Action theory offers well-developed treatments of tool use and of delegation, but agentic AI belongs to neither category. The user specifies an end; the system autonomously selects the means. What, then, secures the connection between the user's intention and outcome that distinguishes intentional action from lucky coincidence when things go right? I examine five candidate approaches - Davidsonian minimalism, principal-agent decomposition, strong practical-knowledge views, permissive intention-plus-success accounts, and extended-agency frameworks - and argue that each draws the line in the wrong place, variously under-crediting supervised use, over-demanding means-level transparency, erasing the competence/luck distinction, or inheriting the bloat and glue problems that plague extended-mind theories. The positive proposal then relocates non-deviance from route-identity to governance-stability. On the know-how account defended here, a user intentionally phi-s through agentic AI when her success manifests the exercise of supervisory know-how - viz., the competence to define an 'acceptable corridor' of outcomes and means, to monitor for departures from that corridor, and to intervene or abort when drift threatens. This framework tolerates the opacity and route-variability characteristic of agentic AI, explains graded attributions across interaction profiles (in/on/out-of-the-loop), and remains neutral on whether the AI itself qualifies as an agent. It thereby vindicates, under specifiable conditions, the ordinary human claim 'I phi-d through an AI agent' as tracking creditworthy governance.

Virtue epistemology has long struggled with the "Creditability Dilemma": how can knowledge gained through deference be creditable to the knower if it primarily depends on others' cognitive work? We propose a novel solution by developing a telic account of doxastic deference as a distinctive kind of social-epistemic performance. On our view, such deference succeeds when a deferrer forms a true belief that p in domain d, which answers their query, on the basis of the fact that a deferee states that p because of their epistemically superior competence in d. Drawing on Sosa's framework of performance assessment, we identify a metaphysical hierarchy of increasingly creditable achievements in deferential belief-formation, from mere accurate doxastic deferral through to fully apt doxastic deferral. This hierarchy reveals that the apparent tension between individual credit and epistemic dependence rests on an "inverse creditability" thesis which fails to apply in cases of deferential belief. Instead, the highest forms of deferential achievement often align with and track the cognitive achievements of those deferred to. This framework also illuminates expert identification and epistemic trespassing as problems of exercising deferential skill under suitably normal conditions.

I reply here to five excellent commentaries on my book, Autonomous Knowledge: Radical Enhancement, Autonomy, and the Future of Knowing (Carter, J. A. 2022. Autonomous Knowledge: Radical Enhancement, Autonomy, and the Future of Knowing. Oxford: Oxford University Press). In what follows, I address objections and constructive proposals from Gloria Andrada, Kevin McCain, Jes & uacute;s Vega-Encabo, Duncan Pritchard, and Nikolaj Jang Lee Linding Pedersen & Peter J. Graham. Topics range from the application of autonomy to know-how and cognitive development, the resilience of evidentialism and virtue epistemology against non-autonomous enhancement cases, the distinction between positive and negative epistemic autonomy, and the axiological foundations of autonomous belief and self-constitution.

This paper develops and argues for a virtue epistemology that includes performance-normative evaluations of interrogative attitudes (IAs) and not just beliefs. In this way, it brings virtue epistemology to interrogative epistemology. The motivating thought behind our proposal is an analogy: as the semantic property of truth is to answering attitudes, so the semantic property of soundness (the property questions have when they admit of true, direct answers) is to questioning attitudes. With this notion of basic success for IAs in hand, we go on to develop notions of competence and aptness for IAs. Roughly, our view says that (1) IAs are competent when they manifest (rightly situated) dispositions to form IAs in sound questions reliably enough, and (2) IAs are apt when their being directed towards a sound question manifests the relevant competence. A complete virtue epistemology that treats both beliefs and IAs is, we argue, better positioned to account for the way that success in questioning sets up yet further success in answering. Moreover, our complete virtue epistemology generates evaluative norms on IAs not yet represented in the emerging zetetic literature.

