PHILOSOPHICAL STUDIES最新文献

I explain and motivate the shutdown problem: the problem of designing artificial agents that (1) shut down when a shutdown button is pressed, (2) don’t try to prevent or cause the pressing of the shutdown button, and (3) otherwise pursue goals competently. I prove three theorems that make the difficulty precise. These theorems suggest that agents satisfying some innocuous-seeming conditions will often try to prevent or cause the pressing of the shutdown button, even in cases where it’s costly to do so. I end by noting that these theorems can guide our search for solutions to the problem.

In Felicitous Underspecification, Jeffrey King draws our attention to a rich and underexplored collection of linguistic data. These are uses of context-sensitive expressions which seem perfectly felicitous despite being such that, on plausible assumptions, the context in which they are used falls short of securing for them a unique semantic value. This raises an immediate puzzle: if, as King argues, these uses of expressions really do lack unique semantic values in context, how can they—as they manifestly do—make contributions to the conversations in which they occur? King answers this question with a novel theory of conversational updating. Here I focus less on this theory than on King’s examples, and consider some ways of accommodating them without positing felicitous underspecification. In some cases I give some reasons for thinking that the alternative explanation is superior. But my aim is less to establish this conclusion than to suggest some new options for thinking about these examples, and hopefully by so doing to advance the conversation about the data to which King has drawn our attention.

Felicitous underspecification—apparently flawless use of context-sensitive words in contexts where they cannot be assigned unique semantic values—is rather common in ordinary speech. King presents a hypothesis about the mechanism conversational participants employ handling felicitous underspecification, one that fits the rich data he surveys well. I will begin by illustrating how King’s account could be put to use in making sense of what happens in a real life conversation. Then I will point out certain shortcomings of the explanation and offer suggestions about how they might be overcome.

This paper illuminates a typically obscured ground for rectificatory obligations: harms justified as ‘lesser evils.’ Lesser-evil harms are not the result of overall morally prohibited acts but of acts permissibly carried out to prevent significantly greater harm. The paper argues that harms caused as unintended side effects of acting on lesser-evil justifications, notably in military rescue operations, may give rise to claims to compensation, even if (1) the military acts that caused the harms in question were justified on lesser-evil grounds and (2) the victims in question are no worse off as a result; they may even owe their survival to the act of rescue. The paper defends three claims. First, being better off as a result of a harmful rescue than one would otherwise have been does not preclude claims to be compensated for harms suffered as a side effect. Second, identifying the relevant counterfactual for purposes of compensatory justice is sometimes a prescriptive, rather than a descriptive, matter. Rather than relying on empirical speculations about what would have happened had a harm not occurred, we must, in certain cases, consider what agents ought to have done. Finally, duties of compensation need not fall on those who caused the to-be-compensated harms. That infringing rights is permissible in certain cases does not imply that no compensation is owed, but merely that it is not necessarily rights-infringers on whom duties of compensation fall.

A key challenge for probabilistic causal models is to distinguish non-causal probabilistic dependencies from true causal relations. To accomplish this task, causal models are usually required to satisfy several constraints. Two prominent constraints are the causal Markov condition and the faithfulness condition. However, other constraints are also needed. One of these additional constraints is the causal sufficiency condition, which states that models must not omit any direct common causes of the variables they contain. In this paper, I argue that the causal sufficiency condition is problematic: (1) it is incompatible with the requirement that the variables in a model must not stand in non-causal necessary dependence relations, such as mathematical or conceptual relations, or relations described in terms of supervenience or grounding, (2) it presupposes more causal knowledge as primitive than is actually needed to create adequate causal models, and (3) if models are only required to be causally sufficient, they cannot deal with cases where variables are probabilistically related by accident, such as Sober’s example of the relationship between bread prices in England and the sea level in Venice. I show that these problems can be avoided if causal models are required to be monotonic in the following sense: the causal relations occurring in a model M would not disappear if further variables were added to M. I give a definition of this monotonicity condition and conclude that causal models should be required to be monotonic rather than causally sufficient.

