mutation

This is often called (Classical) Darwinian Evolution. A mutation of this kind will be adaptive if it results in an adaptive phenotype; either increasing the viability/ fecundity of the organism or increasing both. The expected incidence of the mutation in the next generation will increase and will increase further in future generations as long as the mutation continues to be adaptive.

Driver 2 – Natural environment change

Particular extant alleles of particular genes/ epigenetic-markers will be more adaptive in the changed environment, again resulting in an expected increase in the incidence of this particular allele-set in future generations.

Driver 3 – Behavioural change – including niche construction

Initially, one or a small number of organisms learn a behaviour that proves to be adaptive. This behaviour may or may not affect the physical environment. Manifestation of the adaptive behaviour will often be positively and causally correlated with particular allele-sets that make the organism more likely to manifest the behaviour.  Selection pressure will result in an increase in the expected incidence of the allele-sets’ constituents in the following generation (see Behavioural Genetic Priming). This will, in turn, increase the expected number of organisms manifesting the behaviour in that generation. This positive feedback loop will continue to increase the incidence of the adaptive behaviour in future generations. In due course, all organisms in a population may be, to a greater or lesser extent, genetically primed to manifest the behaviour with a minimal trigger from the environment. I say “to a greater or lesser extent” since there may well be other adaptive behaviours and other sources of selection pressure in play that are underpinned by overlapping gene-sets. This would eventually result in optimal allele configurations for the complete set of adaptive behaviours and other sources of selection pressure but a suboptimal configuration for any particular adaptive behaviour.

Driver 3 can be described as an adaptive acquired characteristic and, although the characteristic itself cannot be assimilated into the genome, its positive association with concomitant, genetically mediated predispositions means that it is ‘able’ to prime the genome to increase manifestation.

If we add back the other drivers of evolution previously mentioned we get:

Driver 4 – Sexual Selection

Driver 5 – Natural Selection within current gene-pool

Driver 6 – Artificial Human Selection

Driver 7 – Hybridisation

Is this an exhaustive set of evolutionary drivers?

(prepared 14 June 2009 by John Jacob Lyons)

The ‘Baldwin Effect’ has been defined as an adaptive trait change in an organism – that occurs as a result of its interaction with its environment- becoming gradually assimilated into its developmental genetic/ epigenetic repertoire. My ‘genetic priming’ hypothesis seeks to clarify the process involved in the Baldwin Effect and to correct several points in the definition above that relate to the effect of this process.

The process consists of an inter-generational positive feedback loop between the adaptive trait and the positively and causally correlated subset of the genome that tends to support the trait. This is described in greater detail in my article in the ‘Evolving Ideas’ blog. As stated there, selective pressure will result in the manifestation of the trait earlier and earlier in the development of the organism.

However, the adaptive trait will never be “ – assimilated into its developmental genetic/ epigenetic repertoire.” What will happen is that selective pressure will result in the allele-configuration of the relevant subset of genes ‘trying’ to change in order to optimize the support given to the trait. In doing so, it may well be in competition with other adaptive traits ‘trying’ to optimize the subset to support them. This will obviously result in sub-optimization for any particular individual trait. The ultimate result will be that the genome will have been, within the aforementioned constraint, primed to support the manifestation of the original adaptive trait. An example will make some of these points clear.

It is well known that, over the human EEA, post-weaning lactose tolerance developed in temperate regions in which cattle were farmed. In this case, the adaptive trait was milk consumption (Vitamin D enables absorption of calcium; particularly adaptive in temperate regions) and the correlated subset of the genome was that involved in controlling lactose tolerance. The positive feedback explained above has resulted, in these regions, in the ubiquitous priming of the human genome toward post-weaning lactose tolerance and not in a genetically assimilated tendency to consume milk! There has been no assimilation of the adaptive behaviour; only genetic priming of the associated subset of the genome.

I suggest that, subject also to many relevant cultural evolutionary factors of course, humans have been genetically primed for language, religiosity, morality and love/ attachment/ empathy in a similar way.

