1 Metascience (2011) 20: DOI /s BOOK SYMPOSIUM Scientific representation: A long journey from pragmatics to pragmatics Bas C. van Fraassen: Scientific representation: Paradoxes of perspective. Oxford: Clarendon Press, 2008, xiv+408pp, HB James Ladyman Otávio Bueno Mauricio Suárez Bas C. van Fraassen Published online: 17 November 2010 Ó Springer Science+Business Media B.V James Ladyman Once again, van Fraassen has given us an incredibly rich, learned and profound book that will be studied and argued about for decades to come. The book is about scientific representation, but from the perspective of Bas van Fraassen s empiricism this means representation of the empirical phenomena (1). He is keen to distance himself from the empiricist tradition that would interpret the latter notion in terms of mental representation, and instead focuses his attention on mathematical and material representations, and one main form of representation in which he is interested is the representational aspect of measurement. Following and extending the line of argument that we find in the work of Poincaré and Weyl among others, van Fraassen argues for the essential indexical (3) in science. This is a radical view that is at odds with the idea of objective inquiry that enables us to transcend our situation and describe the world independently of human beings, which is how many philosophers, especially scientific realists, see science. van Fraassen s J. Ladyman (&) Department of Philosophy, University of Bristol, 9 Woodland Rd, Bristol BS8 1TB, UK O. Bueno (&) Department of Philosophy, University of Miami, Coral Gables, FL 33124, USA M. Suárez (&) Department of Logic and Philosophy of Science, Complutense University of Madrid, Madrid, Spain B. C. van Fraassen (&) Philosophy Department, San Francisco State University, 1600 Holloway Avenue, San Francisco, CA 94132, USA
2 418 Metascience (2011) 20: empiricist and antimetaphysical philosophy permeates his treatment of scientific representation and he repeatedly highlights those aspects of how we draw, describe and measure the phenomena that have to do with our relationship to both the latter and the representations themselves. He argues for subjectivity s place in our understanding of science and denies that science can ever achieve a fully objective representation of the world. In the light of these observations, it is questionable whether the project of describing how scientific representation works can really be largely orthogonal to the realism/antirealism dispute as van Fraassen claims (3). Van Fraassen did a great deal to reinvigorate the scientific realism debate after the post-positivist repudiation of previously popular forms of instrumentalism. In a much-anthologised chapter of The Scientific Image, he defended the idea that the aim of science is to save the phenomena. The latter were understood to be the same as the appearances, but there is an important innovation in his thought expressed by the distinction he now draws between the phenomena and the appearances, where the former are observable entities and the latter are the contents of observation or measurement outcomes (8). So, for example, the measurement of temperature (Chapter 5) is to be understood in terms of the distinction between the phenomena associated with heat and the appearances that are saved by scientific theories mapping to representations of the appearances in the form of experimentally determined values of temperature. As van Fraassen explains, the construction of appropriate appearances thermometry is a rich science in its own right. Here, he cites approvingly the work of Cartwright, Giere and others who have argued that scientific representation must be understood not only at the level of high theory but also in the local domains in which high theory is de-idealised and experimental phenomena are modelled in detail. The distinction between phenomena and appearances is the distinction between what is immediately manifest in experience, and the special kinds of experience that are constructed by scientific experimentation and then codified in the form of measurement outcomes. Clearly, the appearances will be highly theory laden, as in the case of temperature previously mentioned, and in all the appearances of fundamental physics. Van Fraassen stresses that when a theory is said to save the phenomena, what this usually means is saving the appearances. The latter can naturally be understood in terms of structural relationships between representations because the appearances are usually in mathematical form. Consider the case of the planetary orbits and their theoretical representation in Copernicus model of the solar system (Chapter 12). Here, the phenomena are the orbits of the planets and the appearances are the perspectival measurements of the positions of particular heavenly bodies in the night sky. Hence, retrograde motion belongs merely to the appearances and is not a genuine phenomenon. Note however that while the phenomena are supposed to be observable, they are only manifest to us in the form of one or other set of appearances (the positions of points of light in the night sky on Earth seen with the naked eye or through a telescope, as opposed to the positions of lights in the night sky seen from the Moon). So are the phenomena inferred and theoretical (though perhaps not usually as much as the appearances)? Van Fraassen would, I think, say not because he insists that the phenomena are observable entities, events and processes. He denies that we can
3 Metascience (2011) 20: reduce the observation of a galloping horse to a series of mental representations with some characteristic content and structure, and then claim that the horse is inferred as the best explanation. Now of course nobody would deny that in our actual cognition the representation of a continuing entity that moves through space is not a matter of inference. It is in philosophical vein when we ask about the source of our knowledge of the horse and what limits there are to that knowledge that indirect or representative realism is tied to the choices we face among competing explanations of the world we see around us. Returning to the case of the night sky, the appearances were once interpreted in very different terms by people who had no inkling of the distinction between planets and stars. The interpretation of the appearances is up for grabs and is only stabilised as the result of sophisticated inference to orbits based on lots of data. What in principle is different between the case of the orbits and the measurements of them, and cells and how they look through a microscope or particle collisions and the tracks