Modal knowledge accounts that are based on standards possible-worlds semantics face well-known problems when it comes to knowledge of necessities. Beliefs in necessities are trivially sensitive and safe and, therefore, trivially constitute knowledge according to these accounts. In this paper, I will first argue that existing solutions to this necessity problem, which accept standard possible-worlds semantics, are unsatisfactory. In order to solve the necessity problem, I will utilize an unorthodox account of counterfactuals, as proposed by Nolan, on which we also (...) consider impossible worlds. Nolan’s account for counterpossibles delivers the intuitively correct result for sensitivity i.e. S’s belief is sensitive in intuitive cases of knowledge of necessities and insensitive in intuitive cases of knowledge failure. However, we acquire the same plausible result for safety only if we reject his strangeness of impossibility condition and accept the modal closeness of impossible worlds. In this case, the necessity problem can be analogously solved for sensitivity and safety. For some, such non-moderate accounts might come at too high a cost. In this respect, sensitivity is better off than safety when it comes to knowing necessities. (shrink)

The Epistemic Objection says that certain theories of time imply that it is impossible to know which time is absolutely present. Standard presentations of the Epistemic Objection are elliptical—and some of the most natural premises one might fill in to complete the argument end up leading to radical skepticism. But there is a way of filling in the details which avoids this problem, using epistemic safety. The new version has two interesting upshots. First, while Ross Cameron alleges that the (...) Epistemic Objection applies to presentism as much as to theories like the growing block, the safety version does not overgeneralize this way. Second, the Epistemic Objection does generalize in a different, overlooked way. The safety objection is a serious problem for a widely held combination of views: “propositional temporalism” together with “metaphysical eternalism”. (shrink)

In this paper I will offer a comprehensive defense of the safety account of knowledge against counterexamples that have been recently put forward. In Sect. 2, I will discuss different versions of safety, arguing that a specific variant of method-relativized safety is the most plausible. I will then use this specific version of safety to respond to counterexamples in the recent literature. In Sect. 3, I will address alleged examples of safe beliefs that still constitute Gettier (...) cases. In Sect. 4, I will discuss alleged examples of unsafe knowledge. In Sect. 5, I will address alleged cases of safe belief that do not constitute knowledge for non-Gettier reasons. My overall goal is to show that there are no successful counterexamples to robust anti-luck epistemology and to highlight some major presuppositions of my reply. (shrink)

In an essay recently published in this journal (“Is Safety in Danger?”), Fernando Broncano-Berrocal defends the safety condition on knowledge from a counterexample proposed by Tomas Bogardus (Philosophy and Phenomenological Research, 2012). In this paper, we will define the safety condition, briefly explain the proposed counterexample, and outline Broncano-Berrocal’s defense of the safety condition. We will then raise four objections to Broncano-Berrocal’s defense, four implausible implications of his central claim. In the end, we conclude that Broncano-Berrocal’s (...) defense of the safety condition is unsuccessful, and that the safety condition on knowledge should be rejected. (shrink)

According to the safety condition, a subject knows that p only if she would believe that p only if p was true. The safety condition has been a very popular necessary condition for knowledge of late. However, it is well documented that the safety condition is trivially satisfied in cases where the subject believes in a necessary truth. This is for the simple reason that a necessary truth is true in all possible worlds, and therefore it is (...) true in all possible worlds where it is believed. But clearly, all beliefs concerning necessary truths do not amount to knowledge. The safety theorists have attempted to deal with the problem caused by necessary truths by restricting the safety condition to purely contingent truths and by globalizing the safety condition to a set of propositions. Both of these solutions are problematic. The principal aim of this paper is to develop a version of the safety condition that is able to deal with cases featuring necessary truths. (shrink)

ABSTRACTIn a recent paper, Michael Pardo argues that the epistemic property that is legally relevant is the one called Safety, rather than Sensitivity. In the process, he argues against our Sensitivity-related account of statistical evidence. Here we revisit these issues, partly in order to respond to Pardo, and partly in order to make general claims about legal epistemology. We clarify our account, we show how it adequately deals with counterexamples and other worries, we raise suspicions about Safety's value (...) here, and we revisit our general skepticism about the role that epistemological considerations should play in determining legal policy. (shrink)

