Can humans get arbitrarily capable reinforcement learning agents to do their bidding? Or will sufficiently capable RL agents always find ways to bypass their intended objectives by shortcutting their reward signal? This question impacts how far RL can be scaled, and whether alternative paradigms must be developed in order to build safe artificial general intelligence. In this paper, we study when an RL agent has an instrumental goal to tamper with its reward process, and describe design principles that prevent instrumental (...) goals for two different types of reward tampering. Combined, the design principles can prevent reward tampering from being an instrumental goal. The analysis benefits from causal influence diagrams to provide intuitive yet precise formalizations. (shrink)

Causal reasoning and game-theoretic reasoning are fundamental topics in artificial intelligence, among many other disciplines: this paper is concerned with their intersection. Despite their importance, a formal framework that supports both these forms of reasoning has, until now, been lacking. We offer a solution in the form of (structural) causal games, which can be seen as extending Pearl's causal hierarchy to the game-theoretic domain, or as extending Koller and Milch's multi-agent influence diagrams to the causal domain. We then consider three (...) key questions: i) How can the (causal) dependencies in games - either between variables, or between strategies - be modelled in a uniform, principled manner? ii) How may causal queries be computed in causal games, and what assumptions does this require? iii) How do causal games compare to existing formalisms? To address question i), we introduce mechanised games, which encode dependencies between agents' decision rules and the distributions governing the game. In response to question ii), we present definitions of predictions, interventions, and counterfactuals, and discuss the assumptions required for each. Regarding question iii), we describe correspondences between causal games and other formalisms, and explain how causal games can be used to answer queries that other causal or game-theoretic models do not support. Finally, we highlight possible applications of causal games, aided by an extensive open-source Python library. (shrink)

In this paper, we provide a detailed review of previous classifications of 2 × 2 games and suggest a mathematically simple way to classify the symmetric 2 × 2 games based on a decomposition of the payoff matrix into a cooperative and a zero-sum part. We argue that differences in the interaction between the parts is what makes games interesting in different ways. Our claim is supported by evolutionary computer experiments and findings in previous literature. In addition, we provide a (...) method for using a stereographic projection to create a compact 2-d representation of the game space. (shrink)

Causal models of agents have been used to analyse the safety aspects of machine learning systems. But identifying agents is non-trivial -- often the causal model is just assumed by the modeler without much justification -- and modelling failures can lead to mistakes in the safety analysis. This paper proposes the first formal causal definition of agents -- roughly that agents are systems that would adapt their policy if their actions influenced the world in a different way. From this we (...) derive the first causal discovery algorithm for discovering agents from empirical data, and give algorithms for translating between causal models and game-theoretic influence diagrams. We demonstrate our approach by resolving some previous confusions caused by incorrect causal modelling of agents. (shrink)

Abstract: I examine virtue theory, especially as expressed by Rosalind Hursthouse. In its canonical form, the theory claims that living a life of virtue constitutes flourishing, although it also has a possible fall-back claim that a life of virtue is a means to the end of flourishing. I argue that in both interpretations, virtue theory is mistaken. It cannot give any convincing account of how the concepts of wanting, flourishing, and the virtues are connected, nor can it deal adequately with (...) the counter-examples of flourishing by the wicked, and torment for the virtuous. However, I allow that stripped of all its pretensions to universality, there are grounds for some people in some restricted sets circumstances, to follow the path of virtue solely because they will thereby flourish. (shrink)

Abstract I argue that the traditional problem of evil mislocates the problem which confronts the theist. The real problem arises not from the evil in the world, but from the non-perfection of the world. Given that a perfect God could create only a perfect world, and given that the world is not in fact perfect, I construct an argument for atheism. I show that the argument is not open to the objections which theists standardly bring against the traditional objection from (...) evil. (shrink)

