PHILOSOPHICAL STUDIES最新文献

We propose developing AIs whose only final goal is being shut down. We argue that this approach to AI safety has three benefits: (i) it could potentially be implemented in reinforcement learning, (ii) it avoids some dangerous instrumental convergence dynamics, and (iii) it creates trip wires for monitoring dangerous capabilities. We also argue that the proposal can overcome a key challenge raised by Soares et al. (2015), that shutdown-seeking AIs will manipulate humans into shutting them down. We conclude by comparing our approach with Soares et al.'s corrigibility framework.

Jan Narveson has articulated a deeply held, widely shared intuition regarding what moral law has to say about bringing additional people into existence: while we are “in favour of making people happy,” we are “neutral about making happy people.” Various formulations of the Narvesonian intuition (closely related to the person-affecting intuition or restriction) have been widely criticized. This present paper outlines an off-the-beaten-path alternate construction of the intuition—the existence condition—and argues that that particular construction has the resources to avoid some of those criticisms. But still other considerably more widely recognized alternate constructions have been offered as well. Thus John Broome outlines what he calls the neutrality intuition. While Broome finds the underlying intuition “strongly attractive,” he nonetheless argues that the neutrality intuition itself leads us quickly into inconsistency. Wlodek Rabinowicz disagrees. On his view, Broome’s inconsistency argument shows, not that the neutrality intuition is false, but rather that it doesn’t follow, from the fact that the outcome, or possible future or world, that includes the additional person is neither better nor worse than the (otherwise similar) world that excludes that person, that the one world is exactly as good as the other. The better view, according to Rabinowicz, is that, on occasion, and specifically when the coming into existence of additional people is at stake, the one world is incommensurate with the other. What is called the principle of trichotomy is, in other words, false. Difficulties arise, however, when we try to reject that seemingly compelling conceptual principle. This present paper concludes with the argument that the availability of the existence condition—which, together with certain other uncontroversial moral principles and a handful of conceptual principles, forms the existential approach—shows that we can maintain the most intuitive parts of neutrality intuition while avoiding both Broome’s inconsistency worry and Rabinowicz’s commitment to incommensurability. Incommensurability may be correct on other grounds—but not, this present paper argues, on the grounds provided by Broome’s inconsistency argument.

True benevolence is, most fundamentally, a desire that the world be better. It is natural and common, however, to frame thinking about benevolence indirectly, in terms of a desire to make a difference to how good the world is. This would be an innocuous shift if desires to make a difference were extensionally equivalent to desires that the world be better. This paper shows that at least on some common ways of making a “desire to make a difference” precise, this extensional equivalence fails. Where it fails, “difference-making preferences” run counter to the ideals of benevolence. In particular, in the context of decision making under uncertainty, coupling a “difference-making” framing in a natural way with risk aversion leads to preferences that violate stochastic dominance, and that lead to a strong form of collective defeat, from the point of view of betterness. Difference-making framings and true benevolence are not strictly mutually inconsistent, but agents seeking to implement true benevolence must take care to avoid the various pitfalls that we outline.

Several theorists maintain that a consideration is a reason to ϕ (where ϕ-ing is an act-type) if and only if that consideration is a reason to intend to ϕ, and some hold as well that a consideration is a reason not to ϕ if and only if that consideration is a reason to intend not to ϕ. The claims often stem from views about what it is to be a practical reason. Here it is argued that both equivalence claims are false. Although no view of practical reasons is advanced, views that imply either equivalence claim are shown to be mistaken.

An important module in the Belief-Desire-Intention architecture for artificial agents (which builds on Michael Bratman’s work in the philosophy of action) focuses on the task of intention reconsideration. The theoretical task is to formulate principles governing when an agent ought to undo a prior committed intention and reopen deliberation. Extant proposals for such a principle, if sufficiently detailed, are either too task-specific or too computationally demanding. I propose that an agent ought to reconsider an intention whenever some incompatible prospect is sufficiently valuable along some dimension that can be assessed at zero or near-zero computational cost.

How are justified belief and rational belief related? Some philosophers think that justified belief and rational belief come to the same thing. Others take it that justification is a matter of how well a particular belief is supported by the evidence, while rational belief is a matter of how well a belief coheres with a person’s other beliefs. In this paper, I defend the view that justification is a dimension of rationality, a view that can make sense of both of these conflicting accounts. When it modifies belief, ‘rational’ is a multidimensional adjective, as there are multiple dimensions along which a belief can be rational. I will argue that one of these dimensions is justification, an account that can not only explain why philosophers give diverging theories of the relationship between justified belief and rational belief, but can also reveal why rational belief and justified belief are closely related despite being distinct.

The Liar paradox arguably shows that a coherent and self-applicable notion of truth is governed by nonclassical logic. It then seems natural to conclude that classical logic is inadequate for defining a truth theory. In this article, we argue that this is not the case. In the spirit of Reinhardt (Math Logic Formal Syst 94:227, 1985; J Philos Logic 15:219–251, 1986), and in analogy with Hilbert’s program for the foundation of classical mathematics, we will articulate an instrumentalist justification for the use classical logic: it will be argued that classical reasoning is a useful but dispensable instrument, which can yield philosophically adequate truth theories.

Representationalism about consciousness is the view that the phenomenal character of an experience supervenes on the content of that experience. Much of the literature on representationalism concerns putative objections and replies, rather than clarifying the details of the view itself. Defenders of representationalism face a question which has thus far been largely overlooked: what, precisely, is the relationship between phenomenal character and content? The representationalist has three options: mere supervenience, building or metaphysical dependence, or identity. After examining a number of versions of the first two views, I conclude that they all face serious metaphysical difficulties. I argue instead that this relationship is identity, despite the fact that identifying content and phenomenal character requires revising our view of the content of experiences. Identifying content and phenomenal character strengthens the dialectical position of representationalists by providing them with better responses to anti-representationalist objections. In closing, I show how we can accept the implications of this revisionary view of perceptual content.

Intersectionality is the notion that concerns the complexity of the experiences of individuals in virtue of their belonging to multiple socially significant categories. One of its main insights is that the way society is structured around categories such as gender, race, sexuality, class, etc., produces distinctive and specific forms of discrimination and privilege for groups in the intersections. In this paper, we suggest conceiving intersectionality as a general metaphysical framework wherein specific claims to the effect that the experiences of discrimination of Black women, among others, can be fruitfully formulated and examined. The main claim is that intersectional experiences emerge from the conjunction of social categories when social structures make them relevant vis-à-vis discrimination and privilege. We then argue that our view has three main virtues: metaphysical neutrality, explanatory flexibility and methodological openness. Explaining these virtues will allow us to contrast our proposal with alternatives from the recent literature.

Is it good for us if the different parts of our lives are connected to each other like the parts of a good story? Some philosophers have thought so, while others have firmly rejected it. In this paper, I focus on the state-of-the-art anti-narrativist arguments Amy Berg has recently presented in this journal. I argue that while she makes a good case that the best kind of lives for us do not revolve around a single project or theme, the best kind of narrativist views actually encourage us to pursue a variety of different projects, as long as they are mutually supportive. I claim that when interpreted in the most plausible way, prudentially good-making narrative coherence arises precisely out of this kind of unity in diversity. Well-roundedness and narrative coherence are thus not inherently in tension, but are both good-making holistic features of our lives.