Alessio Tacca

The growing integration of Artificial Intelligence (AI) systems into knowledge production, decision-making, and everyday activities has revamped debates about the nature and scope of trust and reliance. The standard view is to distinguish trust from reliance. While trust is an agent-directed attitude involving ethical and normative aspects, reliance does not require them and can also be directed at objects and processes. According to agent-directed accounts of trust, since AI lacks agency, it cannot be trusted. This article examines three alternative frameworks challenging agent-directed accounts of trust and supporting trust in AI. These frameworks better reflect human-AI interactions, align with everyday language, and fit current AI policy discussions. However, some argue, they overlook the core ethical features of trust. While standard views protect these features, they misrepresent actual user engagement with AI. The problem of whether we can truly trust AI systems or merely rely on them remains a central open problem in AI epistemology, with important policy implications.

LLMs’ exceptional capabilities of language translation, sentence completion, sentiment analysis, information retrieval, and much more make them useful in a vast array of contexts, from customer service to scientific research. With these great technological possibilities come many unanswered questions and several concerns. We focus on the use of LLMs in epistemic contexts, broadly construed, meaning those contexts in which an LLM is used by a subject in order to fulfil their epistemic goals, such as acquiring justification, increasing their knowledge or understanding or developing argumentative work. By examining the epistemic context of LLMs, we aim to give an account of some of the epistemological risks associated with their usage.

From epileptic seizures to depressive symptoms, predictive neurotechnologies are used for a large range of applications. In this article we focus on advisory devices; namely, predictive neurotechnology programmed to detect specific neural events (e.g., epileptic seizure) and advise users to take necessary steps to reduce or avoid the impact of the forecasted neuroevent. Receiving advise from a predictive device is not without ethical concerns. The problem with predictive neural devices, in particular advisory ones, is the risk of seeing one’s autonomous choice supplanted by the predictions instead of being supplemented by it. For users, there is a potential shift from being assisted by the system to being over-dependent on the technology. In other terms, it introduces ethical issues associated with epistemic dependency. In this article, we examine the notion of epistemic authority in relation to predictive neurotechnologies. Section 1 of our article explores and defines the concept of epistemic authority. In section 2, we illustrate how predictive devices are best conceived of as epistemic authorities and we explore the subject-device epistemic relationship. In section 3, we spell out the risk of harms interconnected with epistemic deferral. We conclude by stressing a set of preliminary measures to prepare users for the authoritative nature of predictive devices.