A key assumption fuelling optimism about the progress of large language models (LLMs) in accurately and comprehensively modelling the world is that the truth is systematic: true statements about the world form a whole that is not just consistent, in that it contains no contradictions, but coherent, in that the truths are inferentially interlinked. This holds out the prospect that LLMs might in principle rely on that systematicity to fill in gaps and correct inaccuracies in the training data: consistency and (...) coherence promise to facilitate progress towards comprehensiveness in an LLM’s representation of the world. However, philosophers have identified compelling reasons to doubt that the truth is systematic across all domains of thought, arguing that in normative domains, in particular, the truth is largely asystematic. I argue that insofar as the truth in normative domains is asystematic, this renders it correspondingly harder for LLMs to make progress, because they cannot then leverage the systematicity of truth. And the less LLMs can rely on the systematicity of truth, the less we can rely on them to do our practical deliberation for us, because the very asystematicity of normative domains requires human agency to play a greater role in practical thought. (shrink)

The concept of Generative Artificial Intelligence (GenAI) is ubiquitous in the public and semi-technical domain, yet rarely defined precisely. We clarify main concepts that are usually discussed in connection to GenAI and argue that one ought to distinguish between the technical and the public discourse. In order to show its complex development and associated conceptual ambiguities, we offer a historical-systematic reconstruction of GenAI and explicitly discuss two exemplary cases: the generative status of the Large Language Model BERT and the differences (...) between protein structure predictions from AlphaFold 2 and 3. Our analysis shows that there is no unique and unambiguous definition of GenAI based on a purely technical account of the term. Following this conclusion, we argue that the public discourse is not simply a less complex way of speaking, but instead transcends its technical basis. As a means to structure this newly emerging discussion landscape we introduce a non-exhaustive list of four central aspects of GenAI: (multi-)modality, interaction, flexibility, and productivity. These dimensions constitute a first step towards defining GenAI beyond its technical basis. (shrink)

This paper proposes a new normative framework to think about Big Tech reform. Focusing on the case of digital communication, I argue that rethinking the corporate purpose of social media companies is a distinctive entry point to the debate on how to render the powers of tech corporations democratically legitimate. I contend that we need to strive for a reform that redefines the corporate purpose of social media companies. In this view, their purpose should be to create and maintain a (...) free, egalitarian, and democratic public sphere rather than profit seeking. This political reform democratically contains corporate power in two ways: first, the legally enforceable fiduciary duties of corporate boards are reconceptualized in relation to democratic purposes rather than shareholder interests. Second, corporate governance structures should be redesigned to ensure that the abstract purpose is realized through representatives whose incentives align with the existence of a democratic public sphere. My argument complements radical proposals such as platform socialism by drawing a connection between democratizing social media governance and identifying the proper purpose of social media companies. (shrink)

Contextual Integrity (CI) is built to assess potential privacy violations of new sociotechnical systems and practices. It does so by evaluating their respect for the context-relative informational norms at play in a given context. But can CI evaluate new sociotechnical systems that severely disrupt established social practices? In this paper, I argue that, while CI claims to be able to assess privacy violations of all sociotechnical systems and practices, it cannot assess the ones that cause severe changes and disruptions in (...) the norms and values of a given context. These types of technology are known as socially disruptive technologies (SDTs) and this paper argues that they are beyond CI’s scope. It follows that at best, a privacy assessment of those technologies by CI would be useless and, at worst, lead to potential harm, including failure to identify privacy violations and unwarranted legitimisation of privacy-threatening technology. Government actors, policymakers, and academics should refrain from relying on CI to assess this type of technology. (shrink)

There has been a surge of interest in explainable artificial intelligence (XAI). It is commonly claimed that explainability is necessary for trust in AI, and that this is why we need it. In this paper, I argue that for some notions of trust it is plausible that explainability is indeed a necessary condition. But that these kinds of trust are not appropriate for AI. For notions of trust that are appropriate for AI, explainability is not a necessary condition. I thus (...) conclude that explainability is not necessary for trust in AI that matters. (shrink)

