Nikhil Mahant

AI systems are known to induce non-ideal mental states (e.g., false beliefs) in their users while pursuing their objectives. In contemporary discussions, this phenomenon often gets labelled as ‘AI deception.’ But is AI deception really deception? If it is, then how should we make sense of AI deception qua deception given that the capacities of humans seem to be quite different from that of AI systems? This article surveys the space of theoretical alternatives for answering these questions. It also provides some considerations against adopting revisionary accounts of ‘deception’ (that avoid an appeal to mental states in explaining how AI deception can be a type of ordinary deception) and in doing so, makes a case for the ascription of mental states to AI systems as a more promising theoretical alternative. Finally, the article outlines some directions for future research.

Most philosophical analyses assume that for an entity to be morally responsible for an action/outcome, it is necessary that the entity be a moral agent. Call this the ‘Agency Condition’ on moral responsibility. This article provides some methodological considerations that put pressure on the Agency Condition. ‘Agency’ is a rich notion. Some conditions that a thing must satisfy to count as having agency are unrelated to moral responsibility, particularly the ability to form and execute complex plans. Thus, the Agency Condition puts a stricter requirement on moral responsibility than is warranted. One way to get around this problem is to treat ‘moral agent’ as a technical notion, and stipulate the definition of ‘moral agent’ such that moral agents need not be agents. But given that moral responsibility can have distinct kinds, one will need to provide several stipulative definitions of ‘moral agent’, each corresponding to a different kind of moral responsibility. This stipulative move is not methodologically superior to simply abandoning the Agency condition, whose appeal partly lay in its apparent ability to provide a uniform rationale and explanation for why artificial entities cannot be morally responsible.

Metalinguistic views of proper names are often written off on the ground that they are ‘blatantly circular’ (Saul Kripke, Naming and Necessity, 72). This article exonerates metalinguistic views from the charge of circularity. I begin by distinguishing a word from its associated form: while words have semantic properties, forms – which are types of sounds, inscriptions, or gestures – do not. Quotation marks can be used to form a quote-name of either a word or its associated form. I argue that the impression of circularity in metalinguistic views results from the decision to resolve the ambiguity of quotation in the metalinguistic specification of the meaning of a proper name in one way rather than the other – i.e. by taking quotation as forming quote-names of words. Metalinguistic views, however, are not committed to this understanding of quotation. On the contrary, I argue that a plausible and charitable reading of metalinguistic views should take the quotation (in the metalinguistic specification of the meaning of a proper name) to be a quote-name of a form. When metalinguistic views are thus understood, the circularity worry does not even arise.

In recent years, several kinds of non-referential uses of names have received attention within semantics. However, the lack of a principled basis for drawing the distinction between uses that a semantic theory of names must account for (‘literal’ uses) and other uses that it need not explain by itself (‘non-literal uses’) represents an important deficiency in the debate. A prominent objection (the ‘Sceptic’s Challenge’) against semantic views that treat predicative uses as literal exemplifies the kind of problem that can result from this deficiency. This article proposes a general manner of drawing the line between literal and non-literal uses of names. Further, by providing a rationale for treating predicative uses as literal, it also provides a response to the Sceptic’s Challenge.

A distinctive and widely recognized feature of proper names is that, unlike other words, names can be used across languages without modification. Yet, this feature of names—the prevalence and acceptability of their ‘cross‐linguistic’ uses—has been mostly overlooked within philosophy. This article highlights the theoretical importance of the cross‐linguistic uses of names in the debate concerning their syntax and semantics. It identifies an anomalous phonological feature of names in their cross‐linguistic uses and argues that the source of the anomaly is the widespread view that proper names are syntactically simple. It also argues that such uses provide evidence for the syntactic view that the phonological articulation of a name is mentioned and not used in the syntax—a view that is consistent with some but not all semantic views of names. By examining the hitherto overlooked cross‐linguistic uses of names, this article provides new evidence in favour of a certain variety of metalinguistic views concerning the syntax and semantics of names.

Linguistic outputs generated by modern machine-learning neural net AI systems seem to have the same contents—i.e., meaning, semantic value, etc.—as the corresponding human-generated utterances and texts. Building upon this essential premise, Herman Cappelen and Josh Dever's Making AI Intelligible sets for itself the task of addressing the question of how AI-generated outputs have the contents that they seem to have (henceforth, ‘the question of AI Content’). In pursuing this ambitious task, the book makes several high-level, framework observations about how a meta-semantic account of the content of AI-generated outputs should proceed.

The Millian semantic view of names relies on a metaphysical view of names—often given the label ‘common currency conception’ —on which the names of distinct individuals count as distinct names. While even defenders of the Millian view admit that the CCC ‘does not agree with the most common usage’, I will argue further that the CCC makes names exceptional amongst the class of linguistic expressions: if the CCC is correct, then names must have a sui-generis metaphysical nature, distinct from the metaphysics of every other kind of linguistic expression. Such metaphysical exceptionalism would be justified if the Millian view had a clear, uncontested theoretical advantage over its rivals. However, in the context of a semantic debate about names in which the closest competitors of the Millian view—the Predicate view and Indexicalism—do not result in such exceptionalism, it counts as a strike against the Millian view.

I argue that the act-based accounts of propositions, like the one defended by Soames, cannot be used to address Frege’s Puzzle without also giving up the Millian view of names. I begin by identifying two puzzles—both of which have been called Frege’s puzzle—and discuss the act-based theorist’s solution to the first puzzle. I then raise an objection against the solution and argue that it cannot be overcome unless a concession is made. Making the concession, however, would make it impossible for the act-based theorist to solve the second puzzle. I further argue that any attempt to solve the puzzle while also maintaining a commitment to the Millian view will force such theoretical commitments on the act-based theorist which seriously undermine the scope and motivation of the act-based approaches. Thus, any solution to Frege’s puzzle on the act-based account is bound to conflict with the Millian view of names.