This paper presents an argument that certain AI safety measures, rather than mitigating existential risk, may instead exacerbate it. Under certain key assumptions - the inevitability of AI failure, the expected correlation between an AI system's power at the point of failure and the severity of the resulting harm, and the tendency of safety measures to enable AI systems to become more powerful before failing - safety efforts have negative expected utility. The paper examines three response strategies: Optimism, Mitigation, and (...) Holism. Each faces challenges stemming from intrinsic features of the AI safety landscape that we term Bottlenecking, the Perfection Barrier, and Equilibrium Fluctuation. The surprising robustness of the argument forces a re-examination of core assumptions around AI safety and points to several avenues for further research. (shrink)

Under what conditions are two utterances utterances of the same word? What are words? That these questions have not received much attention is rather surprising: after all, philosophers and linguists frequently appeal to considerations about word and sentence identity in connection with a variety of puzzles and problems that are foundational to the very subject matter of philosophy of language and linguistics.1 Kaplan’s attention to words is thus to be applauded. And there is no doubt that his discussion contains many (...) useful insights. Nevertheless, we find his picture deeply flawed for a variety of crosscutting reasons. Our aim in this paper is to further advance an understanding of the nature of words, both by remedying the problems with Kaplan’s account, and also by achieving a suitable perspective on what the metaphysical investigation of word identity can hope to achieve. Our discussion divides into four parts. In Part One, we examine and critique Kaplan’s discussion of a contrast integral to his own account: that between the type-token and the stage-continuant conceptions of words. In Part Two, we present three constraints on any account of words and two further themes in Kaplan’s discussion central to his conception of words – the role of repetition and the constitutive authority of intentions. While these ideas have laudable motivations, we argue they are far from the best way of making good on the insights that drive them. The final two sections take a skeptical turn. In Part Three, we express doubt about Kaplan’s presumption of the importance of what he calls ‘common currency names’, thus raising a suspicion that he may be in pursuit of chimera. Finally, in Part Four, we express pessimism about whether interesting answers to question above will be forthcoming, and suggest that the legitimacy of our word ontology need not depend on the availability of such answers. Along the way, we tease apart a number of metaphysical questions in the vicinity of the topic of word individuation – questions that are often not disentangled – and consider how the discussion of the previous parts bears on them.. (shrink)

Judging by our folk appraisals, then, knowledge and action are intimately related. The theories of rational action with which we are familiar leave this unexplained. Moreover, discussions of knowledge are frequently silent about this connection. This is a shame, since if there is such a connection it would seem to constitute one of the most fundamental roles for knowledge. Our purpose in this paper is to rectify this lacuna, by exploring ways in which knowing something is related to rationally acting (...) upon it, defending one particular proposal against anticipated objections. (shrink)

The need to address our question arises from two sources, one in Kant and the other in a certain type of response to so-called Reformed epistemology. The first source consists in a tendency to distinguish theoretical beliefs from practical beliefs (commitments to the world's being a certain way versus commitments to certain pictures to live by), and to treat theistic belief as mere practical belief. We trace this tendency in Kant's corpus, and compare and contrast it with Aquinas's view and (...) a more conservative Kantian view. We reject the theistic-belief-as-mere-practical-belief view: it is bad descriptive anthropology, it embraces a misguided ideal of a fragmented self unattainable by human beings, and it will deter people from the most desirable sort of faith. The second source consists in the idea that since theistic beliefs function as answers to why-questions, their epistemic status hangs on whether they meet certain distinctively explanatory standards, whatever support they might receive from other sources. We argue that this is a non-sequitur and suggest questions for further research. (shrink)

A very simple contextualist treatment of a sentence containing an epistemic modal, e.g. a might be F, is that it is true iff for all the contextually salient community knows, a is F. It is widely agreed that the simple theory will not work in some cases, but the counterexamples produced so far seem amenable to a more complicated contextualist theory. We argue, however, that no contextualist theory can capture the evaluations speakers naturally make of sentences containing epistemic modals. If (...) we want to respect these evaluations, our best option is a relativist theory of epistemic modals. On a relativist theory, an utterance of a might be F can be true relative to one context of evaluation and false relative to another. We argue that such a theory does better than any rival approach at capturing all the behaviour of epistemic modals. (shrink)