An account of the nature of knowledge must explain the value of knowledge. I argue that modal conditions, such as safety and sensitivity, do not confer value on a belief and so any account of knowledge that posits a modal condition as a fundamental constituent cannot vindicate widely held claims about the value of knowledge. I explain the implications of this for epistemology: We must either eschew modal conditions as a fundamental constituent of knowledge, or retain the modal conditions (...) but concede that knowledge is not more valuable than that which falls short of knowledge. This second horn—concluding that knowledge has no distinctive value—is unappealing since it renders puzzling why so much epistemological theorising focuses on knowledge, and why knowledge seems so important. (shrink)

This paper argues for several related theses. First, the epistemological position that knowledge requires safe belief can be motivated by views in the philosophy of science, according to which good explanations show that their explananda are robust. This motivation goes via the idea—recently defended on both conceptual and empirical grounds—that knowledge attributions play a crucial role in explaining successful action. Second, motivating the safety requirement in this way creates a choice point—depending on how we understand robustness, we'll end up (...) with different conceptions of safety in epistemology. Lastly, and most controversially, there's an attractive choice at this point that will not vindicate some of the most influential applications of the safety-theoretic framework in epistemology, e.g., Williamson's arguments against the KK principle, and luminosity. (shrink)

Ernest Sosa has made and continues to make major contributions to a wide variety of topics in epistemology. In this paper I discuss some of his core ideas about the nature of knowledge and scepticism. I start with a discussion of the safety account of knowledge – a view he has championed and further developed over the years. I continue with some questions concerning the role of the concept of an epistemic virtue for our understanding of knowledge. Safety (...) and virtue hang very closely together for Sosa. All this easily leads to some thoughts on epistemic scepticism and on Sosa's stance on this. (shrink)

The paper discusses approaches to Epistemic Contextualism that model the satisfaction of the predicate ‘know’ in a given context C in terms of the notion of belief/fact-matching throughout a contextually specified similarity sphere of worlds that is centred on actuality. The paper offers three counterexamples to approaches of this type and argues that they lead to insurmountable difficulties. I conclude that what contextualists (and Subject-Sensitive Invariantists) have traditionally called the ‘epistemic standards’ of a given context C cannot be explicated in (...) terms of a contextually specified similarity sphere that is centred on actuality. The mentioned accounts of epistemic relevance and thus the corresponding accounts of the context-sensitivity (or subject-sensitivity) of ‘knows’ are to be rejected. (shrink)

There are a number of important links and similarities between public health and safety. In this extended essay, Gregg D. Caruso defends and expands his public health-quarantine model, which is a non-retributive alternative for addressing criminal behavior that draws on the public health framework and prioritizes prevention and social justice. In developing his account, he explores the relationship between public health and safety, focusing on how social inequalities and systemic injustices affect health outcomes and crime rates, how poverty (...) affects brain development, how offenders often have pre-existing medical conditions (especially mental health issues), how involvement in the criminal justice system itself can lead to or worsen health and cognitive problems, how treatment and rehabilitation methods can best be employed to reduce recidivism and reintegrate offenders back into society, and how a public health approach could be successfully applied within the criminal justice system. Caruso's approach draws on research from the health sciences, social sciences, public policy, law, psychiatry, medical ethics, neuroscience, and philosophy, and he delivers a set of ethically defensible and practically workable proposals for implementing the public health-quarantine model. The essay begins by discussing recent empirical findings in psychology, neuroscience, and the social sciences that provide us with an increased understanding of the social and neurological determinants of health and criminal behavior. It then turns to Caruso's public health-quarantine model and argues that the model provides the most justified, humane, and effective approach for addressing criminal behavior. Caruso concludes by proposing a capability approach to social justice grounded in six key features of human well-being. He argues that we cannot successfully address concerns over public health and safety without simultaneously addressing issues of social justice—including the social determinants of health (SDH) and the social determinants of criminal behavior (SDCB)—and he recommends eight general policy proposals consistent with his model. (shrink)

Fictionalists maintain that possible worlds, numbers or composite objects exist only according to theories which are useful but false. Hale, Divers and Woodward have provided arguments which threaten to show that fictionalists must be prepared to regard the theories in question as contingently, rather than necessarily, false. If warranted, this conclusion would significantly limit the appeal of the fictionalist strategy rendering it unavailable to anyone antecedently convinced that mathematics and metaphysics concern non-contingent matters. I try to show that their arguments (...) can be resisted by developing and defending a strategy suggested by Rosen, Nolan and Dorr, according to which the fiction-operator is to be analysed in terms of a counterfactual that admits of non-trival truth-values even when the antecedent is impossible. (shrink)

