As large language models (LLMs) become increasingly integrated into clinical decision-making, their ethical alignment is as critical as factual accuracy. However, most evaluations focus on correctness while overlooking dimensions that influence patient outcomes and trust. To assess the ethical and contextual performance of three advanced LLMs—ChatGPT, Claude, and Gemini—across intensive care unit (ICU) scenarios and examine how specific ethical dimensions contribute to perceived clinical reliability. In this exploratory cross-model observational study, the models were evaluated using four ethically complex ICU cases. Responses were rated across ten ethical dimensions derived from biomedical ethics and AI ethics. Content analysis identified ethical reasoning patterns. Two expert ICU nurses independently rated clinical accuracy based on diagnosis and management appropriateness, with scores averaged across raters. Ethical dimensions were assessed by three domain experts, yielding 72 evaluations. Regression analysis at the model–scenario level (n = 12) examined associations between ethical scores and perceived clinical accuracy Across 12 model–scenario configurations, autonomy, oversight capability, and transparency were the strongest predictors of perceived clinical accuracy, whereas justice and nonmaleficence showed limited influence. Although findings are exploratory due to the small sample, ethical features accounted for significant variation in perceived reliability. Qualitative analysis showed Gemini produced more transparent and autonomy-supporting responses than Claude or ChatGPT. Ethical quality in AI-generated clinical decisions shapes expert perceptions of accuracy. These findings are hypothesis-generating and should be interpreted cautiously; larger studies are needed to confirm them. Ethical evaluation should be a core component of AI benchmarking in healthcare.
The adoption of new technologies, especially those that affect interpersonal relationships, makes actions possible that were previously impossible to perform. However, with that ability to perform these new actions, new reasonable expectations and even moral duties can emerge. Relatedness-technologies, i.e., communicative technologies aimed at creating the feeling and experience of connection between people across distances, are one such example. These often (and often unintentionally) expand one’s moral duties towards others as they increase the range of our actions, by way of making us more available to others due to their always-on, always-online features. The adoption of such technology, thus, can make us morally duty-bound to be available. Ending the use of a given technology may, thus, come with a higher moral justification than beginning to use it. In this paper, we explore this phenomenon, which we dub the “availability trap”, and elaborate upon its ethical issues, closing with a proposal on how to decrease negative consequences of these technologies for interpersonal relationships.
It has become commonplace to say that digital platforms have displaced traditional associations. Scholars of platform studies often use the phrase “disintermediation” to describe how platforms reshape the infrastructure of our associational life, removing the need for competing intermediaries. Unfortunately, this often fails to describe how platforms change the way we form associations online and offline. This article offers a critique of digital platforms, focusing on how platforms reconfigure our social life. Specifically, we argue that digital platforms undermine our ability to form robust associations online and offline. We first draw on empirical work to explain what makes an association “robust.” We then show that the architecture of platforms structurally undermines our ability to form robust associations. And throughout, we highlight various social goods (trust, loyalty, solidarity, etc.) that might be lost as this process continues unchecked. Our argument is simple: if robust associations are normatively desirable, we have reasons to prevent platformization from undermining them.
The article analyses the capabilities and limitations of large language models’ (LLMs) powering chat interfaces, such as ChatGPT, in the context of their potential use as “psychotherapists”. The authors relate this issue to the principles of psychodynamic psychotherapy, in which interpersonal relationships, unconscious processes, transference, and countertransference are of key importance. Based on a review of the literature and technical analysis, it is indicated that contemporary LLMs do not meet the basic conditions for conducting therapy in this field. They lack awareness, emotionality, and the ability to mentalize, whereas their “empathy” is purely linguistic and simulated. Human interactions with the model can create the illusion of a therapeutic relationship, promote anthropomorphizing and excessive trust in the algorithm. The article highlights the ethical, legal, and purely psychological risks associated with assigning chatbots the role of therapists. The analysis encompasses problems of hallucinations, data bias, and lack of ethical oversight. At the same time, the authors point to the potential benefits of using LLMs as support tools – in psychological education, preliminary diagnosis, stress reduction, or as a support for therapists in the process of analysis. They emphasize the need for an interdisciplinary approach combining psychology, law, and ethics, as well as the need to develop standards for safety and oversight of the use of LLMs in the field of mental health. The article concludes with a call for caution in the face of the growing integration of technology into therapeutic practice.