Around the world, the population is ageing in ways that pose new challenges for healthcare providers. To date these have mostly been formulated in terms of challenges created by increasing costs, and the focus has been squarely on life-prolonging treatments. However, this focus ignores the ways in which many older people require life-enhancing treatments to counteract the effects of physical and mental decline. This paper argues that in doing so it misses important aspects of what justice requires when it comes (...) to older people. (shrink)

Is it possible to prove or disprove God's existence? Arguments for the existence of God have taken many different forms over the centuries: in The Non-Existence of God, Nicholas Everitt considers all of the arguments and examines the role that reason and knowledge play in the debate over God's existence. He draws on recent scientific disputes over neo-Darwinism, the implication of 'big bang' cosmology, and the temporal and spatial size of the universe; and discusses some of the most recent (...) work on the subject, leading to a controversial conclusion. (shrink)

Since 1989, widely circulating statistics on gay teen suicide in the United States have acted as catalysts for institutional reforms, scientific research, and the creation of an identity category “gay youth.” While one figure has been replicated scientifically, these numbers originated not from a scientific research study but as risk estimates developed by a social worker and published in a government document. Many people within the public took up these original numbers, attributing their author the status of scientific researcher. In (...) effect, the numbers became “black boxed,” often traveling without citation. Drawing on Ian Hacking’s “dynamic nominalist” perspective, this article utilizes interviews with the author of these statistics and other key claimants, along with textual analysis, to trace the origins, uptake, and effects of these figures. While making vital policy contributions, the numbers have led to some ironic consequences including the fostering of gay youth identification with suicide as a potential correlate of their identity and the potential antigay redeployment of decontextualized numbers. They have also led to a reaction in the form of “resilience” narratives. (shrink)

How wrong is it to deceive someone into sex by lying, say, about one's profession? The answer is seriously wrong when the liar's actual profession would be a deal breaker for the victim of the deception: this deception vitiates the victim's sexual consent, and it is seriously wrong to have sex with someone while lacking his or her consent.

We apply an online optimization process based on machine learning to the production of Bose-Einstein condensates. BEC is typically created with an exponential evaporation ramp that is optimal for ergodic dynamics with two-body s-wave interactions and no other loss rates, but likely sub-optimal for real experiments. Through repeated machine-controlled scientific experimentation and observations our ’learner’ discovers an optimal evaporation ramp for BEC production. In contrast to previous work, our learner uses a Gaussian process to develop a statistical model of the (...) relationship between the parameters it controls and the quality of the BEC produced. We demonstrate that the Gaussian process machine learner is able to discover a ramp that produces high quality BECs in 10 times fewer iterations than a previously used online optimization technique. Furthermore, we show the internal model developed can be used to determine which parameters are essential in BEC creation and which are unimportant, providing insight into the optimization process of the system. (shrink)

Scientists are seekers of truth; but where science breaks into the everyday world should they be held accountable for the outcome of their actions? The contributors to this volume believe that scientists are more than mere cogs in a machine - science, technology and politics are inseparable. Part 1 describes current scientific practice from three personal perspectives; part 2 looks at the ways in which science, society and the environment could interact given the chance; and part 3 examines the more (...) general aspects of Western science and their relevance to our relationship with the environment, society and all global issues. (shrink)

Healthcare ethics has to date had very little to say about the treatment of chronic illness. That is problematic. Chronic illness differs from other illnesses in that: 1. in most cases it cannot be cured; 2. patients can live with it for many years; and 3. its day to day management is typically carried out, not by healthcare professionals, but by the patient and/or members of their family. These features problematise key distinctions that underlie much existing work in healthcare ethics (...) — including those between beneficence and autonomy, between treatment and prevention, and between the recipient and provider of treatment. They also raise ethical questions that have not to date been addressed in any detail — including what healthcare professionals should do when patients are managing their illness poorly, or when changes occur that affect their patients. For these reasons chronic illness provides a lens through which to challenge the adequacy of existing approaches to healthcare ethics, exposing the gaps and false assumptions built into them. In doing so it also provides the materials to build a more philosophically adequate account of healthcare professionals’ moral obligations than exists to date. Providing such an account is the aim of this monograph. (shrink)

