Natural classification and the dynamics of science

About thirty years ago there was much talk that geologists ought only to observe and not to theorize; and I well remember someone saying that at this rate a man might as well go into a gravel pit and count the pebbles and describe their colours. How odd it is that anyone should not see that all observations must be for or some view if it is to be of any service [Charles Darwin to Fawcett, 1861 (Hull 1973): 9]

… the work of theory and observation must go hand in hand, and ought to be carried on at the same time, more especially if the matter is very complicated, for there the clue of theory is necessary to direct the observer. Though a man may begin to observe without any hypothesis, he cannot continue long without seeing some general conclusion … he is led also to the very experiments and observations that are of the greatest importance … (and) the criteria that naturally present themselves for the trial of every hypothesis. [John Playfair, 1802; Illustrations of the Huttonian theory, Adell and Davies, London, p.524–525. As cited by (Dott 1998): 15].

Traditional philosophy of science, by which of course I mean what I was taught as an undergraduate (best summarised in these two books: Chalmers, and Godfrey-Smith), has it that what science does is develop, test, and argue over theories. Oddly, what a theory is, is rarely discussed*, although there is a consensus that a theory is a formal model with ancillary hypotheses and interpretations of some kind. But the focus has been on theories at least since John Stuart Mill’s A System of Logic in 1843, especially once that work was adopted as the basis for the burgeoning analytic philosophy movement in Britain and America, and the subsequent development of logical positivism and it’s heirs and successors.

Positivism was a linear historical progressivism about science. Comte himself held that societies moved through the theological, the metaphysical and then the positive stages. Likewise, sciences developed this way. This progressivism persisted long after positivism died, or transmuted into logical empiricism. Even as the Baconian idea of sciences developing from masses of naive observation into laws and theories was being abandoned, people still held that there was a constrained historical sequence for the development of sciences. For example, Kuhn’s “normal science/revolutionary science” distinction, and in particular his “evolutionary metaphor”:

Imagine an evolutionary tree representing the development of the modern scientific specialities from their common origins in, say, primitive natural philosophy and the crafts. A line drawn up that tree, never doubling back, from the trunk to the tip of some branch would trace a succession of theories related by descent. Considering any two such theories, chosen from points not too near their origin, it should be easy to design a list of criteria that would enable an uncommitted observer to distinguish the earlier from the more recent theory time after time. Among the most useful would be: accuracy of prediction, particularly of quantitative prediction; the balance between esoteric and everyday subject matter; and the number of different problems solved. – Those lists are not yet the ones required, but I have no doubt they can be completed. If they can, then scientific development is, like biological, a unidirectional and irreversible process. Later scientific theories are better than earlier ones for solving puzzles in the often quite different environments to which they are applied. That is not a relativist’s position, and it displays the sense in which I am a convinced believer in scientific progress. (Kuhn 1970: 205f)

History moves forward. Unfortunately, this is not true of biological evolution, and there is no reason to think it is true of cultural evolution either, so why should it be true in science? Why must science follow a set trajectory? The presumption here is that history is constrained to develop in particular ways. This is just false.

Philosophy of science has assumed that the early stages of a scientific discipline is marked by basically wandering about observing stuff until a theory, hypothesis or law suggests itself. Then it gets tested. We might represent this as a field of possibilities, in which one axis is the axes of conceptual development, and the other of empirical observation. Suppose that constructing a theory is a phase of active conceptualisation. We take our ideas and put them together into a coherent and explanatory structure, and having done so, we run experiments to test this and ensure that what we test is just the theory, and not confounding variables. This is, very roughly, Popperian falsification. And it is, most of the time, precisely not what sciences do. Popper and the positivists were criticised for failing to deliver any vestige of a logic of discovery, despite the English title of Popper’s masterwork. In fact, discovery was regarded as accidental, if anything. The real meat was in the construction and testing of hypotheses, leading to models and thence to theories.

Philosophies of science tend to distinguish between the conceptual and empirical aspects of science. Even views based upon the theory-dependence of observation make the distinction, if only to assert the priority of one over another. Let us take this as first approximation. Conceptual tasks are themselves divided into theoretical and classification tasks, the first being a model or representation of phenomena, and the second supposedly a systematization of the results of the dynamics captured by the theory/model.