Masahito Mori hypothesised that robots that look somewhat but not quite enough like humans are likely to arouse a sense of unease and discomfort. This hugely influential design hypothesis about robot appearance has been dubbed the ‘Uncanny Valley.’ I argue that there is a more general formulation of this principle that can be extended beyond the aesthetic domain of robot appearance. Specifically, I propose that there exists a _Moral Uncanny Valley_, which occurs when the behaviour of artificial agents mimics somewhat (...) but not quite enough the moral behaviour of humans.The question then is: what should we do about the Moral Uncanny Valley? The solution Mori proposed for the appearance of robots is to _avoid_ the Uncanny Valley and, given the inevitability of failing to reach the second peak, advised designers to design machines specifically to aim for the first peak: that is, to look _less_ like humans. Instead of making a parallel recommendation in the moral case, I explore the dangers of being too comfortable around machines and thus the advantages of _embracing_ the Moral Uncanny Valley. The sense of the uncanny brought about through what I call ‘discomfort by design’ can be useful for helping us to avoid making moral mistakes, like over-ascribing moral agency, patiency and responsibility. (shrink)

Each technology advances at its own pace, often indifferent to theoretical and philosophical-scientific conceptualizations. In the case of technologies like artificial intelligence (AI), and especially generative AI (GenAI), developments are so rapid that conceptual and epistemological reflections struggle to keep up, even at the level of basic definitions. Yet these definitions carry significant non-theoretical implications, including social, legal, and policy-related consequences. In this paper, I offer some reflections on the definition of GenAI proposed by Ronge et al. (2025), using it (...) as an opportunity to highlight the most relevant aspects of the ongoing debate sparked by these new AI systems. In doing so, we will seek to explore the implications of generative AI systems in the context of human interaction, particularly in light of their role as active supports rather than passive tools. This shift, coupled with the increasingly anthropomorphic perception of their capabilities, sets GenAI systems apart from any previous technologies, not only in terms of technical features but also in how they are perceived by everyday users. (shrink)

This article assesses the potential of artificial colleagues to help us realise the goods of collegial relationships and discusses its practical implications. In speaking of artificial colleagues, it refers to AI-based agential systems in the workplace. The article proceeds in three steps. First, it develops a comprehensive account of the goods of collegial relationships. It argues that, in addition to goods at the individual level, collegial relationships can provide valuable goods at the social level. Second, it argues that artificial colleagues (...) are limited in their capacity to realise the goods of collegial relationships: at the individual level, they can at best realise some such goods, and at the social level, they can at best support their realisation. This contradicts Nyholm and Smids’ (2020) claim that robots can be good colleagues. The article traces these limitations to particular features of artificial colleagues and discusses to what extent they would hold for radically advanced systems. Third, the article examines the policy implications of these findings. It highlights how the introduction of artificial colleagues, in addition to potentially crowding out human colleagues, will likely impact relations among human colleagues. And it proposes a governance principle that gives strict priority to human collegial relationships. (shrink)

I reply to professor Proudfoot’s reply to my paper ‘The Moral Turing Test: a defense’ (2024).

We investigate the ‘bewitchment’ of understanding interactions between humans and systems based on large language models (LLMs) inspired by Wittgenstein’s later view on language. This framework is particularly apt for analyzing human-LLM interaction as it treats understanding as a public phenomenon manifested in observable communicative practices, rather than as a mental or computational state—an approach especially valuable given LLMs’ inherent opacity. Drawing on this perspective, we show that successful communication requires not merely regularity in language use, but constancy in maintaining (...) reference points through agreement in both definitions and judgments. Crucially, LLMs lack the constancy needed to track negations and contradictions throughout a dialogue, thereby disrupting the reference points necessary for genuine communication. The apparent understanding in human-LLM interactions arises from what we characterize as a ‘bewitchment’: the interaction between LLMs’ statistical adherence to linguistic patterns and humans’ tendency to blindly follow familiar language games. Moreover, when interaction with LLMs is based on stereotyped contexts in which the system seems capable of identifying reference points, we humans automatically apply the practical principle that there is understanding until proven otherwise. The bewitchment becomes thus more profound as LLMs improve in modeling stereotypical aspects of human interaction. This improvement, far from addressing the highlighted limitations, can only deepen the illusion of understanding, raising significant concerns for meaningful control over such systems. (shrink)

Recently, Benjamin Steyn has proposed a novel political spectrum of value where ‘nature’ and ‘technology’ are understood as opposed. The author attempts to provide a metaphysical basis for this spectrum by advocating for the inherent value of technology. However, we regard their argumentation as inadequate, as it fails to substantiate the intrinsic value of technology, instead serves as an interpretive tool for certain images and narratives associated with technologies. From the brief review of the concepts of nature and technology used (...) in the text we offer reasons in favour of the plurality and contingency of technology. (shrink)

We argue for the epistemic and ethical advantages of pluralism in Reinforcement Learning from Human Feedback (RLHF) in the context of Large Language Models (LLMs). Drawing on social epistemology and pluralist philosophy of science, we suggest ways in which RHLF can be made more responsive to human needs and how we can address challenges along the way. The paper concludes with an agenda for change, i.e. concrete, actionable steps to improve LLM development.