Alain Badiou offers a distinctive interpretation of time, anchored in his unique metaphysics and philosophy of the event. For Badiou, time is interventional, demarcating a disjunction between two incommensurable sequences of rupture in human history. The concept of disjunction occupies a pivotal role in Badiou’s philosophy of both love and time. This paper provides a comprehensive examination of the temporality of love within the context of Badiou's philosophy. We argue that, when viewed through Badiou’s philosophical lens, love provides an invaluable (...) vantage point for understanding the very creation and transformation of time. As this chapter progresses, a concise exposition of Badiou's metaphysics and his theory of truth procedure is furnished, emphasizing facets that resonate with his philosophy of love. Time is explored from three vantage points: time between worlds, time within a world, and time as difference-as-such. These distinct conceptions of time, when examined in light of love, provide a fertile ground for further developing his concept of time. (shrink)

Our successful engagement with the world is plausibly underwritten by our sensitivity to affordances in our immediate environment. The considerable literature on affordances focuses almost exclusively on affordances for bodily actions such as gripping, walking or eating. I propose that we are also sensitive to affordances for mental actions such as attending, imagining and counting. My case for this ‘Mental Affordance Hypothesis’ is motivated by a series of examples in which our sensitivity to mental affordances mirrors our sensitivity to bodily (...) affordances. Specifically, subjects perceive opportunities to perform a mental action and their doing so leads, under the right conditions, to the automatic preparation of that action. I conclude by sketching a mental affordance research program that would reinforce my case for the Mental Affordance Hypothesis and establish its ramifications for a number of debates across philosophy and psychology. (shrink)

Nearly everyone prefers pain to be in the past rather than the future. This seems like a rationally permissible preference. But I argue that appearances are misleading, and that future-biased preferences are in fact irrational. My argument appeals to trade-offs between hedonic experiences and other goods. I argue that we are rationally required to adopt an exchange rate between a hedonic experience and another type of good that stays fixed, regardless of whether the hedonic experience is in the past or (...) future. (shrink)

The scope of someone's consent is the range of actions that they permit by giving consent. The Scope of Consent investigates the under-explored question of which normative principle governs the scope of consent. To answer this question, the book's investigation involves taking a stance on what constitutes consent. By appealing to the idea that someone can justify their behaviour by appealing to another person's consent, Dougherty defends the view that consent consists in behaviour that expresses a consent-giver's will for how (...) a consent-receiver behaves. The ultimate conclusion of the book is that the scope of consent is determined by certain evidence that bears on the appropriate interpretation of the consent. (shrink)

You are morally permitted to save your friend at the expense of a few strangers, but not at the expense of very many. However, there seems no number of strangers that marks a precise upper bound here. Consequently, there are borderline cases of groups at the expense of which you are permitted to save your friend. This essay discusses the question of what explains ethical vagueness like this, arguing that there are interesting metaethical consequences of various explanations.

The anglophone philosophy profession has a well-known problem with gender equity. A sig-nificant aspect of the problem is the fact that there are simply so many more male philoso-phers than female philosophers among students and faculty alike. The problem is at its stark-est at the faculty level, where only 22% - 24% of philosophers are female in the United States (Van Camp 2014), the United Kingdom (Beebee & Saul 2011) and Australia (Goddard 2008).<1> While this is a result of the (...) percentage of women declining at each point through-out the standard career trajectory, recent large-scale studies in the United States (Paxton et al. 2012) and Australia (Goddard et al. 2008) have identified a key drop-off point as the transi-tion between taking introductory classes and majoring in philosophy. So why do dispropor-tionately few female students choose to major in philosophy? (shrink)

This text offers an account of how philosophers in the 20th century have challenged the ideas of the modern philosophers of the 17th century on fundamental questions in epistemology. Featuring examples, self-study questions and further readings, the text introduces and critically defines logical analysis, foundationalism and coherentism.

