As Philosophy of Technology is increasingly mobilized to address artificial intelligence, much scholarship—whether utopian, catastrophist, or explicitly anti-singularitarian—remains structured by the ontology and temporality of a future AI threshold event. This paper argues that such futural orientations sustain a potentially hazardous form of “awaiting”: ethical reflection becomes predictive and solutionist, oriented to determinate risks presumed to materialize as future empirical dangers and threats. Against this backdrop, I diagnose a tendency within the “empirical turn” and post-phenomenological AI ethics toward what I call “ontic solutionism”: an approach bracketing ontology, narrowing attention to artifacts and measurable future harms, and thereby obscuring dangers that remain indeterminate in the present. Re-reading Heidegger—despite his exclusion from much post-phenomenological self-definition—I show that neglect of his concept of indeterminacy (Unbestimmtheit) underwrites familiar charges of techno-pessimism and determinism, and blocks key “reversals” (Umkehrungen) needed to grasp AI’s deeper threat: technologically induced nihilism. Developing Heidegger’s crucial distinctions between “awaiting” (erwarten) and “waiting” (warten), and between “advent” (Ankunft) and “event” (Ereignis), I propose a non-predictive orientation to AI as event in the present: a disciplined, non-willing readiness attuned to indeterminacy, which avoids pessimistic resignation in the face of techno-determinism. This stance does not abandon ontic interventions, but re-grounds AI ethics ontologically, clarifying how successful empirical safeguards may intensify the very technological disclosure they seek to manage.
In this paper I investigate how integrating ethics deeply into technology research and development can help developers overcome a major obstacle in the ethical alignment of innovation processes, namely, the “principles-to-practice gap”. By reference to a specific ethics intervention made by Berlin Ethics Lab featuring an AI system under development, I show that the gap is not about a failure to operationalize principles for research and development practice, but rather that it originates in divergent styles of thinking: ethical reflection and design thinking. Based on these insights, I elucidate how ethical reflection can be integrated on a foundational level into research and development processes. An essential starting point is the creation of a space for reflection to explore potential designs, a space dedicated to contextualizing and re-conceptualizing technology. This enables an ethical reframing of the design problem, systematic ethical analysis, and an assessment of potential implications. I conclude by presenting the ways in which this approach of deep ethics integration challenges other approaches to integrating ethics at an early stage (such as embedded ethics, STIR, VCIO, ethics by design, and value sensitive design) and demonstrating what additional benefits it has to offer.
The pursuit of artificial general intelligence (AGI) is widely regarded as the paramount objective of contemporary AI research. This aspiration rests on a seemingly self-evident premise: that general intelligence—the kind of flexible, domain-general cognitive capacity exemplified by Homo sapiens—is extraordinarily valuable. The present paper subjects this premise to critical scrutiny. We first present the intuitive case for the value of general intelligence, acknowledging its genuine strengths, before mounting an evolutionary challenge. We argue that, when measured against the timescales on which biological evolution operates, the adaptive value of general intelligence is far from empirically established. Numerous taxa—from cyanobacteria to horseshoe crabs—have persisted for hundreds of millions, even billions, of years without anything resembling general intelligence, while Homo sapiens has existed for roughly 300,000 years and already faces self-generated existential risks. Mass extinction events, examined as natural experiments, do not preferentially favour cognitively sophisticated species. We argue that general intelligence may be the only biological strategy that generates existential threats to the species that possesses it—an existential risk paradox with no parallel among non-intelligent survival strategies. Unlike the prevailing framing of AI risk, which traces the danger to misalignment, we locate it in the structural features of general intelligence itself, so that even a well-aligned artificial general intelligence would inherit this liability. If the long-term evolutionary value of general intelligence is uncertain or even negative, this raises profound ethical questions about the engineering of AGI systems and, with still greater urgency, about the creation of artificial consciousness—beings that would be both generally intelligent and sentient. Drawing on deontological ethics and the precautionary principle, we argue that this uncertainty imposes a duty of caution: if we bring into existence a new kind of intelligent being, we bear responsibility for ensuring the conditions under which it can flourish.
Artificial intelligence (AI) is widely discussed as either a threat to democracy or a means of strengthening it. But democracy, an essentially contested concept, invokes many different levels at which AI may assist and obstruct the legitimacy of democratic government. Therefore, the aim of this review is to map how the democratic value of AI, whether positive or negative, depends on different democratic models. The review examines what there is currently no examination of: how democracy is conceptualized, what the cross-cutting patterns are, and how we should prescriptively use democracy in relation to assessing how AI supports or undermines democracy. From this, the review maps three dynamics currently at play in the literature. It shows that conceptual unclarity and pessimistic views on AI’s contribution to democracy often coincide, prescribing the need for conceptual clarity. It shows that democratic models are often “inherited” rather than reworked, prompting the need for greater engagement with how AI reshapes democratic theory. And it maps how democratic assumptions shape evaluations of AI’s democratic implications. From these results, we assess the limits of existing scholarship: without greater conceptual clarity regarding democracy, empirical findings and normative arguments risk talking past one another. Taken together, this review suggests that progress in research on democracy and AI depends not only on further empirical investigation or technical refinement, but also on sustained conceptual work that makes democratic assumptions explicit and thus open to scrutiny.
