James Ravi Kirkpatrick

Deepfakes are synthetic media that superimpose or generate someone’s likeness onto pre-existing sound, images, or videos using deep learning methods. Existing accounts of the wrongs involved in creating and distributing deepfakes focus on the harms they cause or the non-normative interests they violate. However, these approaches do not explain how deepfakes can be wrongful even when they cause no harm or set back any other non-normative interest. To address this issue, this paper identifies a neglected reason why deepfakes are wrong: they can subvert our legitimate interests in having authority over the permissible uses of our image and the governance of our identity. We argue that deepfakes are wrong when they usurp our authority to determine the provenance of our own agency by exploiting our biometric features as a generative resource. In particular, we have a specific right against the algorithmic conscription of our identity. We refine the scope of this interest by distinguishing between permissible forms of appropriation, such as artistic depiction, from wrongful algorithmic simulation.

The development of sophisticated artificial intelligence (AI) conversational agents based on large language models raises important questions about the relationship between human norms, values, and practices and AI design and performance. This article explores what it means for AI agents to be conversationally aligned to human communicative norms and practices for handling context and common ground and proposes a new framework for evaluating developers’ design choices. We begin by drawing on the philosophical and linguistic literature on conversational pragmatics to motivate a set of desiderata, which we call the CONTEXT-ALIGN framework, for conversational alignment with human communicative practices. We then suggest that current large language model (LLM) architectures, constraints, and affordances may impose fundamental limitations on achieving full conversational alignment.

There is a standard quantificational view of generic sentences according to which they have a tripartite logical form involving a phonologically null generic operator called 'Gen'. Recently, a number of theorists have questioned the standard view and revived a competing proposal according to which generics involve the predication of properties to kinds. This paper offers a novel argument against the kind-predication approach on the basis of the invalidity of Generic Excluded Middle, a principle according to which any sentence of the form ⌜Either Fs are G or Fs are not G⌝ is true. I argue that the kind-predication approach erroneously predicts that GEM is valid, and that it can only avoid this conclusion by either collapsing into a form of the quantificational analysis or otherwise garnering unpalatable metaphysical commitments. I also show that, while the quantificational approach does not validate GEM as a matter of logical form, the principle may be validated on certain semantic analyses of the generic operator, and so, such theories should be rejected.

This paper is a study of Objectual Quantifier Theory, the view that quantificational noun phrases, such as "every woman" and "some pig", denote generic individuals, such as the arbitrary woman and the indefinite pig. We explore the motivations for this view and various ways of developing it, taking inspiration from and expanding upon Kit Fine’s work on arbitrary objects (Fine, Proceedings of the Aristotelian Society, Supplementary Volumes, 57, 55–77 1983; Journal of Philosophical Logic, 14 (1), 57–107 1985; Reasoning with Arbitrary Objects 1985). Our discussion centres around the task of delivering a compositional treatment of quantifier phrases in natural language. In the first part of the paper, we explore the potential for applying and extending Fine’s framework in the interpretation of quantifier expressions, developing desiderata for a successful theory. In the second part of the paper, we develop our positive theory, which we call Objectual Quantifier Theory, as a theory of quantification in natural language. We argue that Objectual Quantifier Theory is at least as empirically powerful as Generalized Quantifier Theory.

The development of sophisticated artificial intelligence (AI) conversational agents based on large language models raises important questions about the relationship between human norms, values, and practices and AI design and performance. This article explores what it means for AI agents to be conversationally aligned to human communicative norms and practices for handling context and common ground and proposes a new framework for evaluating developers’ design choices. We begin by drawing on the philosophical and linguistic literature on conversational pragmatics to motivate a set of desiderata, which we call the CONTEXT‐ALIGN framework, for conversational alignment with human communicative practices. We then suggest that current large language model (LLM) architectures, constraints, and affordances may impose fundamental limitations on achieving full conversational alignment.

Many philosophers and linguists have been attracted to counterpart theory as a framework for natural language semantics. I raise a novel problem for counterpart theory involving simple declarative sentences with proper names. To resolve this problem, counterpart theorists must introduce the notion of a counterpart in the semantics of the non-modal fragment of language. I develop my preferred solution: a novel theory of proper names as counterpart-theoretic individual concepts. The resulting view highlights a hitherto unnoticed fact: counterpart theorists should formulate their theory, not by modifying the standard semantics for modal operators, but by modifying the standard semantics for names and variables.

Sentences with impersonal pronouns, like 'You do it like this', seem to make both statistical and prescriptive claims, that a certain way of behaving is common and that it is prescriptively good. We argue that these kinds of sentences are closely related to another kind of sentence, namely, indefinite singular generics, like 'A person does it like this'. We propose that there is a single underlying mechanism that allows both kinds of sentences to express mixed statistical/prescriptive readings. We then provide a compositional semantic analysis of these sentences and explain how, together with this single underlying mechanism, it accounts for this phenomenon.