To develop a theoretical framework for drawing moral distinctions between instances of sexual misconduct, I defend the “Ameliorative View” of consent, according to which there are three possibilities for what effect, if any, consent has: “fully valid consent” eliminates a wronging, “fully invalid consent” has no normative effect, and “partially valid consent” has an ameliorative effect on a wronging in the respect that it makes the wronging less grave. I motivate the view by proposing a solution to the problem of (...) characterizing the moral effect of consent that is given in response to minor coercion. (shrink)

Focusing on God's essential attributes of omnipotence, omniscience, being eternal and omnipresent, being a creator and sustainer, and being a person, I examine how far recent discussion has been able to provide for each of these divine attributes a consistent interpretation. I also consider briefly whether the attributes are compatible with each other.

In this essay, I discuss why consent is invalidated by duress that involves attaching penalties to someone's refusal to give consent. At the heart of my explanation is the Complaint Principle. This principle specifies that consent is defeasibly invalid when the consent results from someone conditionally imposing a penalty on the consent‐giver's refusal to give the consent, such that the consent‐giver has a legitimate complaint against this imposition focused on how it is affects their incentives for consenting. The Complaint Principle (...) says that this consent is defeasibly invalid to make room for the Authorization Principle, which allows sincere authorizations to constitute valid consent, even when these are issued by an agent acting under duress. The Authorization Principle has application only in cases of third‐party duress because when the consent‐receiver is the source of the duress, the consent‐giver is not sincerely authorizing the action. (shrink)

In the middle of the last century, it was common to explain the notion of necessity in linguistic terms. A necessary truth, it was said, is a sentence whose truth is guaranteed by linguistic rules. Quine famously argued that, on this view, de re modal claims do not make sense. “Porcupettes are porcupines” is necessarily true, but it would be a mistake to say of a particular porcupette that it is necessarily a porcupine, or that it is possibly purple. Linguistic (...) theories of necessity fell out of favour with the publication of Kripke’s Naming and Necessity, and Quine’s arguments were put aside. In her recent book, Norms and Necessity, Amie Thomasson presents her modal normativism, which is an updated version of the mid-century theories just described. Quine’s arguments are thus relevant once again. We recapitulate Quine’s central argument, in the context of modal normativism. We then criticise Amie Thomasson’s discussion of de re modality. We finish by briefly presenting an alternative account of de re modal statements, which is compatible with modal normativism. (shrink)

This book explores and explains how deep semantics works, how new deep semantics research can be conducted, and how the new scientific method of deep semantics can be used to create new scientific theories and discoveries.

'By identifying the extent to which Aristotle's thinking about ethics was shaped by notions drawn from the crafts Angier has thrown new light on a surprising number of topics and has deepened our understanding of tensions within Aristotle's thought. It is by now a rare achievement to have said something new, true and important about Aristotle.' -- Alasdair MacIntyre, Emeritus Professor of Philosophy, University of Notre Dame, USA.

In this paper we focus on a kind of perceptual states that we call perceptual motivations, that is, perceptual experiences that plausibly motivate us to act, such as itching, perceptual salience and pain. Itching seems to motivate you to scratch, perceiving a stimulus as salient seems to motivate you to attend to it and feeling a pain in your hand seems to motivate actions such as withdrawing from the painful stimulus. Five main accounts of perceptual motivation are available: Descriptive, Conative, (...) Imperativist, Reflexive and Dual Content views. Some of these have been developed in detail and others are natural extensions of existing views. However, we argue that none provides a satisfactory account of the target perceptual states. Driven by the failings of each account, we identify three conditions that any satisfactory theory of perceptual motivation must meet: Direct Motivational Force, Non-Obligatory Force and Intention-Independent Influence. Then we offer an account that attempts to fulfil these conditions: the Affordance Model. This model of perceptual motivation is premised on the claim that each of the target states involves the perception of a specific affordance for action and the triggering of Automatic Motor Initiations. We suggest that AMIs can explain the motivational force of these perceptual states. (shrink)

I argue for the possibility of an extremely intimate connection between the emotional content of the music and the emotional state of the person who produces that music. Under certain specified conditions, the music may not just influence, but also partially constitute the musician’s emotional state.