they cause in cloud chambers? Measurement outcomes in general show not how the phenomena are but how they look. However, if how they are must be inferred from how they look, then ultimately that they are must also be so inferred. This may not go for charging horses, but it surely does for the Moon and public objects like gases such as methane, which we can smell, electromagnetic fields, which we can see in the form of sparks and lightning, and so on. Tracks in particle chambers and images in microscopes are public objects, but van Fraassen classifies them as akin to mirages or rainbows, namely as public images. Phenomenologically, he is correct but the objection I have in mind is that the public objects he takes to be part of the realm of the phenomena are as much inferred as theoretical objects. I think we have good reason to believe that the inferences in the case of everyday objects are ones that we are evolutionarily predisposed to make as we are stimulated by the senses in the course of our development. Furthermore, if someone spends the daily life in a physics laboratory, it is unclear what should we count as the everyday things around him or her. Of course, van Fraassen has in the past set great store in the distinction between the unobservable and the merely unobserved, in the sense that horses galloping too far away to see are nonetheless in the realm of the phenomena. His view in the present work is much the same for the phenomena include all the observable entities events, objects, processes that there are, have been, or will be, whether observed or measured or not (307, his emphasis). I have argued that the modal status of the observable is a problem for van Fraassen s empiricism, but whether or not that is correct, it is worth considering also whether his account of scientific representation has aporia related to modality. Van Fraassen argues that representation must be understood contextually and pragmatically in terms of the use or function it has for someone (23). It seems then that he rules out the naturalisation of representation in evolutionary theory and biology more generally. On this view, RNA and DNA do not represent anything unless used by a genetic engineer to do so, and the state of the visual cortex per se does not represent the environment. This is mysterious, for we are told nothing about how human subjects manage to introduce representation out of nowhere. It is also implausible to make representation and subjectivity all-or-nothing and to confine them to human beings. I see no reason to
4 420 Metascience (2011) 20: deny that animals senses represent their environments to them. Representation may concern aboutness but if the processing in the visual cortex is not about anything, then why suppose that my words are? In van Fraassen s view, intentionality seems to be an all-or-nothing matter of which scientific understanding is impossible. However, the coupling of brain states to behaviour can be investigated scientifically and I see no reason to deny that science can tell us about how we represent the world. More importantly, though it seems that while the modality of natural necessity is rejected as un-empiricist metaphysics, the modality of natural teleology in the form of subjective agency is a vital metaphysical commitment for van Fraassen s view. When it comes to measurement, van Fraassen says that a measurement is at the same time a physical interaction and a meaningful information gathering process (91). However, it is questionable whether we can properly understand measurement without incurring modal commitments again, for to believe that a device is measuring a system it must be interacting with the system, where interaction seems to be causal or modal notion; relatedly, it must be the case that had the system had a different state with respect to the relevant physical property, the measurement device would have been in a correspondingly different state too. Van Fraassen claims that representation can be of something non-existent, and hence that to call something a representation will not imply anything about the reality of what it depicts (35). This is important for his antirealist view of representations of the unobservable in science, for it means that representations of the unobservable need not depict anything real. He emphasises that representations such as maps are useless without knowledge of one s location relative to them and argues that measurement functions as the source of this knowledge. The perspectival nature of spatial measurement is a special case of the general necessity for something indexical to relate theoretical models to empirical situations that van Fraassen approvingly quotes Hermann Weyl calling the ineliminable residue of the annihilation of the ego (87). Van Fraassen s fundamentally Kantian thesis is that scientific representation cannot transcend subjectivity and provide us with a view from above. The relationship between theories and the world is mediated by agents that must be situated with respect to the representation and use it as one for it to be a representation at all. For all the sophistication and persuasiveness of van Fraassen s elaboration of the Kantian arguments concerning the need for the subjects to locate themselves indexically in order to use a map, there seems to be a fundamental objection to the conclusion which he draws from them about the impossibility of objective representation. If I am presented with a map of a region I may for example know that the region is mountainous, or hilly, or flat without knowing where I am in relation to the map. If the contour lines on the map correlate very well with the topography of the region represented, and if the map is just the way it is because that is how the region is, then what grounds do we have for denying that the map represents the region even if the person holding the map has no idea where it is? Van Fraassen would presumably reply that isomorphism between the map and some part of reality is not sufficient for it to be a representation and that for the latter to be the case it is a necessary condition that it be taken by someone to be a
