Luciano Floridi

This article examines the nature of reasoning in current, mainstream Large Language Models (LLMs) that operate within the token-completion paradigm. We explore their stochastic foundations and phenomenological resemblance to human abductive reasoning. We argue that such LLMs generate text based on learned associations rather than performing abductive inferences. When their output exhibits an apparent abductive quality-often reinforced by interface design-this effect is due to the model's training on human-generated texts that encode reasoning structures. We use some examples to illustrate how LLMs produce plausible hypotheses, simulate commonsense reasoning, and provide explanatory answers without grounding them directly in truth, semantics, verification, or understanding, and without any abductive reasoning. This duality, centred on the stochastic core of the models and the abductive appearance of the applications, has important implications for the evaluation and use of LLMs. They can help generate hypotheses and support human reasoning, but their outputs must be critically examined because they cannot discern truth or verify explanations. In the conclusion, we address five objections to our assertions, some limitations of our analysis, and provide a general assessment.

This article introduces a systematic methodology, called red reading, for critical textual analysis that adapts red teaming techniques from computer science and artificial intelligence (AI) into a comprehensive diagnostic protocol, supported by Large Language Models (LLMs), to analyse written documents. As the methodological counterpart of distant writing, which transforms text production through AI-assisted design, red reading expands human analytical capabilities and critical reading by embedding adversarial analysis in a constructive framework powered by LLMs. The approach employs a rigorously tested, diagnostic protocol that stresstests texts across structural, rhetorical, logical, and stylistic dimensions, while strengthening interpretive understanding. The methodology shows how LLM-supported, textual red teaming can provide a rigorous, critical and constructive approach to textual analysis with broad applications across disciplines. We exemplify the method by developing a fifteen-phase diagnostic prompt for red reading. In the Appendix, we apply this protocol to a case study analysing Turing's seminal article, Computing Machinery and Intelligence. We finally discuss the challenges of using red reading and provide some suggestions for future research. In particular, we emphasise the importance of calibration methods for LLMs' critical analysis, along with evidence-based strategies for seamlessly integrating red reading into real-world workflows.

Brain-computer interfaces (BCIs) offer significant therapeutic opportunities for a variety of neurophysiological and neuropsychiatric disorders and may perhaps one day lead to augmenting the cognition and decision-making of the healthy brain. However, existing regulatory frameworks designed for implantable medical devices (IMDs) are inadequate to address the unique ethical, legal, and social risks associated with next-generation networked brain-computer interfaces (BCIs). In this article, we make nine recommendations to support developers in the design of BCIs and nine recommendations to support policymakers in the application of BCIs, drawing insights from the regulatory history of IMDs and principles from AI ethics. We begin by outlining the historical development of IMDs and the regulatory milestones that have shaped their oversight. Next, we summarize similarities between IMDs and emerging implantable BCIs, identifying existing provisions for their regulation. We then use two case studies of emerging cutting-edge BCIs-the HALO (Hardware Architecture for LOw-power BCIs) and SCALO (SCalable Architecture for LOw-power BCIs) computer systems-to highlight distinctive features in the design and application of nextgeneration BCIs arising from contemporary chip architectures, which necessitate reevaluating regulatory approaches. We identify critical ethical considerations for these BCIs, including unique conceptions of autonomy, identity, and mental privacy. Based on these insights, we suggest potential avenues for the ethical regulation of BCIs, emphasizing the importance of interdisciplinary collaboration and proactive mitigation of potential harms. The goal is to support the responsible design and application of new BCIs, ensuring their safe and ethical integration into medical practice.

Political parties strive to understand their electorates, and focus groups are a vital tool in these efforts. AI-enhanced simulation technologies (AESTs) enable synthetic focus groups in a fraction of the time (and cost), raising the question of when and how such simulated evidence can be used in campaign research. This paper develops a decision matrix to help party strategists match research needs to appropriate simulation technologies and to identify when to escalate to hybrid or fully human focus groups. The matrix combines three dimensions: strategic purpose, deployment risk, and empirical grounding of the simulation tool. Strategic purpose is the decisive dimension, as it determines what kind of evidence the focus group is meant to produce: observing how political meanings and identities emerge through interaction (Mode 1) or testing and refining campaign messages (Mode 2). The matrix shows that, given documented failure modes such as sycophancy, persona drift, and the suppression of minority viewpoints, AESTs cannot replace human interaction in Mode 1 at any risk level. Within Mode 2, suitability depends instead on deployment risk and on the empirical grounding. Yet even here, we caution that routine reliance on AESTs may erode the qualitative craft on which sound judgment depends.

