We argue that the trend toward providing users with feasible and actionable explanations of AI decisions—known as recourse explanations—comes with ethical downsides. Specifically, we argue that recourse explanations face several conceptual pitfalls and can lead to problematic explanation hacking, which undermines their ethical status. As an alternative, we advocate that explanations of AI decisions should aim at understanding.

An exploration of the details of counterpart theory, and some applications of the view. In Chapter 1, I set out the view and clarify the most important features: that the counterpart relation is a context dependent similarity relation, and that individuals are world-bound entities. I then set out what I take to be the most promising methods of filling in important details. Chapter 2 is a discussion of an alternative view, lump theory. I attempt to distinguish lump theory from counterpart (...) theory, and argue that several attempt to do so fail. Chapter 3 is an attempt to apply counterpart theory for individuals to properties. I attempt to stay as close to the standard view as possible, but find it necessary to modify the world-bound condition. I suggest a version of property counterpart theory that overcomes the obstacles I have identified. The final chapter is a discussion of a recent attempt to apply counterpart theory to the comparativist view of quantitative properties. I argue that an attempt to rely on counterpart theory to overcome an objection is unsuccessful. (shrink)

This paper considers whether an analogy between distance and dissimilarlity supports the thesis that degree of dissimilarity is distance in a metric space. A straightforward way to justify the thesis would be to define degree of dissimilarity as a function of number of properties in common and not in common. But, infamously, this approach has problems with infinity. An alternative approach would be to prove representation and uniqueness theorems, according to which if comparative dissimilarity meets certain qualitative conditions, then it (...) is representable by distance in a metric space. I will argue that this approach faces equally severe problems with infinity. (shrink)

On a widely suggested knowledge-first account of belief, to believe p is to phi as if one knew p. I challenge this view by arguing against various regimentations of it. I conclude by generalizing my argument to alternative knowledge-first views suggested by Williamson and Wimmer.

Lewis proposed to test the validity of a modal thesis by checking whether its possible-world translation is a theorem of counterpart theory. However, that criterion fails to validate many standard modal laws, thus raising doubts about the logical adequacy of the Lewisian framework. The present paper considers systems of counterpart theory of increasing strength and shows how each can be motivated by exhibiting a suitable intended model. In particular, perfect counterpart theory validates all the desired modal laws and therefore provides (...) a way out of the logical objection. Finally, a weakening of perfect counterpart theory is put forward as a response to some metaphysical objections. (shrink)

I introduce and defend the semantic notion of counterfactual identity, distinguishing it from the metaphysical notion of transworld identity. After showing that Lewis's counterpart theory misconstrues counterfactual identity facts, I outline and motivate a ‘Leibnizian counterpart theory’ where the notion of counterfactual identity is adequately modelled. Finally, I show that counterfactual identity can be characterized without relying on some implausible features of Lewis's theory of conditionals.

Recently, interest has surged in similarity-based epistemologies of possibility. However, it has been pointed out that the notion of ‘relevant similarity’ is not properly developed in this literature. In this paper, I look at the research done in the field of analogical reasoning, where we find that one of the most promising ways of capturing relevance in similarity reasoning is by relying on the predictive analogy similarity relation. This takes relevant similarity to be based on shared properties that have structural (...) relations to the property of interest. I argue that if we base our epistemology of possibility on similarity reasoning on the predictive analogy similarity relation, we require prior knowledge of the specifics of these structural relations. I discuss a number of possible responses to this on behalf of the similarity theorists given their methodological approach to the epistemology of modality more generally. They could either opt for making explicit the metaphysics underlying these structural relations, in which case they need to spell out how we can come to know these relations. Or they could opt for developing a theory that explains why we do not need to have explicit knowledge of these structural relations; for example by suggesting that we make use of epistemic shortcuts. (shrink)

The idea that sentences can be closer or further apart in meaning is highly intuitive. Not only that, it is also a pillar of logic, semantic theory and the philosophy of science, and follows from other commitments about similarity. The present paper proposes a novel way of comparing the ‘distance’ between two pairs of propositions. We define ‘\ is closer in meaning to \ than \ is to \’ and thereby give a precise account of comparative propositional similarity facts. Notably, (...) our definition eschews metric assumptions, which are unrealistic in most applications of interest. (shrink)