Critiques of opaque machine learning models, used to guide consequential decisions, are gaining traction in moral philosophy. They are said to undermine informed self-advocacy, violate duties of consideration to decision-subjects, or reinforce existing power structures, among other things. Even though motivated by different normative frameworks, the received view is that the appropriate amelioration strategy is to make opaque machine learning models explainable. We challenge the received view. Focusing on Vredenburgh’s account of the ‘right to explanation’, we argue that the goals (...) of informed self-advocacy cannot plausibly be satisfied by a combination of causal and rule-based normative explanations. Here we make three arguments in particular: (i) we highlight problems concerning the feasibility of causal explanations of the machine learning output for the purpose of social prediction; (ii) the relevant explanations are likely to be intellectually over-demanding since their correct interpretation will require a combination of domain-knowledge and statistical proficiency that cannot be presumed for laypersons; and (iii) shifting the burden to detect inadequacies to laypeople makes scrutinizing explanations very costly for them. Drawing on the empirical and philosophical literature on informed consent, we argue that it is more appropriate to revise the goal of the right to explanation from an interest in informed self-advocacy to enabling public trust: The public should be justified in believing that it can count on an institution using machine learning models competently, and that the goals for which they are used are morally legitimate. (shrink)

Many dangers of social media are typically framed with images and concepts assuming or employing the paradigm of addiction. The addiction paradigm is valuable descriptively, as a means towards understanding various phenomena of social media, and rhetorically, with regard to public policy. But, the paradigm is limited, and risks reducing the problems of social media to questions of physiology and (brute) animal behavior. This paper focuses on the need to develop distinctively human paradigms for understanding the risks of social media. (...) Fundamentalism is offered as one such uniquely human paradigm. Fundamentalism, abstracted from any specific religion, is proposed as a pathology of reasoning, and specifically of reading. Similarities between fundamentalists and users of social media are articulated, and then an analysis of the form of social media—distinct from their content—is conducted so as to draw out how the form of social media encourages poor habits or pathologies of literacy. Social media are revealed to be less about their content and more about reactions to content, and an argument is made that the kinds of reactions favored by social media erode mature reasoning or rationality in users of the media. (shrink)

This article argues that human intuition, deeply rooted in embodied experience and cultural context, presents fundamental challenges for replication in artificial intelligence systems. Drawing on phenomenology, cognitive science, and contemporary AI research, we demonstrate how genuine intuitive understanding emerges from the integration of embodied experience, phenomenal consciousness, and cultural situatedness in ways that resist computational reduction. Through detailed analysis of embodied skill acquisition, expert judgment, and creative expression, we show that current approaches to artificial intuition fundamentally misconstrue the nature of (...) intuitive understanding. While acknowledging significant advances in AI capabilities, we argue that the irreducibility of human intuition suggests the need to reconceptualize the relationship between human and machine intelligence. Rather than pursuing the replication of human intuition, we propose focusing on the development of complementary forms of intelligence that can work in concert with human intuitive understanding. (shrink)

This commentary builds on Davide Battisti’s proposal of second-person authenticity as a moral standard challenged by AI-mediated romantic communication. While endorsing the ethical concerns raised, I explore whether such mediation—if used reflectively and with a degree of co-authorship—can in fact enhance moral engagement. Drawing on examples and theories from moral philosophy and cognitive science, including Kant’s concept of moral worth and the extended mind framework, I argue that generative AI may support, rather than undermine, relational authenticity under certain conditions.