We think we have lots of substantial knowledge about the future. But contemporary wisdom has it that indeterminism prevails in such a way that just about any proposition about the future has a non-zero objective chance of being false.2, 3 What should one do about this? One, pessimistic, reaction is scepticism about knowledge of the future. We think this should be something of a last resort, especially since this scepticism is likely to infect alleged knowledge of the present and past. (...) One anti-sceptical strategy is to pin our hopes on determinism, conceding that knowledge of the future is unavailable in an indeterministic world. This is not satisfying either: we would rather not be hostage to empirical fortune in the way that this strategy recommends. A final strategy, one that we shall explore in this paper, is one of reconciliation: knowledge of a proposition is compatible with a subject’s belief having a non-zero objective chance of error.4 Following Williamson, we are interested in tying knowledge to the presence or absence of error in close cases, and so we shall explore the connections between knowledge and objective chance within such a framework. We don’t want to get tangled up here in complications involved in attempting to formulate a necessary and sufficient condition for knowledge in terms of safety. Instead, we will assume the following rough and ready necessary condition: a subject knows P only if she could not easily have falsely believed P.5 Assuming that easiness is to be spelt.. (shrink)

Since the release of ChatGPT, there has been a lot of debate about whether AI systems pose an existential risk to humanity. This paper develops a general framework for thinking about the existential risk of AI systems. We analyze a two-premise argument that AI systems pose a threat to humanity. Premise one: AI systems will become extremely powerful. Premise two: if AI systems become extremely powerful, they will destroy humanity. We use these two premises to construct a taxonomy of ‘survival (...) stories’, in which humanity survives into the far future. In each survival story, one of the two premises fails. Either scientific barriers prevent AI systems from becoming extremely powerful; or humanity bans research into AI systems, thereby preventing them from becoming extremely powerful; or extremely powerful AI systems do not destroy humanity, because their goals prevent them from doing so; or extremely powerful AI systems do not destroy humanity, because we can reliably detect and disable systems that have the goal of doing so. We argue that different survival stories face different challenges. We also argue that different survival stories motivate different responses to the threats from AI. Finally, we use our taxonomy to produce rough estimates of ‘P(doom)’, the probability that humanity will be destroyed by AI. (shrink)

Is the fact that our universe contains fine-tuned life evidence that we live in a multiverse? Ian Hacking and Roger White influentially argue that it is not. We approach this question through a systematic framework for self-locating epistemology. As it turns out, leading approaches to self-locating evidence agree that the fact that our own universe contains fine-tuned life indeed confirms the existence of a multiverse. This convergence is no accident: we present two theorems showing that, in this setting, any updating (...) rule that satisfies a few reasonable conditions will have the same feature. The conclusion that fine-tuned life provides evidence for a multiverse is hard to escape. (shrink)

Recent research has identified a tension between the Safety principle that knowledge is belief without risk of error, and the Closure principle that knowledge is preserved by competent deduction. Timothy Williamson reconciles Safety and Closure by proposing that when an agent deduces a conclusion from some premises, the agent’s method for believing the conclusion includes their method for believing each premise. We argue that this theory is untenable because it implies problematically easy epistemic access to one’s methods. Several possible solutions (...) are explored and rejected. (shrink)

Inquiry aims at knowledge. Your inquiry into a question succeeds just in case you come to know the answer. However, combined with a common picture on which misleading evidence can lead knowledge to be lost, this view threatens to recommend a novel form of dogmatism. At least in some cases, individuals who know the answer to a question appear required to avoid evidence bearing on it. In this paper, we’ll aim to do two things. First, we’ll present an argument for (...) this novel form of dogmatism and show that it presents a substantive challenge. Second, we’ll consider a way those who take knowledge to be the aim of inquiry can mount a response. In the course of doing so, we’ll try to get clearer on the normative connections between inquiry, knowledge and evidence gathering. (shrink)

The analysis of desire ascriptions has been a central topic of research for philosophers of language and mind. This work has mostly focused on providing a theory of want reports, that is, sentences of the form ‘S wants p’. In this paper, we turn from want reports to a closely related but relatively understudied construction, namely hope reports, that is, sentences of the form ‘S hopes p’. We present two contrasts involving hope reports and show that existing approaches to desire (...) fail to explain these contrasts. We then develop a novel account that combines some of the central insights in the literature. We argue that our theory provides an elegant account of our contrasts and yields a promising analysis of hoping. (shrink)