5 Metascience (2011) 20: representation. While I believe this may be true, it does not seem to change the fact that there is a representation there to be had in the way that there is not if someone was looking at a blank page. Again, this can be put in modal terms; some physical objects are in virtue of their properties disposed or able to be used as representations even if actually being a representation requires an agent to use them thus. At this point, we run into another one of those arguments that is intended to reduce the idea of a representation per se to absurdity and which van Fraassen discusses at length in Part III. Surely, there is some mapping from any blank sheet of paper to the topography of any region according to which the former can be regarded as structurally similar to the latter? Realists resist this suggestion with the insistence that some mappings are more natural than others. In the end, van Fraassen s way with Putnam s paradox, the Newman problem for structuralism and I suspect with the argument that a rock computes every finite state automaton is to accept that representation and computation are to be individuated in the light of our pragmatic relationships towards certain systems. However, the problems surveyed seem to presuppose extensionalism and perhaps intensionalism about the relevant relational structure might be combined with some form of realist alternative to van Fraassen s pragmatism. Structuralism is an attempt to find objectivity in scientific representation by drawing attention to the product of our abstractions from the content of our concepts about force, mass, extension and so on. The mathematical structures that we now use to express physical concepts like momentum, energy, frequency and so on are far removed from direct connection with the source of those ideas in the experience we have of mechanics and oscillatory phenomena in the manifest image. The idea of understanding representation in terms of isomorphism or structure-preserving maps of some kind is prima facie plausible, given the apparent complexity and richness of the examples of such connections we see in the sciences we have. Mathematics is often said to be the science of structures. Science, as van Fraassen emphasises, centrally involves mathematical representation. We may say at least that it involves logic, since the theoretical part of science has always been associated with the domain of deduction: relationships of entailment among propositions that are expressed by a language that is enriched with general terms and predicates which introduce taxonomic and systematising discipline to our cognition and enable us to cleanly state ceteris paribus laws and universal generalisations. These linguistic resources are extended by stages to become the highly mathematical representations that we now have that describe the behaviour of familiar things in the manifest image, but also of course, of the whole Earth and the rest of the universe. However, much debates about the scientific method lay unresolved, we know clearly enough that something like hypothetico-deductivism is the basic model of much of science, especially theoretical physics. There is an extraordinary connection between the mathematical structure of Hilbert space and concrete phenomena that can be represented by statistical methods as appearances to be embedded in empirical substructures of theories. Clearly, on the purely theoretical side, deductive logic and mathematics (which may include probability theory) exert the only absolute constraints. On the other hand, the connection between our representations and reality is governed by the vagaries of induction and abduction.
6 422 Metascience (2011) 20: Van Fraassen follows Duhem in his belief that the scientific method is not decidable, in the sense that there is no set of rules that we can use to determine what theory-choice is the correct one on any given occasion. However, Duhem thought that this underdetermination is always temporary, and fully expects that the real decision is made with hindsight as the domain in which the underdetermined theories apply is expanded so as to result in new experimental results about which the theories differ. Van Fraassen would, I think, applaud this insistence on the search for more empirical data, for that is what drives science. However, he differs from Duhem in that he rejects the idea that there is some element of natural classification to the relationship of our theories with the world, upon which ontologically different theories are none the less converging. This returns us to the problem of understanding the accord between the logical/mathematical modality of representations, and our considerable success in using them to accurately represent and predict the fine structure of the phenomena. The view that van Fraassen denies has it that there is in the world some causal or nomological structure that is represented by logical and mathematical relationships in our theoretical thought. He (306) approvingly quotes Hertz as follows: the necessary consequences of the pictures in thought are always the pictures of the necessary consequences in nature of the things pictured The pictures which we here speak of are our conceptions of things. With the things themselves they are in conformity in one important respect, namely in satisfying the above requirement. For our purpose it is not necessary that they should be in conformity with the things in any other respect whatever. However, there is a way of reading Hertz here according to which he is giving voice to a form of realism, for the above requirement could be taken to require that the modality in the model represent the modality in the world rather than being so much superfluous structure over and above what is needed to entail the actual extensional regularities in the phenomena. At the end of the book, van Fraassen discusses what he calls the Appearance from Reality criterion according to which science is incomplete until it explains the appearances, where this is understood to require that the latter be not merely predicted but described in terms of the modal constraints or causes that govern them. Van Fraassen argues that much contemporary philosophy of mind has rejected the Appearance from Reality criterion in so far as it is accepted that supervenience without reduction is the relationship between mental states and brain states. He then argues that the Copenhagen interpretation of quantum mechanics also shows us how the criterion is no longer important for science despite its origins in the Scientific Revolution. He here argues that the measurement problem is not a problem for the empiricist. This is interesting because much recent discussion in the philosophy of quantum mechanics has concerned precisely whether the Everett interpretation can be adequate given its alleged failure to make clear how the branching structure of quasi-classical worlds can give rise to the appearance of a definite unique world, where it is often supposed that by contrast, Bohm s theory does solve the appearance from reality problem by positing particles. Van Fraassen will have no truck with either of these approaches of course, but it is not clear that philosophers