In a recent article, Queloz (Philosophy & Technology, 38(1), 2025) proposed that AI cannot rely on the systematicity of truth in normative domains to build comprehensive models. Beyond challenging AI development, the asystematicity of truth in normative domains underscores the essential role of individual human agency in practical deliberation. When faced with difficult choices between conflicting and incommensurable values, individuals must make judgements of importance that cannot be delegated to an algorithm. This commentary expands on the fundamentally personal nature of (...) practical deliberation to argue in favor of generalist AI moral advisors over personalized systems. (shrink)

This paper proposes an enhanced version of the relational approach in the philosophy and ethics of robotics by integrating it with enactivism. The paper begins by providing a concise overview of the relational approach within the field, outlining its key contributions and limitations. It then identifies significant issues in the current version of relationalism, such as its reliance on partial phenomenology, the problem of morality-before-morality, bad inclusion, and the treatment of the relation as a black box. To address these concerns, (...) the paper introduces phenomenologically inspired enactivism and demonstrates how incorporating enactivist principles can resolve these challenges. By applying enactivism, the paper argues for a more robust understanding of human-robot interaction, where relational dynamics are understood as lived, embodied processes that are essential for constructing the otherness of robots. (shrink)

This article provides an overview of the philosophy of Gilbert Hottois, who is usually credited with popularizing the concept of technoscience. Hottois starts from a metaphilosophy of language that diagnoses twentieth-century philosophy as fixated on language at the expense of technology. As an alternative, he developed a philosophy of technoscience that reinterprets science as primarily an intervening and technical activity rather than a contemplative and theoretical one. As I will argue, Hottois articulates the nature of this technicity through a philosophy (...) of time, reflecting on the specific temporality of technoscience as distinct from human history. This temporality of technoscience provoked the need for ethical reflection, since technoscience is constantly changing and transforming the world. This led to Hottois’s engagement with bioethics, in which he sought to develop a framework capable of “guiding” technoscience. Aiming to avoid both total symbolic closure and total technical openness, this guidance is concerned with the preservation of diversity, especially the human capacity for ethics, ethicity. This idea of guidance was later taken up by Dutch philosophers such as Hans Achterhuis and Peter-Paul Verbeek, inspiring their empirical turn in the philosophy of technology. What remains missing in this framework, however, is Hottois’s critical analysis of the different temporalities at work in technology and culture. (shrink)

The goal of this paper is to revisit Baudrillard’s take on artificial intelligence and to present a critique of AI (especially considering large language models). Baudrillard expressed his skepticism about AI already in his essay Xerox and Infinity from 1993. To understand Baudrillard’s argumentation, it is necessary to open up his theoretical body of work, starting from Marxist value theory and leading up to his own concept of simulation theory, which he later became known for. Baudrillard’s main idea throughout his (...) work is how the meaning of a sign, once emerging out of a social bond which he calls symbolic exchange, transfers into a sphere of representative signs that then morph into a life of their own. Baudrillard’s critique falls perfectly in line with the discussion around the dead internet theory and the loss of meaning within the virtual space through the rise of bots and artificially created content, stripped of human origin and connection. Artificial content is, for Baudrillard, the purest form of simulation, with signs that have lost their origin of representation and are now exponentially recreated in a digital sphere. Just like the AI-generated image of “Shrimp Jesus,” which flooded the social media platform Facebook and became a viral hit. “Shrimp Jesus” is a perfect example of an artificial image pushing human-generated content out of the virtual sphere. From a Baudrillardian perspective, that dead internet, the internet dying because of the flood of artificial content, is not an accident, but an inevitable event. (shrink)

The author's response to a commentary on my paper'The Nature Technology Political Spectrum'.

In this paper I explore the potential benefits and harms of avatar or chatbot representations of persons who are in palliative care. Much has been written recently about the benefits and harms of ‘continuing bond’ chatbots: representations of those who have died. Depending on one’s view, continuing bond chatbots are either a useful tool that facilitates the bereaved engaging in conversations with a representation of the deceased or they are an unhealthy block to the bereaved’s ability to move through the (...) stages of grief. But there are other grief contexts in which the needs of the griever are relevantly different to the needs of a person whose loved one has died. Here I consider palliative cases in which a loved one is significantly psychologically altered such that it is no longer easy to experience them as the ‘same person’. I consider the unique problems facing carers in this scenario and how personal avatars might provide some comfort and help with both the grieving and the ongoing caring prospect. I undertake a philosophical exploration of whether and how the arguments for the use of continuing bond avatars in cases of death carry over into palliative care contexts. Ultimately, I propose that a distinctive case can be made for the use of continuing bond avatars to help the grieving process of carers in some palliative contexts. (shrink)

