Abstract

Although logical empiricism is now mostly decried, their naturalist claim that the content of a theory can be read off from its structure, without any philosophical considerations needed, still supports traditional strategies to escape cases of underdetermination. The appeal to theoretical equivalence or to theoretical virtues, for instance, both assume that there is a neutral standpoint from which the structure of the theories can be analyzed, the physically relevant from the superfluous separated, and a comparison made between their theoretical content and virtues. In my dissertation, I examine the presuppositions upon which such strategies depend. I argue that the methodological principle underlying them, according to which theories with no superfluous structure should be preferred, is unpractical, for what constitutes relevant structure is determined by epistemic considerations about the aim of scientific theories. In chapter 1, I analyze the claim that theories with ordinary bosons and fermions are theoretically equivalent to theories with exotic `paraparticles'. I argue that this claim does not do justice to the latter, as the proof is formulated in a vocabulary parochial to the former and thus favors one of the theory while giving an impoverished version of the second. In chapter 2, I examine recent arguments to the effects that any interpretation of Quantum Mechanics that can offer a no-go theorem against paraparticles possesses an explanatory advantage over other interpretations and should, as such, be favored over others. Given that most physicists consider paraparticles as surplus structure whose non-observation does not require an explanation, I evaluate arguments of both sides and suggest a third way to approach the question. Finally, in chapter 3, I turn my attention to methods for excluding another kind of unphysical structure, mathematical artifacts, from rival dark matter models. Simulations are our only window into what rival models predict about the universe’s structure. But for them to play a useful role in generating knowledge, we need to distinguish reliably between real predictions and artifacts. I argue that robustness analysis fails to fulfill this task. I propose in its place another methodology, that of crucial simulations.