Realism, Confirmation, and Explanation: Philosophical Studies in Models and Model-Based Science

Abstract

Over the last several decades, the philosophy of science has been enamored of the role that models play in scientific practice. The methods and applications of models, their role in the production of knowledge, and many other features challenged previously held assumptions in the philosophy of science. That models are poor representations of their targets, that they are limited in scope, and that it is not always seen as a problem that they conflict with other models of the same target caused problems for older accounts of the epistemology of science. For instance, given the features of models, and their prominent role in scientific practice, the classic nomological-deductive (covering-law) account of explanation–where explanation is made by logical deduction from a universal statements–did not seem particularly applicable. The role of many different types of models in various aspects of scientific work, whether explanation, reasoning, prediction, or something else, has taken much of the attention of many philosophers of science. In the chapters that make up this dissertation, I too concern myself with models. The first chapter examines how the “modeling turn” in the philosophy of science impacts the scientific realism debate. The two classic arguments in this debate–the “No Miracles Argument” in favor of realism and the “Pessimistic Meta-Induction” against realism–have often been presented in relation to a view of science as built out of theories. Theories, in the philosophers’ sense, are typically treated like logical languages. Important in this theories-based account of science was that theories were intended to be accurate representations of the world. The NMA claims that theories were generally accurate that this accuracy provided for the success we often attribute to science. The PMI, in response, would point to the graveyard of failed theories in an attempt to show that we had little reason to buy into the truth of our current scientific theories. I argue that these two arguments are not easily applied to a model-based view. Models are not intended to be accurate representations of their targets, often idealizing, abstracting, fictionalizing, or misrepresenting in some way. Further, these misrepresentations are not seen as flaws of the model, like they would be in a theory, but are often central to the goals of the model. Further, models are not discarded because of their lack of truthful representation, but are often discarded because the role that they can play is no longer needed. Given these features of models, the central role that models play in the philosophy of science, and the focus of the NMA and PMI on truth, I argue that the realism debate needs to move beyond these arguments. The second chapter takes up a related concern around the role models play in confirming hypotheses. That models employ idealizations has been a concern for their role in producing knowledge, explanations, and other epistemic achievements. One proposal for getting around the potential concerns introduced by idealizations has been robustness analysis. This is a method of constructing several models with a shared core assumption but different sets of idealizations. If all of these models produce the same result, this is assumed to show that the results of these models are driven by shared assumption and not the idealizations. This, then, is meant to confirm the shared assumption of the models. The confirmatory power of robustness analysis has been questioned in several ways, one of which focuses on the fact that all of the models in a robust set are idealized, and therefore false. Given this, even if all of the models agree, they cannot provide confirmation since none of them accurately reflect the target system. In response, I argue that this is a misunderstanding of the role of some idealizations in models. Idealizations can be incorporated into models for a variety of reason, and sometimes might play the role of controlling some causal influence that is not of interest to the modeler. In this way, idealizations might play a role similar to that of experimental conditions, where laboratory conditions are contrived to control causal influences that might interfere with the causal relation being studied. Robustness, under these conditions, can be analogized with experimental replication. Since it is unclear how laboratory conditions might influence the results of an experiment, replicating the experiment in a variety of ways increases our belief in the result of each experiment. Similarly, it is unclear how a particular idealization, as a means of control, might influence the result of a model, so robustness is needed. Finally, in the third chapter, I take up an account of functional kinds derived from model-based science. Daniel Weiskopf (201a, 2011b, 2017) has argued for the explanatory role of functional kinds and has developed an account of functional kinds derived from model based science. This view, however, has faced criticism from mechanistic-based philosophers. One such criticism is that functional kinds, or at least some of them, cannot be considered true scientific kinds since they are explanatorily weak when compared to similar mechanistic kinds. These arguments, however, often allow that functional kinds might be multiply realized. I argue that, granting this multiple realizability, there is a unique range of explanatory counterfactuals that functional kinds can capture. Given this, I argue that such kinds should count as true scientific kinds.