People sometimes behave differently depending on whether they are interacting online (by email, social media, etc.) vs. interacting in person. Four studies test the hypothesis that when an agent’s behavior is different online vs. in person, people think that the online behavior is less reflective of who the agent truly is deep down. Study 1 found that the very same behavior is regarded as less reflective of the true self when it is performed online. Study 2 showed that this effect (...) is not merely a matter of perceived impression management. Study 3 found that there is a general tendency such that behavior is seen as more reflective of the true self when it is performed in an environment regarded as “natural.” In Study 4, a manipulation that led participants to see online behavior as more natural had a downstream effect on the degree to which this behavior was seen as reflecting the true self. Taken together, these results suggest that people’s judgments about online behavior are not simply a reflection of idiosyncratic facts about the online sphere in particular but are instead driven in part by a far more general psychological process involving perceptions of naturalness. (shrink)

Would a political system where the governance was overseen by an algorithmic system be legitimate? The intuitive answer seems to be no. This paper considers the philosophical effort to justify this intuition that argue for algocracy, a rule by algorithms, being illegitimate. Taking as the paradigmatic example the anti-algocratic argument from Danaher that attempts to ground algocratic illegitimacy in the opacity of algocratic decision-making, it is argued that the argument oversimplfies the matters. Opacity can delegitimise—but not simpliciter. It delegitimises because (...) of the presence of certain downstream violations of obligations and rights of the public that result from the opaque governance. Algocratic decision-makers, however, seem not to be subjects to these normative constraints in the relevant sense. The paper therefore argues that the standards of legitimacy that have been deployed for or against different kinds of human governance do not apply to the algorithmic decision-making systems with quite the same force. New avenues for rooting the illegitimacy of algocratic decision-makers have to be developed. (shrink)

We apply the concept of robustness from the philosophy of science to human–AI collaboration in diagnostic radiology, introducing diagnostic complementarity as a way to understand how radiologists and AI systems can productively work together in the context of double-reading. Diagnostic complementarity refers to the idea that two readers (e.g. radiologist and AI) have different diagnostic strengths and limitations, such that their combined performance exceeds that of either working alone. We argue that state-of-the-art AI diagnostic systems—convolutional neural networks (CNNs)—realize diagnostic complementarity (...) with radiologists, and thus can serve as effective second readers for radiologist first readers in double-reading. In the course of making our argument, we clarify diagnostic complementarity and its epistemic benefits, explain how it can be discovered through theoretical and empirical means, and show how our points can be used to evaluate not only the utility of CNNs as second readers, but also of future, more advanced AI systems. (shrink)

This paper builds on Dewey’s philosophy of education to argue that AI, at least in its current commercial form, is likely to have a negative impact on democratic education. AI and other digital technologies are currently being touted for their potential to “democratise” education, even if it is not clear what this would entail. Adopting Dewey’s notion of democratic education, I emphasise that education needs to provide children with skills and dispositions necessary for democratic living, experience in communication and cooperation, (...) opportunities to codetermine the shape of democratic institutions and education itself, and equal opportunities to participate in learning. This allows me to show that most of the voices discussing AI and democratisation of education, conceptualise it in a narrow sense, focusing primarily on the technology’s potential to increase access to quality education. Consequently, I develop an analysis that investigates the relationship of educational AI to the four aspects inherent to Dewey’s philosophy. By examining today’s commercial AI tools and focusing primarily on Intelligent Tutoring Systems, the most prominent kind of educational AI today, I argue that their emphasis on individualisation of learning, their narrow focus on the mastery of the curriculum, and the drive to automate teachers’ tasks are obstacles to democratic education. I demonstrate that AI deprives children from opportunities to gain experience in democratic living and acquire communicative and collaborative skills and dispositions, while also habituating them to an environment over which they have little or no control, potentially impacting how they will aproach shared problems as democratic citizens. I accompany my analysis with suggestions how AI could be used to better serve the goals and values of democratic education. I highlight the opportunities for experiential learning introduced through simulations and games, note how learning network orchestration could expose students to a wider variety of worldviews, reflect on the use of AI to augment rather than automate teachers’ work, and suggest public development of educational AI. (shrink)

Although there has been much discussion regarding how technology mediates our practical and ethical lives, little has been said about how it mediates phronesis: the skilled deliberative capacity to direct our lives well. With new and emerging technologies like the generative AI of Chat-GPT, mindfulness apps such as Wysa and Headspace, and the datafication of our everyday lives, it becomes necessary to ask how these technologies and practices affect our ability to reason towards and actualize flourishing lives. I argue that (...) (1) phronesis is a technologically mediated capacity whose mediation is best understood from the postphenomenological perspective; (2) through an adaptation of Albert Borgmann’s ideas of commanding and disposable realities, practical wisdom’s mediation can be understood through augmenting, diminishing, and displacing relationships with technology; (3) the rapid proliferation of new and emerging AI technologies threaten to emphasize disposable realities over commanding realities, which leads to a diminishing of phronesis, and by extension human flourishing. If we are interested in creating and using technology in ways that promote human wellbeing, we must first understand how these technologies mediate phronesis and consider what sorts of technological relationships benefit that deliberative skill. (shrink)

