Pittsburgh, part IV

Last report filed from the meeting of the Philosophy of Science Association in Pittsburgh. During the past few days I summarized some of what I thought were the most stimulating sessions for philosophers and scientist (focusing on biology, which is my area of interest), and I’m going to conclude with another spectacular but intense session, on the question of whether evolutionary theory can be understood as a theory of forces, in analogy with, say, Newtonian mechanics in physics. The speakers were Robert Brandon (Duke), Christopher Stephens (British Columbia) and Denis Walsh (Toronto).

Brandon got things started by proposing nothing less than a new law of biology, what he calls the zero force law (ZFEL). Essentially the law says that if there is variation and heredity, and no selection or other evolutionary forces and constraints are acting, then diversity and complexity will increase over time, on average. This is in explicit analogy with Newton’s first law, which gives the default condition in mechanical systems. For Brandon the ZFEL is the natural null hypothesis for evolving systems, and is better than the standard Hardy-Weinberg law in population genetics because it is more generally applicable. (Note that I argued in my book with Jonathan Kaplan, Making Sense of Evolution, that null hypothesis are generally speaking a bad idea for science.)

Brandon doesn’t buy the standard objections to the idea that there are forces in biology, for instance that drift cannot be represented by a vector (it has no direction) and additively summed or subtracted to/from the others (which are selection, mutation, migration and recombination). His solution to the problem of drift is that it is not a force, but a default state of the system, so that there are only four, not five, forces to be reckoned with. These others are real forces because they alter the default state of the system (but so does drift, no?).

As examples of predictions of his new law, Brandon said that he expects the degree of complexity of the eyes of cave animals to increase over time. This is in flagrant contradiction with the observation unless, as Brandon does, one defines complexity as the number of existing states in the system. (Someone during the discussion pointed out that this makes ZEFL essentially a restating of the second principle of thermodynamics, and therefore not a biological law.)

I have to disclose that I have little sympathy for the idea of forces in evolution, and even less so for the idea of laws in biology (again, see my book with Jonathan). Also, I would have hoped that philosophers and biologists (Brandon works with the Duke biologist Dan McShea) would have by now gotten over their physics envy: Ronald Fisher modeled his famous fundamental principle of natural selection (in reality a simple mathematical tautology) after the second principle of thermodynamics, the alleged golden standard of hard science (even though it is actually a probabilistic statement about the world, and thus not really a law). I guess we still have a way to go for people to accept that it is perfectly fine, even better, to do science without laws and in a different way from physics. Hey, in the meantime at least we have elected a black president! But I digress.

Stephens also believes in evolutionary forces, but differs from the classic view presented by Elliott Sober in his landmark 1984 book on the causes of selection. Sober had included drift among the forces, essentially stating that Hardy-Weinberg is a zero-force law. For Stephens that doesn’t sound right (and I agree) because drift definitely is a different kind of beast from the other four, and needs to be accounted for in its own way. One way, as we have seen, is Brandon’s: treat drift as a background condition instead of a cause of evolution, despite the fact that this will sound weird to most biologists. (Actually, despite our disagreement with the force view, Jonathan and I do suggest in our book that drift should be considered a background condition, in a way analogous to Brandon’s, except that we don’t think this amounts to a fundamental law of biology.)

Stephens wants to rescue drift for the force metaphor, and claims that drift does have direction, not just intensity: after all, the effect of drift is to decrease within-population variance, because it brings alleles to fixation. (I pointed out during the discussion that at the exact same time drift also increases between-population variance, so that the overall effect is actually null; this is related to the so-called Wahlund effect in population genetics, for those interested in the technicalities.)

Stephens also discussed Walsh’s “statisticalism” (see below), the only dissenting view about forces presented at the symposium. Stephens doesn’t buy Walsh’s solution for a variety of reasons that range from a disanalogy between drift and coin flipping in terms of their statistical properties (I don’t find this very convincing because the point of the analogy does not depend on such properties) to the fact that Walsh cannot account for actual biological examples of causal separation between selection and drift (there actually are very few such examples, and their interpretation is highly debatable).

Finally, Walsh himself framed his talk as a choice among three models of evolution. The two-factor model is Sober’s classic view, which says that both selection and drift are forces; the one-factor model is Brandon’s, in which selection is a force, but drift is error (or a background condition); and Walsh’s statistical interpretation in which both selection and drift are probabilistic descriptions or outcomes of population change, but have no causal power at all (this sounds very weird, but there is a way to make sense of it, biologically, see below).

For Walsh the first two models claim that the statistical distribution of fitness in a population explains selection, and that the population size explains (as in, it is a measure of) drift. These are considered causes because one can manipulate them experimentally and show that they act independently of each other. In fact, Walsh argues, theoretician John Gillespie showed long ago that every time you change the population size you also alter the distribution of fitness, so the two factors are not actually independent, and one cannot disentangle the two alleged causes. (I don’t buy this. It may be very difficult in practice to disentangle drift from selection, but it is surely possible in principle to construct populations with different sizes that have the same statistical distribution of fitness.)

Walsh’s main point, though, is that population-level statistics about selection and drift cannot be interpreted directly in a causal manner, because they are just probability distributions, which don’t have causal powers. He articulated this through a fairly convoluted analogy with the well known Simpson paradox (look it up, it’s cool), a description that I honestly had a hard time following and that did generate discussion afterwards that made it clear that I wasn’t the only one having trouble.

At any rate, I take it that Walsh doesn’t deny that, at the individual level, selection and random events to occur and are physical and causal, and that his point is that the trouble arises when one wishes to go from the observation of population-level statistical properties to the inference of individual-level causal processes. On that I agree, as -- sorry for being so repetitive today -- Jonathan and I have argued across two chapters of Making Sense of Evolution.

Anyway, the meeting was fun and intellectually stimulating, but I realized that the last few posts on this blog will be read only by very few people with an insane interest in philosophy. So, enough technical posts about philosophy for a while, next week we’ll go back to good ‘ol fashioned discussions based on my musings about the world and how it works. Stay tuned!