Society for Philosophy of Science in Practice, University of Twente, August 2007
Joseph Rouse, Wesleyan University

DRAFT — Please do not cite or quote.

Why does the philosophy of scientific practice matter to philosophy? This question is prompted by dramatic change in the place of philosophy of science within the discipline. Both logical empiricist philosophy of science and its early post-empiricist successors were influential for other philosophical work on mind, language, knowledge, and even ethics. More recent philosophy of science now provides a better understanding of the diverse sciences. Yet these philosophical advances in understanding science remain isolated from the philosophical mainstream in metaphysics, epistemology, and the philosophy of language and mind. In the U.S., where the predominant meta-philosophy is some version of naturalism, this disconnection is ironic. Science sets the horizons for philosophical inquiry, but recent philosophy of science has played a minimal role in shaping the conception of science that other philosophers invoke.

My question about the philosophical significance of the philosophy of scientific practice gains urgency in this context. Will the philosophy of scientific practice merely become a narrower specialist niche within an already isolated sub-discipline of philosophy? Or does attention to scientific practice promise to restore the philosophy of science to a more central place in philosophy?

I believe that the philosophy of scientific practice can indeed make important contributions to philosophy more generally, in part by challenging naive conceptions of science often taken for granted elsewhere. Today I shall only consider a single prominent issue. Kant famously proclaimed that “Concepts without intuitions are empty; intuitions without concepts are blind.” Yet the question of how concepts acquire content from their relation to experience has troubled philosophy ever since Kant’s proclamation. Logical empiricist and early post-empiricist philosophy of science notoriously struggled with this issue. More recently, the question has gained renewed prominence from John McDowell’s influential lectures on Mind and World. McDowell argued that contemporary philosophy has failed to negotiate safe passage between two dangerous attractors. Quine and other empiricists invoke conceptions of experience as “Given” that allow it no bearing upon our conceptual understanding. Davidson, Rorty, and other pragmatists circumvent this Myth of the Given, but only by treating conceptual spontaneity as merely internally coherent. If their views were correct, McDowell argues, conceptual thought could only be a “frictionless spinning in a void,” utterly disconnected from accountability to the world.

It is important not to conflate McDowell’s or Kant’s question with skepticism about the justification of empirical knowledge. Before we can ask about the empirical justification of a claim, we must understand what it claims. Most philosophers distinguish these issues of conceptual articulation and empirical justification by a division of labor. They treat conceptual articulation as an entirely linguistic or mathematical activity of developing and regulating inferential relations among sentences or equations. Experimentation and observation then address only the justification of the resulting claims. Quine succinctly expressed this division of labor in “Two Dogmas of Empiricism.” His famous image depicted scientific theory as a self-enclosed fabric or field that only encounters experience at its periphery. If experience then conflicts with our theoretical predictions, we must go back to make internal adjustments to our theories and try again. Yet that division of labor between internal conceptual development and external empirical testing is the central target of McDowell’s criticism.

How might a philosophy of scientific practice contribute to a better response to McDowell’s and Kant’s concerns? First, it transforms the problem by understanding the sciences’ accountability to the world in terms of experimental and fieldwork practices rather than perceptual receptivity. Second, I shall argue, this transformation then shows that conceptual articulation is not merely a matter of spontaneous thought in language or mathematics, and thus not merely intralinguistic; instead, experimental practice itself can contribute to the articulation of conceptual understanding.

To develop this claim, I begin by asking you to think about a well-known remark by Ian Hacking:

In nature there is just complexity, which we are remarkably able to analyze. We do so by distinguishing, in the mind, numerous different laws. We also do so by presenting, in the laboratory, pure, isolated phenomena. (1983,226)

By “phenomena,” Hacking means events in the world rather than appearances to the mind, and he claims that most phenomena are created in the laboratory rather than found in nature. Experimental work does not simply strip away confounding complexities to reveal underlying nomic simplicity; it creates new complex arrangements as indispensable background to any foregrounded simplicity. Yet I think most philosophical readers have not taken Hacking’s suggested parallel between phenomena and laws as modes of analysis sufficiently seriously. We tend to think only laws or theories allow us to analyze and understand or explain nature’s complex occurrences. Creating phenomena may help discern relevant laws or construct illuminating theories, but they can only indicate possible directions for analysis. Conceptual development must take place “theoretically.” Yet I think this treatment of laboratory phenomena as merely indicative means to the verbal or mathematical articulation of theory is mistaken. It is not enough to acknowledge that experimentation also has its own ends. Experimental practice can be integral rather than merely instrumental to achieving conceptual understanding.