The two conceptual tasks are usually held in opposition, although again some subordinate the one to the other (mostly that theory determines the sorts of categories into which things get sorted. More rarely, that one’s ontology, or classification of possible types of things, determines or constrains theories). Let us visualize each task as a set of goals connected by the common feature of being conceptual, like a dumb-bell. Empirical tasks, similarly, are divided into naive observation and more informed experimental testing, which involves knowledge of the theory. So, on this view of science, the “moments” between which scientific behavior “moves” look like this:

Screen shot 2010-08-06 at 2.39.00 PM.png

Fig. 1 The Bacon and Popper Cycles

The Bacon Cycle is shown as sequence B, while the Popper Cycle is shown as sequence P (being charitable and allowing Popper some classification, which in practice he dismissed the way Rutherford did, as uninteresting stamp collecting). Of course, all views of science hold it to be an iterative process, so if the results are not satisfactory, a movement can be indefinitely repeated. Let us now change the metaphor to a Cartesian graph with two axes: conceptual and empirical. A simple Popper cycle, with theory dependence, might inscribe a quite complex trajectory. If one is trying to work an experiment, or reformulate a model, loops will occur. The permutations for an extended process can become very extensive indeed. But what happens if we allow, as we surely must, that observation can inform classification, or that it can even, as Bacon had it, inform a theory or a model? Let’s fill in all the blanks:

Science movements

Fig. 2 The twelve movements and four moments of scientific processes

This characterizes any autonomous scientific process, from the mental processes of a researcher, to the work of a research group, to a general research program, to the activity of an entire discipline. The influences from one scientific task (“moment”) to another are represented as outputs feeding into inputs (“movements”) (Table 1). The processes within the moments include data analysis and other transformations local to that aspect of science. Experimental techniques, classification sequencing, observational processes and technologies, and theory-building are all aspects of their respective blackboxes (here shown as white circles).

Table 1






3. Experiment can suggest classification category

11. Experiment can restrict or guide “naïve” observations

2. Experiment can restrict theoretical range, or disconfirm theory


4. Classification can suggest things to measure and expect

5. Classification can guide the evidence sought

10. Classification can restrict or guides the ontology of a theory and the explanatory categories used


12. Naïve observation can influence the data used in experiment

6. Classification can be based on naïve observations made pretheoretically

7. Theoretical predictions can fail to be borne out in observation


1. Theory can specify legitimate experimental protocols and approaches

9. Theoretical variables can become classification categories

8. Observation can depend upon the ontology and methodology of a theory

On the B and P cycle and similar views, we have basically ignored three quarters of what it is that science does! There are twelve possible pathways for methodological influence of one task type to lead to results in another, plus the four pathways of self-correction and revision. It seems quite feasible to think that observation might be influenced by theoretical assumptions and expectations, or that we might develop classifications on the basis of our experience and the classification systems, neural and analytic, we apply to such data.

The combinatorial possibilities for any realistic sequence of research are immense, and if we add the possibility that these moves might occur in parallel, and that the moves might be distributed (research groups typically have many brains to do their work on, especially the more pliable brains of doctoral and postdoctoral students), we begin to have an extremely complex dynamic system. I leave it to the reader to work out how complex.

But it doesn’t end here. Many recent treatments of science claim that the history of a science is not simply determined by its internal methodological workings (internalism) but that even the very facts with which it deals, no less than the ways experiments are run, theories are constructed, and classifications developed, is a process influenced entirely or in part by its social and political milieu. Even if we hold that to a minimum, any given discipline, research program, or laboratory is influenced at some or each point by external factors such as funding, engineering and technological resources and equipment, and other disciplines.

The introduction of computers is a case in point; with effects of all three kinds – you have to pay for the hardware, the programmers, and bioinformaticians, and you are limited or freed according to the capabilities (and theoretical developments) outside your particular specialty. This view, if pushed to the limit, is called externalism. Some versions of externalism, known as constructivisms, go so far as to make scientific observations and theories depend entirely upon political, religious and philosophical ideologies. Scientists are just apologists for a worldview and their data are “negotiated” by the community of similar believers. This view goes far beyond the theory-dependence of observation doctrine of Popper towards ideology-dependence of observation stance. It often takes the form that evidence is selected in order to bolster the community’s core doctrines, like finding evidence for natural selection is supposed to shore up capitalist free-market economics.

Data is not the basis of science for constructivists – it is the goal set by prior theoretical commitments based on some wholly external set of beliefs or conditions. We can, I think, reject the strong claims of nai?ve constructivism. But there remains a residual point that cannot be ignored. Social and cultural conditions do indeed influence scientists, theories, research programs, and institutions. Often this influence is exercised by biases in funding and support, but it is also, throughout the history of science, derived from ideas, especially of philosophy. And these influences can be creative and positive, as well as “unscientific”. From Newton onwards, inspiration for hypothesis has been drawn from astrology, alchemy, theology, economics, literature and metaphysics. Sometimes attention is drawn to data that had not been previously examined, or were ignored as anomalies. These influences act as inputs into the scientific process, and we can therefore add to our schematic both these inputs, and the corresponding outputs – the influence of theoretical and taxonomic terms, observations and experiments – on the broader social and cultural systems in which it is situated.