Should you give to charity? Looking at the arguments associated with two recent books it would seem that you are damned if you do and damned if you don’t. But take a second look from an evolutionary perspective and it seems that sins of omission and of commission inhere at a different grain of analysis.

The first book is Peter Singer’s The Life You Can Save: Acting Now to End World Poverty (argument summarised here). Singer is raising consciousness. Writing to an American audience he lays out the facts about how little America actually gives in aid and how much less it gives to the very poorest countries. En route he dispatches arguments that America’s large private sector offsets the lack of public largesse – it doesn’t. But Singer doesn’t stop here. With his utilitarian bent on full display, he compares these aid figures with the amount the US government is willing to pay to save American lives (from road traffic accidents for example) and the amount it costs to save African lives. This gives a rough figure of how much America values the lives of its citizens over others: ten thousand to one. Singer acknowledges the inevitable inequity (the expanding moral circle is not a Boolean guide to action) but questions its degree. He also discusses the effectiveness of aid and points to the effectiveness of at least some interventions while calling for more research (even arguing for prospective control studies – given that there is never enough aid for all who need it). The conclusion is that people in developed countries need to be more generous because their money is more valuable in developing countries than in their own pockets even if some is squandered by the corrupt or unwisely spent by the inept. Acknowledging uncertainty in this way is a clever argumentative strategy on his part and one that seems to turn the tables on opponents of aid who fail to consider the least convenient possible world (or even a minimally inconvenient counterfactual). Thus we are cogently coaxed away from our moral complacency.

But then there is the second book, by economist Dambisa Moyo: Dead Aid: Why Aid Is Not Working and How There is a Better Way for Africa (she is interviewed here). She and Singer don’t seem to have engaged directly so far as I can tell, but her arguments seem effective counters to Singer’s. It would indeed be complacent not to give to charity if at least some good came of it, but what if it did more harm than good? More Africans are poor now than in the past, not just more people, but a vastly higher proportion – this despite a deluge of aid. Aid is sequestered by and sustains corrupt regimes (and yes this still includes Zimbabwe). And while Singer argues that NGOs circumvent this by direct work in the community, she makes the general argument that this damages the credibility of the relatively less effectual government, further weakening democracy. Deeper than this, in fact, aid can directly interefere with the means of production within Africa – she gives the example of 160 mosquito net makers and their dependents who were taken out of business and themselves became dependent on aid as a result of the arrival of nets from abroad. So Moyo wants to shake us from a different sort of complacence – that which favours pity over reason. And she has positive things to say about trade. In a qualified way she praises China for the huge amount it has done for Africa compared with Europe and America simply by trading with Africans (there is an implicit moral point here that traders interact as equal partners while aid leads to dependency). She is a fan of microcredit and recommends this website where you can lend small sums to shop owners, agriculturalists, etc. all over the world who have an effective plan.

So one way of reading this is as a barney (argument) about the effectiveness of aid. But when you attend to the details (especially those in Moyo’s argument) perhaps a “third way” presents itself. What is interesting about microcredit, for example, it that it is somewhere between trade and aid – as a lender you do not charge interest so you are foregoing interest you might have earned. So does the lender take on this opportunity cost out of pity? To me this looks like enlightened generosity – the right mix of sentiment and reasoning on the lender’s part and the right mix of incentives at system level to benefit people in poor countries.

I favour Moyo’s point that aid can be counterproductive although I think this might be because it hasn’t been evidence based and therefore this might be fixable. Of particular concern given the patchwork distribution of aid (and the prevalance of religious charities) is the lack of available contraception. A recent study in PLoS Medicine illustrates this with results that are beautifully congruent with evolutionary theory (or at least the trade-offs assumed in life history theory). Mhairi Gibson and Ruth Mace carried out a retrospective study in which they assessed the effects of the installation of water taps (faucets) in some villages (and not in others) on mortality and health in children and mothers. It seems that the decreased maternal effort associated with obtaining water results in energy being redirected towards increased fecundity. I limn from a multivariate analysis here, but essentially villages with a water supply showed decreases in infant mortality and it seems that fertility increased in mothers with access to taps, but this was associated with no better maternal health and, in fact, with poorer infant health including a greater risk of malnourishment and stunting. The authors point to the growing African population as a severe problem (and here I am reminded of Jared Diamond’s controversial suggestion regarding the causes of the Rwandan genocide) unless interventions are designed to address fertility in concert with more traditional development goals.