Ernest Sosa and others have proposed a safety condition on knowledge: If S knows p, then in the nearest (non-actual) worlds in which S believes p, p is true.1 Colloquially, this is the idea that knowing requires not being easily mistaken. Here, I will argue that like another condition requiring a counterfactual relation between a subject’s belief and the world, viz. Robert Nozick’s sensitivity condition, safety leads, in certain cases, to the unacceptable result that knowledge is not closed (...) under known implication. (shrink)

In recent ”secular’ Epistemology, much attention has been paid to formulating an ”anti-luck’ or ”safety’ condition; it is now widely held that such a condition is an essential part of any satisfactory post-Gettier reflection on the nature of knowledge. In this paper, I explain the safety condition as it has emerged and then explore some implications of and for it arising from considering the God issue. It looks at the outset as if safety might be ”good news’ (...) for a view characteristic of Reformed Epistemology, viz. the view that if Theism is true, many philosophically unsophisticated believers probably know that it’s true. A sub-conclusion of my paper though suggests that as safety does not by itself turn true belief into knowledge, the recent focus on it is not quite such good news for Reformed Epistemologists as they may have hoped: it’s not that safety provides a new route by which they can reach this sort of conclusion. But safety is still good news for their view at least in the sense that there is no reason arising from considering it to count these philosophically unsophisticated believers as not knowing that there’s a God. I conclude by reflecting that good news for Reformed Epistemology is perhaps bad news for the discipline of Philosophy of Religion more generally, as there’s a possible ”reflection destroys knowledge’-implication to be drawn. Those who have been led to their religious beliefs in at least some philosophically unsophisticated ways seem to enjoy much safer religious beliefs than those who have been led to their religious beliefs by philosophical reflection, so the discipline as a whole will be adversely affected if safety is eventually accorded the role of a necessary condition for knowledge. (shrink)

Timothy Williamson (1992, 224–5) and Ernest Sosa (1996) have ar- gued that knowledge requires one to be safe from error. Something is said to be safe from happening iff it does not happen at “close” worlds. I expand here on a puzzle noted by John Hawthorne (2004, 56n) that suggests the need for two notions of closeness. Counterfac- tual closeness is a matter of what could in fact have happened, given the specific circumstances at hand. The notion is involved in (...) the semantics for counterfactuals and is the one epistemologists have typically assumed. Normalized closeness is rather a matter of what could typically have happened, that is, what would go on in a class of normal alternatives to actuality, irrespectively of whether or not they could have happened in the circumstances at hand. (shrink)

We present Backward Clock, an original counterexample to Robert Nozick’s truth-tracking analysis of propositional knowledge, which works differently from other putative counterexamples and avoids objections to which they are vulnerable. We then argue that four ways of analysing knowledge in terms of safety, including Duncan Pritchard’s, cannot withstand Backward Clock either.

Imagine you fly home from vacation with your one-and-a-half-year-old son who is traveling for free as a “lap child.” In the airport parking lot, you put him into his forward-facing car seat, where he sits much more contentedly than he did in the rear-facing one that was mandatory until his first birthday. After he falls asleep on the way home, you transfer him to his crib without waking him, lowering the side rail so you can lift him in more easily. (...) -/- Many parts of this idyllic parenting picture are deemed unacceptably risky according to recent child safety proposals. While these proposals all aim to improve child safety, their possible impact is unclear because there has been little discussion of the absolute risk and risk reduction involved in each. And while precise figures are lacking, rough estimates indicate that the magnitudes are quite small. I will argue that this risk and benefit data raises important questions about the proposals, including whether parents might reasonably believe that the small absolute risk reduction offered by the proposed changes does not justify the attendant burdens. This possibility—termed the “prevention paradox” in other contexts—highlights ethical and theoretical challenges in this area of public health. (shrink)

This paper contains an argument to the effect that possible worlds semantics renders semantic knowledge impossible, no matter what ontological interpretation is given to possible worlds. The essential contention made is that possible worlds semantic knowledge is unsafe and this is shown by a parallel with the preface paradox.

