It occurs to me that I haven’t plugged my own book here. What a failure on my part! It was published in December, so it is really time I did so.
In this book, Malte Ebach and I discuss a topic not often discussed in the philosophy of science: the classification of nature in the absence of a theory that delineates natural kinds. Since when we begin the investigation of a new field, there is no theory of that domain yet, by definition, how do we begin? The standard answer has been that we take the kinds posited by a closely related theory and refine the theory to include that domain. But this doesn’t account for the domains that have no closely related theories that could potentially cover them. For example, classifications of living things began, and were relatively sophisticated, well in advance of anything resembling a theory of life. Instead, researchers refined the folk taxonomies in existence, and approached the domain in a naive empirical fashion. In short, they looked for patterns in the data.
The Standard View is that what we observe is determined by our theories – we literally cannot see what we do not expect. This, the theory-dependence of observation hypothesis, presumes rather bluntly that the observational salience of phenomena is first constructed and then observed. This may be true in cases where there already exists an elaborated theory that is relatively relevant to this new domain. And it is almost dogma that we do not have what Michela Massimi has called “ready-made phenomena”. It is this we take some guarded exception to. There is the following conundrum: if we cannot see until we have a theory, and yet we can learn to see phenomena as children, then “theory” must include not only the formal explanatory models of science, but indeed any disposition to see some things and not others, which makes theory everything we are biased to observe. This is, I think, to attenuate the notion of “theory” so far that it becomes a meaningless term. We see much of what we see because we evolved to see it, so our evolutionary past becomes theory. A concept like that cannot be useful in science. It is much better to separate our dispositions inherited from biology and culture out from the technical apparatus of theory in science, so we capitalise the latter: Theory.
Our brains are wired to find patterns in data, in large part because they are neural networks, and that is what neural networks are good at. We are classifier systems. A classifier system finds regularities in large data sets (including, but not just, measurements using instruments like a pan balance or a thermometer), and once they are found, they call for an explanation. Now, I realise that induction is supposed to be a problem in philosophy, as no finite number of observations of this kind can determine a unique general solution or generalisation, and yet, that is what every child who learns a language or not to touch hot things does. There simply are some ready-made phenomena, even if we cannot justify the regularities deductively. I think we might do so abductively, though, just as Peirce thought.
So, we formulate our classifications and find patterns to explain. We give examples of this in meteorology, pedology (soil science), chemistry, psychiatry, and several other cases. But of course the other kind of natural kinds occur too. It actually is the case that some kinds are formulated by our theories. How do we relate the two? The answer lies, we argue, in the dynamic nature of science. Science is not just a theory-driven enterprise, but when we have theory, we test it and refine it on empirical foundations. If a theory asserts the existence of some kind, and we find that it does so with precision and accuracy, then we have confidence that the kind is real. But if we find the kind is not matching the patterns we identify from direct observation, whether experimental or field observations, then there is something wrong with the theory (i mean here, all the theories used in the investigation of the domain; some theoretical foundations include distal theories as components).
We argue that theory-naive (or just “naive”) classifications are formed by a process of trial and error, to determine the marks that make the classification stable and useful. We call these marks, following the biological practice, homologies, and the marks that are not good discriminata, analogies. In brief, homologies allow projection of our inductive conclusions, while analogies offer no more information than is used in their construction. To give an example, all homologically related organisms in a classification will tend to share the same sets of marks (“characters”), so if most or all of that group have been found to have a mark, any newly discovered kind of organism will very likely have it too, even if that has not been observed. Homological classifications are, as Goodman termed it, projectible.
This isn’t a general solution to the grue problem Goodman formulated, but then classification isn’t about induction. It is about recognising patterns and using those patterns to refine our beliefs. We have to do this in order to survive, as Quine noted a long time ago, but at best a classification is something that is highly defeasible, and the relation of classification to Theory is itself dynamic. A classification is at its best, the beginnings of knowledge. It is a call for Theory to explain what we see.
In one chapter – Monster and misclassifications – we discuss how unnatural classifications are formed and what they tell us. I would summarise it this way: a natural classification formed on homologies, which are usually causal regularities or etiological classifications, tells us two things. It tells us what we, the observers, find salient (as does any statement about the world, for there are an infinite number of things we might say), but it also tells us about the structure of the domain being observed. It is a two-place system, us and it. But a monstrous classification (which includes paraphyletic classifications in biology, for example) tells us only about our salience dispositions. While there are observations being made of the world, the selection of the marks themselves is all about us, and so a monstrous classification is a statement about us and our dispositions only.
Finally, in order to evade the political aspects of classification, we propose a neutral, functional, set of terms to discuss what scientists mean by classificatory terms. Science is done by people, and people, among other things, play political games (in the Wittgensteinian sense of “game”) in order to mark in-group from out-group loyalties. This, while it may seem to be less than admirable from a purely formal perspective, is an irreducible aspect of human science, and indeed it may tend to drive science by motivating, as David Hull called it, the “I’m gonna get that son of a bitch” responses to claims, thus acting as a selective pressure against groupthink. Since all scientists do or ought to take empirical evidence as their starting point, if an opponent can show that a model is not empirically adequate enough, even abstract ideas like “monophyly” in biology can be revised.
I commend our book to my readers. It seems like a silly thing to write about, but I believe the classification of nature has for too long been seen as a merely conventional practice in science. Nobody would deny that there is a conventional aspect to science, just as there is any other human social activity, but it is time to abandon the idea that this is all there is to it.
One final note. This book was written, at the editor’s instructions, to be accessible to both philosophers and scientists, so the language may seem a little untechnical. The result is surely that the scientists will find it too difficult to read, and the philosophers not difficult enough. I beg of my readers not to think that because of this it is either useless or shallow.