Defined narrowly, epistemology is the study of knowledge and justified belief. As the study of knowledge, epistemology is concerned with the following questions: What are the necessary and sufficient conditions of knowledge? What are its sources? What is its structure, and what are its limits? As the study of justified belief, epistemology aims to answer questions such as: How we are to understand the concept of justification? What makes justified beliefs justified? Is justification internal or external to one's own mind? (...) Understood more broadly, epistemology is about issues having to do with the creation and dissemination of knowledge in particular areas of inquiry. This article will provide a systematic overview of the problems that the questions above raise and focus in some depth on issues relating to the structure and the limits of knowledge and justification. (shrink)

Functionalism in the philosophy of mind is the doctrine that what makes something a mental state of a particular type does not depend on its internal constitution, but rather on the way it functions, or the role it plays, in the system of which it is a part. This doctrine is rooted in Aristotle's conception of the soul, and has antecedents in Hobbes's conception of the mind as a “calculating machine”, but it has become fully articulated (and popularly endorsed) only (...) in the last third of the 20th century. Though the term ‘functionalism’ is used to designate a variety of positions in a variety of other disciplines, including psychology, sociology, economics, and architecture, this entry focuses exclusively on functionalism as a philosophical thesis about the nature of mental states. (shrink)

According to a widely held view epistemic justification is a normative notion. According to another widely held assumption, epistemic justification comes in degrees. Given that gradability requires a context-sensitivity that normativity seems to lack, these two assumptions stand in tension. Giving up the assumption of gradability of justification represents a lesser theoretical cost.

As large language models (LLMs) continue to demonstrate remarkable abilities across various domains, computer scientists are developing methods to understand their cognitive processes, particularly concerning how (and if) LLMs internally represent their beliefs about the world. However, this field currently lacks a unified theoretical foundation to underpin the study of belief in LLMs. This article begins filling this gap by proposing adequacy conditions for a representation in an LLM to count as belief-like. We argue that, while the project of belief (...) measurement in LLMs shares striking features with belief measurement as carried out in decision theory and formal epistemology, it also differs in ways that should change how we measure belief. Thus, drawing from insights in philosophy and contemporary practices of machine learning, we establish four criteria that balance theoretical considerations with practical constraints. Our proposed criteria include accuracy, coherence, uniformity, and use, which together help lay the groundwork for a comprehensive understanding of belief representation in LLMs. We draw on empirical work showing the limitations of using various criteria in isolation to identify belief representations. (shrink)

My dissertation is about Bayesian rationality for non-ideal agents. I show how to derive subjective probabilities from preferences using much weaker rationality assumptions than other standard representation theorems. I argue that non-ideal agents might be uncertain about how they will update on new information and consider two consequences of this uncertainty: such agents should sometimes reject free information and make choices which, taken together, yield sure loss. The upshot is that Bayesian rationality for non-ideal agents makes very different normative demands (...) than ideal Bayesian rationality. (shrink)

This paper aims to flesh out the celebrated notion of reflective equilibrium within a probabilistic framework for epistemic rationality. On the account developed here, an agent's attitudes are in reflective equilibrium when there is a certain sort of harmony between the agent's credences, on the one hand, and what the agent accepts, on the other hand. Somewhat more precisely, reflective equilibrium is taken to consist in the agent accepting, or being prepared to accept, all and only claims that follow from (...) a maximally comprehensive theory that is more probable than any other such theory. Drawing on previous work, the paper shows that when an agent is in reflective equilibrium in this sense, the set of claims they accept or are prepared to accept is bound to be logically consistent and closed under logical implication. The paper also argues that this account can explain various features of philosophical argumentation in which the notion of reflective equilibrium features centrally, such as the emphasis on evaluating philosophical theories holistically rather than in a piecemeal fashion. (shrink)

Standard arguments for Bayesian conditionalizing rely on assumptions that many epistemologists have criticized as being too strong: (i) that conditionalizers must be logically infallible, which rules out the possibility of rational logical learning, and (ii) that what is learned with certainty must be true (factivity). In this paper, we give a new factivity-free argument for the superconditionalization norm in a personal possibility framework that allows agents to learn empirical and logical falsehoods. We then discuss how the resulting framework should be (...) interpreted. Does it still model norms of rationality, or something else, or nothing useful at all? We discuss five ways of interpreting our results, three that embrace them and two that reject them. We find one of each kind wanting, and leave readers to choose among the remaining three. (shrink)