Artificial intelligence (AI) assurance is an umbrella term describing many approaches—such as impact assessment, audit, and certification procedures—used to provide evidence that an AI system is legal, ethical, and technically robust. AI assurance approaches largely focus on two overlapping categories of harms: deployment harms that emerge at, or after, the point of use, and individual harms that directly impact a person as an individual. Current approaches generally overlook upstream collective and societal harms associated with the development of systems, such as resource extraction and processing, exploitative labour practices and energy intensive model training. Thus, the scope of current AI assurance practice is insufficient for ensuring that AI is ethical in a holistic sense, i.e. in ways that are legally permissible, socially acceptable, economically viable and environmentally sustainable. This article addresses this shortcoming by arguing for a broader approach to AI assurance that is sensitive to the full scope of AI development and deployment harms. To do so, the article maps harms related to AI and highlights three examples of harmful practices that occur upstream in the AI supply chain and impact the environment, labour, and data exploitation. It then reviews assurance mechanisms used in adjacent industries to mitigate similar harms, evaluating their strengths, weaknesses, and how effectively they are being applied to AI. Finally, it provides recommendations as to how a broader approach to AI assurance can be implemented to mitigate harms more effectively across the whole AI supply chain.

Much of the hype about AI exploits the fact that human cognition, language, and categorization rely on imaginative scaffolding. Some degree of metaphor and conceptual borrowing should not be seen as a methodological failure. Law does not meet AI as a blank slate. It reaches for metaphors and other imaginative patterns that make the unfamiliar legible, channeling attention toward certain features and away from others. These are not superficial rhetorical devices, but imaginative patterns that do real work: They frame risks, allocate authority, and steer doctrinal development. The paper uses cognitive linguistics to trace how imaginative patterns structure the regulatory agenda and create path dependence. It concentrates on three especially influential conceptions that circulate across technical, legal, and public discourse: (artificial) intelligence, black box, and hallucinations. “Intelligence” invites personification and prototype effects that lead to misplaced assumptions about agency, even though what is actually being produced is best described as a simulation of a much narrower capacity associated with intelligence: linguistic competence. “Black box” turns a distributed sociotechnical system into a metaphoric bounded container, misleadingly focusing concerns about transparency and responsibility on the model abstracted from its development and implementation. “Hallucinations” imports from psycho-pathology an inappropriate frame which obscures the fact that many errors are foreseeable products of intentional choices. We suggest, instead, reframing them as deferential hazards that arise from design features and expected interactional dynamics.

Artificial intelligence’s impact on healthcare is undeniable. What is less clear is whether it will be ethically justifiable. Just as we know that AI can be used to diagnose disease, predict risk, develop personalized treatment plans, monitor patients remotely, or automate triage, we also know that it can pose significant threats to patient safety and the reliability (or trustworthiness) of the healthcare sector as a whole. These ethical risks arise from (a) flaws in the evidence base of healthcare AI (epistemic concerns); (b) the potential of AI to transform fundamentally the meaning of health, the nature of healthcare, and the practice of medicine (normative concerns); and (c) the “black box” nature of the AI development pipeline, which undermines the effectiveness of existing accountability mechanisms (traceability concerns). In this chapter, we systematically map (a)–(c) to six different levels of abstraction: individual, interpersonal, group, institutional, sectoral, and societal. The aim is to help policymakers, regulators, and other high-level stakeholders delineate the scope of regulation and other “softer” governing measures for AI in healthcare. We hope that by doing so, we may enable global healthcare systems to capitalize safely and reliably on the many benefits of healthcare AI.