Are there ever good epistemic reasons to neglect base rates? Assuming an empiricist modal epistemology, I argue that we face an interesting tension between some very plausible epistemic norms: a norm requiring us to proportion our beliefs to the evidence may facilitate knowledge of the actual world, whilst inhibiting our acquisition of modal knowledge—knowledge of how things could be, but are not. The potential for this tension in our epistemic norms is a significant result in its own right. It can (...) also rationalize certain forms of demographic base-rate neglect. (shrink)

ABSTRACT In Consciousness and Fundamental Reality, Philip Goff argues that the case against physicalist views of consciousness turns on ‘Phenomenal Transparency’, roughly the thesis that phenomenal concepts reveal the essential nature of phenomenal properties. This paper considers the argument that Goff offers for Phenomenal Transparency. The key premise is that our introspective judgments about current conscious experience are ‘Super Justified’, in that these judgments enjoy an epistemic status comparable to that of simple mathematical judgments, and a better epistemic status than (...) run of the mill perceptual judgments. After presenting the key ideas in the ‘Super Justification Argument’, I distinguish two Super Justification theses, which vary according to the kind of introspective judgments that they take to be Super Justified. I argue that Goff’s case requires ‘Strong Super Justification’, according to which a wide range of introspective judgments about conscious experience are Super Justified. Unfortunately, it turns out that Strong Super Justification is implausible and not well-supported by examples. In contrast, a weaker Super Justification thesis does not require anything like Phenomenal Transparency and, indeed, can be explained by physicalistic accounts of phenomenal concepts. (shrink)

We argue that a semantics for counterfactual conditionals in terms of comparative overall similarity faces a formal limitation due to Arrow’s impossibility theorem from social choice theory. According to Lewis’s account, the truth-conditions for counterfactual conditionals are given in terms of the comparative overall similarity between possible worlds, which is in turn determined by various aspects of similarity between possible worlds. We argue that a function from aspects of similarity to overall similarity should satisfy certain plausible constraints while Arrow’s impossibility (...) theorem rules out that such a function satisfies all the constraints simultaneously. We argue that a way out of this impasse is to represent aspectual similarity in terms of ranking functions instead of representing it in a purely ordinal fashion. Further, we argue against the claim that the determination of overall similarity by aspects of similarity faces a difficulty in addition to the Arrovian limitation, namely the incommensurability of different aspects of similarity. The phenomena that have been cited as evidence for such incommensurability are best explained by ordinary vagueness. (shrink)

Philosophy and Phenomenological Research, EarlyView.

This article analyses the relationship between the concept of single aspect similarity and proposed measures of similarity. More precisely, it compares eleven measures of similarity in terms of how well they satisfy a list of desiderata, chosen to capture common intuitions concerning the properties of similarity and the relations between similarity and dissimilarity. Three types of measures are discussed: similarity as commonality, similarity as a function of dissimilarity, and similarity as a joint function of commonality and difference. Relative to the (...) desiderata, it is found that a measure of the second type fares the best. However, rather than recommend this measure alone as a measure of similarity, it is suggested that there are at least three separate concepts of single aspect similarity, corresponding to the three types of measures. In light of this proposal, three of the eleven measures (and variants of these) are deemed acceptable. (shrink)

Adam Elga has presented an anti-thermodynamic process as a counterexample to Lewis’s default semantics for counterfactuals. The outstanding reaction of Jonathan Schaffer and Boris Kment is revisionary. It sacrifices Lewis’s aim of defining causation in terms of counterfactual dependence. Lewis himself suggested an alternative: «counter-entropic funnybusiness» should make for dissimilarity. But how is this alternative to be spelled out? I discuss a recent proposal: include special science laws, among them the laws of thermodynamics. Although the proposal fails, it serves to (...) uncover the limits of Elga’s example. (shrink)