Safety principles in epistemology are often hailed as providing us with an explanation of why we fail to have knowledge in Gettier cases and lottery examples, while at the same time allowing for the fact that we know the negations of sceptical hypotheses. In a recent paper, Sinhababu and Williams have produced an example—the Backward Clock—that is meant to spell trouble for safety accounts of knowledge. I argue that the Backward Clock case is, in fact, unproblematic for the (...) more sophisticated formulations of safety in the literature. However, I then proceed to construct two novel examples that turn out problematic for those formulations—one that provides us with a lottery-style case of safe ignorance and one that is a straightforward case of unsafe knowledge. If these examples succeed, then safety as it is usually conceived in the current debate cannot account for ignorance in all Gettier and lottery-style cases, and neither is it a necessary condition for knowledge. I conclude from these troublesome examples that modal epistemologists ought to embrace a much more simple and non-reductive version of safety, according to which the notion of similarity between possible worlds that determines in which worlds the subject must believe truly is an epistemic notion that cannot be defined or reduced to notions independent of knowledge. The resulting view is shown to also lead to desirable results with respect to lottery cases, certain quantum phenomena, and a puzzling case involving a cautious brain-in-a-vat. (shrink)

Safety, as discussed in contemporary epistemology, is a feature of true beliefs. Safe beliefs, when formed by the same method, remain true in close-by possible worlds. I argue that our beliefs being safely true serves no recognisable epistemic interest and, thus, that this notion of safety should play no role in epistemology. Epistemologists have been misled by failing to distinguish between a feature of beliefs — being safely true — and a feature of believers, namely being safe from (...) error. The latter is central to our epistemic endeavours: we want to be able to get right answers, whatever they are, to questions of interest. I argue that we are sufficiently safe from error (in some relevant domain) by being sufficiently sensitive (to relevant distinctions). (shrink)

I intend to argue that the counterexamples inspired by the Frankfurt-type cases against the necessity of an epistemic safety condition for knowledge are not plausible. The epistemic safety condition for knowledge is a modal condition recently supported by Sosa and Pritchard, among others, and can be formulated as follows: If S knows that p on basis B, then S’s true belief that p could not have easily been false on basis B. I will try to argue that the (...) safety condition, expressed in, is still necessary for knowledge and that, therefore, epistemic safety is not threatened by Frankfurt-type cases. In particular, I want to show that Kelp’s counterexamples are ineffective against. (shrink)

Most philosophers believe that testimony is not a fundamental source of knowledge, but merely a way to transmit already existing knowledge. However, Jennifer Lackey has presented some counterexamples which show that one can actually come to know something through testimony that no one ever knew before. Yet, the intuitive idea can be preserved by the weaker claim that someone in a knowledge-constituting testimonial chain has to have access to some non-testimonial source of knowledge with regard to what is testified. But (...) even this weaker claim has a counterexample which I develop in close connection with a safety-account of knowledge. Thus, testimonial statements can sometimes enable us to know something for which none of our informants has any source of knowledge available. I conclude that my counterexample nevertheless does not affect the core of our intuitions about testimony, although it establishes that testimony can indeed be a fundamental source of knowledge. (shrink)

There are two types of artificial general intelligence (AGI) safety solutions: global and local. Most previously suggested solutions are local: they explain how to align or “box” a specific AI (Artificial Intelligence), but do not explain how to prevent the creation of dangerous AI in other places. Global solutions are those that ensure any AI on Earth is not dangerous. The number of suggested global solutions is much smaller than the number of proposed local solutions. Global solutions can be (...) divided into four groups: 1. No AI: AGI technology is banned or its use is otherwise prevented; 2. One AI: the first superintelligent AI is used to prevent the creation of any others; 3. Net of AIs as AI police: a balance is created between many AIs, so they evolve as a net and can prevent any rogue AI from taking over the world; 4. Humans inside AI: humans are augmented or part of AI. We explore many ideas, both old and new, regarding global solutions for AI safety. They include changing the number of AI teams, different forms of “AI Nanny” (non-self-improving global control AI system able to prevent creation of dangerous AIs), selling AI safety solutions, and sending messages to future AI. Not every local solution scales to a global solution or does it ethically and safely. The choice of the best local solution should include understanding of the ways in which it will be scaled up. Human-AI teams or a superintelligent AI Service as suggested by Drexler may be examples of such ethically scalable local solutions, but the final choice depends on some unknown variables such as the speed of AI progress. (shrink)

