evolution by drift and the role of fitness concepts in biology
why this is interesting
natural selection is not the only mechanism by which evolution can occur, and this is very important because, to explain the normative concepts (“fit,” “good for,” “healthy,” “flourishing,” “successful”) that pervade biological talk, biologists and folk often appeal to natural selection as if there were no viable alternatives. but there are several other ways of changing the distribution of phenotypes and allele frequencies across populations over time. these ways include migration, genetic mutation, and genetic drift. unfortunately, these concepts often seem to be misunderstood, even by those trained in the discipline.
some have even speculated that their very possibility threatens to spoil any attempt to explain all the biological facts in terms of more fundamental facts from chemistry, physics, and math. if true, that would mean that the normative concepts in biology might need to be taken as primitive in our biological reasoning. they would be bare facts of the biological world, and not shorthand for an explanation in terms of more fundamental facts.
and if that’s true, it could open up more possibility for our own conceptions of what it means to be a successful, healthy, flourishing human being.
a quick primer on drift
genetic drift is usually defined loosely as changes in genetic distribution across a population over time that are due to random chance rather than selective pressure. so for example, infrequent changes in an organism's environment that massively shift the distribution of genes in an environment may explain why certain genes persist while others don't. asteroid strikes may explain why there are fewer dinosaurs around today more effectively than an explanation in terms of evolution by natural selection.
of course the "random" in the loose definition above calls out for more precise explanation. all evolutionary change is the result of some chance, so what makes one kind of chance "random" as opposed to regular such that the definition can be illuminating? presumably, the regular events are supposed to be the frequent ones. but that doesn't settle historical questions about whether the frequent evolutionary changes are the drifty ones or the selective ones. as far as i know it's still an open question whether the majority of evolutionary change has been drift-driven or selection-driven. (philosophers of biology refer to this question as empirical adaptationsim).
this video does a nice job humorously shedding light on what this randomness is intuitively supposed to amount to. things like population migration, or natural disasters, that have big impacts on the distribution of alleles in a population which then persist over time.
another way to get a sense for drift and it’s power is to consider what biologists call “adaptive landscapes.”
skip this next section and just jump straight to the argument if you accept the possibility of widespread non-optimal distributions of alleles in a population; otherwise enjoy this interlude through the adaptive landscape.
an interlude through adaptive landscapes
an illuminating way to think about the importance of drift is to consider what evolutionary biologists call "the adaptive landscape.” an adaptive landscape diagram is like a topographical map where the vertical axis represents the relative fitness that a combination of phenotypes provides an organism in a given environment and the horizontal plane represents all the different possible phenotypic combinations. individual organisms in a population can be plotted along this landscape and their position modeled over time to represent hypotheses about the mechanisms of historical evolution.
Fig 1. Adaptive landcape diagram showing possible distribution of phenotypes and the relative fitness their combination would afford an organism in a given environment. populations of organisms can be plotted on this landscape as dots and their evolution visualized as movement across the landscape over time. Source.
one important insight that adaptive landscape diagrams help bring out is how probabilistically rare it is for organisms to have the best possible fitness for a given environment. the highest possible fitness for a given environment would be the highest peak on the total adaptive landscape. the trouble is, that peak could be very topographically far from a population's current position on the landscape, and evolution by natural selection will tend to cluster populations around local rather than global peaks. this is because once a population has climbed a local fitness peak, evolution by natural selection will not push it down that peak.
however, non-adaptive evolutionary mechanisms can push populations off local fitness peaks into valleys where they can then be pulled by natural selection up on to other, potentially higher, or potentially lower local fitness peaks. migration and drift are examples.
the fitness landscape can also change over time while holding phenotype distribution fixed. adaptive landscapes represent a fitness topology for a fixed environment, so as soon as the environment changes, so too will the landscape's topology. for example, when an ice age descends, the same point in phenotypic space (horizontal plane) for an organism will likely correspond to a very different elevation (fitness) for that organism.
adaptive landscapes also help to illuminate the role that developmental constraints play in evolution. developmental constraints are the limitations to phenotypic malleability imposed by the requirement for organism-level phenotypic coherence. for example, while certain traits considered atomistically may represent higher fitness possibilities for an organism in a given environment, when considered holistically, the evolutionary-developmental pathway towards realizing those traits would require passing through considerable fitness valleys. think of these fitness valleys like moats that prevent populations from moving to global fitness maxima for a given environment.
ultimately, i find adaptive landscapes helpful because they massively complicate our understanding of how organisms could have arrived at their current position in phenotypic space. a simplistic conception would have it that organisms are currently arranged higher in topological fitness space than their ancestors, that they came to be there because of evolution by natural selection rather than evolution by non-selective forces, and that therefore their phenotype combinations are adaptations. the downward movements caused by drift, the stickiness of local maxima, and the shiftiness of the landscape itself all put pressure on that naive idea.
the argument from drift
because drift can cause major changes in the distribution of allele frequences in a population and alter the course of biological evolution, it’s obvious that no explanation of a trait’s presence can simply say it was the result of evolution by natural selection; it could have also been caused by a drifty process. we need some way to answer the following question:
which changes in the distribution of allele frequences are due to drift and which are due to selection?