Understanding the conceptual role of experimental practice requires us to look beyond particular phenomena to consider the development and exploration of what Hans-Jörg Rheinberger (1997) calls “experimental systems.” I long ago described such systems as “microworlds: systems of objects constructed under known circumstances and isolated from other influences so that they can be manipulated and kept track of, ... [allowing scientists to] circumvent the complexity [with which the world more typically confronts us] by constructing artificially simplified ‘worlds’” (1987, 101). Some illustrative experimental systems or microworlds include the Morgan group’s system for mapping genetic mutations in Drosophila melanogaster, the many setups in particle physics that direct a source of radiation toward a shielded target and detector, or the work with alcohols and their derivatives that Ursula Klein argued were the beginnings of experimental organic chemistry. These are not verbal, mathematical or pictorial representations of some actual or possible situation in the world. They are not even physical models, like the machine-shop assemblies that Watson and Crick manipulated to discover 3-dimensional structures for DNA. They are instead novel, reproducible arrangements of some aspect of the world.

Today, I consider a special class of experimental systems. Heidegger, whose writings about science emphasize the practice of scientific research, forcefully characterized the role I am attributing to these systems:

The essence of research consists in the fact that knowing establishes itself as a “forging-ahead” (Vorgehen) within some realm of entities in nature or history. ... Forging-ahead, here, does not just mean procedure, how things are done. Every forging-ahead already requires a circumscribed domain in which it moves. And it is precisely the opening up of such a domain that is the fundamental process in research. (1950, 71; 2002, 59, tr. modified)

What does it mean to open up a scientific domain, and how are such openings related to the construction of experimental systems? Popular presentations of scientific progress often emphasize the replacement of error and superstition by scientific knowledge. Yet in many areas of scientific work, the very phenomena at issue were previously inaccessible. Earlier generations could not be in error about these matters, because they could have little or nothing to say about them. The establishment of new experimental systems opened new possibilities for conceptual articulation, where previously there was, in Hacking’s apt phrase, “just complexity.” Some salient examples of domains opened by new experimental systems include genetics, by the Morgan group’s Drosophila system correlating crossover frequencies with variations in chromosomal cytology (Kohler 1994); quantitative temperature, through the development of intercalibrated practices of thermometry, so nicely described by Hasok Chang (2004); interstellar distances, through Leavitt’s and Shapley’s tracking of period-luminosity relations in Cepheid variables; the functional significance of intra-cellular structure uncovered with ultracentrifuges and electron microscopes (Bechtel 1993, Rheinberger 1995); or sub-atomic structure, first intimated by Rutherford’s targeting of gold leaf with beams of alpha particles. Prior to the development of those experimental practices, these corresponding aspects of the natural world lacked the manifest differences needed to sustain conceptual development. What changed the situation was not just new kinds of data, or newly imagined ways of thinking about things, but new interactions that articulate the world itself differently.

To understand this claim, we must recognize that experimental systems always have a broader “representational” import. It is no accident that biologists speak of the key components of their experimental systems as model organisms, and that scientists more generally speak of experimental models. The cross-breeding of mutant strains of Drosophila with stock breeding populations, for example, was neither interesting for its own sake, nor merely a peculiarity of one species of Drosophila. The Drosophila system was instead understood, rightly, to show something of fundamental importance about genetics more generally; indeed, I shall argue, it constituted genetics as a distinct research field.

As created artifacts, laboratory phenomena and experimental systems have a distinctive aim. Most artifacts, including the apparatus within an experimental system, are used to accomplish some end. The end of an experimental system itself, however, is not what it does, but what it shows. Experimental systems are novel re-arrangements of the world that allow some features that are not ordinarily manifest and intelligible to show themselves clearly and evidently. Sometimes such arrangements isolate and shield relevant interactions from confounding influences. Sometimes they introduce signs or markers into the experimental field, such as radioisotopes, genes for antibiotic resistance, or correlated detectors. Understanding this aspect of experimentation requires that we reverse the emphasis from traditional empiricism: what matters is not what the experimenter observes, but what the phenomenon shows.

Catherine Elgin (1991) develops this point by distinguishing the features or properties an experiment exemplifies from those that it merely instantiates. In her example, rotating a flashlight 90 degrees merely instantiates the constant velocity of light in different inertial reference frames. The Michelson/Morley experiment exemplifies that constancy. Elgin thereby emphasizes the symbolic function of experimental performances, and suggests parallels between their cognitive significance and that of paintings, novels, and other artworks. She claims that a fictional character such as Nora in A Doll’s House, for example, can strikingly exemplify a debilitating situation, which the lives of many actual women in conventional bourgeois marriages merely instantiate.