In this paper, we present two puzzles involving desire reports concerning series of events. What does a person want to happen in the first event – is it the event with the highest expected return, or the event that is the initial part of the best series? We show that existing approaches fail to resolve the puzzles around this question and develop a novel account of our own. Our semantics is built around three ideas. First, we propose that desire ascriptions (...) are evaluated relative to a contextually supplied set of propositions, or alternatives. The semantic value of an ascription ‘S wants p’ is determined by S's preference ordering over these alternatives. Second, we propose that desire reports carry a requirement to the effect that the prejacent of the ascription must be suitably related to the background set of alternatives. Finally, we suggest that desire reports carry a dominance condition concerning the subject's ranking of the alternatives. Overall, we argue that our theory provides us with an elegant resolution of our puzzles, and yields a promising approach to desire. (shrink)

In an anonymous referee report written in 1945, Church suggested a sweeping argument against verificiationism, the thesis that every truth is knowable. The argument, which was published with due acknowledgement by Fitch almost two decades later, has generated significant attention as well as some interesting successor arguments. In this paper, we present the most important episodes in this intellectual history using the logic that Church himself favoured, and we give reasons for thinking that the arguments are less than decisive. However, (...) we present some new arguments against verificationism that we believe to be dialectically effective. (shrink)

This paper models knowledge in cases where an agent has multiple experiences over time. Using this model, we introduce a series of observations that undermine the pretheoretic idea that the evidential significance of experience depends on the extent to which that experience matches the world. On the basis of these observations, we model knowledge in terms of what is likely given the agent’s experience. An agent knows p when p is implied by her epistemic possibilities. A world is epistemically possible (...) when its probability given the agent’s experiences is not significantly lower than the probability of the actual world given that experience. (shrink)

We show that someone who believes in both gunk and the possibility of supertasks has to give up either a plausible principle about where gunk can be located, or plausible conservation principles.

Many defend the thesis that when someone knows p, they couldn’t easily have been wrong about p. But the notion of easy possibility in play is relatively undertheorized. One structural idea in the literature, the principle of Counterfactual Closure (CC), connects easy possibility with counterfactuals: if it easily could have happened that p, and if p were the case, then q would be the case, it follows that it easily could have happened that q. We first argue that while CC (...) is false, there is a true restriction of it to cases involving counterfactual dependence on a coin flip. The failure of CC falsifies a model where the easy possibilities are counterfactually similar to actuality. Next, we show that extant normality models, where the easy possibilities are the sufficiently normal ones, are incompatible with the restricted CC thesis involving coin flips. Next, we develop a new kind of normality theory that can accommodate the restricted version of CC. This new theory introduces a principle of Counterfactual Contamination, which says roughly that any world is fairly abnormal if at that world very abnormal events counterfactually depend on a coin flip. Finally, we explain why coin flips and other related events have a special status. A central take home lesson is that the correct principle in the vicinity of Safety is importantly normality-theoretic rather than (as it is usually conceived) similarity-theoretic. (shrink)

In this paper, we propose a novel account of desire reports, i.e. sentences of the form 'S wants p'. Our theory is partly motivated by Phillips-Brown's (2021) observation that subjects can desire things even if those things aren't best by the subject's lights. That is, being best isn't necessary for being desired. We compare our proposal to existing theories, and show that it provides a neat account of the central phenomenon.

The external world sceptic tells some familiar narratives involving massive deception. Perhaps we are brains in vats. Perhaps we are the victim of a deceitful demon. You know the drill. The sceptic proceeds by observing first that victims of such deceptions know nothing about their external environment and that second, since we cannot rule out being a victim of such deceptions our- selves, our own external world beliefs fail to attain the status of knowledge. Discussions of global external world scepticism (...) tend to focus on the second step, where a number of well-known lines of resistance have been offered. But there has been little attention to the first, seemingly innocuous step. That will be the focus of this paper. Part one – sections 1, 2, and 3 – will explain why these standard narratives are not convincing examples of cases where there is no knowledge of the external world. In part two – section 4 – we shall undertake a useful case study. David Lewis’s ‘Elusive Knowledge’ is often thought of as presenting an epistemological vision that is somewhat friendly to external world scepticism: as Lewis himself presents things, there are contexts where external world knowledge ascriptions are uniformly false, and where true knowledge ascriptions are limited to either axiomatic truths or truths about our inner life. We examine his discussion in the light of the preceding reflections and show that the framework he presents is not so concessionary to global external world scepticism after all. (shrink)