while a simple answer would say: the ones due to selection are the ones that made organisms better adapted to their environments, the mere possibility of drift means that not all organisms who survive are necessarily the ones better adapted to their environments. it’s actually an empirical question whether a given trait is a result of drift or selection. and this turns out to matter for reductive attempts to define terms like “fitness” and “good for” and “purpose” in terms of evolution by natural selection.
here’s why.
consider the following definition of fitness:
to be a fit trait is to be a trait with a higher propensity to persist through subsequent generations than a non-fit trait.
this kind of definition in terms of probabilistic differential rates of reproduction is fairly standard among biologists.
the problem with the definition emerges when considering what i’ll call the mismeasurement conundrum: while we can measure the actual persistence rates over time of traits, when it comes to measuring propensities we have to stake a claim at a particular moment in time about which traits are more likely to persist in subsequent generations so we can then empirically test that claim to see if it’s true by measuring deviation from these predicted propensities to confirm the hypothesis that a trait is fit, per the above definition.
but when deviations do occur, how will we know whether they are the result of drift or simply selection according to different initial probabilities or both? we would need a way to determine whether the initial differences in propensity to persist which we assigned as a hypothesis to confirm were themselves accurate. after all, we could have gotten them wrong and the observed deviations could actually be a reflection of traits with different persistence propensities surviving at the rate those propensities would predict. we could try to find out whether the initial assignments were accurate by going back another generation, attempting to measure fitness, but then we would be off on an infinite regress.
to take an example, suppose we want to test the hypothesis that trait a is a fitter trait relative to trait b. that is, by the definition above, it has a higher propensity to persist across subsequent generations than trait b. to test this hypothesis, we need to make a prediction about how much more likely it is to persist and then see if that comes out right through population census over subsequent generations. suppose we do some estimating and form the hypothesis that it is 50% more likely to persist. but when we do our population census over subsequent generations to test this hypothesis, we find that trait a underperformed our prediction; in fact, both traits persisted at equal rates. we now face the conundrum: was our initial assignment of probabilities accurate, and subsequent generations experienced drift? or was our initial assignment inaccurate, where the true propensities would have predicted even rates of trait persistence? or, the third option: might have both our initial assignment been inaccurate and the subsequent distribution be due to drift rather than reflect the real fitness propensities? there’s no obvious way to answer these questions when we attempt to define fitness purely in terms of probabilities.
simplifying the argument a bit:
to measure the fitness of an organism you have to demonstrate that one or more of its traits makes the organism more likely to produce viable offspring than a comparison trait.
demonstrating this requires measuring actual differential survival rates across generations and comparing them with the survival rates predicted by a fitness hypothesis.
because of the possibility of drift, this can only be done if we can estimate fitness independently of actual differences in survival rates.
but we didn't want to do that, we wanted to define fitness in terms of differential survival rates.
so, fitness cannot be defined solely in terms of propensities
note on point 3 that, if natural selection were the only mechanism of evolution and we found a difference between actual and predicted trait distributions over subsequent generations, we would know that these were due to mismeasuring initial fitness. when drift is a possibility, we lose that recourse.
on the conclusion, note that the argument doesn’t show that fitness cannot be defined in terms of differential reproductive propensities; it just shows that some extra criterion is needed. the challenge is to give an extra criterion that is a concept from math or physics, not one that just smuggle a normative, fitness-like concept back in.
the argument presented here is also distinct from the old triviality objection to the principle of natural selection defined in terms of survival of the fittest. the triviality objection simply claims that defining fitness in terms of actual survival turns out to be un-illuminating, circular, or trivially true. the argument from drift is more ambitious: it aims to show that fitness is not definable in purely probabilistic terms.
to be sure, this drift problem for fitness discussed in this post is in no way settled stuff. philosophers of biology and biologists continue to debate it. nor is it to say that the concept of fitness shouldn’t play a role in our biological thought and talk. it’s just that biologists haven’t obviously succeeded in explaining this concept in terms of more fundamental concepts.
there are a lot of other issues like this in biology like this that we naively take to be uncontroversial but turn out to be really puzzling. to name just a few that i hope to explore here in the future: the role of information-explanations in genetics, the role of development in evolution, the individuation of organisms and species, and many others. to get a sense of how philosophers of biology approach these topics, check out peter godfrey smith’s excellent introduction, philosophy of biology. this interview with john wilkins is great too.
a final point: the argument presented above leaves open the possibility that we can just directly perceive fitness differences. after all, it seems like we can just “see” fitness all around us. the remarkable - often beautiful, sometimes puzzling - “fittedness” of organisms to their environments seems to cry out for explanation. richard dawkins once called it the “big question” for evolutionary biology. so if we find ourselves unable to explain these apparent fitness facts in terms of more basic, non-normative terms, that might appear quite dissatisfying.
on the other hand, it may feel quite liberating. the argument appears to open up the possibility for a picture of ourselves and our neighbors in the biological kingdom as consisting in more than what we get from the picture offered by physics, chemistry, and math. if we can unproblematically take on fitness-talk and other normative concepts in biology as primitives, this may open up interesting space to think about natural purposes without feeling like this talk must ultimately be cashed out in terms of evolution by natural selection.
and this in turn might provide firm intellectual footing against which to unproblematically talk about what is good for human beings in terms of these natural purposes - a kind of “flourishing for homo sapiens” which may become thematic in future posts.