7 Metascience (2011) 20: of physics at least have abandoned the criterion in line with Copenhagen orthodoxy. Whether this speaks well or ill of them depends on the empirical question as to whether the measurement problem and the attempts to solve it contribute to the future development of physics. Otávio Bueno Scientific Representation is an impressive, rich and sophisticated work. It offers an empiricist treatment of representation in science, exploring along the way some lessons that can be gained from representation in art, and integrating the resulting approach with a conception of the role that mathematical representations and structural considerations play in scientific practice. I will focus on two key features of van Fraassen s argument: the connections between scientific representation and microscopy, and the status of mathematical models in an empiricist stance. Scientific representation and microscopy The central feature of constructive empiricism is the characterisation of the aim of science as the construction of empirically adequate theories, that is, theories that correctly describe the observable. According to van Fraassen, the observable is, ultimately, what can be observed with the naked eye. The focus on the observable has been an ongoing source of concern about constructive empiricism, since the proposal was first advanced. Critics have complained, in particular, about what is epistemically special about observation, and how to reconcile the emphasis on the observable (narrowly understood in the constructive empiricist way) with the fundamental role played in contemporary science by a variety of scientific instruments, such as different kinds of microscopes, which aim to allow scientists to visualise what is in fact unobservable to our unaided eyes. In Scientific Representation, van Fraassen provides a very provocative response to this problem (93 113). He suggests that rather than thinking about microscopes as windows to the invisible world (as realists about microscopy insist they are), we can think of them as engines of creation, that is, as instruments that produce phenomena that then need to be explained. Furthermore, the images generated by microscopes, although copy-qualified (that is, they can be thought of as being a copy of some object), are ultimately public hallucinations. The images are public in the sense that several people can share them simultaneously and the images can be photographed. However, they are hallucinations given that there is no object there is no thing that the images are images of. In this respect, microscopes images are similar to rainbows. The latter are also public images (rainbows, after all, can be easily photographed) even though there is no thing there is no object they are about. The same person sees a rainbow in different locations as he or she moves around, and closely located people see a rainbow in different places. Are they seeing the same rainbow? The specification of identity conditions for rainbows is no trivial matter. In fact, it is not even clear whether there are such conditions. But if there is no thing that a rainbow stands for, seeing a rainbow is a form a hallucination although a public one.
8 424 Metascience (2011) 20: By thinking of microscopes images in these terms, van Fraassen can easily motivate his agnostic attitude toward the unobservable. Maybe unobservable objects exist; maybe they do not. The empiricist need not settle the issue. But a difficulty also emerges. It is unclear how to reconcile this agnostic attitude toward the unobservable with the role played by microscopes in contemporary science. Microscopes images are often taken as evidence for the existence of certain unobservable entities. Suppose that, on an image generated by a transmission electron microscope, particulate components of the membranes of cells are seen (Palade 1955). Such particulate components are neither artefacts of the microscope nor of the method of preparation of the sample, given that such components are invariably present despite changes in the preparation method and triangulation with other instruments (such as a centrifuge that indicates different decantation times in the items of the sample). In this situation, it seems almost perverse to insist that the images produced by the instruments are not evidence for the existence of ribosomes, but are only public hallucinations that produce phenomena that need to be explained. What additional explanation is needed in this case but to point out that the ribosomes that are present in the sample, given the interaction with the electron microscope, produced the resulting image? It is unclear how a response much different from that can make sense of the role played by such instruments as sources of visual evidence in scientific practice. Of course, the constructive empiricist could deny that microscopes produce visual evidence for the existence of those unobservable entities that presumably are studied with such instruments. But this is not the move that van Fraassen favours. After all, he advances an agnostic view about the unobservable rather than a sceptical one. As he notes (109): To keep neutrality in this respect [i.e. to think of microscopes images as public hallucinations, thus remaining agnostic about them] does not prevent us from gathering empirically attestable information by means of the microscope, or to base e.g. medical advice on what the microscope shows us. We should recognize that a false contrast is made if we oppose merely producing images to producing something informative about the objects with which the instruments are placed in interaction. The contrast here may indeed be a false one. But the point still remains. Microscopes images do have content: they describe the unobservable objects, processes and events these images are about as being in a certain way. How exactly can an agnostic view about the content of microscope s images that avoids the commitment to the existence of the relevant unobservable objects be reconciled with the idea that such images produce something informative about these objects? Part of the difficulty here is that van Fraassen takes information to be an endorsing term (157). He insists that, despite the communication engineer s neutral usage of information, which to some extent has begun to modify the common usage of this term, it would still be puzzling or provocative to hear We get information from observation, measurement, fictions, lies, and popular astrology. If we insist