Lewis's notion of a "natural" property has proved divisive: some have taken to the notion with enthusiasm, while others have been sceptical. However, it is far from obvious what the enthusiasts and the sceptics are disagreeing about. This paper attempts to articulate what is at stake in this debate.

Inheritance is the principle that deontic `ought' is closed under entailment. This paper is about a tension that arises in connection with Inheritance. More specifically, it is about two observations that pull in opposite directions. One of them raises questions about the validity of Inheritance, while the other appears to provide strong support for it. We argue that existing approaches to deontic modals fail to provide us with an adequate resolution of this tension. In response, we develop a positive analysis, (...) and show that this proposal provides a satisfying account of our intuitions. (shrink)

David Builes presents a paradox concerning how confident you should be that any given member of an infinite collection of fair coins landed heads, conditional on the information that they were all flipped and only finitely many of them landed heads. We argue that if you should have any conditional credence at all, it should be 1/2.

Moral encroachment is the thesis that morality has an effect---unrecognized by traditional epistemology---on which doxastic states are epistemically appropriate. The thesis is increasingly popular among those who, in opposition to Gendler (2011), desire harmony between epistemic and moral demands on belief. This paper has three main goals. First, drawing on attractive structural principles concerning belief and justification, it is shown that a thoroughgoing harmony between moral and epistemic demands is implausible. This weakens the motivation for positing moral encroachment, but a (...) proponent might still hope that moral encroachment can have a significant effect, so that many of the central cases of apparent disharmony can be avoided. The next goal of the paper is to argue that any significant effect of moral encroachment would manifest itself in the credences which are most epistemically appropriate. Finally, it is shown that even a small effect of moral encroachment on credence is in tension with widely accepted basic constraints on epistemically appropriate credence. (shrink)

We present and discuss a counterexample to the following plausible principle: if you know that a coin is fair, and for all you know it is going to be flipped, then for all you know it will land tails.

Famous results by David Lewis show that plausible-sounding constraints on the probabilities of conditionals or evaluative claims lead to unacceptable results, by standard probabilistic reasoning. Existing presentations of these results rely on stronger assumptions than they really need. When we strip these arguments down to a minimal core, we can see both how certain replies miss the mark, and also how to devise parallel arguments for other domains, including epistemic “might,” probability claims, claims about comparative value, and so on. A (...) popular reply to Lewis's results is to claim that conditional claims, or claims about subjective value, lack truth conditions. For this strategy to have a chance of success, it needs to give up basic structural principles about how epistemic states can be updated—in a way that is strikingly parallel to the commitments of the project of dynamic semantics. (shrink)

Our main aims in this paper is to discuss and criticise the core thesis of a position that has become known as phenomenal conservatism. According to this thesis, its seeming to one that p provides enough justification for a belief in p to be prima facie justified (a thesis we label Standard Phenomenal Conservatism). This thesis captures the special kind of epistemic import that seemings are claimed to have. To get clearer on this thesis, we embed it, first, in a (...) probabilistic framework in which updating on new evidence happens by Bayesian conditionalization, and second, a framework in which updating happens by Jeffrey conditionalization. We spell out problems for both views, and then generalize some of these to non-probabilistic frameworks. The main theme of our discussion is that the epistemic import of a seeming (or experience) should depend on its content in a plethora of ways that phenomenal conservatism is insensitive to. (shrink)

To what extent are the answers to theological questions knowable? And if the relevant answers are knowable, which sorts of inquirers are in a position to know them? In this chapter we shall not answer these questions directly but instead supply a range of tools that may help us make progress here. The tools consist of plausible structural constraints on knowledge. After articulating them, we shall go on to indicate some ways in which they interact with theological scepticism. In some (...) cases the structural constraints bear directly on whether one can know answers to theological questions. But the structural considerations are related to theological scepticism in other interesting ways as well; for instance we will also be using them to explore the significance of scepticism, by addressing questions such as ‘To what extent does it matter whether or not we can know the answer to theological questions?’. (shrink)