The call for more research ought to be heeded, but in our current state of partial ignorance I would suggest that Moyo’s argument that aid can be counterproductive is not a feeble excuse for inaction but a reasonable concern.

Postnote: Charlie Rose (as ever) conducts a courteous but thoroughgoing interview with Dambisa Moyo raising Singer’s arguments briefly and many of the most interesting issues discussed here…

I just went to a talk today in my institution by Professor Robert Pennock who spoke a little about education and a little about artificial evolution of intelligence. The latter stole the limelight due to the intrinsic coolness of robots (real or simulated) with goal-directed behaviour. But among his many departmental affiliations at Michigan State is the Philosophy Department so I was interested to hear what he had to say on this score. No doubt because this was a general lecture this content was mostly geared towards understanding why the argument for design is incorrect – familiar territory for evolutionists (though I liked his updating Paley’s argument from watch implies watchmaker to neat iPod app implies SDK programmer). But my parenthetically referring to robots as simulated above was deliberate. He made what I thought was a small but elegant point about this.

Is artificial evolution simulated evolution or an instance of evolution? Pennock argued that the distinction between these descriptions depended partly on pragmatics and partly on what causal processes are of interest. Since the artificial evolution paradigm he discussed modelled the minimal parameters for evolution by natural selection he was happy to describe it as an instantiation even though the mechanisms of metabolism and reproduction were not quite nature-identical or gooey enough. He pointed out that these mechanisms were not known to Darwin and yet we would still call his theory a theory of evolution.

To my mind this is equivalent to the claim that evolution is substrate-neutral. However in so far as parameters necessary for selection are dependent upon genetic mechanisms we might dispute this. In fact the metabolic and reproductive rules in the software he described entailed these parameters (often explicitly – he affirmed that recombination, a source of variation that can break up clonal interference, was modelled in some of his project). So I agree with his implicit point that modelling can be more or less precise along different dimensions such that there is an arbitrariness about the dividing line between simulation/instantiation unless a modeller’s purpose is borne in mind. But I am not sure that specifying what is modelled, and how precisely, is enough. This is because evolution is diverse and is caused by multiple processes at the population genetic level. This is clear in my field of wet-lab experimental evolution in which large populations and strong selection pressures often inflate the perceived important of selection.

Finally he was asked a semantic question about the distinction between machine learning and artificial evolution – a particularly germane one since he was describing the evolution of intelligent systems. Essentially he described the second as an example of the first in which genetical algorithms did the learning by selection. But this calls to mind an earlier distinction I remember hearing (apologies for not attributing) regarding robots that learn about their environment through feedback. An effective robot would understand its environment, i.e., whittle down its set of models, by taking actions that resulted in the maximum discrepancy between the predictions of each model. But when it came time to act in the environment, the best actions were those that minimised the predicted discrepancy between surviving models. (Something like this might be relevant to earlier efforts on this blog to explain the functions of play). I wonder if, in particularly rugged fitness landscapes, this distinction between testing and optimal inference may be important when it comes to design assisted by genetical algorithms. The ability to allow heritable variation in mutation rate may become more important in such instances.

I’ve posted a few times on pleiotropy. Pleiotropy, or the multiple causal connections that a given gene is involved in, is the bête noire of evolution by selection because it increases the probability that a mutation at that gene will be deleterious (even though it may be beneficial within one context). Modularity is required at the gene level to allow functional change to occur by selection and this has motivated some to propose that cis-regulatory elements (CREs) are crucial to the evolution and development of form.