A detailed analysis of joint-contribution of premises and conclusions in classically valid sequents is presented in terms of hypergraphs. In (Saint-Germier, P., Verdée, P., & Villalonga, P. T. (2024). Relevant entailment and logical ground. Philosophical Studies (pp. 1–43). https://doi.org/10.1007/s11098-024-02101-1), this idea of joint-contribution is introduced and motivated as a method for characterizing four kinds of relevant validity, in the sense of selecting the relevantly valid sequents among the classically valid sequents. The account in (Saint-Germier, P., Verdée, P., & Villalonga, P. T. (2024). Relevant entailment and logical ground. Philosophical Studies (pp. 1–43). https://doi.org/10.1007/s11098-024-02101-1) is built on a calculus, called (textsf{GLK}^{hbox {a}}), which proves grounding claims for (enthymematically) valid sequents. In the present paper an adequate representation of (textsf{GLK}^{hbox {a}}) is given in terms of hypergraphs. The hypergraphs are a kind of diagrammatic proofs for Classical Propositional Logic, entirely based on the grounds of premises and conclusions. The hypergraphs and their visualization provide insights into the relations between premises and conclusions and into the way validity is produced by the binding of premises and conclusions via their partial grounds. They visualize the network of elements of the sequent that contribute to its logical validity. Non-contributing (i.e. irrelevant) premises and conclusions are then specified to be those that are disconnected from the network, however one constructs the graphs.

In this paper we introduce the view that realism about a social kind K entails that the grounding conditions of K are difficult (or impossible) to manipulate. In other words, we define social kind realism in terms of relative frame manipulability (RFM). In articulating our view, we utilize theoretical resources from Epstein’s (Epstein, The ant trap: Rebuilding the foundations of the Social Sciences. Oxford University Press, 2015) grounding/anchoring model and causal interventionism. After comparing our view with causal and principle-based (Tahko, Synthese 200(2):1–23, 2022) proposals, we motivate RFM by showing that it accommodates important desiderata about the social landscape (such as recognizing the context-relativity of social properties and the emancipatory dimension of social practice). Finally, we consider three objections. First, we tackle frame-necessitarianism (FN), the view that social kind frames are metaphysically necessary (and thus unmanipulable). Secondly, we engage with what Epstein (Epstein, Philosophy and Phenomenological Research, 99(3):768–781 2019a) calls UNIVERSALITY (the view that social kinds can hold in the absence of anchors) and we argue that it should also be resisted. Finally, we tackle a recent objection from Mason’s (Mason, Philosophical Studies, 178(12):3975–3994) essentialism about social kinds.

The field of AI safety aims to prevent increasingly capable artificially intelligent systems from causing humans harm. Research on moral alignment is widely thought to offer a promising safety strategy: if we can equip AI systems with appropriate ethical rules, according to this line of thought, they’ll be unlikely to disempower, destroy or otherwise seriously harm us. Deontological morality looks like a particularly attractive candidate for an alignment target, given its popularity, relative technical tractability and commitment to harm-avoidance principles. I argue that the connection between moral alignment and safe behavior is more tenuous than many have hoped. In general, AI systems can possess either of these properties in the absence of the other, and we should favor safety when the two conflict. In particular, advanced AI systems governed by standard versions of deontology need not be especially safe.

Drawing on the puzzling behavior of ordinary knowledge ascriptions that embed an epistemic (im)possibility claim, we tentatively conclude that it is untenable to jointly endorse (i) an unfettered classical logic for epistemic language, (ii) the general veridicality of knowledge ascription, and (iii) an intuitive ‘negative transparency’ thesis that reduces knowledge of a simple negated ‘might’ claim to an epistemic claim without modal content. We motivate a strategic trade-off: preserve veridicality and (generalized) negative transparency, while abandoning the general validity of contraposition. We criticize various approaches to incorporating veridicality into domain semantics, a paradigmatic ‘information-sensitive’ framework for capturing negative transparency and, more generally, the non-classical behavior of sentences with epistemic modals. We then present a novel information-sensitive semantics that successfully executes our favored strategy: stable acceptance semantics, extending a vanilla bilateral state-based semantics for epistemic modals with a knowledge operator loosely inspired by the defeasibility theory of knowledge.