9 Metascience (2011) 20: that measurement is information gathering, we mean in part that fictionalizing or speculating or guessing is not measurement (157). If information is, thus, an endorsing term, it is unclear how microscopes can produce something informative about the objects with which they interact, while we remain agnostic about the very existence of these objects. Prima facie, the informativeness of microscopes (understood in an endorsing sense) about the relevant objects seems to undermine the possibility of maintaining an agnostic stance toward the latter. Given that microscopes are intended to yield visually salient information about unobservable objects, that information would be significantly misleading if such objects turn out not to exist. But this is a possibility that the agnostic needs to leave always open. Perhaps, the constructive empiricist has a response here. In the case of scientific theories, the constructive empiricist correctly insists that such theories need not be taken to be true, as long as they are empirically adequate (i.e., roughly, provided that they are true about the observable world). This moves draw a line between, on the one hand, commitment to every aspect of a given scientific theory (if such a theory is true, this includes both the observable and the unobservable aspects of the world) and, on the other hand, commitment to only part of it (if the theory is empirically adequate only, commitment is restricted to the observable). And maybe the constructive empiricist could invoke something like this distinction between truth and empirical adequacy to motivate an agnostic view about microscopes images, while still acknowledging that the latter can be informative. But how exactly can such a distinction be drawn here? What would it mean to state that a microscope s image is empirically adequate? As opposed to a scientific theory, which in order to be tested does include a part that refers to observable objects, no part of a typical microscope s image say, the micrograph of a cell produced by an electron microscope is about something observable in van Fraassen s sense. It is simply unclear how the line between truth and empirical adequacy could be properly drawn in this context. We also face a related concern. There seems to be a tension between the characterisation of microscopes images as being both copy-qualified and public hallucinations. On the one hand, if these images are copy-qualified, presumably there is some object such images are copies of (even if we may be mistaken about which object exactly is involved). On the other hand, if microscopes images are public hallucinations, there is simply no object these images are about, no object that produces such images and which is represented by the latter as being in a certain way. In fact, it s the absence of such object that turns the image into a hallucination. As a result, the alleged hallucinatory nature of microscopes images is now in tension with their copy-qualified status. For in virtue of being copy-qualified, such images have a representational content there is something the images are about, and they represent the relevant objects in a certain manner. But this is precisely the feature that public hallucinations lack. What is needed here is a well-motivated account that specifies the conditions under which one is entitled to take the representational content of images seriously when the relevant microscopes yield images that genuinely represent objects and
10 426 Metascience (2011) 20: relations at the unobservable level and when this is not the case. In this way, a suitable account of the status of microscopes images can be advanced, and the grounds that support an agnostic view, when the latter is justified, can be offered. I think the key condition for the reliability of microscopes is that they are counterfactually dependent on the specimen under study. In particular, two conditions need to be satisfied: (C1) Had the specimen under study been different (within the sensitivity range of the microscope), the resulting image would be correspondingly different. (C2) Had the specimen under study been the same (within the sensitivity range of the microscope), the resulting image would be the same. In this way, it is possible to track the relevant features of the specimen using the microscope. If conditions (C1) and (C2) are satisfied and if we know that they are satisfied the resulting images do provide us with significant information about the specimen. In this case, the images offer us evidence that the relevant features of the sample do obtain. However, if these conditions are not satisfied, clearly the microscope fails to provide adequate evidence. In this case, we are not dealing with a reliable instrument. Finally, if we just do not know whether the conditions are satisfied or not, an agnostic stance is then clearly well motivated. Scientific representation and mathematical models An additional source of concern about constructive empiricism emerges from the status of mathematics within an empiricist framework. Given that mathematical objects are unobservable, and since the constructive empiricist restricts his or her commitments to observable objects, processes and events, it is difficult to reconcile the role of mathematics in science with constructive empiricism. What is needed is an account of mathematics that squares the lack of commitment to mathematical objects with the undeniable role that mathematics plays in empirical science. Van Fraassen (1985) is, of course, aware of the issue. And he insists that there ought to be some feature of mathematical practice that motivates the lack of commitment to mathematical objects despite the need for quantifying over such objects as part of scientific, and philosophical, practice. Interestingly, van Fraassen is sceptical that any form of nominalism ultimately works. I think part of the difficulty concerns the fact that nominalist views tend to provide reconstructions of mathematical discourse and thus are unable to take the latter literally. And a distinctive feature of constructive empiricism, in striking contrast with earlier forms of empiricism in particular, logical positivism is the fact that it takes scientific discourse in a literal way. If according to a given scientific theory, quarks exist, the constructive empiricist will not rewrite that theory so that no commitment to quarks is forthcoming. The empiricist will simply take such a theory to be empirically adequate rather than true. It is then perfectly understandable why the constructive empiricist is not moved by standard forms of nominalism that end up doing to mathematical theories precisely what the empiricist tries to avoid with scientific theories: to rewrite them.
