Evolving Thoughts

A new paper has been published in the History and Philosophy of the Life Sciences, entitled “Pattern Cladism, Homology, and Theory-Neutrality” by Christopher Pearson. Either the journal has done something horrible to the text, or the author doesn’t know the difference between Willi Hennig and William Hennig, or between Gareth Nelson and Garrett Nelson.

But this is not the worst aspect of the paper. It states “save for a few notable exceptions in John Beatty (1982), David Hull (1988), and N.R. Scott-Ram (1990), pattern cladism has avoided philosophical scrutiny”. This leaves me aghast. Apart from anything else (the extensive debates of a philosophical nature during the 80s and 90s in journals like Cladistics and Systematic Biology and its predecessors), there is the work of the late lamented Ron Brady, who debated pattern cladism as a philosopher (1979, 1982, 1985). Pearson mentions Brady later in the paper, but not his defence of pattern cladism in general.

The three philosophers (well, two philosophers and one philosophically minded systematist) who Pearson mentions all opposed pattern cladism (PattC), which is the point I wish to draw out here. There has been a strong mythologising of the “debate” over pattern cladism over the years, begun by its opponents (especially Steve Farris, and Mark Ridley), accusing it of being “anti-evolutionary” or “anti-Darwinian” or even “creationist”. At some point in the late 80s, editors got sick of the topic, and it no longer was discussed. Now, if you submit a paper on the matter (and I have been part of a collective that has tried), you are told either that the issue was resolved in the 80s, or that it represents an outmoded metaphysics. But debates, especially philosophical debates, are not resolved because editors tire of them, nor are the issues any less (or for that matter, more) important because nobody talks about it any more. Perhaps Brady is no longer cited because he was pro-PattC.

Pearson has actually focused on an interesting issue: theory-neutrality. I’ll get back to that in a minute. First, let me make a few points. Pattern cladism is, in the mythology that Ridley others set in play, essentialistic and antievolutionary. This is rather odd, given that one of the leading lights of the movement, Colin Patterson, wrote a book on evolution which is (in its second edition) still one of the best introductions to the topic I know. But the mythology is strong, and few scientists think much past that propaganda. What pattern cladism actually was, although like any movement of ideas the originators are sometimes less than clear on the matter, was the claim that before one can work through the history of taxa, one first has to make a relationship scheme that can test that history. In short, classification and historical reconstruction are distinct activities.

The alternative form of phylogenetic systematics (the correct term for “cladistics”) in effect conceded this point by denying that we did classification at all. Phylogenetics was only, and always, historical reconstruction, a view espoused to philosophers by Elliot Sober’s Reconstructing the Past. I call this process cladism, a term also used by Marc Ereshefsky. I won’t argue this here as I have often done so before. Before I proceed, I should note that I am neither pro-PattC nor anti-PattC. However, I do think that PattC arguments have a philosophical bite that has never been successfully dealt with either by the process cladists nor by philosophers; that is, the indefinitely large number of reasonable histories that a single cladogram supports. If reconstructing history is based on our cladograms, then the one to very many mapping of cladograms to histories means that at best any history is only something that is likely on the basis of prior assumptions and models. Of course, that is fine, so long as that is what we understand that we are doing here. It rarely is.

But let’s examine the question of theory-neutrality. This is an interesting problem. Clearly Hennig thought that some theory was essential to systematics. And those who in systematics played the theory-free or objectivity card most often were the numerical taxonomists (NumTax), the so-called pheneticists. So pattern cladism, which seeks to be both phylogenetic in some sense, and theory-free, appears to be in conflict with both styles of systematics; but this is only the case if you question-beggingly define phylogenetic systematics as process cladism, and theory-free classification as phenetics.

So, can systematics be theory-free? Another way to ask that question is to ask if systematics can be objective. Most now agree that the older style of systematics, as practised under the rubric of “Linnean systematics”, was often arbitrary and subjective, leading to splitting and lumping based on the predilections of the systematist. But that isn’t really the issue now. Instead, it has to do with arcane issues in the philosophy of science itself.

Pearson recounts the usual story against essentialism, in the “species-as-individuals” account that is now the consensus in philosophy of biology (almost). He notes that this is not a necessary conclusion, however. Still, the argument here is that if one is a pattern cladist, one is essentialist, and if evolution mandates the view that taxa are individuals, then pattern cladism is false.