A compilation of cutting-edge, comprehensive insights into digital ethics from leading scholars As digital technologies shape every aspect of today's society, ethical considerations have never been more pressing. In A Companion to Digital Ethics, editors Luciano Floridi and Mariarosaria Taddeo bring together leading experts to analyse key ethical challenges posed by artificial intelligence, privacy, cybersecurity, cyberwarfare, sustainability, digital consent, and many other topics. With a multidisciplinary approach, this authoritative volume introduces all the relevant topics in digital ethics clearly and accessibly, integrating insights from philosophy, law, computer science, and the social sciences. Unlike other collections that focus on specialised research or introductory overviews, this Companion strikes a critical balance—offering a systematic, comprehensive, simple yet rigorous examination of digital ethics that is both conceptually rich and practically relevant. Each chapter can be read as part of the whole or independently. Throughout the book, the contributing authors equip the reader with diverse perspectives for navigating ethical dilemmas in the evolving digital landscape, with an essential list of references and further readings. Integrating real-world examples and case studies throughout, A Companion to Digital Ethics: Explores the ethical value and implications of digital technologies and their applications Offers clear and precise analysis of both current and potential ethical challenges Includes discussions of ethical issues in contexts of technology governance and digital law Helps professionals and policymakers implement ethical principles in AI and digital technologies Contains authoritative and cutting-edge chapters, edited by two leading experts in digital ethics Whether used in academic courses or professional discussions on responsible digital innovation, A Companion to Digital Ethics is an essential guide to understanding and addressing the ethical questions of the digital age. It is a must-have resource for undergraduate and postgraduate students in any academic discipline, as well as policymakers and practitioners seeking to implement ethical frameworks for emerging technologies.

What is the ultimate nature of reality? This paper defends an answer in terms of informational realism (IR). It does so in three stages. First, it is shown that, within the debate about structural realism (SR), epistemic (ESR) and ontic (OSR) structural realism are reconcilable by using the methodology of the levels of abstractions. It follows that OSR is defensible from a structuralist-friendly position. Second, it is argued that OSR is also plausible, because not all related objects are logically prior to all relational structures. The relation of difference is at least as fundamental as (because constitutive of) any relata. Third, it is suggested that an ontology of structural objects for OSR can reasonably be developed in terms of informational objects, and that Object Oriented Programming provides a flexible and powerful methodology with which to clarify and make precise the concept of “informational object”. The outcome is informational realism, the view that the world is the totality of informational objects dynamically interacting with each other..

The article introduces the concept of “semantic pareidolia” - our tendency to attribute consciousness, intelligence, and emotions to AI systems that lack these qualities. It examines how this psychological phenomenon leads us to perceive meaning and intentionality in statistical pattern-matching systems, similar to seeing faces in clouds. It analyses the converging forces intensifying this tendency: increasing digital immersion, profit-driven corporate interests, social isolation, and AI advancement. The article warns of progression from harmless anthropomorphism to problematic AI idolatry, and calls for responsible design practices that help users maintain critical distinctions between simulation and genuine consciousness. It is the English translation and adaptation of an article originally published in Italian in Harvard Business Review Italia, June 2025.

This article introduces the concept of “distant writing”, a novel literary practice in which authors act as designers, employing Large Language Models (LLMs) to generate narratives, while retaining creative control through precise prompting and iterative refinement. Unlike Moretti’s distant reading, which uses computational analysis to interpret large corpora of existing texts, distant writing harnesses computational tools (LLMs) to author new narratives, reshaping the literary production process. By examining theoretical frameworks and practical consequences, and running an experiment in distant writing called Encounters, this article argues that distant writing represents a significant evolution in authorship, not replacing but expanding human creativity within a design paradigm. The distinction between writing (close) and “wrAIting” (distant) reveals how LLM-assisted creativity can generate narrative possibilities previously inaccessible, transforming literature’s modal space while challenging traditional notions of authorship, creativity, and literary production. This emerging practice merits critical attention as it shapes future literary landscapes and reconfigures relationships between human creativity and artificial intelligence.

