Works in Progress
A common view of epistemic norms is alethic consequentialism: the view that accurate beliefs are the sole source of epistemic value, and that the aim of epistemically rational beliefs is to promote that value. Since the goal of inquiry is to learn truths, it’s natural to think of the norms of inquiry (“zetetic norms”) as alethic consequentialist norms as well. This paper argues against alethic consequentialism for both epistemic and zetetic norms. The general claim is this: we can acquire accurate beliefs in two distinct ways: by discovering truths and by creating truths. Creating truths isn’t a source of positive epistemic or zetetic value. Indeed, it’s sometimes intuitively a source of positive zetetic disvalue. I consider two possible explanations for why creating truths may be zetetically problematic: that it’s incompatible with conducting inquiry, and that it blocks off access to sources of genuine zetetic value.
It's often treated as a truism that we objectively epistemically ought to believe, or have maximal credence in, the truth. This claim is open to several interpretations. I explore a variety of prima facie plausible deontic semantics for the claim, and argue that each generates profoundly implausible predictions. (The objections I raise are independent of human cognitive limitations.) I conclude that there's no genuine truth norm on belief: the objective 'ought' doesn't extend to epistemology. I extend the argument to weaker putative objective epistemic norms, according to which one ought to hold beliefs or credences that approximate the truth as closely as possible. While we can order possibilities in terms of their alethic value, we can't generate norms on this basis. This has significant ramifications in other areas of epistemology; I discuss the example of accuracy-first epistemology.
Pace its main proponents, epistemic utility theory (EUT) is not a form of epistemic consequentialism and is not a decision theory. There are necessary conditions on being a decision theory that EUT doesn’t satisfy: a decision theory must evaluate acts’ choiceworthiness in terms of the goodness or rightness of their outcomes, in the loosest possible sense of "outcome". This undermines all extant interpretations and justifications of EUT. But, I argue, EUT, best understood, still does crucial work in the theory of epistemic rationality: it explains the relation between accuracy and evidential support. I argue that our epistemic duties are not consequentialist: not, e.g., to maximize accuracy. Instead, they are nonconsequentialist: we're required to conform our beliefs to what our evidence supports. I show that on this interpretation, epistemic rationality is guaranteed not to be consequentializable.
Publications
According to metanormativists, normative uncertainty can affect how a rational agent ought to act, and what acts are blameworthy. Metanormativism is a controversial view. The aim of this paper is not to provide a positive argument for metanormativism, but to show how it can avoid a problem that is sometimes taken to be insuperable: the problem of intertheoretic value comparisons. In other work, I defend a means of avoiding the problem in cases where moral value is representable with precise utilities. But what about agents who give positive credence to merely ordinal theories? Call this the hard problem of intertheoretic utility comparisons. The challenge is to construct a decision theory that could prescribe rational choices for uncertainty about the measure-theoretic metaphysics of utility. Such a decision theory would allow for decision problems where some outcomes have cardinal utility and some have merely ordinal utility. I argue that a specific formulation of imprecise utilities is suitable for the job. I then show that traditional decision theories for imprecise utilities face a puzzle when applied to normative uncertainty: these decision theories require reifying elements of the model that don't correspond to anything in moral reality. I discuss two ways of dissolving the puzzle, and their comparative perks and drawbacks.
Ideal epistemologists are concerned with questions about what perfectly rational, cognitively idealized, computationally unlimited fictional believers would believe. Nonideal epistemologists are concerned with questions about epistemic norms that are satisfiable by most humans much of the time. Ideal epistemology faces a number of challenges, both in its substantive commitments and in its value as a research program. This paper explores the relation between ideal and nonideal epistemology, with the aim of justifying ideal epistemology. I proceed by exploring the meaning and purpose of epistemic evaluations. I offer a unified semantics for the ideal and nonideal epistemic "ought", "rational", and so on. The fundamental difference between ideal and nonideal epistemology is that only the nonideal epistemic "ought" implies any substantive "can".