Clowes, G & auml;rtner, and Hip & oacute;lito (2021) describe the Mind-Technology Problem as a new constellation of philosophical problems about the nature of mind generated by advances in technology we increasingly rely on to meet both theoretical and practical aims. We agree with Clowes, G & auml;rtner, and Hip & oacute;lito that the problems they identify frame a timely and worthwhile new research programme. We aim to contribute to this research programme by motivating and canvassing the key contours of four different Mind-Technology problems that arise for, and which are distinctive to, approaches to knowledge-how that give abilities a central theoretical role in explaining intelligent states and behaviour. The result is an agenda in the theory of know-how aimed at reconciling prevailing anti-intellectualist thinking about know-how apt abilities and their exercise with the datum that much of our know-how is increasingly supported by our use of smart technological scaffolding. We don't fully answer the questions we raise; our objective is to put them on the map and show why we think they matter.

Moore and Russell thought that perceptual knowledge of the external world is based on abductive inference from information about our experience. Sosa maintains that this ‘indirect realist’ strategy has no prospects of working. Vogel disagrees and thinks it can and does work perfectly well, and his reasoning (and variations on that reasoning) seem initially promising, moreso than other approaches. My aim, however, will be to adjudicate this dispute in favor of Sosa’s pessimistic answer, and in doing so, to better uncover the important role abductive inference does have in a wider theory of perceptual knowledge, even if it doesn’t feature in any promising vindication of (anti-skeptical) indirect realism.

Abstract Implicit and deliberative trust differ, but under what conditions is one kind of trust more appropriate than the other, and what kinds of considerations determine this? This chapter takes up this question, and in doing so, distinguishes between two distinct though complementary domains of normativity of interest in the philosophy of trust. The guiding analogy this chapter develops is that telic assessment of beliefs stands to intellectual ethics as the telic assessment of trust stands to the ethics of cooperation.

Abstract A good knife is a sharp knife. In virtue of what is trust good, as such, when it is? Standard accounts of the nature of trust suggest that good trusting ought to be assimilated in some way to good believing, or to good affect, or to good conation. This chapter raises doubts for all three of these main answer types, and the criticisms given then set the scene for motivating (in Chapter 2) a new performance-theoretic approach to trust and its evaluative normativity, which has the resources to do better.

We explore a cluster of epistemic vices that lead us to refrain from trusting when we should.

Abstract We’ve up to this point seen what convictively apt trust demands (Chapter 5); this chapter explores what it permits. Our guiding question is: What kind of risks to the first-order aptness of trust can the convictively apt trustor non-negligently ignore? An answer inspired by Ernest Sosa’s answer to a generalized version of this question is canvassed and criticized, and a different answer—one that gives de minimis risk a central place—is developed and defended.

Abstract This concluding chapter briefly summarizes the key contours of the telic theory of trust as it has been developed across Chapters 1–9; a short list of research topics and questions, both theoretical (§2.1) and applied (§2.2), is then outlined as an agenda for further work.

Knowledge, like other things of value, can be faked. According to Hawley (2011), know-how is harder to fake than knowledge-that, given that merely apparent propositional knowledge is in general more resilient to our attempts at successful detection than are corresponding attempts to fake know-how. While Hawley's reasoning for a kind of detection resilience asymmetry between know-how and know-that looks initially plausible, it should ultimately be resisted. In showing why, we outline different ways in which know-how can be faked even when a given performance is successful, and in doing so, we distinguish how know-how can be faked (no less than know-that) via upstream and downstream indicators of its presence, and within each of these categories, we'll distinguish (in connection with detection resilience) both faking symptoms and (various kinds of) criteria. The unappreciated resilience of faked knowledge-how to successful detection highlights a largely overlooked dimension of social-epistemic risk--risk we face not just in our capacity as recipients of testimony, but in our capacity as both (would-be) apprentices and clients of knowledge-how.