What conditions on a person’s knowledge must be satisfied in order for them to be morally responsible for something they have done? The first two decades of the twenty-first century saw a surge of interest in this question. Must an agent, for example, be aware that their conduct is all-things-considered … Continue reading Epistemic Conditions of Moral Responsibility →.

In this book, Tom Cochrane develops a new control theory of the emotions and related affective states. Grounded in the basic principle of negative feedback control, his original account outlines a new fundamental kind of mental content called 'valent representation'. Upon this foundation, Cochrane constructs new models for emotions, pains and pleasures, moods, expressive behaviours, evaluative reasoning, personality traits and long-term character commitments. These various states are presented as increasingly sophisticated layers of regulative control, which together underpin the architecture of (...) the mind as a whole. Clearly structured and containing numerous diagrams and examples to illustrate the discussion, this study draws on the latest research from fields including philosophy, psychology and neuroscience, and will appeal to readers interested in the philosophy and cognitive science of emotion. (shrink)

In this paper we focus on a kind of perceptual states that we call perceptual motivations, that is, perceptual experiences that plausibly motivate us to act, such as itching, perceptual salience and pain. Itching seems to motivate you to scratch, perceiving a stimulus as salient seems to motivate you to attend to it and feeling a pain in your hand seems to motivate actions such as withdrawing from the painful stimulus. Five main accounts of perceptual motivation are available: Descriptive, Conative, (...) Imperativist, Reflexive and Dual Content views. Some of these have been developed in detail and others are natural extensions of existing views. However, we argue that none provides a satisfactory account of the target perceptual states. Driven by the failings of each account, we identify three conditions that any satisfactory theory of perceptual motivation must meet: Direct Motivational Force, Non-Obligatory Force and Intention-Independent Influence. Then we offer an account that attempts to fulfil these conditions: the Affordance Model. This model of perceptual motivation is premised on the claim that each of the target states involves the perception of a specific affordance for action and the triggering of Automatic Motor Initiations (AMIs). We suggest that AMIs can explain the motivational force of these perceptual states. (shrink)

I consider the claim that analytic statements are "true in virtue meaning", giving the claim a ground-theoretic interpretation.

It is common to focus on the duties of the wrongdoer in cases that involve injustice. Presumably, the wrongdoer owes her victim an apology for having wronged her and perhaps compensation for having harmed her. But, these are not the only duties that may arise. Are other beneficiaries of an injustice permitted to retain the fruits of the injustice? If not, who becomes entitled to those funds? In recent years, the Connection Account has emerged as an influential account that purports (...) to explain cases such as Embezzlement. This account holds that benefiting from injustice can give rise to a corrective duty - that is, a duty of compensation - owed specifically to the victim of the injustice from which the recipient benefits. This duty is grounded in the connection between the victim and the beneficiary of a given injustice. This paper has two aims. First, I show that we must reject the Connection Account on the grounds that it risks failing correctly to identify those who become entitled to the fruits of injustice. I achieve this by developing and defending the fairness objection. Second, I offer an alternative account: the Moral Taintedness Account. This account states that, when identifying who is entitled to the fruits of injustice, the cause and the degree of the harm suffered by a victim are both relevant considerations, though it does not matter whether the victim is the victim of the injustice that gave rise to the fruits in question. This account avoids the problem associated with the Connection Account, and yields intuitive conclusions in an important range of test cases. (shrink)