Skip to content

Tautology 6: A resolution

Last updated on 22 Jun 2018

If, as Byerly and Michod think, along with many others such as Brandon (1990), there are two kinds of “fitness”; actual and expected, and only with actual fitness is there any possibility of a tautology, we need to work out what expected fitness is. Brandon calls it “adaptedness” proper; a genotype is expected to be fit to the extent that it will differentially reproduce in a given environment. Even so, this looks like it involves logical tautologies. I am going to say, yes, it does. So what?
Explanations in science are of the rough form thus: some model or set of models M have a formal structure. In those models are a number of definitions T. We think that M explains the observable phenomena O if there is some interpretation of M that maps onto O in a non-arbitrary manner.

In simpler terms, the explanation is when we can say the Model descrIbes the Observations.

Here’s a diagram that explains what I mean.

Tautology map.png

As Brandon says, the PNS is devoid of empirical content; but the overall models are not. Each instantiation of the model of evolution by NS (ENS) has a significant amount of explicit and implicit biological content. For a start, in genetics, the ENS has content about genes. Since genes are not defined into existence, but are understood as the result of considerable empirical and experimental work, population genetics, which includes ENS, is not a tautology. Of course, the PNS is a tautology within that model. Other tautologies in ENS include such things as the Hardy-Weinberg Equilibrium, and genetic drift.

Now science is not just the making of formal models, or else we’d all be happy with making computer simulations all day long (and some people do this, but I rather doubt they think that is the whole of science). If you have a model, you need interpretation rules to make it apply to the domain or group of organisms to which you want it to apply. Where do these come from? What makes the ENS “about” the biology? And how do tautologies in the model explain anything in the biological (that is to say, physical) world?

To take one example, ENS applies only to populations, not individuals (cf. the discussion in Forber 2005), which is to say it explains only the temporal dynamics of populations. Hence, we need to be able to identify what a population is, independently of the ENS. This we usually can do, but there are problem cases, as in so-called “superorganisms” (is the ant colony one organism or many?), so what happens is that the subdiscipline applies the models in various ways until a reasonable fit is found. Oddly, this is like a conceptual process of NS. Trial and error is another way to express the PNS without reference to organisms or genes, etc. But we can’t say that the PNS is what gives the PNS its empirical purchase, even if it is in another domain (the cultural), or can we?

Each subdiscipline, discipline and domain of biology is the end result of a long tradition of practices, institutions, teacher-student lineages, and techniques. The reason why a substantial number have been retained is that they are effective at learning about the biological world, which we assess independently (by, for example, doing field observations, breeding organisms or curing diseases). These are the interpretation rules of the discipline, and it is from this that we test and embed the models in scientific practice.

Given that the PNS itself is empirically empty, it forms a kind of schematic explanation:

P1. PNS (+ ancillary principles, making up M)

P2. Empirical data (O)

+ Interpretation Rules (I)

NS Prediction (or retrodiction)

It is not a true explanation until interpreted in a particular case, what Brandon refers to as “instantiating” the theory. Each part of the schema is substituted by some actual data or organisms, processes and parts. Then, it is an explanation. For example, while the same model may apply to a fungus or a fig, there is no physical property that they share uniquely. Without that, explanations by NS are merely promissory notes that there may be an account in terms of the ENS model. And given that until the model is physical and instantiated, any set of population dynamics might have turned out to be the one that best represented the particular case, it is hardly a tautology in practical terms. It may have been that the trait under investigation in that population is undergoing no selection at all. It may be drifting, or be hitchhiking on some trait with which it is linked that is undergoing selection.

In the final post I will tidy up some loose ends, so point them out in the comments…


Brandon, Robert N. 1990. Adaptation and environment. Princeton, N.J.: Princeton University Press.

Forber, Patrick. 2005. On the Explanatory Roles of Natural Selection. Biology and Philosophy 20 (2):329-342.


  1. Garamond Lethe Garamond Lethe

    Some of us spend our days making computer simulations of computer simulations. It’s a living…

    • But until you actually run the computer simulations that you are simulating, how do you know that you have correctly simulated them? No, really, I want to know…

      • Friar Broccoli Friar Broccoli

        In case you are asking the question seriously, we don’t know. Scratch that we actually do know that our simulations are wrong, we just don’t know how wrong until we have run them against the original using the same data.

  2. jeff jeff

    NS is tautological when fitness is determined post-selection. That is true in a model as well as in the real world. Presumably, expected fitness avoids the tautology by determining fitness pre-selection. But if there is not a significant correlation between expected fitness and post-selection fitness, then it is a relatively meaningless concept that predicts very little. It may avoid the tautology in a model, but not in the real world.

    In a GA (and I’ve written many of them in my day), you have some kind of fitness function that is computed before a selection method (elite, tournament, or whatever) is invoked. The combination of the fitness function and the selection method determines the actual fitness. There may be some stochastics in either of them (which are actually deterministic in practice), but in general the actual fitness is quite predictable. I can see no way to map this model to the real world, where the GA concepts of fitness and selection are not only crude approximations, but are also inextricably intertwined (hence the tautology). Can you really use expected fitness to predict how populations evolve in real world (not lab-constricted) conditions?

    • arjunsajip arjunsajip

      Ironic, really – “inextricably intertwined” is itself a tautology ๐Ÿ˜‰

      • No it isn’t. Something can be extricably intertwined.

        • TomS TomS

          As well as being inextricably in some other way than being intertwined. Inextricably buried, for example.

  3. bob koepp bob koepp

    I’m not sure what loose ends you have in mind to tie off, since I think you’ve done a pretty good job here of exploring how the “tautology charge” relates to the theory of natural selection. So I’ll just add my two cents; i.e., I think the sense that there is a logical flaw in the theory of natural selection usually arises when people are not careful to distinguish conceptual, empirical and methodological questions. If we are very careful about our assumptions and inferences, there’s no problem at all.

Comments are closed.