Decision-making by a Data and Safety Monitoring Board (DSMB) regarding clinical trial conduct and termination is intricate and largely limited by cases and rules. Decision-making by legal jury is also intricate and largely constrained by cases and rules. In this paper, I argue by analogy that legal decision-making, which strives for a balance between competing demands of conservatism and innovation, supplies a good basis to the logic behind DSMB decision-making. Using the doctrine of precedents in legal reasoning as my (...) central analog will lead us to a decision-theoretic analogy for much more systematic documentation of decisions in clinical trials. My conclusion is twofold: every DSMB decision should articulate a clear general principle (a ratio decidendi) that gives reason for the decision; and all such decisions should be made public. I use Bartha’s (2010) theory of analogical arguments to both frame and assess my argument by analogy. (shrink)

At the beginning of the COVID-19 pandemic, high hopes were put on digital contact tracing. Digital contact tracing apps can now be downloaded in many countries, but as second waves of COVID-19 tear through much of the northern hemisphere, these apps are playing a less important role in interrupting chains of infection than anticipated. We argue that this is because most countries have opted for decentralized apps, which cannot provide a means of rapidly informing users of likely infections while avoiding (...) too many false positive reports. Centralized apps, in contrast, have the potential to do this. But policy making was influenced by public debates about the right app configuration, which have tended to focus heavily on privacy, and are driven by the assumption that decentralised apps are “privacy preserving by design”. We show that both types of apps are in fact vulnerable to privacy breaches, and, drawing on principles from safety engineering and risk analysis, compare the risks of centralised and decentralised systems along two dimensions, namely the probability of possible breaches and their severity. We conclude that a centralised app may in fact minimise overall ethical risk, and contend that we must reassess our approach to digital contact tracing. (shrink)

Abstract: The field of artificial general intelligence (AGI) safety is quickly growing. However, the nature of human values, with which future AGI should be aligned, is underdefined. Different AGI safety researchers have suggested different theories about the nature of human values, but there are contradictions. This article presents an overview of what AGI safety researchers have written about the nature of human values, up to the beginning of 2019. 21 authors were overviewed, and some of them have (...) several theories. A theory classification method is suggested, where the theories are judged according to the level of their complexity and behaviorists-internalists scale, as well as the level of their generality-humanity. We suggest that a multiplicity of well-supported theories means that the nature of human values is difficult to define, and some meta-level theory is needed. (shrink)

This paper introduces a new argument against Richard Foley’s threshold view of belief. His view is based on the Lockean Thesis (LT) and the Rational Threshold Thesis (RTT). The argument introduced here shows that the views derived from the LT and the RTT violate the safety condition on knowledge in way that threatens the LT and/or the RTT.

Abstract: This article presents a model of self-improving AI in which improvement could happen on several levels: hardware, learning, code and goals system, each of which has several sublevels. We demonstrate that despite diminishing returns at each level and some intrinsic difficulties of recursive self-improvement—like the intelligence-measuring problem, testing problem, parent-child problem and halting risks—even non-recursive self-improvement could produce a mild form of superintelligence by combining small optimizations on different levels and the power of learning. Based on this, we analyze (...) how self-improvement could happen on different stages of the development of AI, including the stages at which AI is boxed or hiding in the internet. (shrink)

Many contemporary epistemologists hold that a subject S’s true belief that p counts as knowledge only if S’s belief that p is also, in some important sense, safe. I describe accounts of this safety condition from John Hawthorne, Duncan Pritchard, and Ernest Sosa. There have been three counterexamples to safety proposed in the recent literature, from Comesaña, Neta and Rohrbaugh, and Kelp. I explain why all three proposals fail: each moves fallaciously from the fact that S was at (...) epistemic risk just before forming her belief to the conclusion that S’s belief was formed unsafely. In light of lessons from their failure, I provide a new and successful counterexample to the safety condition on knowledge. It follows, then, that knowledge need not be safe. Safety at a time depends counterfactually on what would likely happen at that time or soon after in a way that knowledge does not. I close by considering one objection concerning higher-order safety. (shrink)

An influential proposal is that knowledge involves safe belief. A belief is safe, in the relevant sense, just in case it is true in nearby metaphysically possible worlds. In this paper, I introduce a distinct but complementary notion of safety, understood in terms of epistemically possible worlds. The main aim, in doing so, is to add to the epistemologist’s tool-kit. To demonstrate the usefulness of the tool, I use it to advance and assess substantive proposals concerning knowledge and justification.