But modularity can also occur at the level of genetic networks and it is often asserted that such networks can be co-opted in their entirety to fulfill novel (non-homologous) functions. In the context of testing this idea, this recent article addresses the consequences of this for the pleiotropy of CREs. If network co-option really is that common, pleiotropic CREs will also be common at least among those genes that tend to be found within networks rather than upstream. This seems to be a nice example of entrenchment in evolution.

Post Author: John Jacob Lyons

In their book, “The Four Dimensions of Evolution”, the Waddington ‘canalization’ explanation of the genetic assimilation of adaptive acquired characteristics is referenced (p.262) and tacitly accepted by Eva Jablonka and her co-author, Marion Lamb. I don’t find this explanation at all convincing and want to propose my own explanation. I suggest that adaptive acquired characteristics are always positively and causally correlated with concomitant, genetically generated propensities. These propensities gradually become more prevalent in the gene-pool because of the success of the positively correlated acquired characteristic. In time, the organism will appear to be primed to acquire the adaptive characteristic. It is suggested that examples in humans are language and religion.

Suppose that a particular mutation (M) that appears at generation n increases the capacity to learn an adaptive behaviour (AB). AB will have a selective advantage and consequently the relative frequency of M in the population will increase in generation n+1. This will, in turn, increase the frequency of AB in this generation. So long as AB remains adaptive, this positive feedback loop will, over evolutionary time, lead to all organisms in the population having mutation M and exhibiting adaptive behaviour AB. Additionally, selective pressure will result in AB appearing earlier and earlier in the lifetime of organisms. In due course, it will appear that all organisms in the species are primed to acquire the AB.

As stated, I believe that two examples of this process in humans are language and religion. This would account for the innate ‘Language Acquisition Device’ hypothesized by Noam Chomsky and Precocious Religious Belief hypothesized and empirically demonstrated by, among others, Justin Barrett (Centre for Anthropology and Mind, Oxford University). I don’t believe that an adaptive acquired characteristic is ever genetically assimilated as proposed by Waddington. In other words, I don’t accept that the Weismann Barrier between somatic and germ cells is ever crossed in these circumstances. Rather it is as if the constituents of the genetic soil, as it were, are gradually optimized to promote the germination and growth of the AB seed. In the case of religion, the seed of belief/ faith may be provided by the parent explaining to the child that their sadly expired pet kitten, Tiddles, is now “with god in heaven” and reinforced by similar references later on. In language, the innate universal grammar proposed by Chomsky and others, may be characterized in a similar way with the heard phonemes, words, syntax and grammatical exceptions of the native language providing the seeds.

It is also suggested that the niche construction and extensions to phenotype seen in many species of animal may also have a similar origin. These could well have originated as behaviours that proved to be adaptive and that, eventually, resulted in the concomitant, positively correlated and genetically mediated allele-sets becoming ubiquitous in the species. These would then have primed the young organism to reproduce the behaviour with minimal exposure to the behaviour by others.

I recently attended a workshop at my institution covering a broad range of evolutionary topics (including many exciting hominin fossil finds in the offing it seems). The last talk of the day was Adam Siepel of Cornell University who gave a talk about positively selected genes reflecting this recent paper in PLoS Genetics. His group used well-sequenced genomes of human, chimp, macaque, mouse, rat and dog to infer positive selection in protein-coding genes using likelihood ratio tests. Bayesian methods were  used to establish likely selection histories (which suggested frequent state changes along the phylogeny implying that positive selection occurs and re-occurs over short intervals).

Interestingly positively selected genes tended to be expressed at lower levels and with more tissue-specific expression patterns. This mirrors inverse/positive correlations found elsewhere between substitution rate and expression level/tissue-specificity. More subtly, though, the inverse correlation between substitution rate and expression level was more pronounced in genes not subject to positive selection indicating that responses to purifying selection might be responsible for most of the trend. Thus: “It appears that genes may be more likely to come under positive selection if they are in a state of evolutionary flexibility brought on by reduced or tissue-specific expression, but once positive selection has taken hold their subsequent evolutionary course is not strongly dependent on their expression patterns.”