Pearson says “Patterns in nature will be recognized as patterns only if the observer is armed with the relevant theory to recognize them as patterns.” This claim, which is crucial to the theory-ladenness hypothesis (I can’t call it a truism, nor can I think of it as a theory in its own right) is faintly absurd, ranging through to completely overblown. Did nobody observe before we had theories? Or do we all have theories, in which case why does theory-ladenness have any special purchase in science (it’s like saying that when we throw an object we calculate the differential equations needed to find the optimal trajectory)?

Theory-ladenness only means anything when the very act of observing is itself theoretically charged, such as identifying the meaning of particle tracks in cloud chambers (the classical example). However, if there is any act of observation we are able to make in the absence of any theory (apart from the adaptiveness of our evolved sensory capacities), it is in the observation of most macro-organisms. There are theories that explain how we do that, yes. They were worked out as we developed physiological and psychological accounts of the mechanisms of our senses. But before we learned or developed those theories, we did not need theory to observe. There is a logical or category error here akin to the use-mention distinction in philosophy of language: a “use-account” distinction. That there is a theory of how telescopes work (the theory of optics) does not mean that I need to know that theory to use the telescope. Indeed, as Hacking argued, I can iteratively refine my confidence in the fact that telescopes work by testing it initially with things I can inspect with the naked eye, so that I am sure of what it does even if I do not know why. Presumably there is an account of how it is that we identify taxa through observation (in the normal conditions that we evolved to do so). But until that account is formulated and tested, nobody knew it, and did not depend upon it to be able to identify taxa.

[Parenthetically: in my opinion, that account is a complex of the properties of neural nets as classifier systems, along with cultural and economic interests in getting the classifications right. Basically, if you are an animal and you hunt for food and evade predators and dangers, then you will tend to evolve sensory systems that correctly classify. Some call this evolvedness “theory”; I think that it beggars the meaning of the term “theory” to call biological adaptedness theoretical.]

Why did the PattC proponents and the numerical taxonomists think that we needed to be theory-free in order to classify objectively? This has a history that relates to the course of positivistic thinking in the first half of the twentieth century. Basically, when in the philosophy of science the post-positivists asserted the primacy of theory in science to the exclusion of observation sentences and sensory data “clips”, they insisted that to observe relied upon prior theory. PattC and NumTax arose at a time when this was the brave new idea in philosophy.

Now there was an ambiguity in that claim. It is one thing to assert that logically, if one is to justify an act of observation, one must have an account of observing because the act of observing is constituted by some prior capacities that need explaining. Nobody, I think, objects to this. Animal A observes its world, and we need to explain why; that account relies upon a theory T. But does the animal employ T in observing? Almost all animals do not, and yet they manage quite satisfactorily to evade predators, find food, and in some cases, learn quite sophisticated facts about their world.

Scientists are theoretical animals par excellence, but one may reasonably doubt that they employ theory in every act they make as a scientist. But for the theory-dependence claim to work, this is exactly what must be true. Not just that there is a theory of how observation is being done, but that the scientist must necessarily employ a theory (whether or not it is the right one) in observing. Taxonomy, more than most disciplines, rests on the observation of traits of things, but seeing a spider’s pedipalp (the mobile sensory appendages on its head) is untheoretical, even if identifying these appendages as pedipalps, or at least naming them thus, is not. Pearson uses a similar point: the abdominal spinneret on spiders, which David Hull famously (among systematists) used as a reductio of theory-free observation in the context of PattC. But there is a theory-free way to identify a spinneret: look to where the silk comes out on the lower abdomen. A five year old could do it, unless you think identifying an abdomen or silk needs prior theory.

Pedipalps in a Striped lynx spider, from Wikipedia. Do you really need a theory to see them? Could you identify these in another spider without theory?

Returning to Pearson’s argument, let us ask why systematists insisted upon theory-free observation; it is the solution to a long standing goal in systematics: to have a natural classification that is logically prior to theory, and which is also not subjective. For a long time, systematics was driven not so much by theory as by authority. My colleague Malte Ebach and I call this, somewhat unfairly to German-speaking science, the German Authority Syndrome. Sociologically, for a long time it was the case that if an Authority had worked on a group, nobody else could work on that group while the Authority was alive. This was especially true in German traditions, where authority is widely respected, but it exists even now in most systematics across cultures. This meant that the personal preferences of individuals who held senior positions could become established for wildly contingent reasons, such as the psychological dispositions or the individual to lump or split, or because of professional exigencies like unifying or splitting departments or funding.