Many philosophers since Hume have accepted that imagining/conceiving a scenario is our prime guide to knowing its possibility. Stephen Yablo provided a more systematic criterion: one is justified in judging that p is possible if one can imagine a world which one takes to verify p. I defend a version of Yablo’s criterion against van Inwagen’s moderate modal scepticism. Van Inwagen’s key argument is that we cannot satisfy Yablo’s criterion because we are not in a position to spell out far-fetched (...) possible scenarios in relevant detail. Van Inwagen’s argument can be applied to the use of conceivability for everyday possibility claims, leaving us with the spectre of pervasive modal scepticism. In order to answer the sceptical threat, I combine van Inwagen’s main example with general considerations about the nature of metaphysical modality to motivate a version of Yablo’s criterion and show that it does not lead to scepticism. One structural condition of p being metaphysically possible is that it coheres with a complete reality. This condition gives rise to Yablo’s criterion. However, for the criterion to be of any avail, we have to disregard details we are not in a position to specify. To account for our practice of doing so, I use Yablo’s distinction between imagining a world as determinate and imagining it determinately. I present a condition when we may simply disregard details as determinate. The condition results from integrating analogical reasoning into the conceivability test. (shrink)

People frequently entertain counterfactual thoughts, or mental simulations about alternative ways the world could have been. But the perceived plausibility of those counterfactual thoughts varies widely. The current article interfaces research in the philosophy and semantics of counterfactual statements with the psychology of mental simulations, and it explores the role of perceived similarity in judgments of counterfactual plausibility. We report results from seven studies (N = 6405) jointly supporting three interconnected claims. First, the perceived plausibility of a counterfactual event is (...) predicted by the perceived similarity between the possible world in which the imagined situation is thought to occur and the actual world. Second, when people attend to differences between imagined possible worlds and the actual world, they think of the imagined possible worlds as less similar to the actual world and tend to judge counterfactuals in such worlds as less plausible. Lastly, when people attend to what is identical between imagined possible worlds and the actual world, they think of the imagined possible worlds as more similar to the actual world and tend to judge counterfactuals in such worlds as more plausible. We discuss these results in light of philosophical, semantic, and psychological theories of counterfactual thinking. (shrink)

Multidimensional adjectives are ubiquitous in natural language. An adjective F is multidimensional just in case whether F applies to an object or pair of objects depends on how those objects stand with respect to multiple underlying dimensions of F-ness. Developing a semantics for multidimensional adjectives requires us to address the problem of dimensional aggregation: how do the application conditions of an adjective F in its positive and comparative forms depend on its underlying dimensions? Here we develop a semantics for multidimensional (...) adjectives that incorporates aggregation functions. We then explore an analogy between dimensional aggregation and preference aggregation, bringing results from social choice theory to bear on the number and kind of aggregation functions that are admissible in a context. These results suggest that, for any given adjective, there will often be multiple aggregation functions admissible, meaning that multidimensional comparatives are often vague. (shrink)

Our ordinary judgments and our metaphysical theories share a common commitment to facts about resemblance. The nature of resemblance is, however, a matter of no small controversy. This essay examines some of the pressing questions that arise regarding the status and structure of resemblance. Among those to be discussed in what follows: what kinds of resemblance relations are there? Can resemblance be analyzed in terms of the sharing of properties? Is resemblance an objective or subjective matter? What, if any, resemblance (...) facts are fundamental? How does resemblance relate to naturalness? (shrink)

According to Satosi Watanabe's "theorem of the ugly duckling", the number of predicates satisfied by any two different particulars is a constant, which does not depend on the choice of the two particulars. If the number of predicates satisfied by two particulars is their number of properties in common, and the degree of resemblance between two particulars is a function of their number of properties in common, then it follows that the degree of resemblance between any two different particulars is (...) also constant, which is absurd. Avoiding this absurd conclusion requires questioning assumptions about infinity in the proof or interpretation of the theorem, adopting a sparse conception of properties, or denying degree of resemblance is a function of number of properties in common. After arguing against both the first two options, this paper argues for a version of the third which analyses degree of resemblance as a function of properties in common, but weighted by their degree of naturalness or importance. (shrink)

Big decisions in a person’s life often affect the preferences and standards of a good life which that person’s future self will develop after implementing her decision. This paper argues that in such cases the person might lack any reasons to choose one way rather than the other. Neither preference-based views nor happiness-based views of justified choice offer sufficient help here. The available options are not comparable in the relevant sense and there is no rational choice to make. Thus, ironically, (...) in many of a person’s most important decisions the idea of that person’s good seems to have no application. (shrink)