This paper examines how information environments shape individuals in the construction of what we dub epistemic selfhood. Using euthanasia as an illustration, we argue that end-of-life decision-making potentially suffers from this process when epistemic selfhood is malformed leading to decisions that lack proper justification. Our argument draws on Ian Hacking’s constructivist account of selves via public conceptualizations of human kinds, and then moves to the case of euthanasia to show that normalization dynamics create the kind of person who is thereby hindered in the capacity to make rational end-of-life choices. This is brought into stark relief where legislative frameworks seek to expand the range of cases, e.g., in Canada, where eligibility criteria for mental illness alone is under consideration. Under these conditions vulnerable individuals, lacking epistemic security, may make poorly grounded, misguided, decisions. Thus, we should promote information environments that support prudent practical reasoning, including at life’s end.
This article introduces the concept of virtual trespassing, a novel privacy violation enabled by photographic virtual reality (VR). Photographic VR refers to VR experiences that allow users to inhabit real captured spaces through 360° photography or video. We argue that the privacy risks posed by access in photographic VR exceed those associated with analogous forms of access in traditional 2D photography like Google Street View. Our argument extends Helen Nissenbaum’s (2004) contextual integrity theory of privacy beyond informational privacy to include sensorial privacy. Contextual integrity can be violated not only through the inappropriate flow of one’s personal information, but also through inappropriate access to one’s sensorial personal space (Véliz, 2018). Photographic VR makes possible unique kinds of contextually inappropriate access to one’s sensorial personal space through its capacity to create a sense of presence, a feeling of “being there.” We draw on empirical research on presence in VR to support our argument before outlining different forms of virtual trespassing and identifying an inherent value-privacy tradeoff in photographic VR. Specifically, the same capabilities that make photographic VR valuable (presence, authenticity, and unrestricted access) also enable privacy violations. We conclude by suggesting practical privacy measures grounded in contextual integrity theory that can mitigate virtual trespassing while preserving the technology’s unique value proposition.
This paper analyses an ethical problem emerging at the intersection of two recent developments: the deployment of anticipatory companion AI (ACAI) as continuously integrated personal agents, and the articulation of Agentic AI Optimisation (AAIO) as a paradigm for shaping how such agents engage with their digital environment. The paper’s central diagnosis is the fiduciary illusion: the configuration in which the fiduciary appearance of ACAI is preserved as a surface phenomenon while two interlocking illusions (a categorical illusion that suppresses recognition of the relationship as one calling for critical evaluation, and a contamination illusion that hides commercial influence within the system’s reasoning) operate beneath the surface to violate the two conditions of legitimate suggestion that the appearance purports to satisfy. We argue that the correctives proposed by Floridi et al. do not jointly address this configuration, with the gaps traceable to anticipating engagement and companion-mode presence, and that existing frameworks on online manipulation, relational autonomy, and the right against mental interference, while indispensable to the diagnosis, require extension at the specific points the configuration exposes. A moral right to vanilla AI is advanced as a normative response, articulated in two layers corresponding to the two illusions. The right’s minimal form secures availability of vanilla engagement; how the conditions the diagnosis identifies could be secured for users who choose ACAI remains an urgent open question.
This paper develops a novel account of privacy rights grounded in beneficial ownership. While tech companies and regulators often frame privacy in terms of data ownership, the nature of property rights in personal data remains unclear. I argue that treating personal data as fully owned property fails to capture the contextual sensitivity of privacy rights and conflicts with basic intuitions about privacy. Instead, I propose understanding privacy rights through the model of beneficial ownership, where data subjects retain rights to use and benefit from their data while others may control and transfer it only as trustees. This “Beneficial Ownership Model” better aligns with our privacy interests, which stem from our distinctly social form of agency and our need for selective self-presentation. The model has significant implications for the attention economy, suggesting reforms to how platforms manage user data and pointing toward new institutional arrangements for personal data governance. Beyond its practical applications, the model illuminates why personal data, like our physical bodies, inherently resists full ownership while still requiring robust protections.