The article analyses the integration of Large Language Model (LLM)-based interfaces (editorial LLMs or eLLMs) in scholarly publishing workflows, focusing specifically on their growing role in editorial screening, manuscript preparation, and peer-review processes. It assesses the benefits eLLMs offer, including efficiency gains, improved compliance with journal guidelines, enhanced objectivity, and reduced editorial workload; and the risks, especially algorithmic biases, false positives and negatives, data privacy concerns, and potential opacity in automated decision-making. The article then offers some design recommendations for eLLMs that prioritise transparency, fairness, and user-centredness, ensuring human oversight remains integral to the editorial process. It concludes by encouraging a proactive and thoughtful engagement with these technologies to enhance scholarly publishing rather than undermine its values.

This article analyses the epistemological asymmetry between remembering and forgetting Good and Evil and their ethical implications. While philosophy has traditionally focused on the metaphysical asymmetry (the privation theory that Evil exists only as an absence of Good), this article addresses the uneven moral distribution arising from memory practices: remembering Good is morally good, whereas forgetting Good, forgetting Evil, and remembering Evil may each bear morally negative implications. The article argues that this asymmetry should be rectified through differentiated memory practices: celebration for Good and closure for Evil. Celebration involves active commemoration that perpetuates Good’s presence, while closure enables remembering Evil without causing further Evil. The article then explores practical applications of this framework to physical memory artefacts, including monuments and “cancel culture” debates, and digital memory technologies, arguing that current designs often fail to differentiate adequately between these types of “moral memory”. It concludes by suggesting some principles and requirements for designing ethical memory architectures that properly support celebration and closure, contributing to improved human flourishing through more nuanced approaches to remembering the past.

This article examines potential regulatory pathways for AI in the United States following the Trump administration's 2025 revocation of the Biden-era AI Executive Order. We outline two competing governance scenarios: decentralized state-level regulation (with minimal federal oversight) and centralized federal dominance (through legislative pre-emption). We critically evaluate each model's policy implications, constitutional challenges, and practical trade-offs, particularly regarding innovation and state autonomy. We argue that AI's technological characteristics and context-dependent nature complicate achieving regulatory coherence amid competing federal and state interests. As a result, even under the Trump administration's broader deregulatory agenda, targeted federal intervention may remain necessary.

This work fulfills the need for a conceptual and technical framework to improve understanding of Information Quality (IQ) and Information Quality standards. The meaning and practical implementation of IQ are addressed, as it is relevant to any field where there is a need to handle data and issues such as accessibility, accuracy, completeness, currency, integrity, reliability, timeliness, usability, the role of metrics and so forth are all a part of Information Quality. In order to support the cross-fertilization of theory and practice, the latest research is presented in this book. The perspectives of experts from beyond the origins of IQ in computer science are included: library and information science practitioners and academics, philosophers of information, of engineering and technology, and of science are all contributors to this volume. The chapters in this volume are based on the work of a collaborative research project involving the Arts and Humanities Research Council and Google and led by Professor Luciano Floridi, University of Oxford. This work will be of interest to anyone handling data, including those from commercial, public, governmental and academic organizations. The expert editors’ contributions introduce issues of interest to scientists, database curators and philosophers, even though the issues may be disguised in the language and examples common to a different discipline.

This theme issue has the founding ambition of landscaping Data Ethics as a new branch of ethics that studies and evaluates moral problems related to data (including generation, recording, curation, processing, dissemination, sharing, and use), algorithms (including AI, artificial agents, machine learning, and robots), and corresponding practices (including responsible innovation, programming, hacking, and professional codes), in order to formulate and support morally good solutions (e.g. right conducts or right values). Data Ethics builds on the foundation provided by Computer and Information Ethics but, at the same time, it refines the approach endorsed so far in this research field, by shifting the Level of Abstraction of ethical enquiries, from being information-centric to being data-centric. This shift brings into focus the different moral dimensions of all kinds of data, even the data that never translate directly into information but can be used to support actions or generate behaviours, for example. It highlights the need for ethical analyses to concentrate on the content and nature of computational operations — the interactions among hardware, software, and data — rather than on the variety of digital technologies that enables them. And it emphasises the complexity of the ethical challenges posed by Data Science. Because of such complexity, Data Ethics should be developed from the start as a macroethics, that is, as an overall framework that avoids narrow, ad hoc approaches and addresses the ethical impact and implications of Data Science and its applications within a consistent, holistic, and inclusive framework. Only as a macroethics Data Ethics will provide the solutions that can maximise the value of Data Science for our societies, for all of us, and for our environments.