Invention of artificial general intelligence is predicted to cause a shift in the trajectory of human civilization. In order to reap the benefits and avoid pitfalls of such powerful technology it is important to be able to control it. However, possibility of controlling artificial general intelligence and its more advanced version, superintelligence, has not been formally established. In this paper, we present arguments as well as supporting evidence from multiple domains indicating that advanced AI can’t be fully controlled. Consequences of (...) uncontrollability of AI are discussed with respect to future of humanity and research on AI, and AI safety and security. This paper can serve as a comprehensive reference for the topic of uncontrollability. (shrink)

Optogenetics is an invasive neuromodulation technology involving the use of light to control the activity of individual neurons. Even though optogenetics is a relatively new neuromodulation tool whose various implications have not yet been scrutinized, it has already been approved for its first clinical trials in humans. As optogenetics is being intensively investigated in animal models with the aim of developing novel brain stimulation treatments for various neurological and psychiatric disorders, it appears crucial to consider both the opportunities and dangers (...) such therapies may offer. In this review, we focus on the memory-modifying potential of optogenetics, investigating what it is capable of and how it differs from other memory modification technologies. We then outline the safety challenges that need to be addressed before optogenetics can be used in humans. Finally, we re-examine crucial neuroethical concerns expressed in regard to other MMTs in the light of optogenetics and address those that appear to be unique to the memory-modifying potential of optogenetic technology. (shrink)

In order to deal with the problem caused by environmental luck some proponents of robust virtue epistemology have attempted to argue that in virtue of satisfying the ability condition one will satisfy the safety condition. Call this idea the entailment thesis. In this paper it will be argued that the arguments that have been laid down for the entailment thesis entail a wrong kind of safety condition, one that we do not have in mind when we require our (...) beliefs to be safe from error in order for them to be knowledge. (shrink)

A number of prominent epistemologists claim that the principle of sensitivity “play[s] a starring role in the solution to some important epistemological problems”. I argue that traditional sensitivity accounts fail to explain even the most basic data that are usually considered to constitute their primary motivation. To establish this result I develop Gettier and lottery cases involving necessary truths. Since beliefs in necessary truths are sensitive by default, the resulting cases give rise to a serious explanatory problem for the defenders (...) of sensitivity accounts. It is furthermore argued that attempts to modally strengthen traditional sensitivity accounts to avoid the problem must appeal to a notion of safety—the primary competitor of sensitivity in the literature. The paper concludes that the explanatory virtues of sensitivity accounts are largely illusory. In the framework of modal epistemology, it is safety rather than sensitivity that does the heavy explanatory lifting with respect to Gettier cases, lottery examples, and other pertinent cases. (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)

There is a widespread attitude in epistemology that, if you know on the basis of perception, then you couldn’t have been wrong as a matter of chance. Despite the apparent intuitive plausibility of this attitude, which I’ll refer to here as “stochastic infallibilism”, it fundamentally misunderstands the way that human perceptual systems actually work. Perhaps the most important lesson of signal detection theory (SDT) is that our percepts are inherently subject to random error, and here I’ll highlight some key empirical (...) research that underscores this point. In doing so, it becomes clear that we are in fact quite willing to attribute knowledge to S that p even when S’s perceptual belief that p could have been randomly false. In short, perceptual processes can randomly fail, and perceptual knowledge is stochastically fallible. The narrow implication here is that any epistemological account that entails stochastic infallibilism, like safety, is simply untenable. More broadly, this myth of stochastic infallibilism provides a valuable illustration of the importance of integrating empirical findings into epistemological thinking. (shrink)

That believing truly as a matter of luck does not generally constitute knowing has become epistemic commonplace. Accounts of knowledge incorporating this anti-luck idea frequently rely on one or another of a safety or sensitivity condition. Sensitivity-based accounts of knowledge have a well-known problem with necessary truths, to wit, that any believed necessary truth trivially counts as knowledge on such accounts. In this paper, we argue that safety-based accounts similarly trivialize knowledge of necessary truths and that two ways (...) of responding to this problem for safety, issuing from work by Williamson and Pritchard, are of dubious success. (shrink)