Everybody thought this was a mistake, but nobody knew how to avoid it. As theory took hold as the arbiter of cognition in mid-twentieth century philosophy of science (particularly Kuhn’s view of paradigms constraining normal science), it became fashionable among philosophically inclined scientists to assert that theory drove observation, and this was therefore a way to avoid German Authority Syndrome. And the obvious theory that could unify all observation and determine what was and what wasn’t good taxonomy in biology was obviously evolution, and particularly theories of phylogeny (biological diversity and its development over time).

Pattern cladism objected to this. Again it depends on ambiguities in terms. By “the theory of evolution” do we mean the theory that all things have evolved by branching descent, or instead the hypothesis of the evolution of this group of organisms? This was not always kept clear and distinct in the debates of the 70s and 80s. PattCists held that we cannot, logically, presume a given mode of evolution or a given history in order to classify if we then expect that we can use phylogenetic classifications to test our hypotheses of relatedness. So, they asserted that some theory-free observation, particularly of homologs, was possible, to kick-start the process. Have observation sentences from the positivists been revived?

I suspect they never died, really, but the reality is much more complex than simple nomological deductive accounts of explanation and inference in science indicate. There is what I call a domain-relativity in play. That is, sure, theory is sometimes used in observation, but in a classificatory activity like systematics, the theory you use had better not be question-beggingly the theory you are trying to test or support with your classification. That is, you can use theories from outside the domain under investigation. If you were a Bayesian, this would set up some of the priors you use in testing the present hypothesis. An example might be using the ways a subgroup of spiders spin webs to set up their relations, in the context of wider views about how spiders in general spin webs. You do not come to the issue or domain ignorant of spiders, or, if you do, you might not be doing systematics. Nobody addresses any issue knowing nothing to begin with. But a lot of our knowledge is not theory-driven in any reasonable sense. I do not avoid large rapidly moving objects because I have a theory of cars, for example. I don’t even have a theory of large moving objects. I have learned that large moving things hurt me when they hit me, from experience (one broken leg and a crushed foot later…).

Observations can be theory-free for the domain investigated, even if they are theory-driven (that is, the theory is employed by the observer) by extra-domain theories. Moreover, the domain and its theory are in temporal relations. What one knows at one time tests what one previously knew, and what one knows now will test theories in that domain in the future that may, if successful, end up revising observations (“you were mistaken to think this was a spinneret; we know now it is a different structure not homologous to spinnerets”). Science is an iterative refining of observations and inference. Hennig, borrowing a term from stemmatics (the science of manuscripts) called “reciprocal illumination”, made the same point.

I have used Pearson’s paper as a jumping off point, and not given it the treatment it deserves, but I wish to end this post on a comment he makes almost en passant. He writes:

Indeed, as Scott-Ram (1990) argues at length, the very idea of a natural hierarchy by which pattern cladists seek to categorize the living world is problematic in the absence of evolutionary theory. For Scott-Ram, either the pattern cladists must recognize the role of evolutionary theory for a cladistic taxonomy or accept a Platonic view of the natural world. A Platonic view is, of course, unsupportable for the biologist, thus evolutionary theory should be recognized.

We need to challenge this major premise, that a natural hierarchy is inexplicable in non-evolutionary terms. Historically, nothing could be further from the truth. The natural hierarchy preceded evolutionary theory, and was, indeed, the main motivation for Darwin to develop it. If history has any weight in this, then a natural hierarchy is exactly as the pattern cladists think: it is prior to, and tests hypotheses of, common descent. I do not know what a “Platonic view of the world” might be – many claims have been made that evaporate on closer inspection – but when people have been scientific Platonists they tend to think things that have little to do with the progress over time of physical objects, rather than trying to make an account of the physical world in terms of ideas or forms. It is a phantasm rarely encountered. And to assert that pattern cladism is this mythological chimera is poisoning the well in a major way.

I think that systematists should be local pattern cladists, even when they are global evolutionary theorists. Just like Patterson was.