The modern abundance and prominence of data have led to the development of “data science” as a new field of enquiry, along with a body of epistemological reflections upon its foundations, methods, and consequences. This article provides a systematic analysis and critical review of significant open problems and debates in the epistemology of data science. We propose a partition of the epistemology of data science into the following five domains: (i) the constitution of data science; (ii) the kind of enquiry that it identifies; (iii) the kinds of knowledge that data science generates; (iv) the nature and epistemological significance of “black box” problems; and (v) the relationship between data science and the philosophy of science more generally.

The article discusses the process of “conceptual borrowing”, according to which, when a new discipline emerges, it develops its technical vocabulary also by appropriating terms from other neighbouring disciplines. The phenomenon is likened to Carl Schmitt’s observation that modern political concepts have theological roots. The authors argue that, through extensive conceptual borrowing, AI has ended up describing computers anthropomorphically, as computational brains with psychological properties, while brain and cognitive sciences have ended up describing brains and minds computationally and informationally, as biological computers. The crosswiring between the technical languages of these disciplines is not merely metaphorical but can lead to confusion, and damaging assumptions and consequences. The article ends on an optimistic note about the self-adjusting nature of technical meanings in language and the ability to leave misleading conceptual baggage behind when confronted with advancement in understanding and factual knowledge.

International regulation of autonomous weapon systems (AWS) is increasingly conceived as an exercise in risk management. This requires a shared approach for assessing the risks of AWS. This paper presents a structured approach to risk assessment and regulation for AWS, adapting a qualitative framework inspired by the Intergovernmental Panel on Climate Change (IPCC). It examines the interactions among key risk factors—determinants, drivers, and types—to evaluate the risk magnitude of AWS and establish risk tolerance thresholds through a risk matrix informed by background knowledge of event likelihood and severity. Further, it proposes a methodology to assess community risk appetite, emphasizing that such assessments and resulting tolerance levels should be determined through deliberation in a multistakeholder forum. The paper highlights the complexities of applying risk-based regulations to AWS internationally, particularly the challenge of defining a global community for risk assessment and regulatory legitimization.

In information societies, operations, decisions and choices previously left to humans are increasingly delegated to algorithms, which may advise, if not decide, about how data should be interpreted and what actions should be taken as a result. More and more often, algorithms mediate social processes, business transactions, governmental decisions, and how we perceive, understand, and interact among ourselves and with the environment. Gaps between the design and operation of algorithms and our understanding of their ethical implications can have severe consequences affecting individuals as well as groups and whole societies. This paper makes three contributions to clarify the ethical importance of algorithmic mediation. It provides a prescriptive map to organise the debate. It reviews the current discussion of ethical aspects of algorithms. And it assesses the available literature in order to identify areas requiring further work to develop the ethics of algorithms.

Research on the ethics of algorithms has grown substantially over the past decade. Alongside the exponential development and application of machine learning algorithms, new ethical problems and solutions relating to their ubiquitous use in society have been proposed. This article builds on a review of the ethics of algorithms published in 2016, 2016). The goals are to contribute to the debate on the identification and analysis of the ethical implications of algorithms, to provide an updated analysis of epistemic and normative concerns, and to offer actionable guidance for the governance of the design, development and deployment of algorithms.

Accountability is a cornerstone of the governance of artificial intelligence (AI). However, it is often defined too imprecisely because its multifaceted nature and the sociotechnical structure of AI systems imply a variety of values, practices, and measures to which accountability in AI can refer. We address this lack of clarity by defining accountability in terms of answerability, identifying three conditions of possibility (authority recognition, interrogation, and limitation of power), and an architecture of seven features (context, range, agent, forum, standards, process, and implications). We analyse this architecture through four accountability goals (compliance, report, oversight, and enforcement). We argue that these goals are often complementary and that policy-makers emphasise or prioritise some over others depending on the proactive or reactive use of accountability and the missions of AI governance.