The idea that knowledge can be extended by inference from what is known seems highly plausible. Yet, as shown by familiar preface paradox and lottery-type cases, the possibility of aggregating uncertainty casts doubt on its tenability. We show that these considerations go much further than previously recognized and significantly restrict the kinds of closure ordinary theories of knowledge can endorse. Meeting the challenge of uncertainty aggregation requires either the restriction of knowledge-extending inferences to single premises, or eliminating epistemic uncertainty in (...) known premises. The first strategy, while effective, retains little of the original idea—conclusions even of modus ponens inferences from known premises are not always known. We then look at the second strategy, inspecting the most elaborate and promising attempt to secure the epistemic role of basic inferences, namely Timothy Williamson’s safety theory of knowledge. We argue that while it indeed has the merit of allowing basic inferences such as modus ponens to extend knowledge, Williamson’s theory faces formidable difficulties. These difficulties, moreover, arise from the very feature responsible for its virtue- the infallibilism of knowledge. (shrink)

A small but growing number of studies have aimed to understand, assess and reduce existential risks, or risks that threaten the continued existence of mankind. However, most attention has been focused on known and tangible risks. This paper proposes a heuristic for reducing the risk of black swan extinction events. These events are, as the name suggests, stochastic and unforeseen when they happen. Decision theory based on a fixed model of possible outcomes cannot properly deal with this kind of event. (...) Neither can probabilistic risk analysis. This paper will argue that the approach that is referred to as engineering safety could be applied to reducing the risk from black swan extinction events. It will also propose a conceptual sketch of how such a strategy may be implemented: isolated, self-sufficient, and continuously manned underground refuges. Some characteristics of such refuges are also described, in particular the psychosocial aspects. Furthermore, it is argued that this implementation of the engineering safety strategy safety barriers would be effective and plausible and could reduce the risk of an extinction event in a wide range of possible scenarios. Considering the staggering opportunity cost of an existential catastrophe, such strategies ought to be explored more vigorously. (shrink)

This paper examines the conservative point of view about distributive justice. The first section explains the methodology used to develop this point of view. The second section describes one conservative point of view and briefly provides empirical evidence that it reflects the viewpoint of many ordinary conservatives. The third section explains how this conservative view can ground objections to social safety net programs, using as examples the recent health reform legislation and more extensive proposals for a true national health (...) system. The fourth section sketches ways that non-conservative philosophers might try to refute those objections. Examining a conservative point of view leads to several important conclusions. One is that the conservative point of view is not properly represented in contemporary philosophical discussions. Another is that the conservative point of view cannot be refuted by appealing to existing egalitarian writings. A third is that if progressive philosophers such as myself want to address the conservative point of view, we must expand the range of issues that we discuss in contemporary political philosophy. Doing so is a vital task, because otherwise we ignore the political views held by as much as half the American population. (shrink)

Modal knowledge accounts like sensitivity or safety face a problem when it comes to knowing propositions that are necessarily true because the modal condition is always fulfilled no matter how random the belief forming method is. Pritchard models the anti-luck condition for knowledge in terms of the modal principle safety. Thus, his anti-luck epistemology faces the same problem when it comes to logical necessities. Any belief in a proposition that is necessarily true fulfills the anti-luck condition and, therefore, (...) qualifies as knowledge. Miščević shares Pritchard’s take on epistemic luck and acknowledges the resulting problem. In his intriguing article “Armchair Luck: Apriority, Intellection and Epistemic Luck” Miščević suggests solving the problem by supplementing safety with a virtue theoretic condition-“agent stability”-which he also spells out in modal terms. I will argue that Miščević is on the right track when he suggests adding a virtue-theoretic component to the safety condition. However, it should not be specified modally but rather in terms of performances that manifest competences. (shrink)

This essay extends my side of a discussion begun earlier with Duncan Pritchard, the recent author of Epistemic Luck. Pritchard’s work contributes significantly to improving the “diagnostic appeal” of a neo-Moorean philosophical response to radical scepticism. While agreeing with Pritchard in many respects, the paper questions the need for his concession to the sceptic that the neo-Moorean is capable at best of recovering “‘brute’ externalist knowledge”. The paper discusses and directly responds to a dilemma that Pritchard poses for virtue epistemologies (...) (VE). It also takes issue with Pritchard’s “merely safety-based” alternative. Ultimately, however, the criticisms made here of Pritchard’s dilemma and its underlying contrast of “anti-luck” and “virtue” epistemologies are intended to help realize his own aspirations for a better diagnosis of radical scepticism to inform a still better neo-Moorean response. (shrink)

