LLMs have dramatically improved in capabilities in recent years. This raises the question of whether LLMs could become genuine agents with beliefs and desires. This paper demonstrates an in principle limit to LLM agency, based on their architecture. LLMs are next word predictors: given a string of text, they calculate the probability that various words can come next. LLMs produce outputs that reflect these probabilities. I show that next word predictors are exploitable. If LLMs are prompted to make probabilistic predictions (...) about the world, these predictions are guaranteed to be incoherent, and so Dutch bookable. If LLMs are prompted to make choices over actions, their preferences are guaranteed to be intransitive, and so money pumpable. In short, the problem is that selecting an action based on its potential value is structurally different then selecting the description of an action that is most likely given a prompt: probability cannot be forced into the shape of expected value. The in principle exploitability of LLMs raises doubts about how agential they can become. This exploitability also offers an opportunity for humanity to safely control such AI systems. (shrink)

Must rational thinkers have consistent sets of beliefs? I shall argue that it can be rational for a thinker to believe a set of propositions known to be inconsistent. If this is right, an important test for a theory of rational belief is that it allows for the right kinds of inconsistency. One problem we face in trying to resolve disagreements about putative rational requirements is that parties to the disagreement might be working with different conceptions of the relevant attitudes. (...) My aim is modest. I hope to show that there is at least one important notion of belief such that a thinker might rationally hold a collection of beliefs (so understood) even when the thinker knows their contents entail a contradiction. (shrink)

We examine two leading theories of rational belief, the Lockean view and the explanationist view. The first is appealing because it fits with some independently plausible claims about the ways that rational persons pursue their aims. The second is appealing because it seems to account for intuitions that cause trouble for the Lockean view. While fitting the intuitive data is desirable, we are troubled that the explanationist view seems to clash with our theoretical beliefs about what rationality must be like. (...) We think that upon further examination, the intuitive appeal of the explanationist view starts to diminish. We also think that these further intuitions that spell trouble for the explanationist spell trouble for any theory that is not expectationist. We propose a novel expectationist theory of rational belief that improves upon the Lockean and the explanationist views. We think that recent defences of the Lockean view contain an important insight. A substantive theory of rational response should be based on a suitable theory of prizes and a suitable theory of how we should pursue prizes in the face of uncertainty. Most theories of rational belief typically take for granted a truth-centred picture of epistemic prizes (e.g., that epistemic desirability and undesirability can be fully understood in terms of accuracy) and then differ in terms of how they recommend pursuing prizes so understood. We think the Lockeans embrace plausible principles of how prizes should be pursued. We trace the difficulties that this view faces to veritistic assumptions about prizes. We suggest that some prizes are epistemically loaded in that a complete description of the prize will itself make reference to our epistemic states or standards. We argue that knowledge matters to rational belief and choice because in the epistemic domain, knowledge is the prize. We see this in practical domains, too. In some choice settings, what's desired is desired, in part, because it involves a kind of connection to reality only knowledge provides. (shrink)

We know a lot about what the world is like. We know less, it seems, about what we know about what the world is like. According to a common thought, it is easier for us to come to know about the state of the world than to come to know about the state of our own knowledge. What explains this gap? An attractively simple hypothesis is that our ignorance about what we know is explained by our ignorance about the world. (...) There are things we fail to know about what we know about the world because there are things we fail to know about the world. This hypothesis is often motivated by the idea that knowledge requires a margin‐for‐error. In this paper, I'll argue that this simple hypothesis is inadequate. Not all our ignorance of our knowledge can be explained by our ignorance about the world. In this sense, at least some of our ignorance about what we know is brute. (shrink)

In this paper I investigate whether there are any cases in which it is rational for a person to hold inconsistent beliefs and, if there are, just what implications this might have for the theory of epistemic justification. A number of issues will crop up along the way – including the relation between justification and rationality, the nature of defeat, the possibility of epistemic dilemmas, the importance of positive epistemic duties, and the distinction between transitional and terminal attitudes.

It seems that different epistemic norms can come into conflict and so we might wonder what happens when they do impose incompatible requirements upon us. According to the dilemmic view, they might sometimes generate sets of requirements that cannot be satisfied, ensuring that there is no rationally acceptable way for a thinker to deal with the predicament she’s in. After reviewing the case for the dilemmic view, I introduce an alternative framework that accounts for the appearance of dilemma-like conflicts without (...) committing us to the possibility of genuine dilemmas. I argue that this alternative view has some important explanatory virtues that the dilemmic view lacks and can account for the problematic intuitions that seem to support the dilemmic view. (shrink)

This article develops a modal theory of justification, according to which a belief is justified if it is more possible that it amounts to knowledge than that it does not. The core of the theory is neutral between internalism and externalism and it solves two problems that extant modal accounts of justification suffer from. In developing the theory, an account of comparative possibility is provided to yield degrees of justification.

We argue that there is nothing that can do the work that normative reasons are expected to do. A currently popular view is that in any given situation, a set of normative reasons (understood as a set of facts, typically about the agent’s situation) always determines the ways we prospectively should or should not respond. We discuss an example that we think shows no such collection of facts could have this normative significance. A radical response might be to dispense with (...) reasons and explain the facts about the ways we should respond using something else entirely (e.g., expected value). We think a more fruitful approach is to explore the possibility of using decision-theoretic tools in a reasons-centred framework. (shrink)

Several authors have recently argued that knowledge is not the aim of inquiry since it can make sense to inquire into a question even though one knows the answer. I argue that this a faulty diagnostic for determining whether one has met the constitutive standard of success of an activity type. The constitutive standards of success tell us when an activity is successful, but such standards can be exceeded and exceeding them can be reasonable. To back this up I develop (...) an account of zetetic supererogation and argue that subjects who continue to inquire into a question while knowing the answer do what is zetetically supererogatory. (shrink)

When should you believe testimony that you receive from your social media feeds? One natural answer is suggested by non-reductionism in the epistemology of testimony. Avoid accepting social media testimony if you have an undefeated defeater for it. Otherwise, you may accept it. I argue that this is too permissive to be an adequate epistemic policy because social media have some characteristics that tend to facilitate the efficacy of misinformation on those platforms. I formulate and defend an alternative epistemic policy (...) for belief on social media, which is inspired by reductionism in the epistemology of testimony. (shrink)

In this article I draw on existing theory in a project of synthesis and integration to develop a terminological, conceptual, and graphic model for understanding the scope of critical thinking. I begin by showing why the idea of critical scope matters. I then develop a model that maps the scope of critical thinking across individual, sociocultural, and existential domains, in conjunction with the ontological, epistemological, emotional, and political dimensions. Building on this, I outline the relationships between contexts, critical frameworks, and (...) metacritique as essential to understanding the scope of critical thinking. The model also includes a means for organising the attributes (character traits) of critical thinkers. This project provides resources to encourage and guide critical thinking at a more expansive scope in education, society, and daily life by helping people to understand the contours and potential confines of criticality. This has profound implications for education and society, helping to improve the rigour, reach, and diversity of critical thinking. This model is far from complete, and designed to invite ongoing engagement, including critique. It is intended as a rubric to prompt and inspire, not as a device to ‘measure’ or ‘assess’ critical thinking. Educationally, the intention is to provide resources for understanding, negotiating, and facilitating critical thinking at an expansive scope. (shrink)