Fig. 3 The scientific system in a social context.

From the perspective of the social milieu, science is an active and dynamic process that transforms data inputs and resources into representational outputs and products. Science, at any level from researcher to discipline, is a dynamic cognitive system. It is a formal analogue to any other cognitive system, such as a classifier or heuristic system implemented on a computer.

The metaphor of the dance floor

If we think of this field of possibilities as a dance floor, we can get a handle on how it is that individual scientists, laboratories, research programs, schools of thought, institutions and disciplines behave. If there is no simple or singular trajectory around this dance floor, we might expect that individuals and all other agents in science right up to disciplines will tend to congregate closest to the action, but where there is a reasonable chance of success. In other words, they will look for some elbow room, and rewards. If the choice of partners is so empty in one region that it won’t pay to go to that corner, then the science will tend to congregate in the areas where there is a choice of partners. When the density of dancers to partners is so high that your chances are low, you will move to a part of the floor where there is a better choice and chance.

Consequently, a new scientist, or discipline, or school of thought may embark on their science in a region not traditionally discussed by the philosophers, and one of those is the corner in which one does naive observation (passive observation in the field, usually), and classification. Or one might classify on the basis of experimental work, or make theory on the basis of the prior passive observation, and so on. Theory may lead to observation of either kind (“go out and see if guppies in Jamaica match the model of selective balance for mate attraction and predator avoidance”, as Endler did). A novel field, however, lacks dance partners to speak of, and so naive classification may be all one can do, to begin with. Our physics-fascination has led us to think that all science is theory and experiment, when even in physics that is not true (consider the periodic table).

Classification sets up the problems that theory has to deal with. A classification is the explanandum, that which is to be explained; theory is the explanans. However, theories and classifications are also historical objects; they change over time, and so a theory at t can influence a classification at t + n, and the classification can then influence the theory at t + n + m, and so on. Science is an iterative process over time.


Dott, Robert H. 1998. What is unique about geological reasoning? GSA Today October 1998:15-18.

Hull, David L., ed. 1973. Darwin and his critics; the reception of Darwin’s theory of evolution by the scientific community. Cambridge, Mass.: Harvard University Press.

Kuhn, Thomas S. 1970. The structure of scientific revolutions. 2nd enl. ed, International encyclopedia of unified science. Foundations of the unity of science; v. 2 no. 2. Chicago: University of Chicago Press.

* I have other posts on theory: Laws, theories and models, What is a theoretical object?, Why do scientific theories work? The inherent problem, and What philosophy of science and “postmodernism” have in common.

12 thoughts on “Natural classification and the dynamics of science

  1. Okay, now I am thoroughly confused. I don’t see any difference between theory/model-building and classification. What is that distinction supposed to be doing? I’m not trying to denigrate classification. Classifying stuff is important work. But I don’t see why it isn’t just theorizing/model-building of a particular sort. Example: The standard model of particle physics says there are 12 varieties (or classes) of Fermions. That classification is part of the theory. Example: Suppose I am interested in the effect of class size on student success. I want to control for confounds, so I collect information about, among other things, family income. I don’t want to think about family income as continuous, so I discretize it (common practice) into five categories 0-25k, 25k-50k, 50k-100k, 100k-250k, and 250k+. This classification scheme is part of my model.

    Maybe a better thing to do would be to blow up the theorizing/model-building moment and look inside, since there will be all sorts of craziness in there!

    Also, where do you locate things like instrument- and technique-development? Is that supposed to be external to science?


    1. I, too, would consider both model-building and classification the same sort of thing. But I wouldn’t consider classification the same as building a model; rather, I think both a model and a classification are representations.

      And, being a pragmatist, I also care a lot about things like instrument- and technique-development (and use!).

      So, all together, I guess I’m inclined to collapse the two `theoretical’ circles into one `representation’ circle and add a `pragmatics’ circle.


  2. FWIW, it would seem that psychologist Paul Rozin has been arguing that psychology is too hypothesis driven and needs more descriptive work. From a post by Hugo Mercier over at ICCI:

    Rozin’s second important contention is that psychology has become too much hypothesis-driven and that it pays too little attention to the simple study and reporting of phenomena. In order to drive this point home, he engages in a comparison of psychology with her big sister, biology. Rozin points out that when biology was developing, in the 18th and 19th centuries, the study of phenomena—natural history—was playing the major role, while hypothesis driven research only kicked in much later. Of course, someone could point out that psychology does not have to replicate what may have been the errors of a young science and that modern psychology is perfectly justified in adopting the latest research methods from biology. But this criticism would miss the mark, as even modern biology is much less hypothesis-driven than modern psychology.