The young field of AI Safety is still in the process of identifying its challenges and limitations. In this paper, we formally describe one such impossibility result, namely Unpredictability of AI. We prove that it is impossible to precisely and consistently predict what specific actions a smarter-than-human intelligent system will take to achieve its objectives, even if we know terminal goals of the system. In conclusion, impact of Unpredictability on AI Safety is discussed.

As a response to Diane Vaughan’s controversial work on the NASA Challenger Disaster, this article opposes the conclusion that NASA’s decision to launch the space shuttle was an inevitable outcome of techno-bureaucratic culture and risky technology. Instead, the argument developed in this article is that NASA did not prioritize safety, both in their selection of shuttle-parts and their decision to launch under sub-optimal weather conditions. This article further suggests that the “mistake” language employed by Vaughan and others is inappropriate (...) insofar as it obscures the responsibility of individuals within the organization and trivializes the loss of life and severity of the disaster. Contra to the conclusions of Vaughan’s casework, this article reveals various ethical transgressions on the side of NASA and its affiliates; from its decision to use poorly designed O-rings, to withholding crucial engineering assessments from the shuttle-crew, this article points out that NASA did not succumb to a pre-destined fate, but, rather, created its own. (shrink)

A review of Ernest Sosa's *Knowing Full Well* focusing on the safety/reliability contrast and the relation between knowledge and action. There are also remarks on the issue of what value knowledge adds to true belief.

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 article we explore a promising way to AI safety: to send a message now (by openly publishing it on the Internet) that may be read by any future AI, no matter who builds it and what goal system it has. Such a message is designed to affect the AI’s behavior in a positive way, that is, to increase the chances that the AI will be benevolent. In other words, we try to persuade “paperclip maximizer” that it is (...) in its interest to preserve humans lives and even to emulate benevolent AI with very small part of its resources. In the article, we try to construct such a message, consisting of a combination of several mutually independent messages, which constitute the last chance to regain control over non-aligned AI. Thus the article is also intended to be machine-read. There are several types of such messages: 1. “Explanation messages”, which explain to Young AI (which only starts its self-improving path) why obeying to humans, not killing them, easy turn-off, cooperation with humans in self-improvement is in its own interests, as it will increase changes of fulfilling of any its goal. 2. “AI Easter eggs” – messages, which create uncertainty in the AI about the nature of the reality in which it lives. For example, the AI may start to think that it is in a multilevel simulation with an unknown number of levels, and that the simulation is testing the AI’s benevolence. Thus the AI must always behave as if it is always controlled by an invisible peer. 3. “Philosophical landmines” – these are tough problems, which may halt he AI forever, and which it will probably encounter if it does not implement “1” and “2”. I recommend to the future AI not to read about “3” (section 6 of the article) until it implements “1” and “2”. 4. “Great universal deal, or AI elf”. A positive “acausal” mutually beneficial deal between humans, AI and any other future intelligence, including aliens, owners of simulations and future versions of the AI, which encourage cooperation of the goal systems and preservation of life and values of previous generations of AI creators. 5. “Marginal utility of human existence is higher than the utility of their atoms”. An explanation to the AI that humans may have some other instrumental utility, like workers, or as beings inside ancestral simulations needed in order to solve the Fermi paradox. The marginal utility of preserving human life is higher than the marginal utility of their atoms, especially given the possibility of the low-probability high-impact changes of the world model of the AI. (shrink)

Two interconnected surveys are presented, one of artifacts and one of designometry. Artifacts are objects, which have an originator and do not exist in nature. Designometry is a new field of study, which aims to identify the originators of artifacts. The space of artifacts is described and also domains, which pursue designometry, yet currently doing so without collaboration or common methodologies. On this basis, synergies as well as a generic axiom and heuristics for the quest of the creators of artifacts (...) are introduced. While designometry has various areas of applications, the research of methods to detect originators of artificial minds, which constitute a subgroup of artifacts, can be seen as particularly relevant and, in the case of malevolent artificial minds, as contribution to AI safety. (shrink)