Elgin’s distinction between actual experiments and fictional constructions gives priority to instantiation over exemplification. Nora’s life is fictional, and is therefore only metaphorically constrained. Light within the Michelson interferometer, by contrast, really does travel at constant velocities in orthogonal directions. The constancy of light’s velocity is already ‘there’ in the world, awaiting only the articulation of concepts that allow us to recognize it. Unexemplified and therefore unconceptualized features of the world would then be like the statue of Hermes that Aristotle thought exists potentially within a block of wood. Their emergence awaits only the sculptor’s (or scientist’s) trimming away of extraneous surroundings.

In retrospect, with a concept clearly in our grasp (or better, with us already in the grip of that concept), the presumption that it applies to already-extant features of the world is unassailable. Of course there were mitochondria, spiral galaxies, polypeptide chains and tectonic plates before anyone discerned them, or even conceived their possibility. Yet this retrospective standpoint, in which the concepts are already articulated and the only question is where they apply, crucially mislocates important aspects of scientific research. In Kantian terms, researchers initially seek reflective rather than determinative judgments. Scientific research must articulate concepts with which the world can be perspicuously described and understood, rather than simply apply those already available. To be sure, conceptual articulation does not begin de novo. Yet in science, one typically recognizes such prior articulation as tentative and open-textured, at least in those respects that the research aims to explore.

When a domain of research has not yet been conceptually articulated, the systematic character of experimental operations becomes especially important. Domain-constitutive systems must have sufficient self-enclosure and internal complexity to allow relevant features to stand out through their mutual interrelations. That scientific experimentation typically needs an interconnected experimental system is now widely recognized in the literature. Yet the importance of experimental systematicity is still commonly linked to questions of justification. Thus, Ludwik Fleck (1979) long ago claimed that, “To establish proof, an entire system of experiments and controls is needed, set up according to an assumption or style and performed by an expert” (Fleck 1970, 96, my emphasis). Epistemic justification was likewise the issue for Hacking’s (1992) discussion of the “self-vindication” of the laboratory sciences. Their self-vindicating stability, he argued, is achieved in part by the mutually self-referential adjustment of theories and data.

I am making a different claim: typically, new domains are opened to contentful conceptual articulation at all by creating systematically intraconnected “microworlds”. “Genes,” for example, changed from merely hypothetical posits to the locus of a whole field of inquiry (“genetics”) by the Morgan group’s correlations of cross-over frequencies of mutant traits with visible transformations in chromosomal cytology, in flies cross-bred to a standardized breeding population. As a different example, Ursula Klein showed that carbon chemistry likewise became a domain of inquiry, distinct from the merely descriptive study of various organically-derived materials, through the systematic, conceptually articulated tracking of ethers and other derivatives of alcohol (Klein 2003). Leyden jars and voltaic cells played similar roles for electricity. What is needed to open a novel research domain is typically the display of an intraconnected field of reliable differential effects: not merely creating phenomena, but creating an experimental practice.

This constitution of a scientific domain accounts for the conceptual character of the distinctions that function within the associated scientific field. Consider what it means to say that the Drosophila system developed initially in Morgan’s laboratory at Columbia was about genetics. We need to be careful here, for we cannot presume the identity and integrity of genetics as a domain. The word ‘gene’ predates Morgan’s work by several years, and the notion of a particulate, germ-line “unit” of heredity emerged earlier from the work of Mendel, Darwin, Weismann, Bateson and others. Yet the conception of genes as the principal objects of study within the domain of genetics marks something novel. Prior conceptions of heredity did not and could not distinguish genes from the larger processes of organismic development in which they functioned. What the Drosophila system initially displayed, then, was a field of distinctively genetic phenomena. The differential development of organisms became part of the experimental apparatus that articulated genes, by connecting relative chromosomal locations, characteristic patterns of meiotic crossover, and phenotypic outcomes.

What the Drosophila system thus did was to allow a much more extensive inferential articulation of the concept of a gene. Concepts are marked by their availability for use in contentful judgments, whose content is expressed inferentially. For example, a central achievement of Drosophila genetics was the identification of phenotypic traits with chromosomally-located “genes.” Such judgments cannot simply correlate an attributed trait to what happens at a chromosomal location, because of their inferential interconnectedness. Consider the judgment in classical Drosophila genetics that the Sepia gene is not on chromosome 4. This judgment does not simply withhold assent to a specific claim; it has the further content that either the Sepia gene has some other chromosomal locus, or that Sepia mutants vary in more (or less) than one “gene”. Such judgments, that is, indicate a more-or-less definite space of alternatives. Yet part of the content of the “simpler” claim that Sepia is on chromosome 3 is the consequence that it is not on chromosome 4. Any single judgment in this domain presupposes the intelligibility of an entire conceptual space of interconnected traits, loci, and genes (including the boundaries that delimit that space).