The rapid proliferation of AI-mediated digital afterlife technologies, from chatbots trained on personal data to voice clones and posthumous avatars, has generated a substantial body of ethical literature identifying the moral risks of posthumous simulation. Yet this growing consensus has not been matched by frameworks capable of translating ethical principles into operational design constraints. This paper addresses that gap from the perspective of ethical design and governance. We introduce a nine-dimensional taxonomy of digital afterlife technologies, mapping the features that carry independent moral weight in any such system: timing, consent, data sources, interaction modality, fidelity and disclosure, purpose, audience and access, governance and ownership, and autonomy and behavioral agency. Building on this taxonomy, we derive a two-tier structure of design constraints: three threshold constraints — consent, fidelity/disclosure, and purpose — that function as near-absolute conditions of permissibility (Tier 1), and six contextual dimensions that modulate ethical risk profiles without individually determining permissibility (Tier 2). A system that fails any single Tier 1 constraint is impermissible regardless of its Tier 2 configuration. We argue that the ethical constraints derived from the taxonomy are not merely evaluative, but auditable, regulatable, and actionable by designers, governance actors, and legislators. By locating normative assessment at the level of design configuration rather than stated intent alone, the framework offers a concrete bridge between existing ethical consensus and the governance of AI-mediated digital afterlives.
Automated decision making (ADM) is increasingly prevalent in our world. Alongside the pursuit of efficiency, it is crucial to ensure that ADM is sustainable from a human-centric standpoint, i.e., with a long-term positive impact. To facilitate this development, we propose a new theoretical framework to assess and guide ADM processes, entitled the “human-centric digital decision making assessment framework”. For its development, state-of-the-art interdisciplinary integration techniques were used to combine relevant concepts from psychology and law. As such, the framework integrates the human and legislative perspective in the digital transformation, emphasizing three key considerations: fairness of the decision-making process, well-being of decision makers utilizing ADM, and the development of sustainable relationships with decision recipients. The human-centric digital decision making framework can be used for research, and as a tool for evaluating AMD in practice.
The harm caused by artificial agents, such as autonomous machines, can coincide with the problem that it is difficult to attribute responsibility to any party. Certain harms may stem from the sociotechnical system that has arisen from the deployment of such artifacts. I argue that to fully address these types of harms and appoint responsibility, we can make use of political responsibility and structural injustice. In these terms, it is the ecosystem itself which is erected and perpetuated, for which someone needs to carry responsibility. This can be done by looking at power, influence, privilege gained from the structure erected and narrating that those agents who have the greatest accumulation of these factors are in fact responsible. As such, we propose that forward-looking responsibility, the responsibility for the avoidance or creating of a certain state-of-affairs, may be a welcome addition to describe who may be responsible in cases where clear causal links aren’t easily found.
This paper intervenes in the debate over whether ChatGPT and other similar large language models (LLMs) are bullshit and answers that they are not. LLMs, however, don’t avoid being bullshitters in the way that humans do, i.e., by caring about the truth and avoiding deception. Rather, LLMs are not the kinds of things that can even count as bullshit. Attributing the property of being a bullshitter to LLMs is a category mistake.
Infodemics fuel the proliferation of misinformation and disinformation, often outpacing the effectiveness of existing countermeasures. Current interventions—such as debunking and psychological inoculation—typically address either specific instances of false or misleading content or the known manipulative tactics through which such content spreads. While these approaches are effective and scalable, most remain inherently corrective and struggle to anticipate shifts in the disinformation landscape. In this paper, we argue for a theoretical shift in paradigm towards a proactive framework to address disinformation: we propose a content-agnostic model grounded in the identification of recurring linguistic, narrative, logical, and critical thinking patterns that characterize manipulative disinformation prone to virality. By identifying these structural “fingerprints” of disinformation, our framework aims to open new avenues for a research agenda that could inform new inoculation strategies to promote individual and collective resilience. If inoculation provides the methodological foundation for building resistance to manipulation, then identifying structural “fingerprints” provides the core “antigens” that such intervention can target. This, in turn, could enable people to recognize and resist manipulation regardless of the topic, medium, or context through which manipulatory content is delivered. We argue that this approach may offer a sustainable path to building long-term resilience in the face of an evolving and increasingly complex and challenging information ecosystem.
