As part of my occasional series devoted to my own technical papers (forgive the self indulgence, but I’d like that body of work to be known to, if not appreciated by, a slightly wider audience than my academic colleagues), let’s discuss the nature of evolutionary biology as a scientific discipline.
The occasion is provided by a paper I published in a collection put together by K. Kampourakis, entitled The Philosophy of Biology: A Companion for Educators. (The book, here, is pretty expensive; my chapter can be downloaded for free here.)
For some reason, the scientific status of evolutionary theory has always been questioned, and I don’t mean just by creationists. Darwin’s famous “long argument” laying out the foundations of the new field of evolutionary biology, was seen as ill-fitting with both of the major philosophical views of how science works that were being debated in late nineteenth century Victorian England. This is known as the great induction debate, and featured a who’s who of early philosophy of science, with John Stuart Mill and William Whewell (the man who coined the term “scientist” in 1834).
Darwin got in trouble with both Mill and Whewell because they both regarded Darwin’s work as an example of deductive reasoning, not of induction (whitch they both favored as a scientifically-sound means of reasoning). According to Mill and Whewell’s understanding, what Darwin had done was to arrive at a hasty generalization based on a small number of observations, proceeding then (deductively) to derive consequences from them, and finally collecting data for decades to back up his hasty conclusions.
Needless to say, Darwin was taken aback by all this, particularly by Whewell’s criticism, since he had been the latter’s student. Indeed, in apparent frustration, Darwin wrote to a friend: “How odd it is that anyone should not see that all observation must be for or against some view if it is to be of service!” — a sentiment that is ironically perfectly consonant with Whewell’s idea of inference to the best explanation.
The next big battle for the soul of evolutionary biology was the famous, decade-long, debate between two of the founding fathers of population genetics, Ronald Fisher and Sewall Wright. The history of this debate, and the role it played in population genetics, is both complex and fascinating. Fisher was convinced of the absolute preeminence of natural selection in shaping organismal evolution, so much so that he consciously modeled his famous fundamental theorem after the second principle of thermodynamics, one of the most successful laws formulated within the dominant experimental science, physics. Wright, on the other hand, was attracted by the complexity and messiness of biology, and his emphasis was always on non-linear, non-additive genetic effects (epistasis, or gene-gene interaction, and pleiotropy, the fact that the same gene can affect a number of traits). Most famously, Wright was interested in the role of genetic drift in evolution, a stochastic process of population sampling that he regarded as countering, and some times even nullifying, the role of natural selection.
Back in 2006, Jonathan Kaplan and I have suggested a way to conceptualize the balance between natural selection and drift, an approach that might be helpful as part of a general understanding of the respective roles of chance and necessity in evolution. Instead of thinking of drift as a force antagonistic to selection, we can visualize it it as a measure of the “error” surrounding the expected evolutionary change caused by selection. According to this way of thinking, if the target of selection is a particular phenotypic value of a given trait (in a certain environment, of course), then there is a probability distribution that tells us how likely the population actually is to land on that phenotypic target. The smaller the population, the larger the error around the selective target, i.e., the larger the drift. This means that the efficacy of natural selection is directly proportional to the size of the population, just like Fisher maintained. But it is also true, as Wright thought, that if the population is small enough, then the stochastic effect of the drift “error” may completely override natural selection’s push.
The latter part of the twentieth century saw the opening of yet another front in the seemingly perennial discussion about the relative role of chance events in evolutionary biology, as well as about the status of the discipline as historical and yet scientifically fully mature. The main charge was led by Stephen Gould and his associates, with a series of papers that unleashed decades of debates and new research. The opening salvo by Niles Eldredge and Gould himself was the famous 1972 paper on “punctuated equilibria,” where the standard Darwinian view of gradual evolution was challenged and, by implication, the role of natural selection in shaping long-term evolution somewhat curtailed. The paper argued that at the least in a number of cases evolution proceeds by long periods of stasis — with no detectable phenotypic change — “punctuated” by “sudden” (geologically speaking) bursts of activity. More than four decades later, my sense is that paleontologists are by and large in agreement that Eldredge and Gould basically got it right; neontologists (i.e., largely population biologists), however, are of the opposite opinion.