I argue that only ideal epistemic evaluations are, in an important sense, normatively robust: they are non-conventional and not seriously context-sensitive. Preserving substantive "ought"-implies-"can" principles leaves nonideal epistemic evaluations normatively non-robust: they exhibit a high degree of both conventionality and serious context-sensitivity. For this reason, nonideal epistemic evaluations won't characterize a cohesive notion of epistemic rationality. Nonideal epistemic rationality depends not merely on what's epistemically valuable and how to effectively pursue it, but also on modally contingent epistemic conventions and contextually contingent assumed constraints on what we "can" do. If we want a normatively robust theory of epistemic rationality, ideal epistemology is the only game in town. Ideal epistemology therefore has an important role to play in epistemic theorizing.
How should an agent act under normative uncertainty? We might extend the orthodox theory of rational choice to the case of uncertainty between competing normative theories. But this requires that the values assigned by different normative theories be comparable. This paper defends a strategy for avoiding the need for intertheoretic value comparisons: instead of comparing competing moral theories, I argue that values can be represented in terms of a de dicto specification of value. I provide a decision theory for de dicto values that generalizes expected utility theory and compare the proposal with alternative strategies for avoiding the problem of intertheoretic comparisons.
Accuracy-first epistemology aims to show that the norms of epistemic rationality can be derived from the effective pursuit of accuracy. This paper explores the prospects within accuracy-first epistemology for vindicating “modesty”: the thesis that ideal rationality permits uncertainty about one’s own rationality. I argue that accuracy-first epistemology faces serious challenges in accommodating three forms of modesty: uncertainty about what priors are rational, uncertainty about whether one’s update policy is rational, and uncertainty about what one’s evidence is. I argue that the problem stems from the representation of epistemic decision problems. The appropriate representation of decision problems, and corresponding decision rules, for (diachronic) update policies should be a generalization of decision problems and decision rules for (synchronic) coherence. I argue that extant accounts build in conflicting assumptions about which kinds of information about the believer should be used to structure epistemic decision problems. In particular, extant accounts of update build in a form of epistemic consequentialism. Related forms of epistemic consequentialism have been shown to generate problems for accuracy-first epistemology’s purported justifications of probabilism, conditionalization, and the principal principle. These results are vindicated only with nonconsequentialist epistemic decision theories. I close with suggestive examples of how, with a fully nonconsequentialist representation of epistemic decision problems, accuracy-first epistemology can allow for rational modesty.
Does rationality require imprecise credences? Many hold that it does: imprecise evidence requires correspondingly imprecise credences. I argue that this is false. The imprecise view faces the same arbitrariness worries that were meant to motivate it in the first place. It faces these worries because it incorporates a certain idealization. But doing away with this idealization effectively collapses the imprecise view into a particular kind of precise view. On this alternative, our attitudes should reflect a kind of normative uncertainty: uncertainty about what to believe. This view refutes the claim that precise credences are inappropriately informative or committal. Some argue that indeterminate evidential support requires imprecise credences; but I argue that indeterminate evidential support instead places indeterminate requirements on credences, and is compatible with the claim that rational credences may always be precise.
This paper is on the foundations of Bayesianism, a theory of rational belief and inference that has a central place in mathematical statistics and is widely applied in the sciences, engineering, computer science, medicine, law, sports, and beyond. Bayesianism makes highly specific mathematical prescriptions. The aim of epistemic utility theory is to provide a precise philosophical rationale for why adherence to Bayesian dictates is a requirement of rational evaluation of evidence. It does so by revealing a necessary connection between Bayesian rationality and truth, providing formal proofs that adherence to Bayesian dictates maximizes a reasoner’s degree of accuracy in forecasting. In other words, in order to pursue the end of true belief most effectively, an agent must satisfy Bayesian constraints. My paper proves that these results are only true in a highly restricted range of cases: those where the agent’s predictions are certain to have no causal or probabilistic impact on facts in the world. For other cases (for example, public election forecasts that potentially causally impact voter turnout), Bayesian predictions are provably less accurate than some non-Bayesian alternatives. But all is not lost for the Bayesian: I argue that while Bayesianism need not maximize accuracy, it characterizes a maximally evidence-responsive approach to forecasting.
On an attractive, naturalistically respectable theory of intentionality, mental contents are a form of measurement system for representing behavioral and psychological dispositions. This paper argues that a consequence of this view is that the content/attitude distinction is measurement system relative. As a result, there is substantial arbitrariness in the content/attitude distinction. Whether some measurement of mental states counts as characterizing the content of mental states or the attitude is not a question of empirical discovery but of theoretical utility. If correct, this observation has ramifications in the theory of rationality. Some epistemologists have argued that imprecise credences are rationally required, while others have argued that precise credences are rationally required. If the measure theory is correct, however, then neither imprecise credences nor precise credences can be rationally required.