The inverse correlation with expression may relate to selection against protein-misfolding but the picture with tissue-specificity is about the costs imposed by pleiotropy and is interesting to me because I recently ran an argument about escaping pleiotropy in a recent paper with Samir Wadhawan and my advisor Anton Nekrutenko about the mammalian Gnas locus. The Gnas locus is formidably complex. Crudely it consists of multiple alternative transcripts which share downstream exons but which differ in first exons. These transcripts are subject to complex patterns of tissue-specific and imprinted gene expression with the non-canonical transcripts showing tissue-specificity. We observed an increased rate of evolution (by likelihood ratio testing of dN/dS) among exons unique to these alternative tissue-restricted and imprinted transcripts. I argued that the selection pressures commonly invoked to explain the evolution of imprinting were not sufficient to explain sustained elevated rates and that an escape from pleiotropy was also required in the explanation.

My point was that pleiotropic genes disproportionately attract imprinting (because are more likely to have phenotypes relevant to asymmetric kin interactions: see manuscript), while imprinting of widely expressed transcripts imposes heavier phenotypic costs, which may be avoided by imprinting of alternative tissue-specific forms or acquisition of tissue-specific imprinted expression (demonstrated by the canonical, near-ubiquitously expressed Gnas locus transcript). I used this to motivate an argument for the complexity of imprinted loci separate from Wilkins’ competitive signal discrimination argument for regulative complexity. From Siepel’s data it may be that the tissue-specificity alone is then sufficient to explain elevated rates (contra my thought that patrilineal/matrilineal arms races were involved at this stage). I wonder if this dynamic is generalisable – perhaps pleiotropic genes attract various forms of intragenomic conflict, which then favour the development of baroque modifications?

Siepel’s study had other interesting things to say about the functions of genes subject to positive selection (with some unsurprising targets such as immune system genes), but all-in-all the dynamical aspects seem very interesting to me.

I recently attended two talks by Professor Ford Doolittle of Dalhousie University. Doolittle’s argument is about the validity of phylogenetic trees. For prokaryotes specifically, his claim is that trees do not map onto natural kinds. Or more starkly the idea that there is a tree of life is a hypothesis which has now shown to be incorrect for the majority of life on earth.

In his first talk, he began by discussing classification as a common and useful practice and mentioned a paper by Jared Diamond (this one I think) showing that zoological classification by tribal peoples can correspond with scientific classifications. This inter-subjectivity seems to confirm the existence of real phenotypic gaps, but these days we are wont to think of our modified Linnean hierarchical system as reflecting something else: descent with modification (to use Darwin’s term). Creatures assigned to one taxon cannot also be assigned to another and higher level taxa are inclusive of lower level taxa.

Doolittle’s core point is that the prevalence of horizontal gene transfer (HGT) means that descent with modification needn’t produce a branching structure so much as a reticulate one. Perhaps the best known example of HGT is one Doolittle has been pivotal in establishing as a biological fact: the invasion of early eukaryotes by alpha-proteobacteria giving rise to mitochondria, but many more examples are now well known and these are mediated in bacteria by the processes of transformation, transduction and conjugation (loosely these are the uptake of naked DNA, viral DNA transfer and plasmid exchange, respectively). The key point is that the larger the segment of time considered, the smaller the proportion of a bacterial genome that has not been subject to HGT. He provided data in support of this.

In his second talk Doolittle gave details of his own research in the field of metagenomics where the concept of a metagenome is significant. Gene-level phylogenies frequently contrast with organismal phylogenies demonstrating HGT and in his study systems he provided evidence in support of exchange between halophylic archaea and eubacteria (organisms in different domains of life) with the former transferring salt-tolerant genes to the latter and metabolic genes flowing the other way. He gave some reasons for thinking that HGT can be recurrent in structured environments and that significant rearrangement of functional units can take place over periods of decades: i.e., that bacterial macroevolution is reticulate.

