Doing History & Philosophy of Science like Paleontology

Adrian Currie writes...

On Extinct, we discuss all kinds of odd creatures. Today’s odd creature is the history and philosophy of science, or HPS. As it says on the box, HPS involves an ‘H’ – history of science – and a ‘P’ – philosophy of science. Why would these two disciplines have their own club, with specific departments, dedicated journals, and so forth? During the 20th Century many philosophical arguments seemed to turn crucially on how the history of science went (how do theories change? Does science progress from less to more accurate theories?), and some figures—easily the most famous being Thomas Kuhn—began straddling the gap between philosophy and history in response. This had the promise of a historicised philosophy of science and, for that matter a philosophical history of science.

I want to suggest that HPS should be done more like paleontology. This is a surprising (and at this point super ambiguous!) claim, but stick with me and see what you think.

Disclaimer: I am—rather obviously—on the P side of HPS, and so it is likely that my discussion is skewed towards philosophical interests: I’d be pretty interested in hearing what those on the H take the philosophy to be doing for them. Further, although I’m framing this in terms of HPS, it could just as easily be framed in terms of philosophy of science in general, or perhaps philosophy of science in practice. I’m pretty free with labels in this sphere…

Is the H and P’s marriage going well? Should we expect it to go well? On one view, HPS is in effect just two disciplines co-occupying the same academic space: the philosophers do the philosophy, the historians do the history. On another view, the two are integrated: the HPS scholar is the perfect amalgam of the historian and the philosopher. In practice people fall in various places along the continuum. And, to be honest, they typically cluster on the extremes. This unclarity about what HPS is has, unsurprisingly, led to more than a little methodological hand-wringing on the part of the practitioners. In this post, I want to talk about one of the lynchpins of that hand-wringing: the use of case studies.

HPS is often carried out using case-studies: that is, particular episodes in science’s history. Here’s a caricature of a typical HPS paper. We begin with a few preliminary philosophical points, puzzles or arguments, then launch into a detailed analysis of some scientific event, debate, theory or whatever—an analysis which takes up the vast majority of the paper—and then somehow draw some philosophy out of it. Despite my tone, I approve of such papers (I’ve written a few myself, although in my case we’d have to take ‘history’ to mean ‘some science from last week’); but there’s a big problem: one I’ve previously called the curse of the case study. It goes like this.

On the face of it, philosophers make general claims about science (say, we should be realists about science, or whatever), yet they cite a single piece of evidence—the case study—to do so. Why is this problematic? Well, if you’re going to draw a conclusion about a class of things by looking at a sample of that class, at least two things are crucial to determining your conclusion’s warrant. First, the size of your sample compared to the class. Second, the expected heterogeneity of the class. If you’ve good reason to think the class is likely to be homogenous—that all the bits in it will be the same—then you don’t need so big a sample. If you’ve got a big enough sample, then even for heterogenous classes you can be assured that patterns in the sample are likely representative of the class.

However, a case study is, basically by definition, a single instance—a sample of one. And further, we’ve very good reason to expect science to be quite heterogenous. Why? Well, for lots of reasons. At base, science is a messy, human kind of thing. And messy, human kinds of things are often heterogeneous. As Jim Lennox has said,

[ ... ] the foundations of a particular scientific field are shaped by its history, and to a much greater degree than many of the practitioners of a science realize. There is more conceptual freedom in the way  theories—even richly confirmed theories—may be formulated and revised than is usually realized.

(Lennox 2001, 657)

Science is historically contingent: how it turns out depends on where it has been before. And historical contingency generates differences between cases. This doesn’t mean that there are no general claims to be had about science. Rather, it means that we should expect heterogeneity and thus should require a large sample set to warrant general conclusions. This suggests an argument concerning the use of case studies:

(1)    If case studies play a legitimate role in the philosophy of science, they must justify our claims.

(2)    Case studies cannot justify claims in the philosophy of science.