Reasoning from inconclusive evidence, or ‘induction’, is central to science and any applications we make of it. For that reason alone it demands the attention of philosophers of science. This Element explores the prospects of using probability theory to provide an inductive logic, a framework for representing evidential support. Constraints on the ideal evaluation of hypotheses suggest that overall support for a hypothesis is represented by its probability in light of the total evidence, and incremental support, or confirmation, indicated by (...) an increased probability given the evidence than otherwise. This proposal is shown to have the capacity to reconstruct many canons of the scientific method and inductive inference. Along the way, significant objections are discussed, such as the challenge from inductive scepticism and the objection that the probabilistic approach makes evidential support arbitrary. (shrink)

This chapter is concerned with the distinction that most contemporary epistemologists express by distinguishing between “propositional” and “doxastic” justification. The goal is to develop an account of this distinction that applies, not just to full or outright beliefs, but also to partial credences—and indeed, in principle, to attitudes of all kinds. The standard way of explaining this distinction, in terms of the “basing relation”, is criticized, and an alternative account—the “virtue manifestation” account—is proposed in its place. This account has a (...) certain kinship with the epistemological approach that is known as “reliabilism”, but it also has certain key differences—above all, it is compatible with an internalist view of rationality. (shrink)

In response to two trenchant objections, radical subjective Bayesianism has been widely rejected. In this paper, I seek, if not to rehabilitate subjectivism, at least to show its critic what is attractive about the position. I argue that what is at stake in the subjectivism/anti-subjectivism debate is not, as is commonly thought, which norms of rationality are true, but rather, the conception of rationality that we adopt: there is an alternative approach to the widespread telic approach to rationality, which I (...) call the poric approach, on which subjectivism is an attractive position. I then argue that the poric approach—and therefore subjectivism—cannot be easily dismissed. (shrink)

This is a three-part exchange on the relationship between belief and credence. It begins with an opening essay by Roger Clarke that argues for the claim that the notion of credence generalizes the notion of belief. Julia Staffel argues in her reply that we need to distinguish between mental states and models representing them, and that this helps us explain what it could mean that belief is a special case of credence. Roger Clarke's final essay reflects on the compatibility of (...) the previously discussed views with dualist and monist interpretations of our mental states and our ways of modeling them. (shrink)

Some epistemologists think that the Bayesian ideals matter because we can approximate them. That is, our attitudes can be more or less close to the ones of our ideal Bayesian counterpart. In this paper, I raise a worry for this justification of epistemic ideals. The worry is this: In order to correctly compare agents to their ideal counterparts, we need to imagine idealized agents who have the same relevant information, knowledge, or evidence. However, there are cases in which one’s ideal (...) counterpart cannot have one’s information, knowledge, or evidence. In these situations, agents cannot compare themselves to their ideal counterpart. (shrink)

The framework of Solomonoff prediction assigns prior probability to hypotheses inversely proportional to their Kolmogorov complexity. There are two well-known problems. First, the Solomonoff prior is relative to a choice of Universal Turing machine. Second, the Solomonoff prior is not computable. However, there are responses to both problems. Different Solomonoff priors converge with more and more data. Further, there are computable approximations to the Solomonoff prior. I argue that there is a tension between these two responses. This is because computable (...) approximations to Solomonoff prediction do not always converge. (shrink)

Structural requirements of rationality prohibit various things, like having inconsistent combinations of attitudes, having means-end incoherent combinations of attitudes, and so on. But what is the distinctive feature of structural requirements of rationality? And do we fall under an obligation to be structurally rational? These issues have been at the heart of significant debates over the past fifteen years. Some philosophers have recently argued that we can unify the structural requirements of rationality by analyzing what is constitutive of our attitudes (...) (Lee, 2020; Worsnip, 2018a). It has also been suggested that, in the course of good first-personal deliberation, agents should treat structurally irrational combinations of attitudes as off-limits (Worsnip, 2021). In this paper, I raise a worry for these two theses concerning structural rationality. Roughly, I argue that some imperfect epistemic agents (like us) can be disposed to have structurally irrational combinations of attitudes. Also, when these imperfect agents deliberate, they should not treat structurally irrational combinations of attitudes as off-limits. Given our imperfections, being structurally irrational can very well be the best option we have. More generally, these observations reveal that structural rationality should not always be theorized independently from more “substantive” norms, like responsiveness to reasons or expected value optimization. (shrink)