11 Metascience (2011) 20: However, an account then needs to be offered of how exactly mathematical theories can be used in science while being taken literally without the resulting commitment to mathematical entities ever emerging. In Scientific Representation, van Fraassen attempts to dissolve the problem, at least in a particular context. It is not clear to me, however, that the proposed solution works. After examining van Fraassen s solution, I will suggest another that seems to satisfy the demands he correctly imposed on the problem. The concept of empirical adequacy, as van Fraassen characterises it, involves reference to unobservable entities, such as models and isomorphic mappings. Even if the concept of isomorphism can be expressed in a way that does not involve commitment to abstract entities (e.g. it can be formulated with the resources of a second-order language), the statement that a given scientific theory is empirically adequate involves reference to models namely, the empirical substructures of a given scientific theory which are abstract. But perhaps this commitment can be avoided. As van Fraassen points out (243), Snow is white is true only in virtue of snow being white, not in virtue of the existence of a universal, whiteness, to which snow bears some relation. And he continues: The same point applies to isomorphism. Let s take a particular case of a concrete physical, observable object: this table top is metrically isomorphic to a Euclidean square. That is true, but simply because this table top is square c est tout! It is true because the top s sides are of equal length and the angles between them are right angles. It could be paraphrased as the table top instantiates the Euclidean square Form, but the cash value of the assertion carries no metaphysical commitment: it is just that the table top is square (249). The point is quite right in the particular case of concrete objects. But it is not clear how exactly it can be extended when we are asserting that a given scientific theory is empirically adequate. For in this case, as opposed to the case of a concrete object, the isomorphism in question connects two abstract structures: an empirical substructure (a structure from a given theory that describes the relevant observable objects, processes, and events) and the surface model (the mathematical structure that represents the data that have been obtained empirically). Here, we cannot say that the cash value of the assertion carries no metaphysical commitment, since we are asserting that two abstract objects have the same structure (that is, there is a oneto-one mapping between the empirical substructure and the surface models that preserves the relations in each of them). Is there a way out? One possibility is to insist that there is no need to interpret the quantifiers in an ontologically loaded way. We can, of course, quantify over objects in whose existence we have no reason to believe, such as average moms who have 2.4 children. Commitment to the existence of something is only indicated by a suitable existence predicate, whose sufficient (but not necessary) conditions include having a form of access to objects that is robust, can be refined, and that allows us to track the relevant objects in space and time. (Interestingly, these conditions are all satisfied by the output of scientific instruments that meet the two counterfactual conditions
12 428 Metascience (2011) 20: mentioned earlier.) In this way, the constructive empiricist is in a position to take mathematical discourse literally, quantify over mathematical objects in the formulation of scientific theories, but refrain from being committed to such objects (for details, see Bueno 2005, 2009). Thus, it is possible to make sense of the indispensable role mathematical models and various morphisms (such as isomorphism or partial isomorphism) play in scientific representation so insightfully described by van Fraassen without overstepping the boundaries of constructive empiricism. In the end, scientific representation and empiricism can live in quiet harmony. Mauricio Suárez Scientific Representation is the latest stage in Van Fraassen s sustained and profound defence of empiricism against scientific realism over the years. The book s key message may be summarised as follows: constructive empiricism the antirealist view first defended in the celebrated The Scientific Image (1980) needs to be modified in some substantial ways to make room for an appropriate notion of representation. The modification yields empiricist structuralism or, as I shall call it, structural empiricism. The shift is meant to preserve the central epistemic commitments of the old view while bringing in some new advantages. Van Fraassen provides reasons for the change and offers arguments for the new position. But like any shift this is a gamble. If the reasons for change are convincing but the arguments for the alternative are not, we end up in no man s land. Or in someone else s land. I think, we end up in the land of pragmatism. A good way to express the reasons for change is by reference to an old paper of mine discussed critically in what appear to be two key passages of the book (25 26 and ). Since I do not think my views are described or identified correctly, perhaps some extensive discussion and self-quotation will be forgiven in the first part of the review. The second part critically reviews the arguments for structural empiricism. In the third and final part, a more positive assessment is provided, which hopefully will make it clear that I think Scientific Representation is nonetheless a very good book. Its main merit is to point towards a thoroughly pragmatist conception of science. Moving away from constructive empiricism There is a standard line on constructive empiricism that we all have been taught and go on to teach our students. It takes as central a commitment to the aim of science, encapsulated in the famous slogan: Science aims to give us theories which are empirically adequate, and acceptance of a theory involves as belief only that it is empirically adequate. The definition of empirical adequacy then enables us to extract out of the slogan the nuanced and rich position that is constructive empiricism. As is well known, Van Fraassen takes the empirical adequacy of a theory to crucially turn on a distinction between observable and unobservable