To open such a conceptual space, experimental systems need not be typical or representative of the domain. Consider once more Drosophila melanogaster as an experimental organism. As the preeminent model system for classical genetics, Drosophila was quite atypical. As a human commensal, it is relatively cosmopolitan and genetically less-diversified than alternative model organisms. More important, Robert Kohler has shown that for D. melanogaster to function as a model system, its atypical features had to be artificially enhanced. Much of its residual “natural” genetic diversity had to be removed from experimental breeding stocks (Kohler 1994, ch. 1, 3, 8). Drosophila is even more anomalous in its recently acquired role as a model system for evolutionary-developmental biology. Drosophila is now the textbook model for the development, developmental genetics, and evolution of animal body plans generally (Carroll, et al. 2001, esp. ch. 2-4). Yet the long syncytial stage of Drosophila development is extraordinary even among Arthropods. In fact, however, domain-constitutive systems need not even instantiate the features they exemplify. The Michelson-Morley experiment, after all, exemplifies what happens to light in inertial reference frames, but the experiment itself is gravitationally accelerated rather than inertial.

Experimental systems mediate the empirical accountability of verbally articulated concepts to the world, which allows the use of those concepts to be more than just McDowell’s “frictionless spinning in a void” (1994). Mary Morgan and Margaret Morrison (1999) have compellingly characterized theoretical models as partially autonomous mediators between theories and the world. I am claiming that scientific understanding is often doubly mediated; experimental systems mediate between the kinds of models Morgan and Morrison describe, and the circumstances to which scientific concepts ultimately apply. The explication of these models within the microworld of an experimental system is what allows them to have intelligible applications elsewhere. Moreover, in many cases, the experimental model has to come first. It introduces relatively well-behaved circumstances that can be tractably modeled in other ways, such as a Drosophila chromosome map.

To understand the significance of this claim, we need to ask what “well-behaved circumstances” means here. Nancy Cartwright (1999, 49-59) has raised similar issues by talking about mediating models in physics or economics as “blueprints for nomological machines.” Nomological machines are arrangements and shielding of various components, so that their capacities reliably interact to produce regular behavior. I want to expand her conception to include not just regular behavior, but conceptually articulable behavior more generally.

I nevertheless worry about her metaphors of blueprints and machines. The machine metaphor suggests an already determinate purposiveness, something the machine is a machine for. With purposes specified, Cartwright’s normative language in discussing nomological machines becomes straightforward: she speaks of successful operation, running properly, or of arrangements that are fixed or stable enough. Yet where do the purposes and norms come from? The need for such norms is the most basic reason to think about experimental systems as opening research domains by mediating between theoretical models and worldly circumstances. They help articulate the norms for circumstances to be “well-behaved,” and for nomological machines (or experiments with them) to run “properly” or “successfully.” Scientific concepts, then, both articulate and are accountable to norms of intelligibility, expressed in these notions of proper behavior and successful functioning.

For theoretical models and the concepts they employ, Cartwright and Ronald Giere (1988) seek to regulate their normativity in terms of either their empirical adequacy, or their “resemblance” to real systems. In discussing the domain of the concept of ‘force’, for example, Cartwright (1999, 28) claims that,

When we have a good-fitting molecular model for the wind, and we have in our theory ... systematic rules that assign force functions to the models, and the force functions assigned predict exactly the right motions, then we will have good scientific reason to maintain that the wind operates via a force.

Giere in turn argues that theoretical models like those for a damped harmonic oscillator only directly characterize fictional, abstract entities of which the models are strictly true, whose relation to real systems is one of relevant similarity:

The notion of similarity between models and real systems ... immediately reveals— what talk about approximate truth conceals— that approximation has at least two dimensions: approximation in respects, and approximation in degrees. (1988, 106)

For my concerns, however, considerations of similarity or empirical adequacy come too late. What is at issue are the relevant respects of possible resemblance, or what differences in degree are degrees of. These matters could be taken for granted in mechanics, because the relevant experimental systems were long ago established and stabilized, and mutually adjusted with the relevant idealized models. The pendula, springs, falling objects, and planetary trajectories that make up the domain of mechanics were mostly already in place.