Multidimensional concepts are everywhere, and they are important. Examples include moral value, welfare, scientific confirmation, democracy, and biodiversity. How, if at all, can we aggregate the underlying dimensions of a multidimensional concept F to yield verdicts about which things are Fer than which overall? Social choice theory can be used to model and investigate this aggregation problem. Here, we focus on a particularly thorny problem made salient by this social choice-theoretic framework: the underlying dimensions of a given concept might be (...) measurable on different types of scales—e.g., some ordinal and some cardinal. An underappreciated impossibility theorem due to Anna Khmelnitskaya shows that seemingly plausible constraints on aggregation across scale types are inconsistent. This impossibility threatens to render the notion of overall Fness incoherent. We attempt to defuse this threat, arguing that the impossibility depends on an overly restrictive conception of measurement and of how measurement constrains aggregation. Adopting a more flexible—and, we think, more perspicuous—conception of measurement opens an array of possibilities for aggregation across disparate scale types. (shrink)

This chapter discusses the application of formal methods from social choice theory to the metasemantic question of whether radical interpretation is possible. Radical interpretation involves deducing semantic truths from non-semantic truths by appeal to certain a priori principles or criteria, such as the principle of charity. A familiar view is that the intended interpretation is the one that best meets a combination of constraints. It is suggested that this situation can be modelled as follows: each constraint determines a binary relation (...) on the set X of interpretations that is transitive and complete. The radical interpreter’s task is to determine an overall ordering as a function from the profile of individual orderings. The application of Arrow’s theorem in this context is discussed. (shrink)

Social choice theory is the study of collective decision processes and procedures. It is not a single theory, but a cluster of models and results concerning the aggregation of individual inputs (e.g., votes, preferences, judgments, welfare) into collective outputs (e.g., collective decisions, preferences, judgments, welfare). Central questions are: How can a group of individuals choose a winning outcome (e.g., policy, electoral candidate) from a given set of options? What are the properties of different voting systems? When is a voting system (...) democratic? How can a collective (e.g., electorate, legislature, collegial court, expert panel, or committee) arrive at coherent collective preferences or judgments on some issues, on the basis of its members' individual preferences or judgments? How can we rank different social alternatives in an order of social welfare? Social choice theorists study these questions not just by looking at examples, but by developing general models and proving theorems. (shrink)

Divers (2014) argues that a Lewisian theory of modality which includes both counterpart theory and modal realism cannot account for the truth of certain intuitively true modal sentences involving cross-world comparatives. The main purpose of this paper is to defend the Lewisian theory against Divers’s challenge by developing a response strategy based on a degree-theoretic treatment of comparatives and by showing that this treatment is compatible with the theory.

Scientific models are important, if not the sole, units of science. This thesis addresses the following question: in virtue of what do scientific models represent their target systems? In Part i I motivate the question, and lay out some important desiderata that any successful answer must meet. This provides a novel conceptual framework in which to think about the question of scientific representation. I then argue against Callender and Cohen’s attempt to diffuse the question. In Part ii I investigate the (...) ideas that scientific models are ‘similar’, or structurally morphic, to their target systems. I argue that these approaches are misguided, and that at best these relationships concern the accuracy of a pre-existing representational relationship. I also pay particular attention to the sense in which target systems can be appropriately taken to exhibit a ‘structure’, and van Fraassen’s recent argument concerning the pragmatic equivalence between representing phenomena and data. My next target is the idea that models should not be seen as objects in their own right, but rather what look like descriptions of them are actually direct descriptions of target systems, albeit not ones that should be understood literally. I argue that these approaches fail to do justice to the practice of scientific modelling. Finally I turn to the idea that how models represent is grounded, in some sense, in their inferential capacity. I compare this approach to anti-representationalism in the philosophy of language and argue that analogous issues arise in the context of scientific representation. Part iii contains my positive proposal. I provide an account of scientific representation based on Goodman and Elgin’s notion of representation-as. The result is a highly conventional account which is the appropriate level of generality to capture all of its instances, whilst remaining informative about the notion. I illustrate it with reference to the Phillips-Newlyn machine, models of proteins, and the Lotka-Volterra model of predator-prey systems. These examples demonstrate how the account must be understood, and how it sheds light on our understanding of how models are used. I finally demonstrate how the account meets the desiderata laid out at the beginning of the thesis, and outline its implications for further questions from the philosophy of science; not limited to issues surrounding the applicability of mathematics, idealisation, and what it takes for a model to be ‘true’. (shrink)