The development of AI tools, such as large language models and speech emotion and facial expression recognition systems, has raised new ethical concerns about AI’s impact on human relationships. While much of the debate has focused on human-AI relationships, less attention has been devoted to another class of ethical issues, which arise when AI mediates human-to-human relationships. This paper opens the debate on these issues by analyzing the case of romantic relationships, particularly those in which one partner uses AI tools, (...) such as ChatGPT, to resolve a conflict and apologize. After reviewing some possible, non-exhaustive, explanations for the moral wrongness of using AI tools in such cases, I introduce the notion of second-person authenticity: a form of authenticity that is assessed by the other person in the relationship (e.g., a partner). I then argue that at least some actions within romantic relationships should respect a standard of authentic conduct since the value of such actions depends on who actually performs them and not only on the quality of the outcome produced. Therefore, using AI tools in such circumstances may prevent agents from meeting this standard. I conclude by suggesting that the proposed theoretical framework could also apply to other human-to-human relationships, such as the doctor-patient relationship, when these are mediated by AI; I offer some preliminary reflections on such applications. (shrink)

System thinking is widespread in technology development approaches such as “system engineering” and “system design.” We argue that postphenomenology, as a broadly accepted and essential philosophy of technology, has individual intentionality as a core foundational concept and, therefore, struggles to describe system thinking. We start by indicating that some contemporary postphenomenology scholars discuss system-related concepts such as intentional structures of human experience. We then turn to the fundamentals of postphenomenology to better understand how individual intentionality can be related to system (...) thinking via consciousness. We discuss the classical system thinking concept of autopoiesis as an intentionality structure relevant to individual consciousness of psychic systems and communication for social systems. However, the relation to the world is underdeveloped in these classical system theories, so we turn to contemporary thinkers who stress both the systemic and world aspects of phenomenology in sympoiesis. This helps us formulate three system-thinking recommendations for system phenomenology related to system-environment difference, structure versus process, and sympoiesis. We provide two illustrations of how system phenomenology can help understand technological design. We conclude with a plea for further conceptual development and practical application of system phenomenology in postphenomenology, system design, and system engineering. (shrink)

In view of the dramatic advancements in the development of artificial intelligence technology in recent years, it has become a commonplace to demand that AI systems be trustworthy. This view presupposes that it is possible to trust AI technology in the first place. The aim of this paper is to challenge this view. In order to do that, it is argued that the philosophy of trust really revolves around the problem of how to square the epistemic and the normative dimensions (...) of trust. Given this double nature of trust it is possible to extract a threefold challenge to the defenders of the possibility of AI trust without presupposing any particular trust theory. They have to show (1) how trust in AI systems is more than mere reliance; (2) how AI systems can become objects of normative expectations; and (3) how the resulting attitude gives human agents reassurance in their interactions with AI systems. In order to demonstrate how difficult this task is, the threefold challenge is then applied to two recent accounts that defend the possibility of trust in AI systems. By way of conclusion it is suggested that instead of trusting AI systems, we should strive to make them reliable. (shrink)

Medical artificial intelligence (AI) systems are value-laden technologies that can simultaneously encourage and discourage conflicting values that may all be relevant for the pursuit of justice. I argue that the predominant theory of healthcare justice, the Rawls-inspired approach of Norman Daniels, neither adequately acknowledges such conflicts nor explains if and how they can resolved. By juxtaposing Daniels’s theory of healthcare justice with Axel Honneth’s and Nancy Fraser’s respective theories of justice, I draw attention to one such conflict. Medical AI may (...) improve the distribution of opportunity qua health while simultaneously mis-recognizing patients and thereby reducing their self-respect. I argue that justly resolving this conflict will at times require greater inclusion of those mis-recognized in deliberation about medical AI, and consider what such inclusion may entail. (shrink)