Mariusz Mazurek’s commentary offers a careful and constructive reconstruction of my proposal of functional intelligence. I welcome his reading because it clarifies the intermediate space that my article intended to open between Floridi’s (2023, 2025) Artificial Agency and human normative understanding. In this reply, I argue that Mazurek’s contribution strengthens the project in three respects: by sharpening the distinction between functional semantics and mere predictive correlation, by proposing a useful taxonomy of agency, and by identifying the unresolved role of embodiment in artificial cognition. I also suggest two qualifications. First, the proposed three level taxonomy need not be interpreted as a vertical ontological hierarchy, but may be better understood as a plurality of levels of abstraction. Second, the difference between biological predictive control and artificial predictive control remains epistemically open: current evidence establishes the sufficiency of biological embodiment for cognition, not its necessity.
The emergence of LLMs necessitates a re-evaluation of communicative agency in social systems. Integrating Habermas’s theory of communicative action with speech act theory, this paper investigates whether LLMs can engage in communicative action, produce speech acts, and function as social agents. We advance the novel claim that LLMs are “asymmetrical communicative agents” that satisfy behavioral but not intentional conditions for communicative action. While LLMs cannot yet generate genuine illocutionary acts, they produce significant perlocutionary effects. We adopt a gradualist position, arguing that LLMs lack a full-fledged “cognitive-volitional” complex, making them susceptible to co-option for strategic action even as they intervene in canonical communicative scenarios. This dual-status ontology suggests LLMs are quasi-social agents capable of advancing validity claims and influencing social norms. After anchoring LLMs within a “digital lifeworld,” we analyze their evolving roles across the strategic-communicative spectrum and their implications for the construction of social reality.
The increasing integration of artificial intelligence into decision-making processes has intensified questions about how responsibility should be understood within socio-technical systems. Although debates surrounding the responsibility gap have identified important challenges for existing models of responsibility, they often leave underexamined the internal structure of responsibility itself, treating it as a one-dimensional concept despite comprising multiple analytically distinct normative relations. This paper argues that a more adequate account can be developed by drawing on Gary Watson’s distinction between attributability and accountability, together with his later elaboration of responsibility in terms of attributability, accountability, answerability, and culpability. These distinctions provide the basis for analysing responsibility as a multi-layered structure composed of distinct but related normative dimensions. This perspective makes it possible to examine responsibility under conditions of technological mediation without attributing moral agency or moral responsibility to artificial intelligence systems. Building on this framework, I develop eight sub-dimensions of blame designed to capture how responsibility relations are differentiated and reconfigured within socio-technical systems. The resulting account reconceives the responsibility gap not as a single problem, but as a set of structurally distinct layers between responsibility relations. The paper’s contribution is to provide a structured analytical framework for mapping these relations with greater precision, thereby clarifying how responsibility is organised in socio-technical environments while preserving human and institutional actors as the primary bearers of accountability and culpability.
Generative AI exposes the historical fragility of Romantic myths of the sovereign, self-transparent writer without inaugurating a crisis of authorship itself. Tracing a genealogy from symbolic AI to large language models, we show how creativity has always depended on distributed infrastructures, archives, and labour that the figure of the solitary author conceals. Drawing on Barthes, Foucault, Butler, Haraway, Hayles, and recent legal and bibliometric debates, we argue that “AI authorship” is a category mistake: statistical systems cannot occupy positions of accountability, vulnerability, and justificatory dialogue. They operate as powerful catalysts within socio-technical assemblages whose conditions of possibility lie in data extraction, platform governance, and planetary logistics. On this basis, we advance a relational conception of authorship as ethical and epistemic stewardship over hybrid writing systems, contending that even under conditions of distributed agency and technical opacity, identifiable human agents must remain answerable for AI-mediated outputs. We conclude by sketching implications for copyright doctrine, contributorship taxonomies, and research ethics in an era of pervasive generative automation.
What is real has always been something we find, not something we make—or so philosophy has assumed. This paper argues otherwise. Characterizing reality through resistance rather than substance (the ways the world refuses a subject’s mastery), I distinguish three modalities correlative to epistemic, judgmental, and practical mastery: Substrate (matter’s resistance to representation), Contingency (the forceful givenness of experience that resists revision by judgment), and the Inexorable (structures’ resistance to intervention). Treating virtual environments, AI agents, and blockchain smart contracts not as proofs but as revelatory cases, I show that technology now extends the latter two modalities, Contingency and the Inexorable, artificially. The result is the paper’s central concept, Artificial Externality: human-made structures whose resistance to intervention is deliberately engineered to be practically insurmountable, even for their creators, and that thereby acquire an externality once attributed only to nature. Absoluteness, traditionally found, can now be produced. I close by drawing out the stakes for consciousness: our criteria for what counts as real quietly shape our criteria for what counts as conscious.