(3)    Therefore, the use of case studies in the philosophy of science is illegitimate.

Larry Laudan and company said this with a sting back in the ‘80s:

Nothing resembling the standards of testing that these very authors [postpositivist philosophers of science] insist upon within science has ever been met by any of their theories about science. Those of us who claim some modest expertise in the logic of empirical inference have been notably indifferent about subjecting our own theories to empirical scrutiny, even though our own philosophies of science suggest that without such scrutiny we might well be building castles in the air.

(Laudan et al. 1986, p 142).

They attempted to reform how the philosophy of science was done: articulate philosophical hypotheses as empirical claims about the history of science, and then collate a series of historical episodes to test them. We abandon the use of case studies and instead treat the history of science as an inductive base for drawing general philosophical conclusions and testing hypotheses.

Another—much more popular—move also accepts the conclusion but, instead of reforming practice in HPS, it pulls back its ambition. Instead of aiming for general conclusions, we aim for understanding the case studies for their own sake. Jim Lennox’s ‘phylogenetic approach’ captures this pretty nicely. He describes his work as “… the activity of understanding foundational problems in biology through a study of the historical origins and development of those problems” (657). We unravel contemporary conceptual confusion by unweaving their history.

I’ve no problem with people following either of these options (although so far as I know Laudan’s program has been effectively abandoned): HPS should be diverse, I reckon. But, I think there are other ways of viewing the use of case studies. In previous work, I’ve denied the second premise: acting as a base for enumerative induction is not the only way that they might justify claims in the philosophy of science. Back then, I went explicitly pragmatic. Philosophers construct conceptual and epistemic tools, and case studies play a crucial role in developing those tools, and articulating their various uses (I think Hasok Chang’s view is similar).

But today I want to try something else.

I think we’ve abandoned the idea that case studies aren’t a source of evidence for developing philosophical accounts of science too quickly. This doesn’t requires us going full Laudan, but it does involve taking the evidential role of case studies seriously. I’ve come to this view based on thinking about historical science. So, let’s take a step away from method in HPS, and towards method in paleontology.


Way back in 1998 Aviezer Tucker pointed out that explaining unique events is really tricky. To explain an event, we typically need to model that event. But to build a good model of an event, we need evidence. In particular, we often would like a bunch of other events which we can use to build a model of our target. This is what we’re doing when we examine asteroid strikes generally to help us understand that infamous one that hit the Earth took 66 million years ago. If I have a lot of cases of asteroid strikes, I can use enumerative induction to support my hypotheses about a particular asteroid strike. But if an event is unique, that means we simply don’t have other examples, so can’t build a model—there’s no enumerative induction to be had. Further, because history is so contingent, we should expect unique events.

So, how do historical scientists tackle the apparently unique? I reckon they do it using what I call the Exquisite Corpse method. I’ve explained what that is in an earlier post, so here I’ll just sketch the case and the view (if you want the full story check out chapter 8 of Rock, Bone and Ruin: available now for pre-order oooh!). Check out these extinct mammals:


On the right is Smilodon fatalis, the poster-cat for North American sabre-toothed lions. On the left is Thylacosmilus atrox and it isn’t a cat at all: it’s a South American almost-marsupial (notice how consciously thylacine-like the reconstruction is!). T. atrox has some unique properties. In particular, it appears to have a weirdly weak bite-force. We’ve good reason to think T. atrox was a fairly major predator (although it did share an environment with Terror Birds…), and predators tend to have strong bites. What’s going on?

It turns out that T. atrox doesn’t bite the way most other carnivores do. Instead of its jaw muscles doing most of the work, its neck muscles take up the majority of the strain. How did we figure that out? Well, the hypothesis was developed by looking at a whole bunch of different lineages. Ancient synapsid sabre tooths are examined to help understand the function of sabre-teeth, typical cats such as panthers provide a contrast to T. atrox’s bite-mechanics, bears are drawn on to suggest morphological adaptations to deal with weak jaw muscles, marmosets are examined to figure out what kinds of morphological adaptations might mitigate T. atrox’s enormous gape. By navigating between a bunch of imperfect analogies, we’re able to develop a pretty reasonable empirical model of how that apparently unique critter does its biting.