Achieving trustworthy AI is increasingly considered an essential desideratum to integrate AI systems into sensitive societal fields, such as criminal justice, finance, medicine, and healthcare, among others. For this reason, it is important to spell out clearly its characteristics, merits, and shortcomings. This article is the first survey in the specialized literature that maps out the philosophical landscape surrounding trust and trustworthiness in AI. To achieve our goals, we proceed as follows. We start by discussing philosophical positions on trust and (...) trustworthiness, focusing on interpersonal accounts of trust. This allows us to explain why trust, in its most general terms, is to be understood as reliance plus some “extra factor”. We then turn to the first part of the definition provided, i.e., reliance, and analyze two opposing approaches to establishing AI systems’ reliability. On the one hand, we consider transparency and, on the other, computational reliabilism. Subsequently, we focus on debates revolving around the “extra factor”. To this end, we consider viewpoints that most actively resist the possibility and desirability of trusting AI systems before turning to the analysis of the most prominent advocates of it. Finally, we take up the main conclusions of the previous sections and briefly point at issues that remain open and need further attention. (shrink)

The “impossibility results” in algorithmic fairness suggest that a predictive model cannot fully meet two common fairness criteria – sufficiency and separation – except under extraordinary circumstances. These findings have sparked a discussion on fairness in algorithms, prompting debates over whether predictive models can avoid unfair discrimination based on protected attributes, such as ethnicity or gender. As shown by Otto Sahlgren, however, the discussion of the impossibility results would gain from importing some of the tools developed in the philosophical literature (...) on feasibility. Utilizing these tools, Sahlgren sketches a cautiously optimistic view of how algorithmic fairness can be made feasible in restricted local decision-making. While we think it is a welcome move to inject the literature on feasibility into the debate on algorithmic fairness, Sahlgren says very little about what are the general gains of bringing in feasibility considerations in theorizing algorithmic fairness. How, more precisely, does it help us make assessments about fairness in algorithmic decision-making? This is what is addressed in this Reply. More specifically, our two-fold argument is that feasibility plays an important but limited role for algorithmic fairness. We end by offering a sketch of a framework, which may be useful for theorizing feasibility in algorithmic fairness. (shrink)

In a recent article in this journal, Baron (2025) argues that we can appropriately trust unexplainable artificial intelligence (AI) systems, so explainability is not necessary for AI’s trustworthiness. In this commentary, I argue that Baron is wrong. I first offer a positive argument for the claim that explainability is necessary for trustworthiness. Drawing on this argument, I then show that Baron’s argument for thinking otherwise fails.

This paper provides four insights relating to policy-making, focusing on the complex relationship between the abstract concept of trust—with its numerous empirical expressions—and the concept of opacity in AI technologies. First, we set the ground by discussing the nature of trust as it evolves from interpersonal to technological realms (§ 2), examining the plethora of measurement methods and objects that reflect the concept’s rich diversity. We then investigate the concept of opacity in AI systems (§ 3), challenging the conventional wisdom (...) that less opacity invariably leads to greater trust, and illustrate this relationship’s nuanced and context-dependent reality. The conclusion (§ 4) bridges these theoretical discussions with pragmatic policy-making considerations, crystallizing abstract ideas into actionable directives. We present four key insights for policy-makers: First, policies must mediate between the abstract concept of trust and the concrete outcomes of AI interactions. Second, they should mirror societal values. Third, policy-making must be adaptive, evolving alongside technological advancements to effectively address both technical and policy issues relating to trust and opacity. Lastly, we anticipate that academic discourse will inform future policies, suggesting the creation of a standardized trust index for AI systems, transparency guidelines based on opacity levels, and explainability protocols for high-stakes applications. These insights aim to ensure that AI-based technologies are governed in a manner that upholds societal values and fosters the common good. (shrink)

Aylsworth and Castro’s recent paper, Should I Use ChatGPT to Write My Papers?, argues that students in the humanities have a moral obligation to refrain from using AI tools such as ChatGPT for writing assignments. Their claim is that writing is an autonomy-fostering activity, essential for intellectual growth and critical reflection, and that every agent has the moral duty to respect their own autonomy. While the authors raise significant ethical concerns, the paper lacks the identification of which specific features of (...) writing promote autonomy. Writing is a multifaceted process which involves distinct activities, not all of which equally contribute to fostering autonomy. Furthermore, writing can be a shared endeavour, for example when co-authoring a paper with a colleague. This prompts the question of which features or activities exactly are the ones to support autonomy. The usage in AI in education is a double-edged sword: while some scholars emphasize the limitations of generated texts, others have reported positive outcomes from using AI tools in educational settings. This invites a deeper examination of how technology can be designed or used to support, rather than mine, students’ autonomy. (shrink)