This paper offers a philosophical and ethical critique of artificial intelligence (AI) and platform capitalism in virtual reality (VR) environments, examining how digital colonialism and algorithmic sovereignty reshape identity, agency, and justice in immersive lifeworlds. Drawing on political theory, decolonial scholarship, and digital phenomenology, the analysis traces three interrelated transformations. First, it examines how data colonialism operates within platforms such as Meta’s Horizon Worlds, where behavioral data extraction and algorithmic systems structure perception and interaction, producing forms of cognitive colonization embedded in immersive experience. Second, it analyzes the rise of algorithmic sovereignty—exemplified by environments such as Decentraland—through which governance is exercised via privatized, often opaque infrastructures regulating participation, visibility, and cultural production. Third, it highlights resistance movements, including Indigenous-led initiatives such as Te Awa Tupua, which propose alternative models of immersive governance grounded in relationality, collective stewardship, and epistemic plurality. Across these transformations, the paper develops the concept of immersive justice as a normative framework for evaluating and redesigning immersive environments. It examines the ethical implications of immersive technologies, including algorithmic asymmetries, ecological costs, and the transformation of interaction through avatars and synthetic agents. Engaging cyberfeminist and virtue ethics perspectives, the analysis shows how immersive systems can reinforce extractive logics or support more inclusive, participatory, and sustainable forms of collective life. The paper argues that immersive environments must be governed as shared infrastructures of experience, embedding accountability, plurality, and responsibility within their design and operation.
Large Language Model (LLM) based chatbots are increasingly prominent in the digital landscape, promising users a range of experiences, everything from human-like conversation to intimate companionship. But where exactly is the human in such technology? And what is its impact on the humanity of those persons engaging with it? Taking as its focus the popular LLM-based chatbot Replika, this article contends that - positive use cases notwithstanding - emerging generative AI technology carries with it a novel and more subtle form of dehumanisation. We combine an autoethnographic investigation with a multi-disciplinary critique (human-computer interaction; moral philosophy; law) to argue that such technology risks dehumanisation by inviting, demanding or conditioning human emotional and attentional engagement in circumstances where reciprocity is structurally impossible. We argue further that existing conceptual, philosophical and regulatory systems will need to be adapted to help diminish or erase the risks and harms presented by these emerging technologies.
To create an environment means thinking about, or rethinking, its structure so as to make it as inclusive as possible for individuals who use it. Recent advances in extended reality (XR) technologies have shown tremendous promise in this respect. Blended environments, those that include physical and digital elements, are especially promising because they allow developers and users to co-create social spaces that empower users to represent themselves and others with an eye toward inclusivity. This is a paper about how not to create such spaces. I will argue that the psychology of XR, especially the psychology of XR embodiment, should help us understand the limits of these technologies for enabling environments. First I will explain what XR technologies and how they enable a form of embodiment, XR embodiment. I’ll then explain the psychology of XR embodiment. XR embodiment should be taken seriously as a new possibility for the creation of selves but it has limitations. Virtual reality (VR) is often praised for its power to enhance empathy and thus to be a powerful tool for enabling environments. I argue that these claims are mistaken. While XR embodiment can be a transformative tool for identity formation, it cannot enhance empathic understanding of others. I then explore the social and legal implications of XR embodiment. If XR bodies are rightfully a component of a person’s self then if we want to create more inclusive and enabling environments we need to revisit the legal and social norms around bodies. Current laws (e.g., sexual harassment law) fail to take XR embodiment into account and thus fail to enable inclusive environments. I end the paper by suggesting that XR embodiment poses fundamental metaphysical questions about the relationship of the self to the physical body. These questions need to be addressed before enabling environments are possible.