This chapter provides a selective survey of prominent theories of the semantics of deontic modals in logic and natural language. Understanding the logic and truth-conditions of normative language is one of the core areas of metaethics. It informs our understanding of normative arguments and normative reasoning. Some forms of normative language don’t allow for the inferences that classical logic trains philosophers to expect. Understanding what inferences are valid for normative language should impact our understanding of how we reason, and should reason, about the normative.
Can we extend accuracy-based epistemic utility theory to imprecise credences? There's no obvious way of proceeding: some stipulations will be necessary for either (i) the notion of accuracy or (ii) the epistemic decision rule. With some prima facie plausible stipulations, imprecise credences are always required. With others, they’re always impermissible. Care is needed to reach the familiar evidential view of imprecise credence: that whether precise or imprecise credences are required depends on the character of one's evidence. I propose an epistemic utility theoretic defense of a common view about how evidence places demands on imprecise credence: that your spread of credence should cover the range of chance hypotheses left open by your evidence. I argue that objections to the form of epistemic utility theoretic argument that I use will extend to the standard motivation for epistemically mandatory imprecise credences.
It’s been argued that there are no diachronic norms of epistemic rationality. These arguments come partly in response to certain kinds of counterexamples to Conditionalization, but are mainly motivated by a form of internalism that appears to be in tension with any sort of diachronic coherence requirements. I argue that there are, in fact, fundamentally diachronic norms of rationality. And this is to reject at least a strong version of internalism. But I suggest a replacement for Conditionalization that salvages internalist intuitions, and carves a middle ground between (probabilist versions of) conservatism and evidentialism.
Epistemology should take seriously the possibility of rationally evaluable changes in conceptual resources. Epistemic decision theory compares belief states in terms of epistemic value. But it's standardly restricted to belief states that don't differ in their conceptual resources. I argue that epistemic decision theory should be generalized to make belief states with differing concepts comparable. I characterize some possible constraints on epistemic utility functions. Traditionally, the epistemic utility of a total belief state has been understood as a function of the epistemic utility of individual (partial) beliefs. The most natural ways of generalizing this account generate a kind of repugnant conclusion. I characterize some possible alternatives, reflecting different epistemic norms.
The subjective deontic ought generates counterexamples to classical inference rules like modus ponens. It also conflicts with the orthodox view about modals and conditionals in natural language semantics. Most accounts of the subjective ought build substantive and unattractive normative assumptions into the semantics of the modal. I sketch a general semantic account, along with a metasemantic story about the context sensitivity of information-sensitive operators.
Ecumenical views in metaethics hold that normative utterances express hybrid mental states, states which include both a cognitive and a conative component. The ecumenicist can have her cake and eat it too: the view reaps the benefits of both cognitivist and non-cognitivist theories of normative judgement. The conative component of normative judgements accounts for their necessary link with motivation and rational action. The cognitive component makes it possible for the ecumenicist to endorse expressivism without facing the most difficult Frege-Geach challenges. Ridge (2014) provides a defence of ecumenical expressivism about practical normativity that is both ambitious and compelling. I discuss and try to challenge two (unrelated) parts of the view. In the first section, I’ll raise a question for Ridge’s strategy for addressing the Frege-Geach problem, and in particular his proposed recipe for how to interpret normative assertions of arbitrary logical complexity. Then I discuss Ridge’s defence of cognitivism about rationality. I try to highlight some considerations that make expressivism about theoretical and practical rationality attractive.
I discuss the so-called "Zvolenszky problem": that possible worlds semantic accounts of modals, combined with the widely accepted restrictor analysis of conditionals, validate the following schema: If P, ought P. But this schema isn't valid. For example, it's false to say that if you beat up elderly people, you ought to beat up elderly people. I consider two inadequate solutions to this problem and show how they can be combined to generate a more adequate solution. Then I offer a puzzle case for the new account which suggests that we might need a less conservative amendment to the standard semantics. I offer two semantic accounts that make sense of the puzzle case and explain these accounts' merits and risks.