The standard view about generic generalizations is that they have a tripartite quantificational logical form involving a phonologically null quantificational expression called ‘Gen’. However, proponents of the cognitive defaults theory of generics have forcefully rejected this view, instead arguing that generics express the default generalizations of our cognitive system, and, as such, they are different in kind from quantificational generalizations. While extant criticism of the cognitive defaults theory has focused on the extent to which it is supported by the empirical evidence, there has been little discussion of a neglected, albeit essential, theoretical argument in its defence, namely, that generics cannot be quantificational because they lack a central logical property of quantifiers: isomorphism invariance. This paper addresses this lacuna by considering and rejecting this argument. Consequently, an essential argument in favour of the cognitive default theory is found wanting.

It is a familiar point that we can use generic sentences to express generalisations that are tolerant to exceptions and then go on to state those exceptions explicitly. It is a less familiar point that switching the order of the generics has deleterious effects on their felicity. For example, the sequences ‘Ravens are black, but albino ravens aren’t’ is perfectly felicitous and judged to be true, whereas its reverse ‘Albino ravens aren’t black, but ravens are’ is infelicitous and contradictory-sounding. This paper argues that such sequences pose a problem for extant theories of generics: while they have no trouble predicting the felicity of the first sequence of generics, they are unable to explain why reversing the order results in infelicity. I propose to account for these observations by adopting a dynamic semantic theory of generic sentences.

There is a standard quantificational view of generic sentences according to which they have a tripartite logical form involving a phonologically null generic operator called 'Gen'. Recently, a number of theorists have questioned the standard view and revived a competing proposal according to which generics involve the predication of properties to kinds. This paper offers a novel argument against the kind-predication approach on the basis of the invalidity of Generic Excluded Middle, a principle according to which any sentence of the form ⌜Either Fs are G or Fs are not G⌝ is true. I argue that the kind-predication approach erroneously predicts that GEM is valid, and that it can only avoid this conclusion by either collapsing into a form of the quantificational analysis or otherwise garnering unpalatable metaphysical commitments. I also show that, while the quantificational approach does not validate GEM as a matter of logical form, the principle may be validated on certain semantic analyses of the generic operator, and so, such theories should be rejected.

This paper reevaluates Kaplan’s (Themes from Kaplan, Oxford University Press, Oxford, pp 481–563, 1989b) infamous ‘compulsive talker’ objection to Reichenbach’s (Elements of symbolic logic, AQ1 Macmillan, New York, 1947 ) token-reflexive theory of indexicals. It argues that Kaplan’s objection depends on the modal status of Reichenbachian tokens. On one interpretation, Kaplan’s objection stands. But on another, equally plausible interpretation, the following points hold: (i) Reichenbach’s theory effectively preempts contemporary discussion of rigid definite descriptions, (ii) Kaplan’s own analysis of indexicals in terms of dthat-terms comes extremely close to Reichenbach’s own analysis, and (iii) Kaplan’s ‘compulsive talker’ objection should be rejected.

It has been argued that the primary acquisition of genericity in early child speech poses a problem for standard quantificational approaches to generics and instead motivates the claim that generics give voice to an innate, default mode of generalising. This article argues that analogous puzzles involving the acquisition of A‐quantifiers undermine the empirical support for a purely cognition‐based approach to generics. Instead, these acquisition puzzles should be solved by generalising the core insight of the cognitive defaults theory to these expressions, reconciling formal semantic approaches with the role that cognitive development plays on lexical competence.

Generic sentences involving phrasal conjunctions present a prima facie problem for the standard theory of generics according to which they express quasi-universal generalisations about what is characteristic for members of a particular kind. For example, the sentence ‘Elephants live in Africa and Asia’ is true, even though it is uncharacteristic for an elephant to live in both Africa and Asia. In response to this problem, theorists have recently proposed radical departures from the standard view. This paper argues that such departures are unwarranted: not only do they fail to fully accommodate the data involving generic conjunctions, their scope is overly narrow, since the phenomena in question also arise in non-generic contexts. I propose a new theory of generics that aims to account for generic conjunctions in a principled manner and which sheds new light on the mereological commitments of natural language.

Kevin Dorst has recently pointed out an apparently puzzling consequence of denying epistemic luminosity: given some natural-sounding bridging principles between knowledge, credence, and indicative conditionals, the denial of epistemic luminosity licenses the knowledge and assertability of abominable-sounding conditionals of the form ⌜If I don’t know that ϕ, then ϕ⌝. We provide a general and systematic examination of this datum by testing Dorst’s claim against various semantics for the indicative conditional in the setting of epistemic logic. Our conclusion is that, regardless of whether knowledge is luminous, the knowability of these conditionals is highly sensitive to the correct semantic analysis of the indicative conditional. Moreover, standard pragmatic resources can explain away the infelicity of such assertions. As it stands, the datum does not tell against epistemic non-luminosity.

John Broome argues that fairness requires that claims are satisfied in proportion to their strength. Broome holds that, when distributing indivisible goods, fairness requires the use of weighted lotteries as a surrogate to satisfy proportionally each candidate's claims. In this article, we present two arguments against Broome's account of fairness. First, we argue that it is almost impossible to calculate the weights of the lotteries in accordance with the requirements of fairness. Second, we argue that Broome rules out those methods whose use might provide some resolution to this problem. From these arguments, we conclude that, contra Broome, fairness does not require the proportional satisfaction of claims