Doolittle’s principle philosophical claim flowing from this, and made with Eric Baptiste in this paper, is that while evolutionary biology now admits of multiple evolutionary processes besides natural selection, it needs to adopt a similarly pluralistic stance about the patterns produced by these. He calls those who accept (even implicitly) the tree of life idea, pattern monists. And he cited S. J. Gould’s criticism of process monists: those who attribute all of life’s complexity to natural selection alone.

This made me think of (and ask) a question. In this blog I have previously defended a kind of monism by emphasising that I am monist with respect to adaptation not the origin of species or diversity in general. I think most evolutionists would accept that selection (natural or sexual) is the only process able to generate adaptive change (though the details of any particular adaptive walk may involve drift or recombination as a key component). In brief I have defended this notion by pointing to the need for some sort of discrimination (something not provided in Lamarck’s alternative evolutionary theory for example). So I asked if monism/pluralism disputes about pattern might similarly be resolved by an analogous change of focus. Is the tree concept valid for some lower level unit? Let’s call it an atom. He suggested I might be able to wriggle in that direction but that he was betting against monism at any level of organisation (at least in prokaryotes; one questioner raised the interesting point that sex in eukaryotes might be a way of regulating HGT to largely within-group transfers).

Now an obvious candidate for the atom of analysis mentioned is the gene, particularly when you consider that HGT is often demonstrated using discordance between organismal and gene-level phylogenetic trees. But the trees used here are still arguably only instrumental and homologous recombination between similar genomes (which might also be counted a form of HGT) may be seen to occur across functional genes. This forces our atom definition into a unit of zero recombination or more generally and accurately anything that selection or linkage maintains in linkage disequilibrium. Ultimately we are referring to the selfish genetic element concept a la Dawkins and Trivers and much of this seems unsurprising from a neo-Darwinian perspective (contra Doolittle’s claim that his thesis represents a radical departure and, even more provocatively, one required to create a creationism-resistant evolutionary biology – I take issue with this as there is already evidence enough for evolution and for the processes posited to explain it).

I think that the timescale matters and that during microevolution it is reasonable to describe a tree of selfish genes, which include hyper-motile elements such as integrons. At the organismal level these trees may move in and out of phase with each other as the coalitions we call organisms shift about, but we can still validly describe a tree of life at lower levels. At the macroevolutionary level beyond the population dynamics and lifetime of particular selfish genetic elements, I am not so sure and there may be parts of the genome that are evolutionarily modular at a different scale from others (how this relates to function is interesting). The problem is that the birth and death of these elements seems to invalidate any claim to a universal tree of life. Fine, but remember this has the possible and interesting connotation that there is no last universal common ancestor (LUCA; which is not a statement about a unitary origin of life – this is about backward coalescence and all chemistries not extant are excluded).

Where I think I depart from Doolittle (if I represent his claim fairly) is in his saying, in answer to a question, that even the genetic code has no last common ancestor (LCA). There is basically one genetic code so far known with a few minor variants and given the combinatorial architecture of the transcription and translation machinery this amounts to a vanishingly small segment of the possible code space. It therefore seems likely that common descent explains code identity across the natural world. There are two objections to this: 1. the code may be the only optimal code and we have it thanks to selection and 2. it may itself be modular and hence derived from multiple lineages. With respect to point 1 it is well known that the code is highly optimised but it is not perfect and a better code is known to exist in a distant point of code space. With respect to point 2, it is reasonable to suppose gradual recruitment of tRNAs and competition between alternatives, but it is hard to imagine the de novo assembly of these sets of tRNAs into independent but identical codes in separate lineages. Once derived that would seem to be it.

In summary each of the major transitions in evolution may have fashioned a new set of replicators and the tree of life familiar to us in the eukaryote world is approximately correct but we must acknowledge its roots among the prokaryotes are highly reticulate.