That is not the case when scientists begin to formulate and explore a new domain of phenomena. For example, Mendelian ratios of inheritance obviously predated Morgan, but spatialized “linkages” between heritable traits were novel. The discovery that the white-eyed mutation was a “sex-linked” trait was an anchoring point within the emerging field of mutations, much as the freezing and boiling points of water helped anchor the field of temperature differences. Yet as Chang (2004, ch. 1) has shown in the latter case, these initially “familiar” phenomena could not be taken for granted; to serve as anchors for a conceptually articulated space of temperature differences, the phenomena of boiling and freezing required canonical specification. Such specification required practical mastery of techniques and circumstances as much or more than explicit definition. Indeed, my point is that practical and discursive articulation of the phenomena had to proceed together. Likewise with the development and refinement of the instruments through which such phenomena could become manifest, such as thermometers for temperature differences or breeding stocks for trait-linkages.

Recognizing that domain-opening experimental systems help constitute the norms for their own assessment may seem to raise the spectre of idealism. Do they merely stipulate standards for the application of a concept, without being accountable to further normative considerations? No. Indeed, that is why recognizing the role of experimental practice in articulating scientific concepts and norms is especially important. Chang’s (2004) study of thermometry practices illustrates one important reason why such norms are not merely stipulated. There are many ways to produce regular and reliable correlates to changes in heat under various circumstances. Much work went into developing mercury, alcohol, or air thermometers along with their analogues at higher and lower temperatures. Yet it not enough just to establish a reliable, reproducible system for the thermal expansion or contraction of some canonical substance, and use it to stipulate degrees of heat (or cold). The substantial variations in measurement among different standard systems suggested a norm of temperature independent of any particular measure, however systematic and reproducible it was. Such a norm, once it was coherently articulated, introduced order into these variations by establishing a standard for assessing its own correctness. That the development of a standard is itself normatively accountable is clear from the possibility of failure: perhaps there would have been no coherent, systematic way to correlate the thermal expansion of different substances within a single temperature scale.

The most dramatic display of the defeasibility of experimental domain constitution, however, comes when domain-constituting systems are abandoned or transformed by constitutive failure, or forced re-conceptualization. Consider the abandonment in the 1950's of the Paramecium system as a model organism for microbial genetics. Paramecium was dealt a double blow. Its distinctive advantages for the study of cytoplasmic inheritance became moot when the significance of supposed differences between nuclear and cytoplasmic inheritance dissolved. More important from my perspective, however, was the biochemical reconceptualization of genes. The identification of genes as enzyme-makers emerged from the study of biochemically-deficient mutants. Such mutations could only be displayed in organisms like Neurospora that could grow on a variable nutrient medium. Despite extensive effort, Paramecium would not grow on a biochemically controllable medium, and hence could not display auxotrophic mutations. In Elgin’s terms, the cytogenetic patterns in Paramecium could now only instantiate the distinctive manifestations of genes, and could no longer exemplify them.

A different kind of failure occurs when the “atypical” features of an experimental system become barriers to adequate conceptual articulation. For example, standardization of genetic background made the D. melanogaster system the exemplary embodiment of chromosomal genetics. Yet this very standardization blocked any display of population-genetic variations and their significance for evolutionary genetics. Theodosius Dobzhansky had to adapt the techniques of Drosophila genetics to a different species to display the genetic diversity of natural populations (Kohler 1994, ch. 8). For a currently controversial example, Jessica Bolker (1995) has argued that the very features that recommend the standard model organisms in developmental biology may be systematically misleading. Laboratory work encourages using organisms with rapid development and short generations; these features in turn correlate with embryonic prepatterning and developmental canalization. The choice of experimental systems thereby materially conceives development as a relatively self-contained process. A reconceptualization of development as ecologically-mediated may therefore require its exemplification in different experimental practices, which will likely employ different organisms.

I have been arguing that attention to scientific practice provides important new resources for philosophy of language and mind. Recurrent philosophical anxieties about the groundlessness of conceptual thought are undercut once we recognize how scientific thinking is embedded within experimental practice. Conceptual domains in science typically emerge from new practical articulations of the world itself. Yet reconnecting the study of scientific practice to philosophy more generally may also transform philosophy of science in constructive ways. Our dominant cultural and philosophical images of science still place the retrospective justification of scientific knowledge at the forefront. Attending to the prospective role of scientific research in articulating the world conceptually and practically could replace this conception with a provocative and promising new “scientific image.” Defending that claim, however, must be left for another occasion.