How should a group with different opinions (but the same values) make decisions? In a Bayesian setting, the natural question is how to aggregate credences: how to use a single credence function to naturally represent a collection of different credence functions. An extension of the standard Dutch-book arguments that apply to individual decision-makers recommends that group credences should be updated by conditionalization. This imposes a constraint on what aggregation rules can be like. Taking conditionalization as a basic constraint, we gather (...) lessons from the established work on credence aggregation, and extend this work with two new impossibility results. We then explore contrasting features of two kinds of rules that satisfy the constraints we articulate: one kind uses fixed prior credences, and the other uses geometric averaging, as opposed to arithmetic averaging. We also prove a new characterisation result for geometric averaging. Finally we consider applications to neighboring philosophical issues, including the epistemology of disagreement. (shrink)

Seth Yalcin has pointed out some puzzling facts about the behaviour of epistemic modals in certain embedded contexts. For example, conditionals that begin ‘If it is raining and it might not be raining, … ’ sound unacceptable, unlike conditionals that begin ‘If it is raining and I don’t know it, … ’. These facts pose a prima facie problem for an orthodox treatment of epistemic modals as expressing propositions about the knowledge of some contextually specified individual or group. This paper (...) develops an explanation of the puzzling facts about embedding within an orthodox framework. (shrink)

The principle of universal instantiation plays a pivotal role both in the derivation of intensional paradoxes such as Prior’s paradox and Kaplan’s paradox and the debate between necessitism and contingentism. We outline a distinctively free logical approach to the intensional paradoxes and note how the free logical outlook allows one to distinguish two different, though allied themes in higher-order necessitism. We examine the costs of this solution and compare it with the more familiar ramificationist approaches to higher-order logic. Our assessment (...) of both approaches is largely pessimistic, and we remain reluctantly inclined to take Prior’s and Kaplan’s derivations at face value. (shrink)

The problem of evil is the most prominent argument against the existence of God. Skeptical theists contend that it is not a good argument. Their reasons for this contention vary widely, involving such notions as CORNEA, epistemic appearances, 'gratuitous' evils, 'levering' evidence, and the representativeness of goods. We aim to dispel some confusions about these notions, in particular by clarifying their roles within a probabilistic epistemology. In addition, we develop new responses to the problem of evil from both the phenomenal (...) conception of evidence and the knowledge-first view of evidence. (shrink)

Some have argued for a division of epistemic labor in which mathematicians supply truths and philosophers supply their necessity. We argue that this is wrong: mathematics is committed to its own necessity. Counterfactuals play a starring role.

“There are no gaps in logical space,” David Lewis writes, giving voice to sentiment shared by many philosophers. But different natural ways of trying to make this sentiment precise turn out to conflict with one another. One is a *pattern* idea: “Any pattern of instantiation is metaphysically possible.” Another is a *cut and paste* idea: “For any objects in any worlds, there exists a world that contains any number of duplicates of all of those objects.” We use resources from model (...) theory to show the inconsistency of certain packages of combinatorial principles and the consistency of others. (shrink)

Certain passages in Kaplan’s ‘Demonstratives’ are often taken to show that non-vacuous sentential operators associated with a certain parameter of sentential truth require a corresponding relativism concerning assertoric contents: namely, their truth values also must vary with that parameter. Thus, for example, the non-vacuity of a temporal sentential operator ‘always’ would require some of its operands to have contents that have different truth values at different times. While making no claims about Kaplan’s intentions, we provide several reconstructions of how such (...) an argument might go, focusing on the case of time and temporal operators as an illustration. What we regard as the most plausible reconstruction of the argument establishes a conclusion similar enough to that attributed to Kaplan. However, the argument overgenerates, leading to absurd consequences. We conclude that we must distinguish assertoric contents from compositional semantic values, and argue that once they are distinguished, the argument fails to establish any substantial conclusions. We also briefly discuss a related argument commonly attributed to Lewis, and a recent variant due to Weber. (shrink)

Recent work on the evidential argument from evil offers us sundry considerations which are intended to weigh against this form of atheological arguments. By far the most provocative is that on a priori grounds alone, evil can be shown to be evidentially impotent. This astonishing thesis has been given a vigorous defense by Keith Yandell. In this paper, we shall measure the prospects for an a priori dismissal of evidential arguments from evil.