This opinionated and philosophical article about the financial crisis by Nassim Nicholas Taleb (writer of “The Black Swan”; see my “Induction and scepticism” post) has caught my attention. With the benefit of hindsight we can attribute the current crisis, in broad terms, to the existence of incentives to misrepresent risk and the otherwise beneficial fact that the risk is distributed in the financial system. This story tells us whom to be angry with and why: the brokers for not being honest and the regulators for not regulating. (What’s more there is the moral intuition that if the public bails out companies we ought to impose costly regulation on them).

But Taleb has a different target for his ire: those who claim to understand the financial system! In the course of mapping out how we mistake our ignorance for knowledge he makes two interesting points about: 1. the inverse problem and 2. functional redundancy.

1. The Inverse Problem

This is the problem of using data to create a model (accommodating data) and can be contrasted with using data to test the predictions of an a priori hypothesis. Considering this contrast, we can ask the philosophical question: is accommodation worse than prediction?  The answer to this is not as obvious as it seems, but for a clear and excellent description I recommend this article. To summarise, the problem is that ad hoc hypothesising is bad because it increases the chance that the person (or computer) doing it will violate parsimony (and for why this is bad see my post: “Parsimony is probabilistic”). While the correctness of an hypothesis is strictly independent of how and when it is formulated, this argument does, under partial ignorance, give reasonable grounds for suspecting error.  What Taleb misses is that this suspicion can be overcome by paying attention to theoretical virtues (such as simplicity) as well as a priori information from multiple sources (for another suggestion see this article).

But Taleb makes a new and more practical argument. This begins with the fact that the inverse problem is intensified for rare events because the backward time slice required for modelling is larger so, for a given time slice, sample sizes are smaller and inference weaker. When rare events have big consequences (e.g., leading to banks losing more money than they have ever made), we are in trouble. Adding to this abstract point he claims to demonstrate the negative point that future rare events are unpredictable from past rare events using a über-large dataset.

Into further detail I shall not/cannot descend, but I’d like to add that, by logic, the sorts of phenomena that we model rather than predict are likely to be the rarer ones. I think Taleb has therefore made a general and valuable point about the limits of knowledge.

This admission of ignorance is rather sobering for those who want globalisation to work better and are not too ideologically blinkered about how. If we are ignorant of rare destructive events, applying more regulation against specific threats isn’t likely to help us (quite apart from the neoliberal defence that regulation can itself produce perverse incentives). An interesting answer might lie in redundancy.

2. Functional redundancy

Taleb argues that part of the problem is that markets are short-term optimisers. Rare spikes in demand are more likely to lead to catastrophic events like the NY electricity blackout if the system matches supply with demand too efficiently.

But how does natural selection do it? The short answer would seem to be with more time and with lineage-level selection. By analogy a free-marketeer might need to argue that whole economies should go under to bring the benefits of the market to bear (an unpatriotic sentiment). But how is the solution to rare events instantiated in the design of living organisms? Robustness through redundancy might be the answer achieved for example by having multiple copies of a gene. This is a controversial topic in biology and one I need to learn more about. With my ignorance noted, I might argue that gene duplication is incidental to robustness because: 1. it doesn’t always lead to it – in true redundancy a population could always drift to having only one active copy by Muller’s Ratchet and 2. duplication is something that happens anyway and isn’t selected for because it increases robustness.

This second point is key. Perhaps the evolution of redundancy has less to do with selection per se and more to do with its substrate. The relevant property is that the substrate is modular and is therefore subject to selection at a lower level. In finance perhaps regulators should be looking at the design of the system itself with an eye to building incentives for redundancy in?

Is this a real example of how knowing more about biology might inform politics?

I have recently had an interesting e-mail conversation with Tom Meli. We were discussing part of a recent paper by Tom Dickins (TD) and I. The overall thesis TD and I advanced was that recent attempts to construct heterodox evolutionary theories, at odds with the New Synthesis, are failing because natural selection is both necessary and sufficient to explain design. In our view, many forms of heterodox argumentation founder because they tacitly assume that selection is not necessary (see our paper for detail). But the existence of all sorts of complex ecological interactions, which might be expected to create novel evolutionary outcomes, seems to undermine our claim that natural selection is sufficient. Taking an orthodox example, the Red Queen hypothesis shows how host-parasite interactions can provide selection pressures required for the evolution of sex.

