Herman Cappelen

This paper presents an argument that certain AI safety measures, rather than mitigating existential risk, may instead exacerbate it. Under certain key assumptions - the inevitability of AI failure, the expected correlation between an AI system's power at the point of failure and the severity of the resulting harm, and the tendency of safety measures to enable AI systems to become more powerful before failing - safety efforts have negative expected utility. The paper examines three response strategies: Optimism, Mitigation, and Holism. Each faces challenges stemming from intrinsic features of the AI safety landscape that we term Bottlenecking, the Perfection Barrier, and Equilibrium Fluctuation. The surprising robustness of the argument forces a re-examination of core assumptions around AI safety and points to several avenues for further research.

On a standard understanding of externalist metasemantic theories, such theories require a speaker to defer to other speakers in order to share content with them. We argue that this standard understanding is mistaken, and that, on a proper understanding of externalism, sharing content does not depend in any way on deference, either to experts, or one’s linguistic community. We defend a version of externalism that we call ‘pure externalism’, and we argue that the idea that shared content requires deference is a residue of internalism/descriptivisism to which externalists ought to be opposed. We also argue that, despite common belief to the contrary, several of the originators of externalism, including Tyler Burge, do not think that deference is in any way relevant, let alone required, for shared externalist content.

This is an open access title available under the terms of a CC BY-NC-ND 4.0 International licence. It is free to read on Oxford Academic and offered as a free PDF download from OUP and selected open access locations. If we don't know what the words 'democracy' and 'democratic' mean, then we don't know what democracy is. This book defends a radical view: these words mean nothing and should be abandoned. The argument for Abolitionism is simple: those terms are defective and we can easily do better, so let's get rid of them. According to the abolitionist, the switch to alternative devices would be a significant communicative, cognitive, and political advance. The first part of the book presents a general theory of abandonment: the conditions under which language should be abandoned. The rest of the book applies this general theory to the case of 'democracy' and 'democratic'. Cappelen shows that 'democracy' and 'democratic' are semantically, pragmatically, and communicatively defective. Abolitionism is not all gloom and doom. It also contains a message of good cheer: we have easy access to conceptual devices that are more effective than 'democracy'. We can do better. These alternative linguistic devices will enable us to ask better questions, provide genuinely fruitful answers, and have more rational discussions. Moreover, those questions and answers better articulate the communicative and cognitive aims of those who use empty terms like 'democracy' and 'democratic'.

Language Turned on Itself examines what happens when language becomes self-reflexive; when language is used to talk about language. Those who think, talk, and write about language are habitual users of various metalinguistic devices, but reliance on these devices begins early: kids are told, 'That's called a "rabbit"'. It's not implausible that a primitive capacity for the meta-linguistic kicks in at the beginning stages of language acquisition. But no matter when or how frequently these devices are invoked, one thing is clear: they present theorists of language with a complex data pattern. Herman Cappelen and Ernest Lepore show that the study of these devices and patterns not only represents an interesting and neglected project in the philosophy of language, but also carries important consequences for other parts of philosophy. Part I is devoted to presenting data about various aspects of our metalinguistic practices. In Part II, the authors examine and reject the four leading metalinguistic theories, and offer a new account of our use of quotation in a variety of different contexts. But the primary goal of this book is not to promote one theory over another. Rather, it is to present a deeply puzzling set of problems and explain their significance.

In Insensitive Semantics (2004), we argue for two theses – Semantic Minimalism and Speech Act Pluralism. In this paper, we outline our defense against two objections often raised against Semantic Minimalism. To get to that defense, we first need some stage setting. To that end, we begin with five stage setting sections. These lead to the first objection, viz., that it might follow from our view that comparative adjectives are context insensitive. We defend our view against that objection (not, as you might expect, by denying that implication, but by endorsing it). Having done so, we address a second objection, viz., that Semantic Minimalism makes it difficult to see what role semantic content plays in communicative exchanges. We respond and end with a reversal, i.e., we argue that even though the second objection fails against us, it works against those who raise the objection. In particular, we show that Recanati ends up with a notion of communicated content that fails various tests for psychological reality.

In our book The Inessential Indexical we argue that the various theses of essential indexicality all fail. Indexicals are not essential, we conclude. One essentiality thesis we target in the third chapter is the claim that indexical attitudes are essential for action. Our strategy is to give examples of what we call impersonal action rationalizations, which explain actions without citing indexical attitudes. To defeat the claim that indexical attitudes are essential for action, it suffices that there could be even one successful impersonal action rationalization. In what follows we bolster our case against an essential connection between action and the de se (or indexicality), first by developing a range of new action models and secondly by responding to challenges from Dilip Ninan, Stephan Torre, and José Luis Bermúdez.