What is the most general common set of attributes that characterises something as intrinsically valuable and hence as subject to some moral respect, and without which something would rightly be considered intrinsically worthless or even positively unworthy and therefore rightly to be disrespected in itself? This paper develops and supports the thesis that the minimal condition of possibility of an entity's least intrinsic value is to be identified with its ontological status as an information object. All entities, even when interpreted as only clusters of information, still have a minimal moral worth qua information objects and so may deserve to be respected. The paper is organised into four main sections. Section 1 models moral action as an information system using the object-oriented programming methodology (OOP). Section 2 addresses the question of what role the several components constituting the moral system can have in an ethical analysis. If they can play only an instrumental role, then Computer Ethics (CE) is probably bound to remain at most a practical, field-dependent, applied or professional ethics. However, Computer Ethics can give rise to a macroethical approach, namely Information Ethics (IE), if one can show that ethical concern should be extended to include not only human, animal or biological entities, but also information objects. The following two sections show how this minimalist level of analysis can be achieved. Section 3 provides an axiological analysis of information objects. It criticises the Kantian approach to the concept of intrinsic value and shows that it can be improved by using the methodology introduced in the first section. The solution of the Kantian problem prompts the reformulation of the key question concerning the moral worth of an entity: what is the intrinsic value of x qua an object constituted by its inherited attributes? In answering this question, it is argued that entities can share different observable properties depending on the level of abstraction adopted,and that it is still possible to speak of moral value even at the highest level of ontological abstraction represented by the informational analysis. Section 4 develops a minimalist axiology based on the concept of information object. It further supports IE's position by addressing five objections that may undermine its acceptability.

Three classic distinctions specify that truths can be necessary versus contingent,analytic versus synthetic, and a priori versus a posteriori. The philosopher reading this article knows very well both how useful and ordinary such distinctions are in our conceptual work and that they have been subject to many and detailed debates, especially the last two. In the following pages, I do not wish to discuss how far they may be tenable. I shall assume that, if they are reasonable and non problematic in some ordinary cases, then they can be used in order to understand what kind of knowledge the maker’s knowledge is. By this I mean the sort of knowledge that Alice enjoys when she holds the information that Bob’s coffee is sweetened because she just put two spoons of sugar in it herself. The maker’s knowledge tradition is quite important but it is not mainstream in modern and analytic epistemology and lacks grounding in terms of exactly what sort of knowledge one is talking about. My suggestion is that this grounding can be provided by a minimalist approach, based on an information-theoretical analysis. In the article, I argue that we need to decouple a fourth distinction, namely informative versus uninformative, from the previous three and, in particular, from its implicit association with analytic versus synthetic and a priori versus a posteriori; such a decoupling facilitates, and is facilitated by, moving from a monoagent to a multiagent approach: the distinctions qualify a proposition, a message, or some information not just in themselves but relationally, with respect to an informational agent; the decoupling and the multiagent approach enable a re-mapping of currently available positions in epistemology on these four dichotomies; within such a re-mapping, two positions, capturing the nature of a witness’ knowledge and of a maker’s knowledge, can best be described as contingent, synthetic, a posteriori, and uninformative and as contingent, synthetic, weakly a priori, and uninformative respectively. In the conclusion, I indicate why the analysis of the maker’s knowledge has important consequences in all those cases in which the poietic intervention on a system determines the truth of the model of that system.

This paper outlines a quantitative theory of strongly semantic information (TSSI) based on truth-values rather than probability distributions. The main hypothesis supported in the paper is that the classic quantitative theory of weakly semantic information (TWSI), based on probability distributions, assumes that truth-values supervene on factual semantic information, yet this principle is too weak and generates a well-known semantic paradox, whereas TSSI, according to which factual semantic information encapsulates truth, can avoid the paradox and is more in line with the standard conception of what generally counts as semantic information. After a brief introduction, section two outlines the semantic paradox implied by TWSI, analysing it in terms of an initial conflict between two requisites of a quantitative theory of semantic information. In section three, three criteria of semantic information equivalence are used to provide a taxonomy of quantitative approaches to semantic information and introduce TSSI. In section four, some further desiderata that should be fulfilled by a quantitative TSSI are explained. From section five to section seven, TSSI is developed on the basis of a calculus of truth-values and semantic discrepancy with respect to a given situation. In section eight, it is shown how TSSI succeeds in solving the paradox. Section nine summarises the main results of the paper and indicates some future developments.