13 Metascience (2011) 20: domains of the world. The theory is true if it literally describes the world accurately; and it is empirically adequate if it describes the observable part of the world accurately. Hence, a false theory may be empirically adequate; and the commitments of realism and constructive empiricism come apart. Of course, we must remind ourselves all along that the language of truth, description and says is, taken literally, fundamentally inappropriate. It is a façon de parler, inherited from a syntactic, or language-based, view of theories. On the appropriate semantic conception, a theory is rather a set of mathematical structures: To present a theory is to specify a family of structures, its models; and secondly, to specify certain parts of those models (the empirical substructures) as candidates for the direct representation of observable phenomena. (Van Fraassen 1980, 64). We may then say that the theory is true if it is isomorphic to the world, and empirically adequate if some of its substructures are isomorphic to the observable part of the world. Truth and empirical adequacy are then structural surrogates of the linguistic truth and empirical adequacy. So, what is the problem then? It turns out that this conception of a theory s truth and empirical adequacy is in tension with a proper account of scientific representation. The not-so-often-quoted sentence right after the passage above reads as follows: The structures which can be described in experimental and measurement reports we call appearances; the theory is empirically adequate if it has some model such that all appearances are isomorphic to empirical substructures of that model. This sentence leaves little doubt that the empirical adequacy of a theory is to be understood as the isomorphism of the phenomenological and the empirical substructures of the theory (Suárez 2005). In conjunction with the previous quote, it seems to render the view that scientific representation is isomorphism; and that constructive empiricism stipulates that the acceptance of a theory implies the belief that the theory structurally embeds, i.e. represents, the observable phenomena. It is not surprising then to find Van Fraassen s commentators and critics over the years assuming that his version of the semantic conception requires an isomorphism conception of scientific representation. Among the commentators, I include my own (Suárez 1999b), which might have unfortunately started the trend. At least it led Steven French (French 2003) to respond by arguing for isomorphism as a necessary and sufficient condition on representation in general. Scientific Representation makes it clear however that this is not Van Fraassen s considered view. Rather he agrees that representation is not definable by means of any necessary or sufficient conditions: it is not the sort of notion that stands in need of any theory. Instead representation is linked to use; the hauptsatz is: there is no representation except in the sense that some things are used, made, or taken, to represent some things as thus or so (23). But the recognition of the fundamental role of use in scientific representation poses problems for constructive empiricism which I flagged over a decade ago. Unfortunately, Van Fraassen seems to have misunderstood what I wrote then (although he clearly recognises the tension). He writes: In a comment on similar intentional views of what constitutes representation, Mauricio Suárez suggests that it will hamstring the idea that theories represent (26). And he then quotes the following sentence: on the intentional conception of representation a theory cannot
14 430 Metascience (2011) 20: represent a phenomenon that hasn t yet been observed (Suárez 1999b, 81). But the extract has been taken out of context. Far from criticising intended use, my paper set out explicitly to critique what I took it to be the received understanding of Van Fraassen s conception of representation as isomorphism and I argued against it precisely by showing it to ignore intended use! My own view is of course that representation is essentially linked to representational use, and I have always accepted the hauptsatz above. (Even though the hauptsatz is really very thin and provides little by way of understanding the relevant practices). This is precisely the sense in which theories do not always represent their representational uses are not built in. The claim van Fraassen ascribes to me in no way follows from this. An intended use conception does not hamstring the idea that theories represent at all. It rather shows that theories sometimes represent, and sometimes do not, and whether they do or not depends on nothing other than use. In particular, it does not depend on isomorphism. (In criticising isomorphism, partial isomorphism, etc., I took myself to be working in the tradition of Goodman. It is a cruel irony of life that van Fraassen (349, n. 1) appears to ascribe the application of Goodman s exemplification to