What role does ordinary language play in philosophical theorizing today? One might think: little. After all, analytic philosophy has moved past its “ordinary language” phase; in metaphysics, for example, few would think that attending to “time” and related words has anything to teach us about the nature of, and how we persist through, time. The aim of this paper, however, is to argue that contemporary analytic philosophy pays more attention to ordinary language use than the implicit historiography mentioned above suggests. Moreover, and moving from description to evaluation, the paper argues that this is philosophically defensible. In many areas of philosophy, to learn about a phenomenon we ought to pay at least some attention to how ordinary language users speak about it.

Hva vil det si å ha kunnskap om noe? Hva slags ting finnes i verden, og hva går det an å vite noe om? Hvordan bør vi leve? Hva er verdifullt? I denne boka søkes svar på disse spørsmålene gjennom eksempler fra filosofi- og vitenskapshistorien og fra etikk. Boka har 20 kapitler fordelt på tre deler. Del 1 – vite – stiller spørsmål om kunnskap. Del 2 – være – tar utgangspunkt i hva som er virkelig. Del 3 – gjøre – er viet etikk. Hvert kapittel har en introduksjon samt én til to originaltekster som er skrevet av tenkere fra vår samtid og fra filosofi- og vitenskapshistorien. Hvert kapittel har også studiespørsmål og en opplisting av det du bør ha lært, etter å ha lest kapittelet. [Dette er den eneste godkjent boken som dekker pensum til examen philosophicum ved Universitetet i Oslo]

What does it mean to possess knowledge about something? What kinds of things exist and can be known? How should we live? What matters and what has value? In this book, we seek to answer these questions by examining contributions from the history of philosophy, science and ethics. The book has 20 chapters divided into three parts. Part 1, Knowing, asks questions about knowledge. Part 2, Being, starts by asking about what is real. Part 3, Doing, is devoted to ethics. Every chapter has an introduction and one or in some cases two original texts written either by contemporary thinkers or by important figures from the history of philosophy and science. Each chapter is accompanied by study questions and learning goals. The aim of this book is twofold: to teach you to become a critical reader, who can identify positions, arguments, and structures in a scientific text; and to introduce you to some important positions and ideas in order to enable you to discuss the big questions raised in the book. [Note: This book has been specifically developed as a textbook for the so-called 'Examen Philosophicum" course; it exists also in Norwegian as 'Vite, være, gjøre']

Can humans and artificial intelligences share concepts and communicate? Making AI Intelligible shows that philosophical work on the metaphysics of meaning can help answer these questions. Herman Cappelen and Josh Dever use the externalist tradition in philosophy to create models of how AIs and humans can understand each other. In doing so, they illustrate ways in which that philosophical tradition can be improved. The questions addressed in the book are not only theoretically interesting, but the answers have pressing practical implications. Many important decisions about human life are now influenced by AI. In giving that power to AI, we presuppose that AIs can track features of the world that we care about (for example, creditworthiness, recidivism, cancer, and combatants). If AIs can share our concepts, that will go some way towards justifying this reliance on AI. This ground-breaking study offers insight into how to take some first steps towards achieving Interpretable AI.

This paper examines two attempts to justify the way in which intuitions about specific cases are used as evidence for and against philosophical theories. According to the concept model, intuitions about cases are trustworthy applications of one’s typically tacit grasp of certain concepts. We argue that regardless of whether externalist or internalist accounts of conceptual content are correct, the concept model flounders. The second justification rests on the less familiar belief model, which has it that intuitions in philosophy derive from one’s beliefs. Although more promising than the concept model, the belief model fails to justify traditional philosophical use of intuitions because it is not clear a priori that the beliefs at issue are true. The latter model may, however, legitimize a less a prioristic approach to intuitions.

I call the activity of assessing and developing improvements of our representational devices ‘conceptual engineering’.¹ The aim of this chapter is to present an argument for why conceptual engineering is important for all parts of philosophy (and, more generally, all inquiry). Section I of the chapter provides some background and defines key terms. Section II presents the argument. Section III responds to seven objections. The replies also serve to develop the argument and clarify what conceptual engineering is.

Conceptual engineering and conceptual ethics are branches of philosophy concerned with questions about how to assess and ameliorate our representational devices (such as concepts and words). It's a part of philosophy concerned with questions about which concepts we should use (and why), how concepts can be improved, when concepts should be abandoned, and how proposals for amelioration can be implemented. Central parts of the history of philosophy have engaged with these issues, but the focus of this volume is on applications to work in contemporary philosophy of language and mind, epistemology, gender and race theory, ethics, philosophy of science, and philosophical logic. This is the first volume devoted entirely to conceptual engineering and conceptual ethics. The volume explores the possibilities, benefits, problems, and applications of conceptual engineering and conceptual ethics. It consists of twenty chapters written by leading philosophers.