The original Red Queen hypothesis is named after the character in Lewis Carroll’s “Through the Looking Glass” who had to run constantly to stay in the same place, and posits a cyclical change between two or more resistance alleles in a population of hosts that is caused by and causes a lagged cycle of cognate virulence alleles in parasites. In a cyclical “arms race” of this sort either participant (hosts or parasites) can benefit by producing varied offspring and sex is one way to achieve this. Modern versions include the idea that, since parents bequeath parasites to offspring, there is an added incentive for producing varied offspring: dissimilarity from the parents abrogates what would otherwise be a head-start for an evolving parasite population in any given offspring (Agrawal, 2006). Now while selection may be necessary in this example it is surely illegitimate to state that it is sufficient. Host-parasite interactions, and perhaps a kind of extra-genetic inheritance, are also required (see our paper for a discussion of the role of non-genetic inheritance).

Tom Meli’s question also relates to our sufficiency claim. He wanted to know why we were critical of the niche constructionist perspective. Crudely niche construction is described as a kind of feedback process in which an organism’s impact on its environment alters conditions in such a way that the selection pressures on the organism change. In common with the Red Queen, this does not shun a necessary role for selection and the claim that niche construction is a “neglected process in evolution” (Odling-Smee et al., 2003) doesn’t in itself amount to the claim that natural selection is unnecessary. No indeed, although, to be fair, TD and I did believe this was implied; but I believe our sufficiency claim is justified and that niche construction falls foul of it.

The reason for this is that we are talking about necessity and sufficiency of selection for design. Here design means the adaptive fit between an organism and its environment. In the Red Queen, the environment is a rapidly (and cyclically) changing one and selection provides a solution: sex. Sex should only be considered a design in respect of those particular selection pressures to which it represents a solution and not outside this context. We are not claiming that selection is necessary and sufficient for any biological feature in any context. Indeed half the explanatory battle is to reveal the sorts of dynamics in response to which one may consider a feature a design.

This argument can be extended by considering the nature of niche construction in more detail. Niche construction is not a kind of selection, but perhaps it is a kind of ecological process. TD and I are sceptical of this. The dynamics captured by niche construction can be split into different kinds. This is also true of the Red Queen, but these dynamics are instrumentally lumped together as those that create an advantage for high genetic variance in offspring (note that the Red Queen is not the only theory and that broader lumping is perhaps more productive: see West et al., 1999). The evolutionary consequences of niche construction seem more varied than this and it is not clear what it causes selection for.

Another comparison is to sex itself. While both sex and niche construction are consequences of selection, they also lead to different kinds of selection, the former being a pre-requisite for sexual selection and the latter, for natural selection. Here is a hierarchy of concepts with sex and niche construction at a higher level and sexual and natural selection at the lower level. Sex is easily defined as the presence of two or more self-incompatible mating types and, among other things, this can lead to a kind of selection that is characterised by a different mechanism from natural selection. Traditionally female choice is the agent of selection in sexual selection and differential survival, in natural selection. Because both kinds of selection entail differential reproductive success they can be lumped together, but they do show differences in mechanism. Although selection resulting from niche construction is different from sexual selection it is not clear how it mechanistically differs from any other kind of natural selection resulting from, say, high temperature or overcrowding.

So niche construction is hard to define in itself (cf. sex), in terms of the kind or mechanism of selection that results (cf. sex) or in terms of what it causes selection for (cf. the Red Queen). This is not to say that the various dynamics lumped together under niche construction do not have consequences for evolutionary outcomes. It is just the lumping we object to, as it doesn’t appear to serve any useful purpose. We believe the fact that a set of difficult to model feedback processes are important in evolution does not justify their elevation to the status of a singular and neglected process.

Postnote: This is an extension of a debate with Tom Meli, so please read the comments to hear his side of the argument…