– We offer a new motivation for imprecise probabilities. We argue that there are propositions to which precise probability cannot be assigned, but to which imprecise probability can be assigned. In such cases the alternative to imprecise probability is not precise probability, but no probability at all. And an imprecise probability is substantially better than no probability at all. Our argument is based on the mathematical phenomenon of non-measurable sets. Non-measurable propositions cannot receive precise probabilities, but there is a natural (...) way for them to receive imprecise probabilities. The mathematics of non-measurable sets is arcane, but its epistemological import is far-reaching; even apparently mundane propositions are liable to be affected by non-measurability. The phenomenon of non-measurability dramatically reshapes the dialectic between critics and proponents of imprecise credence. Non-measurability offers natural rejoinders to prominent critics of imprecise credence. Non-measurability even reverses some of the critics’ arguments—by the very lights that have been used to argue against imprecise credences, imprecise credences are better than precise credences. (shrink)

Formal epistemologists often claim that our credences should be representable by a probability function. Complete probabilistic coherence, however, is only possible for ideal agents, raising the question of how this requirement relates to our everyday judgments concerning rationality. One possible answer is that being rational is a contextual matter, that the standards for rationality change along with the situation. Just like who counts as tall changes depending on whether we are considering toddlers or basketball players, perhaps what counts as rational (...) shifts according to whether we are considering ideal agents or creatures more like ourselves. Even though a number of formal epistemologists have endorsed this type of solution, I will argue that there is no way to spell out this contextual account that can make sense of our everyday judgments about rationality. Those who defend probabilistic coherence requirements will need an alternative account of the relationship between real and ideal rationality. (shrink)

In this paper, we ask: how should an agent who has incoherent credences update when they learn new evidence? The standard Bayesian answer for coherent agents is that they should conditionalize; however, this updating rule is not defined for incoherent starting credences. We show how one of the main arguments for conditionalization, the Dutch strategy argument, can be extended to devise a target property for updating plans that can apply to them regardless of whether the agent starts out with coherent (...) or incoherent credences. The main idea behind this extension is that the agent should avoid updating plans that increase the possible sure loss from Dutch strategies. This happens to be equivalent to avoiding updating plans that increase incoherence according to a distance-based incoherence measure. (shrink)

What is the relevance of ideals for determining virtuous argumentative practices? According to Bailin and Battersby (2016), the telos of argumentation is to improve our cognitive systems, and adversariality plays no role in ideally virtuous argumentation. Stevens and Cohen (2019) grant that ideal argumentation is collaborative, but stress that imperfect agents like us should not aim at approximating the ideal of argumentation. Accordingly, it can be virtuous, for imperfect arguers like us, to act as adversaries. Many questions are left unanswered (...) by both camps. First, how do we conceptualize an ideal and its approximation? Second, how can we determine what is the ideal of argumentation? Third, can we extend Stevens and Cohen’s anti-approximation argument beyond virtue theory? In order to respond to these questions, this paper develops a second-best perspective on ideal argumentation. The Theory of the Second Best is a formal contribution to the field of utility (or welfare) optimization. Its main conclusion is that, in non-ideal circumstances, approximating ideals might be suboptimal. (shrink)

In Unsettled Thoughts, Julia Staffel argues that non-ideal thinkers should seek to approximate ideal Bayesian rationality. She argues that the more rational you are, the more benefits of rationality you will enjoy. After summarizing Staffel's main results, this paper looks more closely at two issues that arise later in the book: the relationship between Bayesian rationality and other kinds of rationality, and the role that outright belief plays in addition to credence. Ultimately, I argue that there are several roles that (...) outright belief might play, and I explore different ways that these roles for belief might fit together. (shrink)

Which rules should guide our reasoning? Human reasoners often use reasoning shortcuts, called heuristics, which function well in some contexts but lack the universality of reasoning rules like deductive implication or inference to the best explanation. Does it follow that human reasoning is hopelessly irrational? I argue: no. Heuristic reasoning often represents human reasoners reaching a local rational maximum, reasoning more accurately than if they try to implement more “ideal” rules of reasoning. I argue this is a genuine rational achievement. (...) Our ideal rational advisors would advise us to reason with heuristic rules, not more complicated ideal rules. I argue we do not need a radical new account of epistemic norms to make sense of the success of heuristic reasoning. (shrink)