the assessment of the accuracy of scientific representation to of all people Steven French!) The misunderstanding might have its source in a conflation of intentionality and intended use conceptions of representation. From the start, I set out to defend an intended-use conception: Velazquez s portrait of Innocent X and Picasso s Guernica are explicitly advanced to illustrate that use is the key to establish representational contents or targets in general. As a reaction to a draft of my paper, van Fraassen sent me a draft of the paper that eventually appeared as his (van Fraassen 2000), which continues his (van Fraassen 1994) development of an intentional conception. Since this view is also at least prima facie incompatible with the reduction to isomorphism, the difference between intended use and intentional in general did not matter to my purpose then which was to criticise and reject the reduction to isomorphism. So in the rejoinder I wrote in response, intended use and intentional were run together. Later on, I did make clear that the differences do matter for the appropriate conception of representation (Suárez 2003). Chapter 2 of Scientific Representation is to my knowledge van Fraassen s first attempt to disentangle these two different views. Despite of some residual hesitations on pp , he also seems to be ultimately inclined towards an intended-use conception. I realise this is indulging in some self-referential score-keeping, so let us quickly take stock before we move on. I once made the following claim: there is tension between an intentional notion of representation and van Fraassen s constructive empiricism (Suárez 1999b, 81). With the appropriate caveat to read intentional as intended use, I continue to endorse it. It actually nicely summarises the main reasons that have moved van Fraassen towards structural empiricism. The arguments for structural empiricism The most original material in Scientific Representation is contained in part II, which reveals great attention to the practices of measurement and instrumentation. Many
15 Metascience (2011) 20: years ago, the Stanford school urged philosophers to turn towards an experimental philosophy of science. Some of us embraced the call, while others resisted. The resistance often presented van Fraassen as a leading theory-monger and employed him as a banner for the entire movement. This no longer seems credible: Scientific Representation is a turning point for admirably taking stock of the literature on instrumentation and experimentation of the last two decades. The autonomy of the experimental sciences is fundamentally taken to heart, which is a wonderful development for at least two reasons. First, it is refreshing to find a senior and distinguished figure in the field taking honest stock of a distinct and even contrary tradition. Second, constructive empiricism suffered from lack of detail in the experimental side of the theory experiment dichotomy that it promoted; and this was undermining it in the contemporary context of new experimentalism. The most outstanding outcome of this encounter with experimentalism is a new distinction between the observable phenomena and the appearances, which comes to the fore in Part IV of the book. The introduction of a 3-layer model (theoryphenomena-appearances) is motivated by both the autonomy of experiment and the use-based conception of representation. The observable phenomena underlies (and presumably causes) the appearances (Van Fraassen of course refrains from using any causal language). These are the outcomes of measurement procedures, recorded in various data models, and routinely employed to represent the phenomena. But the representation at this level cannot be isomorphism since while the data models are mathematical structures, the phenomena are real entities and processes in the physical world. Rather, the appearances represent the phenomena in the prescribed use-based sense of representation in the given context and for the purposes required. No proof of isomorphism or similarity are needed: it just makes no sense for users of the data model to deny that the model represents the phenomenon since the claim is built into their very use ( ). The theory is then empirically adequate if it embeds the appearances and this no longer carries the implication that a substructure of the theory must be shown to be isomorphic to the phenomena. Saving the phenomena has turned into embedding the appearances that represent the phenomena. The adoption of the 3-layer model and consequent rejection of the theory-phenomena dichotomy turns constructive empiricism into structural empiricism. The rest of the book is a defence of structural empiricism against (a) objections to structuralism in general (part III) and (b) the requirement that the phenomena ought to productively explain the appearances (part IV). While I have some sympathy with the rebuttal of (b), the arguments in part III don t strike me as successful. Van Fraassen points out similarities between Newman s argument against Russell s structuralism, and Putnam s model-theoretic argument against metaphysical realism. Roughly, both Newman and Putnam show that the statement of isomorphism between theory and world picks out nothing but the cardinality of the domains. Van Fraassen s resolution to the problem resembles Goodman s solution of the new riddle of induction in that it appeals to entrenched facts of our linguistic practice in particular the inevitable use of indexicals within a familiar language in setting up the right coordination or correspondences between theory and world. Goodman s entrenchment move has been considered ad hoc or circular as a justification of