Virtual Reality (VR) precipitates a fundamental crisis in moral reasoning by dismantling its chronological foundations. While existing normative frameworks—such as Virtue Ethics and Consequentialism—predicate moral agency on linear, irreversible time (Chronos), the VR engine operationalizes a temporal logic that is programmable, iterative, and recursive. We diagnose this operational contradiction as the “Death of Chronos”: a state where the diachronic continuity required for traditional ethical identity is shattered. Against this backdrop, we explicitly reject a blind anthropocentrism that seeks merely to “humanize” the simulation by imposing physical temporal constraints onto the digital. Instead, we argue for the necessity of rethinking the ethical subject itself—one capable of dwelling within the “Programmable Flow”, the distinct techno-ontological milieu of virtual action. We propose a theoretical framework to reconstruct an instantaneous ethical subject: a Agambenian form-of-life that derives normative weight from the sheer intensity of the absolute present. Ultimately, this reconstruction does not lead to solipsism, but establishes a new foundation for co-existence: enabling the subject, anchored in the virtual “now”, to guide both the self and the community toward a shared ethical horizon.
As virtual reality and virtual worlds in general take up more and more space in people’s everyday lives, the question of how virtual actions should be morally evaluated becomes increasingly pressing. One of the main challenges in the ethics of virtual worlds, in our view, is to find a normative framework, that is not overly restrictive but still accounts for common intuitions, such as the immorality of discrimination or harassment in virtual realms. In this paper we ask which criterion could be suitable to allow for a more nuanced evaluation of various potentially immoral virtual actions. First, we introduce a case to distinguish possible harms and immoral behaviors in virtual worlds. We then present a concept of consent inspired by principlism in research ethics and discuss criteria for valid consent and its limits. In applying this to our case, we find that consent can render many allegedly immoral actions in virtual worlds – such as virtual killing or theft – morally permissible, while it cannot be used to justify discrimination or violations of personal identity. A consent-based approach is thus shown to successfully reconstruct common moral intuitions, like those on the gamers’ dilemma. Finally, we address some objections to our argument, namely objections from deontological ethics and virtue ethics. The result of the paper is a framework for the moral evaluation of virtual actions, that allows for a more balanced approach and takes the particular purposes of virtual worlds and the interests of participating individuals into account.
This article explores the normative significance of experiments in living within immersive social Virtual Reality (VR) environments. Building on John Stuart Mill’s insights into the value of experimentation, we argue that social VR platforms allow users to critically engage with their values, revise commitments, and explore with alternative ways of conducting one’s life under conditions of reduced social and material constraint. We distinguish between individual and collective experiments in living and contend that immersive virtual environments can scaffold both types by reducing feasibility constraints and enabling the enactment of alternative identities and community structures. Special attention is given to the role of avatars, customization practices, and the architectural features of virtual environments that support moral inquiry and self-exploration. Finally, we raise concerns about the symbolic marginalization of virtual experimentation, particularly when such practices are framed as socially deviant or escapist. We conclude that social VR should be recognized as a legitimate site of moral exploration and self-formation, with implications for how we understand agency, recognition, and the ethics of virtual life.
Since 1999, scholars have queried whether Immanuel Kant’s moral philosophy would render problematic, wrong, or impermissible the performance of gratuitous violence and non-consensual sexual acts available in videogames and multiplayer virtual realms. Some contend that it does, some that it does not, and some ponder whether Kant provides grounds to prohibit legal access. This article posits that the scholarship has either misread Kant or read him incompletely, and a defensible reading of Kant reveals better strategies to address the issue than what has been proposed. Part I asserts that the commission of virtual taboos cannot be legally prohibited unless it raised the legitimate concern that someone else’s freedom might be hindered. Part II argues the commission of virtual taboos does not quite fall afoul of Kant’s “supreme principle of morality” when applied only to the perfect, negative duties gameplayers have to themselves. Part III argues that they do fall afoul of the supreme principle of morality when applied to the imperfect, positive duties gameplayers have to their self-perfection and membership in a “moral world.” Part IV provides further evidence from Kant’s aesthetics that renders virtual taboos impermissible.