    I’m quite sympathetic to that notion, especially if one regards literature — poems, plays, stories, oral or written — as a component of culture. There I believe that better description is perhaps the most important thing we can do. Of course, few would regard the study of literature as being scientific.

    But then, in what sense is the study of psychology scientific? If one takes a Kuhnian view, it would appear that psychology is pre-scientific. For the field is rife with competing schools organized around competing sets of models and hypotheses. The field as a whole hasn’t arrived at a paradigm. In fact, I think the Kuhnian view won’t quite do.


  3. “FWIW, it would seem that psychologist Paul Rozin has been arguing that psychology is too hypothesis driven and needs more descriptive work. From a post by Hugo Mercier over at ICCI:”

    I’m not that convinced that there is really such a conflict between descriptive and hypothesis-driven science in most fields of biology these days; I think this is being confused with the reality that there’s simply a range of strengths and focus of hypotheses from the very broad to the very specific (as there surely must be), and that sometimes the hypothesis need not be written down because it automatic follows from our investigative actions (what we’re looking for).

    It’s not like the old days when new purely descriptive observations could be made while out for a stroll. Now, to observe something new usually requires you to have been actively searching for something in the first place. So while it’s not uncommon to stumble upon an interesting observation while testing a hypothesis about a a separate phenomenon, in nearly all situations a competent scientist will nevertheless proceed to investigate that new observation with a fresh hypothesis and further experiments. Sometimes that hypothesis might simply be an affirmation of the observation, with subsequent experiments merely to test reproducibility and to rule out a chance event (which is incredibly important if you don’t want some other poor buggers wasting public funds on the basis of published findings that are actually bunk).

    Paleontologists and cosmologists are often put forward as examples of purely descriptive scientists, but when you look closely you realise that this is not true for the most part. Cosmologists are no more likely to spend tremendous amounts of time scanning the sky for shits and grins than paleontologists organise field trips on the basis of, “We’re just gonna wander around such-and-such a place and see what turns up”. Both go out looking for something (in part because, as Wilkins asserts, funding is a crucial influence on the scientific process, and it generally demands clearly stated research goals with their attending hypotheses and predictions).


  4. It gets worse. Whether we emphasize theory, classification, or observation, we mostly think of science as a cognitive enterprise; but science is also a process in the real world that not only yields an understanding of the properties of things but exploits these actual properties in the course of discovering them. It isn’t just the graduate students who don’t get proper credit in scientific articles, though I admit it would be a bit cumbersome to acknowledge the contributions of Andy the mouse and Betty the mouse and Cathy the mouse… And even that act of generosity would leave out the role of the reagents and instruments that not only must be assumed to follow the rules but actually follow them.

    Gaston Bachelard was good on this issue. He pointed out, for example, that the synthesis and purification of chemical species, something often done by craft people, was as critical to the emergence of chemistry from alchemy as observation, classification, or theory. I think the better part of what George LaTour’s has had to say also, to invoke the title of a Francois Ponge poem, involves Taking the Part of Things.

    We don’t just construct theoretical objects, we make physical substances and things that are good to think with and experiment on–think of where contemporary biology would be without the immense herds of pure-bread lab rats. Admittedly, once you start thinking of science this way, what was already sufficiently difficult becomes down right daunting, but as the black lady insisted in the old comedy record, “Sure learning about modern science is like having a swarm of bees in your head, but there they are.”


    1. Might that old comedy record be “We’re all bozos on this bus”?

      Follow the yellow rubber line…

      Hey man, he broke the president…


  5. Great post. Science is indeed iterative, but by no means follows a simple cyclic path. I think the context within which the research takes place and the history of both internal and external influences can be seen as a set of constraints on the scientific process. Not all trajectories may be allowed, or some may have a low probability associated with them as a result. The influence of these constraints is vividly apparent to those who have written a review article in any particular field.

    Would you suggest that novel classification, though, is a way to relax these constraints a bit?


    1. I think it is one of many ways in which a discipline that has gotten stuck on a local suboptimal peak in the conceptual landscape can shift off it. Another is a radical theory that divides the world up differently.


  6. “Science, at any level from researcher to discipline, is a dynamic cognitive system.”

    That may be true, but it is cognition restricted to analysis, generally excluding synthesis (technology, art, culture, etc). Analysis applied to synthesis will always lag behind.


  7. I suspect your B and P labels in Figure 1 are messed up, somehow. Neither produces a cycle.


  8. Hi John,

    I just wanted to let you know that I’ve included this essay in the latest Scientia Pro Publica carnival over on my blog. Do visit when you have a moment.



Leave a Reply to John S. Wilkins Cancel reply