Much in the above mentioned discussions — like any debate on the nature of science — reflects a certain philosophy of what it means to be doing experimental or historical science. It is therefore a good idea to explicitly deal with that philosophy. An excellent framework for this is provided by a paper published by Carol Cleland in 2002, on the epistemic differences between historical and experimental sciences. Cleland’s pivotal idea is that the two types of science are separated by what she calls an asymmetry of overdetermination. Building on previous work by David Lewis, she explains that “the basic idea is that localized present events overdetermine their causes and underdetermine their effects.”
She elucidates what this means by considering the example of a crime being investigated. Once committed, a crime leaves a number of historical traces, no matter how careful the perpetrator was in erasing as many of them as possible. All it takes for a criminal investigator to figure out what happened is a relatively small number of traces that clearly enough point toward a particular sequence of events. The investigator would then be using a type of induction known as inference to the best explanation (first articulated by Whewell) to pinpoint the culprit. Conversely, the simple act of not committing the crime obviously instantly erases the possibility of any historical trace to be left around. Few currently available clues overdetermine a past event, while so many futures are possible given a particular current state of things that the latter underdetermines the range of futures.
Cleland cashes in this asymmetry of overdetermination by arguing that — contrary to popular wisdom (and to the opinion of many practicing scientists) — there is nothing inherently epistemically superior about experimental over historical science. This is because of two consequences of overdetermination. On the one hand, while experimental scientists have the ability to strictly control the conditions of their experiments, it is that very strictness that limits the scope of applicability of their results: as soon as one widens the settings of a given experiment, different factors begin to interact with each other in complex ways, quickly leading to a large number of possible future outcomes; in other words, predictability is purchased at the expense of generality. On the other hand, while historical traces constantly decay through time, and may disappear forever, the historical scientist often needs only a small amount of them to arrive at a sufficiently accurate reconstruction of what happened — just like the criminal investigator in the hypothetical example of the impossibility of a perfect crime.
Cleland’s account makes sense of some surprising limitations of experimental science, as well as some spectacular successes of historical science. In the first case, it is notable, for instance, that non-equilibrium thermodynamics (for example in its applications to atmospheric physics and climate science) quickly reaches a limit in terms of predictive ability, where complex mathematical models are incapable of generating more than very approximate statistical predictions about the future behavior of complex systems, predictions often accompanied by rapidly expanding margins of error.
In the second case, however, we have for example the success of paleontologists in determining that an extraterrestrial body of massive proportions hit the Earth 65 million years ago, contributing to the extinction of countless numbers of species, chief among them the dinosaurs. The impact was suspected once geologists discovered a worldwide thin layer of iridium in rocks datable to the K/T (Cretaceous/Tertiary) boundary. This led to a search for a crater, the remnants of which were eventually identified off the Yucatan peninsula via satellite imagery. From there, geologists could calculate the size and direction of the impact, and therefore make fresh predictions concerning additional historical traces, for instance those left by the tsunamis that must have hit the western coast of Mexico as a result of the asteroid crush. Sure enough, those traces were found, leading to even more confidence in the conclusion that “the crime” had indeed taken place in the way it had been hypothesized.
Despite, or perhaps because, of all the controversies, these are exciting times for evolutionary biology, both in terms of empirical discoveries (such as those resulting from the so-called “evo-devo” approach, or from comparative genomics) and conceptual advances (e.g., discussions of new ideas, like evolvability, emergent complexity, and others). And while it is true, as Nobel winner Jacques Monod said, that “even today a good many distinguished minds seem unable to accept or even to understand that from a source of noise natural selection could quite unaided have drawn all the music of the biosphere,” that understanding is getting richer and deeper, but it still hinges on taking seriously the dichotomy and interaction between stochasticity (“drift”) and determinism (natural selection).