This article introduces vector theory as a critical approach for understanding the shift from symbolic to probabilistic computation in contemporary AI systems. The paper argues that the digital turn organised meaning through discrete bits, Boolean logic, and hierarchical structures, in contrast large language models (LLMs) and diffusion architectures operate through high-dimensional vector spaces, cosine similarity, and probability manifolds. The three sections of the article examine the geometry of meaning, the dynamics of stochastic flow, and the political economy of the vector turn. These sections connect concepts such as vectors, tokenisation and generative AI to critical traditions from Marx through the Frankfurt School to contemporary media theory. Drawing on and expanding Kittler’s media materialism, Stiegler’s grammatisation, and Deleuze’s notion of smooth space, the article argues that existing approaches, developed under the paradigm of discrete digitality and symbolic logic, generate too many explanatory anomalies. The vector paradigm represents a new stage in the real subsumption of cognitive and linguistic labour, where capital reconstitutes language as geometry within proprietary vector space. The article connects this to notions of cognitive anaesthesia and the systematic “smoothing” of social friction, arguing that this potentially threatens the tacit dimension of critical thought, the very faculties required to diagnose the computational regime that produces it.
Artificial intelligence (AI) has renewed interest in the possibility of central economic planning. Scholars and technology leaders argue that sufficiently advanced AI systems could replace decentralized market processes by optimizing production and allocation decisions. This paper contends that such proposals overlook a significant epistemic function performed by markets. I first distinguish five epistemic functions of markets: economic calculation, knowledge aggregation, error detection, decentralized experimentation, and entrepreneurial discovery. I then focus on entrepreneurial discovery as analyzed by Israel Kirzner. Markets coordinate economic activity not only by processing existing information but also by generating new knowledge through entrepreneurial alertness to previously unnoticed opportunities. Drawing on research in embodied cognition and phenomenology, I contend that this discovery process relies crucially on embodied tacit knowledge. Entrepreneurs recognize opportunities through practical sensorimotor and affective engagement with tools, physical environments, and social practices. Contemporary AI systems, such as large language models and AI agents, operate through disembodied computational processes and lack sensory and affective coupling through which such knowledge arises. Because entrepreneurial discovery requires embodiment, AI systems cannot replicate this epistemic function of markets. The paper concludes by considering whether future developments in embodied AI could approximate entrepreneurial discovery and what this possibility implies for AI-assisted economic planning.
There has been considerable debate over both the ontology and ethical permissibility of so-called ‘deathbots,’ ‘thanabots,’ or ‘Interactive Personality Constructs of the Dead’ (IPCDs) - AI-driven software programs that emulate the conversational style of a person who has died. This paper explores two possible candidates for the ontological status of IPCDs: ‘remnant personhood,’ as recently argued for by Jurgis Karpus and Anna Strasser, and ‘animated persona,’ a category recently articulated by Masahiro Morioka. Both candidates are shown to pick out important features of IPCDs, while also having distinctive drawbacks. I conclude by exploring the possibility that these categories, instead of being exclusive, might be complementary lenses through which we can consider IPCDs – and that this dual focus helps us to better understand where the ethical and psychological dangers of this technology may lie.
In a recent article in this journal, Ari Deller has argued that there are some severe epistemic costs to super-persuasive AI. In this commentary I explain why, while I agree with Deller about the epistemic costs of super-persuasive AI, I disagree with him about the likely source of the problem. Where Deller focuses on the super-persuasive power of future AI technologies, I focus on the enormous increase in persuasive capacity that is provided by a technology that can produce very large amounts of persuasive content at minimal cost.
When interacting with social AI systems (SAIs), we routinely speak of what they ‘believe’, ‘want’, or ‘know’. With some exceptions, philosophers tend to treat such anthropomorphism as a single phenomenon that risks one kind of error: mistaken ontological commitment to machine minds and mental states. This paper challenges this monistic assumption. I distinguish two modes of anthropomorphic attribution—metaphysical and pragmatic—and identify two corresponding kinds of possible anthropomorphic error. In the metaphysical mode, speakers commit themselves to the existence of machine mental states, risking straightforward ontological error. In the pragmatic mode, speakers adopt the intentional stance without ontological commitment, yet still risk error when another interpretive strategy would better serve their purposes. I defend Mixed Anthropomorphism: both modes are common. This pluralist account reveals that the current debate’s focus on whether users ‘really mean it’ obscures the pragmatic dimension of anthropomorphic ascription (and its risks). Even ontologically innocent anthropomorphism can constitute a mistake because it employs the wrong interpretive tool for the task at hand. Understanding these distinct error types matters both theoretically, for clarifying the nature of human-AI interaction, and practically, for designing systems that encourage and scaffold appropriate interpretive strategies.