How should we characterize this practice? None of those lineages are pure analogues of T. atrox—they’re not a base for a traditional enumerative induction like those asteroids strikes I mentioned earlier—however, each has a property or a set of properties which can be drawn on to build a composite model of our target. So long as such imperfect analogues are around (or can be constructed!), then there is empirical juice available for developing well-confirmed, robust models and hypotheses concerning unique events. (this of course relies of events being unique insofar as that combination of properties is unique, as opposed to the properties themselves being unique – but I’d bet that the latter case is *way* more rare). I’ve called the practice of navigating between various imperfect analogues to construct composite models of particular lineages the Exquisite Corpse method.

So what does this have to do with HPS? Easy: the thought underwriting the argument against the use of case studies was that science is too heterogeneous for inductions to be made on the basis of a few cases. This is pretty close to what is going on in paleontology: historical contingency breeds uniqueness, which is really just to say that history will be heterogeneous.

So, if historical scientists can use exquisite corpse methods, why don’t the folk in HPS? I think we just haven’t tried. There doesn’t seem to be any in-principle reason why I can’t use particular episodes in science to inform composite models targeting particular episodes. This doesn’t mean that HPS would suddenly be targeting great big general claims about the fundamental nature of a unified science—far from it—rather we would be developing local models which partially integrate a series of scientific practices, disputes, or whatever. We might find ourselves making some pretty general claims (I have a few in my back pocket, but I’ll leave you in suspense as to what they are), but they’ll be built from the bottom up. It also doesn’t mean that HPS needs to eschew the particularism and concern for differences which is so distinctive of the ‘H’ in the equation. In fact, composite models make the very uniqueness of particular episodes much clearer to see. Understanding T. atrox’s unique bite by comparison with big cats and other sabre-teeth didn’t merely allow us to infer what the bite was like, but see clearly what made it unique.

The central claim here is that we needn’t trap ourselves between a generalist rock or a particularist hard-place. Like paleontologists, we can use partial analogues to construct composite models to help us understand particular cases.

An Objection (owing to Edouard Machery): but Adrian, surely a stark disanalogy between HPS and paleontology is that paleontology can make use of well-established, well-verified theories like biomechanics, molecular phylogenetics, and other fancy things. There’s nothing like that for HPS!

I’m not convinced. First, we shouldn’t over-emphasize the role which those quantified theories play in the kinds of paleontological reconstructions I’m discussing. Undoubtedly, biomechanical analysis matters for understanding T. atrox’s bite, but I don’t think it is privileged: the warrant of the neck-bite hypothesis is tied together by a wide variety of empirical and theoretical threads, from narrative links, to consideration of ecological context, to fine-grained fossil analysis, and so forth: fancy theories are there, to be sure, but it’s a joint effort. Further, fancy theories aren’t incorporated into paleontology in a pure form: they must be calibrated and altered in order to be relevant to the cases they are pointed at. Often, this involves them being transformed: they not so general after all. Second, are we so sure that there aren’t analogues of biomechanics or molecular phylogenetics for science or its history—have we really looked for them in the right way? The sense I get is that when historians or philosophers have thought about generalities across science, they have fallen into the rock-and-hard-place between particularism and generalism I mentioned above. It may be that perfectly good fodder for exquisite corpse method was abandoned because it wasn’t general enough.

This is all just a sketch and a suggestion which could definitely do with being filled out with examples, but nonetheless I think a case can be made for still thinking about case studies in HPS as playing evidential roles. If we do HPS like a paleontologist, that is.

I presented some of these ideas at a recent workshop run by the When Experts Disagree folk at UCDublin. Thanks for having me and some useful discussion!