Do people tend to be overconfident? Many think so. They’ve run studies on whether people are calibrated: whether their average confidence in their opinions matches the proportion of those opinions that are true. Under certain conditions, people are systematically ‘over-calibrated’—for example, of the opinions they’re 80% confident in, only 60% are true. From this empirical over-calibration, it’s inferred that people are irrationally overconfident. My question: When and why is this inference warranted? Answering it requires articulating a general connection between being (...) rational and being right—something extant studies have not done. I show how to do so using the notion of deference. This provides a theoretical foundation for calibration research, but also reveals a flaw: the connection between being rational and being right is much weaker than is standardly assumed, since rational people can often be expected to be miscalibrated. Thus we can’t test whether people are overconfident by simply testing whether they are over-calibrated; instead, we must try to predict the rational deviations from calibration, and then compare those predictions to people’s performance. I show how this can be done—and that doing so complicates the interpretation of robust empirical effects. (shrink)

This précis gives a brief summary of the key points of Julia Staffel’s book Unsettled Thoughts.

Formal representations drawn from rational choice theory have been used in a variety of ways to fruitfully model the way in which actual agents are approximately rational. This analysis requires bridging between ideal normative theory, in which the mechanisms, representations, and other such internal parts are in an important sense interchangeable, and descriptive psychological theory, in which understanding the internal workings of the agent is often the main goal of the entire inquiry. In this paper, I raise a problem brought (...) on by this gap: for almost every theory of approximate rationality, there will be an empirically indistinguishable alternative that individuates the parts of the agent in a significantly different way. (shrink)

Ideal epistemologists investigate the nature of pure epistemic rationality, abstracting away from human cognitive limitations. Non-ideal epistemologists investigate epistemic norms that are satisfiable by most humans, most of the time. Ideal epistemology faces a number of challenges, aimed at both its substantive commitments and its philosophical worth. This paper explains the relation between ideal and non-ideal epistemology, with the aim of justifying ideal epistemology. Its approach is meta-epistemological, focusing on the meaning and purpose of epistemic evaluations. I provide an account (...) on which the fundamental difference between ideal and non-ideal epistemic evaluations is that only the non-ideal epistemic ‘ought’ implies any substantive ‘can’. I argue that only ideal epistemic evaluations are ‘normatively robust’: they are neither conventional nor seriously context-sensitive. Non-ideal epistemic evaluations are normatively non-robust, exhibiting both conventionality and serious context-sensitivity from an interesting variety of distinct sources. For this reason, non-ideal epistemic evaluations won’t characterize the fundamental nature of epistemic rationality. Non-ideal epistemic rationality depends, not merely on what’s epistemically valuable, but also on modally contingent epistemic conventions and contextually contingent constraints on epistemic options. If we want a normatively robust theory of epistemic rationality, ideal epistemology is the only game in town. (shrink)

Recently some epistemologists have approached the question whether epistemic justification comes in degrees from a linguistic perspective. Drawing insights from linguistic analyses of gradable adjectives, they investigate whether epistemic occurrences of ‘justified’ are gradable and if yes what type of gradability they involve. These authors conclude that the adjective passes standard tests for gradability, but they classify it as belonging to different categories: as either an absolute or a relative gradable adjective. The aim of this paper is to further clarify (...) the question of what kind of gradability is instantiated by epistemic uses of ‘justified’, and to investigate the consequences that this may have on epistemological theorizing. In particular, we challenge the alleged evidence provided by Siscoe (2021: 517–59) for the relative gradability of ‘justified’ and we provide positive arguments for the claim that ‘justified’ is a specific kind of absolute gradable adjective (a totally closed scale absolute with default minimal reading). We consider some important philosophical implications of these results. We also argue that the scale type of ‘justified’ is ordinal, and we consider which other properties typically associated to justification possess a similar type of scale. We argue that several popular contemporary models of graded justification, such as probabilistic and reliabilist accounts and various kinds of Lockean views, do not fit well with our results about the gradability of ‘justified’. Linguistic data rather favour normalcy and plausibility accounts of graded justification. (shrink)