This article introduces the notion of a kind of inference called black box measurement and argues that it is both historically and philosophically significant. Thinking about certain classic cases of underdetermination using this notion can give us a better understanding of how these cases are resolved. I take the main philosophical problem of black box measurement to be the justification of assumptions that are needed in order to make these measurements. I sketch some ways in which such enabling assumptions might be justified.

Halvorson argues that the semantic view of theories leads to absurdities. Glymour shows how to inoculate the semantic view against Halvorson’s criticisms, namely, by making it into a syntactic view of theories. I argue that this modified semantic-syntactic view cannot do the philosophical work that the original “language-free” semantic view was supposed to do.

Debiasing procedures are experimental methods aimed at correcting errors arising from the cognitive biases of the experimenter. We discuss two of these methods, the predesignation rule and randomization, showing to what extent they are open to the experimenter’s regress: there is no metarule to prove that, after implementing the procedure, the experimental data are actually free from biases. We claim that, from a contractarian perspective, these procedures are nonetheless defensible since they provide a warrant of the impartiality of the experiment: we only need proof that the result has not been intentionally manipulated for prima facie acceptance.

Agreement between “independent” measurements of a theoretically posited quantity is intuitively compelling evidence that a theory is, loosely speaking, on the right track. But exactly what conclusion is warranted by such agreement? I propose a new account of the phenomenon’s epistemic significance within the framework of Bayesian epistemology. I contrast my proposal with the standard Bayesian treatment, which lumps the phenomenon under the heading of “evidential diversity.”

Testing a point null hypothesis is a classical but controversial issue in statistical methodology. A prominent illustration is Lindley’s Paradox, which emerges in hypothesis tests with large sample size and exposes a salient divergence between Bayesian and frequentist inference. A close analysis of the paradox reveals that both Bayesians and frequentists fail to satisfactorily resolve it. As an alternative, I suggest Bernardo’s Bayesian Reference Criterion: (i) it targets the predictive performance of the null hypothesis in future experiments; (ii) it provides a proper decision-theoretic model for testing a point null hypothesis; (iii) it convincingly addresses Lindley’s Paradox.

An evolutionary basis for Bayesian rationality is suggested, by considering how natural selection would operate on an organism’s ‘policy’ for choosing an action depending on an environmental signal. It is shown that the evolutionarily optimal policy, as judged by the criterion of maximal expected reproductive output, is the policy that, for each signal, picks an action that maximizes conditional expected output given that signal. This suggests a possible route by which Bayes-rational creatures might have evolved.

Does the strength of a particular belief depend upon the significance we attach to it? Do we move from one context to another, remaining in the same doxastic state concerning p yet holding a stronger belief that p in one context than in the other? For that to be so, a doxastic state must have a certain sort of context-sensitive complexity. So the question is about the nature of belief states, as we understand them, or as we think a theory should model them. I explore the idea and how it relates to work on imprecise probabilities and second-order confidence.

Multiarm bandit problems have been used to model the selection of competing scientific theories by boundedly rational agents. In this article, I define a variable-arm bandit problem, which allows the set of scientific theories to vary over time. I show that Roth-Erev reinforcement learning, which solves multiarm bandit problems in the limit, cannot solve this problem in a reasonable time. However, social learning via preferential attachment combined with individual reinforcement learning, which discounts the past, does.

Game theory is often used to explain behavior. Such explanations often proceed by demonstrating that the behavior in question is a Nash equilibrium. Agents are in Nash equilibrium if each agent’s strategy maximizes her payoff given her opponents’ strategies. Nash equilibriums are fundamentally static, but it is usually assumed that equilibriums will be the outcome of a dynamic process of learning or evolution. This article demonstrates that, even in the most simple setting, this need not be true. In two-strategy games with just a single equilibrium, a family of imitative learning dynamics does not lead to equilibrium.

Traditionally, cognitive values have been thought of as a collective pool of considerations in science that frequently trade against each other. I argue here that a finer-grained account of the value of cognitive values can help reduce such tensions. I separate the values into groups, minimal epistemic criteria, pragmatic considerations, and genuine epistemic assurance, based in part on the distinction between values that describe theories per se and values that describe theory-evidence relationships. This allows us to clarify why these values are central to science and what role they should play, while reducing the tensions among them.

We argue that the analysis of cognitive attitudes should play a central role in developing more sophisticated accounts of the proper roles for values in science. First, we show that the major recent efforts to delineate appropriate roles for values in science would be strengthened by making clearer distinctions among cognitive attitudes. Next, we turn to a specific example and argue that a more careful account of the distinction between the attitudes of belief and acceptance can contribute to a better understanding of the proper roles for values in a case study from paleoanthropology.

The argument from inductive risk attempts to show that practical and ethical costs of errors should influence standards of evidence for accepting scientific claims. A common objection charges that this argument presupposes a behavioral theory of acceptance that is inappropriate for science. I respond by showing that the argument from inductive risk is supported by a nonbehavioral theory of acceptance developed by Cohen, which defines acceptance in terms of premising. Moreover, I argue that theories designed to explain how acceptance can be guided exclusively by epistemic values suffer from difficulties that do not afflict Cohen’s theory.

Proponents of the value ladenness of science rely primarily on arguments from underdetermination or inductive risk, which share the premise that we should only consider values where the evidence runs out or leaves uncertainty; they adopt a criterion of lexical priority of evidence over values. The motivation behind lexical priority is to avoid reaching conclusions on the basis of wishful thinking rather than good evidence. This is a real concern, however, that giving lexical priority to evidential considerations over values is a mistake and unnecessary for avoiding the wishful thinking. Values have a deeper role to play in science.

There are two senses of ‘what scientists know’: An individual sense (the separate opinions of individual scientists) and a collective sense (the state of the discipline). The latter is what matters for policy and planning, but it is not something that can be directly observed or reported. A function can be defined to map individual judgments onto an aggregate judgment. I argue that such a function cannot effectively capture community opinion, especially in cases that matter to us.

Many scientific models lack an established representation relation to actual targets and instead refer to merely possible processes, background conditions, and results. This article shows how such models can be appraised. On the basis of the discussion of how-possibly explanations, five types of learning opportunities are distinguished. For each of these types, an example—from economics, biology, psychology, and sociology—is discussed. Contexts and purposes are identified in which the use of a model offers a genuine opportunity to learn. These learning opportunities offer novel justifications for modeling practices that fall between the cracks of standard representationalist appraisals of models.

The multiple-models approach, which has its origins in Levins’s work, is gaining broader acceptance among philosophers. Levins asserted that there is a trade-off between modeling desiderata, which justified the multiple-models approach through two separate justificatory paths. Some attention has been paid to the trade-off thesis, culminating in a paper by Matthewson and Weisberg. However, no attention has been paid to how the trade-off is supposed to justify the multiple-models approach. I argue that a trade-off between generality and precision cannot support one of Levins’s justificatory paths, and I consider what that might mean for the multiple-models approach.