This paper canvasses the various dimensions along which theories of truth may disagree about the extent to which truth is minimal.

Some central epistemological notions are expressed by sentential operators O that entail the possibility of knowledge in the sense that 'Op' entails 'It is possible to know that p'. We call these modal-epistemological notions. Using apriority and being in a position to know as case studies, we argue that the logics of modal epistemological notions are extremely weak. In particular, their logics are not normal and do not include any closure principles.

It is common these days for theists to argue that we aren’t justified in believing atheism on the basis of evil. They claim that neither facts about particular horrors nor more holistic considerations pertaining to the magnitude, kinds and distribution of evil can ground atheism since we can't tell whether any evil is gratuitous.1 In this paper we explore a novel strategy for shedding light on these issues: we compare the atheist who claims that there is no morally sufficient reason (...) for certain evils with the physicist who claims that there is no causally sufficient reason for some events. (shrink)

A critique of Plantinga's free will defense. For an updated version of this critique, with a reply to objections from William Rowe and Alvin Plantinga, see my "The logical problem of evil: Plantinga and Mackie," in Justin P. McBrayer & Daniel Howard‐Snyder (eds.), The Blackwell Companion to the Problem of Evil, Wiley-Blackwell, 2013, pp. 19-33.

This is an extended review of Juhani Yli-Vakkuri & John Hawthorne's book: Narrow Content (Oxford: Oxford University Press, 2018)..

This chapter addresses concerns that pragmatic encroachers are committed to problematic knowledge variance. It first replies to Charity Anderson and John Hawthorne’s new putative problem cases, which purport to show that pragmatic encroachment is committed to problematic variations in knowledge depending on what choices are available to the potential knower. It argues that the new cases do not provide any new reasons to be concerned about the pragmatic encroacher’s commitment to knowledge-variance. The chapter further argues that concerns about knowledge-variance (...) are not limited to the pragmatic encroacher, but come up for traditional purist invariantism as well. (shrink)

‘Why does language matter to philosophy?’ is the name of a minor classic by Ian Hacking published in 1975. It’s a good question. Among the many charges laid against academic philosophy, one of the more familiar is that it concerns itself excessively with verbal or ‘merely semantic’ questions, at the expense of the real questions of philosophy. And yet those who have made a serious attempt to engage with philosophical problems quite soon finds themselves grappling with the very words they (...) use to formulate the problems. (shrink)

This is a review of *Knowledge, Belief and God: New Insights in Religious Epistemology* (edited by Matthew Benton, John Hawthorne, and Dani Rabinowitz). The review briefly discusses the contributed essays by Benton and Isaac Choi.

In the first chapter of his Knowledge and Lotteries, John Hawthorne argues that thinkers do not ordinarily know lottery propositions. His arguments depend on claims about the intimate connections between knowledge and assertion, epistemic possibility, practical reasoning, and theoretical reasoning. In this paper, we cast doubt on the proposed connections. We also put forward an alternative picture of belief and reasoning. In particular, we argue that assertion is governed by a Gricean constraint that makes no reference to knowledge, and (...) that practical reasoning has more to do with rational degrees of belief than with states of knowledge. (shrink)

In “Knowledge and epistemic necessity,” John Hawthorne gives a defense of what he rightly calls the “standard approach” to epistemic possibility against what he calls a new “competing idea” presented by Dougherty and Rysiew which he notes has been “endorsed and elaborated upon” by Fantl and McGrath. According to the standard approach, roughly, p is epistemically possible for S if S doesn’t know that not-p. The new approach has it that p is epistemically possible if p has a non-zero (...) epistemic probability. Both approaches, he notes, would explain the oddness of CKAs, utterances of the form “p, but possibly not p.” However, he offers a number of arguments designed to show that the standard approach has other advantages. In this paper, we undermine Hawthorne’s reasons for favoring the standard approach over Dougherty and Rysiew’s alternative approach. (shrink)