Conceptual engineering is now a central topic in contemporary philosophy. Just 4-5 years ago it wasn’t. People were then engaged in the engineering of various philosophical concepts (in various sub-disciplines), but typically not self-consciously so. Qua philosophical method, conceptual engineering was under-explored, often ignored, and poorly understood. In my lifetime, I have never seen interest in a philosophical topic grow with such explosive intensity. The sociology behind this is fascinating and no doubt immensely complex (and an excellent case study for those interested in the dynamics of academic disciplines). That topic, however, will have to wait for another occasion. Suffice it to say that if Fixing Language (FL) contributed even a little bit to this change of focus in philosophical methodology, it would have achieved one of its central goals. In that connection, it is encouraging that the papers in this symposium are in fundamental agreement about the significance and centrality of conceptual engineering to philosophy. That said, the goal of FL was not only to advocate for a topic, but also to defend a particular approach to it: The Austerity Framework. These replies have helped me see clearer the limitations of that view and points where my presentation was suboptimal. The responses below are in part a reconstruction of what I had in mind while writing the book and in part an effort to ameliorate. I’m grateful to the symposiasts for helping me get a better grip on these very hard issues.

This paper assembles examples and considerations bearing on such questions as the following. Are statements to the effect that someone is too young or that someone is old enough always to be understood in terms of someone's being too young or too old for such-and-such-for example, for them to join a particular organization? And when a 'such-and-such' has been specified, is it always at least tacitly modal in force-in the case just given, too young or old enough to be able to join the organization? These questions are explored by means of a critical examination of the answers given to them by Eric Nelson in a 1980 paper on the subject, with part of the intention being to rescue Nelson's thoughtful discussion from the oblivion into which it appears to have fallen, judging by more recent contributions on the subject by semanticists.

Paul Saka, in a recent paper, declares that we can use, mention, or quote an expression. Whether a speaker is using or mentioning an expression, on a given occasion, depends on his intentions. An exhibited expression is used, if the exhibiter intends to direct his audience’s attention to the expression’s extension. It is mentioned, if he intends to draw his audience’s attention to something associated with the exhibited token other than its extension. This includes, but is not limited to, an orthographic form, a phonic form, a lexical entry, and an intension.

This paper evaluates arguments presented by John Perry (and Ken Taylor) in favor of the presence of an unarticulated constituent in the proposition expressed by utterance of, for example, (1):1 1. It's raining (at t). We contend that these arguments are, at best, inconclusive. That's the critical part of our paper. On the positive side, we argue that (1) has as its semantic content the proposition that it is raining (at t) and that this is a location-neutral proposition. According to the view we propose, an audience typically looks for a location when they hear utterances of (1) because their interests in rain are location- focused: it is the location of rain that determines whether we get wet, carrots grow, and roads become slippery. These are, however, contingent facts about rain, wetness, people, carrots, and roads – they are not built into the semantics for the verb 'rain'.

There are at least four varieties of quotation, including pure, direct, indirect and mixed. A theory of quotation, we argue, should give a unified account of these varieties of quotation. Mixed quotes such as 'Alice said that life is 'difficult to understand'', in which an utterance is directly and indirectly quoted concurrently, is an often overlooked variety of quotation. We show that the leading theories of pure, direct, and indirect quotation are unable to account for mixed quotation and therefore unable to provide a unified theory. In the second half of the paper we develop a unified theory of quotation based on Davidson's demonstrative theory. 'Language is the instrument it is because the same expression, with semantic features (meaning) unchanged, can serve countless purposes.' (Davidson 1968).

One of the three central issues in Lloyd Humberstone's ‘Sufficiency and Excess’ is what he calls ‘the Complete Thought Issue’ (CTI, for short). This is the question of whether some declarative sentences have proposition radicals, rather than full-blown propositions, as their semantic values. My focus in this reply is exclusively on Humberstone's comments about CTI and on CTI more generally. The goal of Humberstone's discussion of CTI is to defend ‘[Kent] Bach's claim against Cappelen and Lepore's attacks’ (Humberstone, 2006, p. 316). These ‘attacks’ on Bach are found in Cappelen and Lepore (2004). In section one I present CTI, in section two I evaluate Humberstone's defence of Bach, and in section three I discuss two solutions to CTI.

Much work in the philosophy of language assumes that a semantic theory T, for a language L should assign p as the semantic content of an utterance u, by A, of a sentence S in L, if and only if “A said that p” is true. This assumption is mistaken. More generally, the aim of semantics cannot be to capture the extension of English expressions such as “meaning” or “what was said”. This provides support for Davidson’s paratactic theory of indirect speech and for the view that a semantic theory should take the form of a truth-theory.

One of Szabo's central objections is his ‘reservations about the alleged slide from moderate to radical contextualism’. First, some background: the argument Szabo expresses doubt about is essential both to the critical part of our book and to its positive part. Our argument against what we call moderate contextualism depends on the assumption that it collapses into radical contextualism. Our positive view depends on the assumption that for any utterance, we can trigger the intuition that many different propositions are said (this is at the core of our speech act pluralism). So Szabo is attacking one of the key assumptions in IS – if his 'reservations' are well justified, most of what we have to say collapses.