As Bryan Norton notes in the opening of his contribution to this volume, the term “model” is “very elastic.” So, apparently, is a volume on “model-based reasoning.” Readers expecting a volume dedicated to a unified set of issues will be disappointed, especially if they harbor expectations that the subtitle of the book, “Science, Technology, Values,” will provide any clue as to what they will find. What the book offers, instead, is a representative sample of the kind of work going on in cognitive science, computer science, philosophy of science, and their intersections that studies science from the point of view of the role of models. Anyone whose interests intersect with the topics included under that elastic term will find something to like among the twenty-one articles. Since my space is limited, and I couldn't hope to summarize all of what is offered, I will instead discuss a few themes that unify some of the work on display here.
Several contributions discuss the role of analogy in scientific discovery and representation. The range of topics covered is itself a vivid illustration of the degree of cross-pollination that occurs in the sciences. Mark Johnson discusses models from optics and biology and their analogous application in the science of cognitive psychology. Dedre Gentner explores Kepler's use of sunlight as an analogy for his famous vis motrix. Daniela Bailer-Jones and Coryn Bailer-Jones show how models from neuroscience, evolutionary biology, and statistical mechanics have provided much of the conceptual backbone of contemporary thinking in data analysis. All of these examples serve not only to illustrate the wide variety of disciplines across which analogies function, but also the variety of functions that analogies themselves can play—cognitive, epistemic, and social—in the generation, exploration, and dissemination of scientific tools and ideas.
Another important topic is the role that models play in reasoning and cognition. A growing literature in philosophy of mind and cognitive science focuses on expanding the boundaries of cognitive agency by emphasizing that cognition is embodied. Contributions by Ron Giere and Lorenzo Magnani provide a nice link between this literature and recent thinking about scientific models. Magnani focuses on what he calls “manipulative abduction,” in which cognitive agents generate or select hypotheses by manipulating external representations—in this case, primarily diagrams. Giere uses the language of “distributed cognitive systems” to emphasize how external representations—the usual zoo of models with which all readers of Giere are now familiar (scale models, analog models and even abstract models)—are not just the subject matter of cognition, but are themselves a part of the cognitive systems that reason about the world.
This is a provocative claim. It becomes especially provocative when we recognize that most of the relevant models of modern science are abstract models. And as Giere himself admits, we don't even really know what abstract models are. A discussion of this question, Giere tells us, would “take us into the arid land of the philosophy of mathematics.”(236) And so, while Giere makes it plausible that the diagrams, pictures, physical models, and perhaps even tokens of equations that serve to characterize abstract models might be part of cognitive systems, it is not altogether clear how abstract models themselves can be. After all, it is perhaps plausible that tokens of numerals inscribed on a blackboard are part of a distributed cognitive system, but I'm not sure it's plausible that numbers could play such a role.
Finally, a cluster of papers explores the distinction between descriptive or representational devices on the one hand, and that which is described or represented on the other. Several of the contributors to this volume can be seen as challenging, or blurring, this distinction. For example, there is Ryan Tweney's case study of Michael Faraday's exploration of the properties of thin films of metalic gold. Tweney labels the thin films that Faraday produced “epistemic artefacts” and argues that in their most refined form, they became, in and of themselves, models of the very phenomena they embody.
The distinction can also be challenged from the other side—by looking at cases where abstract models play the role of the very thing being investigated. Francesco Guala, for example, urges us to recognize that models, experiments, and simulations are all “tokens of the same kind.” Models are not that different from experimental targets; both can function as objects of investigation. When used in this context, models and simulations are not a step further removed from the world than are experimental setups because all three require “external validity hypotheses” in order to inform. Both Guala and Mary Morgan provide examples of how mathematical models can function as experimental apparati. These models, they both argue, sit somewhere between the world and representations of the world. It is commonly thought that simulations are a “poor man's” experiment, whose results are always less reliable than those of “real” experiments, but Gaula and Morgan make the case that the epistemological relationship between experiments and simulations is murkier than one might think.
This is certainly true. Simulationists and experimentalists face equally relevant challenges when it comes to establishing that the results of their simulation or experiment are informative about the real world. But it is one thing to point this fact out, and it is another to understand how those challenges are overcome, under differing circumstances, and in varying contexts. It is here that Marcel Boumans’ contribution becomes especially valuable. He presents an example from economics in which a mathematical model performs the role, not of a representational entity, but of a data sensor. Boumans argues, and I concur, that the manner in which such models are assessed is particularly interesting. They cannot be assessed merely by being confronted with facts about the world, since these models are themselves used in generating the relevant data about the phenomena in question. The relevant strategy for assessing these models is calibration. In other words, rather than being held side by side with the relevant bit of the world, the models are held up against other instruments that are antecedently believed to be reliable sources of data.
This is an area of research on models—looking at the variety of ways in which models are assessed—that deserves much more attention in philosophy of science. Philosophers rarely think the assessment of models is any different from the assessment of theories. But this is a provincial view of models. In a vast range of applications, models are created to represent phenomena for which data are conspicuously scarce—for example in the case of a model of the inner convective structure of a star—or where the model itself is central in creating the data, as in Boumans’ example. In such cases, the standard story, which relies on concepts like similarity, goodness of fit or isomorphism, is inapplicable. It has nothing to say about the relevant criteria of assessment. Philosophers have much to learn about the various ways in which models are assessed under circumstances in which the traditional story does not apply. Some of the contributions to this volume offer a good start.