This study adopts a hermeneutic, practice-based approach to Responsible Innovation to explore how a reflective and proactive attitude can be implemented in a start-up context. We hypothesised that a moral hermeneutics framework - rooted in post-phenomenology and theories on technology-induced value change - could provide a way to understand how practitioners in a start-up make sense of the different kinds of responsibilities in their work, balancing professional demands and standards of excellence with broader ecological and social commitments. Using in-depth interviews (...) with the team members of a start-up R&D laboratory, we explored how they interpret their responsibilities-as-(moral)-obligations. Our findings suggest that the syntactical ways team members make sense of the relationship between these responsibilities can be useful for understanding how reflexivity can surface in this environment. We conclude by proposing that less conciliatory interpretations of conflicting responsibilities may lead to a collective search for practical solutions addressing these tensions, as long as it is embedded in a collective dialogue involving the other members’ moral perspectives and technical expertise. (shrink)

Creating a legacy is often seen as a way to circumvent mortality. At the very least, it is a way to ensure that someone’s ideas live on and their influence on others extends beyond their own lifetime. Common examples of such legacies are cultural products, such as books, music, or art, that one creates and leaves behind. In light of rapid advances in artificial intelligence research, it is conceivable that it will soon become possible – and, to some extent, it (...) already is – to create a new type of legacy in the form of a virtual entity, such as a chatbot or an avatar, that acts as a digital replica of someone. In this paper, we investigate in what sense, if any, one could say that the original person continues to exist in their digital replica. Taking Derek Parfit’s work on the nature of personal identity as our inspiration and starting point, we suggest under what circumstances one could (and should) perceive such replicas as genuine extensions of one's personhood. (shrink)

Digital duplicates reduce the scarcity of individuals and thus may impact their instrumental and intrinsic value. I here expand upon this idea by introducing the notion of collective scarcity, which pertains to the limitations faced by social groups in maintaining their size, cohesion, and function.

Digital duplicates reduce the scarcity of individuals and thus may impact their instrumental and intrinsic value. I here expand upon this idea by introducing the notion of collective scarcity, which pertains to the limitations faced by social groups in maintaining their size, cohesion and function.

This paper asks three fundamental questions on the nature of trust: What is trust? What is trustworthiness? When is trust warranted? These discussions are then applied to the context of Human-Robot Interaction (HRI), asking whether we can trust social robots, whether they can be trustworthy, and, lastly, whether we should trust them. After revising the literature on the nature of trust and reliance on one hand, and on trust in social robots, considering both properties-based and non-properties-based views, on the other (...) hand, this paper defends that, given the current state of technology, we can be subjects of a paradoxical scenario in which there is trust without trustworthiness, i.e., human users that interact with social robots can develop something resembling interpersonal trust towards an artificial entity which cannot be trustworthy. This occurs because we perceive and treat social robots as trustworthy entities, while they seem to lack certain properties that would make them capable of being trustworthy (as well as untrustworthy). Understanding our psychology in HRI and trying to discern what social robots are (and are not) is capital when confronted with ethical issues. Some of the ethical issues that arise in the context of trust without trustworthiness will be considered to address the debate about if we should trust social robots. This paper concludes that we should, at least for now, not trust social robots, given the potential harms that can be done and the responsibility gaps that might appear when these harms are to be repaired. (shrink)

Artificial Intelligence (AI) Agents are often compared to psychopaths in popular news articles. The headlines are ‘eye-catching’, but the questions of what this analogy means or why it matters are hardly answered. The aim of this paper is to take this popular analogy ‘seriously’. By that, I mean two things. First, I aim to explore the scope of this analogy, i.e. to identify and analyse the shared properties of AI agents and psychopaths, namely, their lack of moral emotions and their (...) capacity for instrumental rationality. Second, I aim to examine the impact of the analogy. I argue that both agents, as ‘amoral calculators’, present the perfect candidates to revisit two long-standing debates on moral and criminal responsibility, regarding the necessity of moral emotions for ‘moral-agent-capacity responsibility’ and the necessity of ‘moral-agent-capacity responsibility’ for criminal responsibility. Finally, cross-examining the debates on the moral and criminal responsibility of psychopaths and AI agents is instructive and revealing. Instructive since the moral and legal treatment of psychopaths can be telling about the future treatment of AI agents (and vice versa) and revealing since it makes explicit our often-implicit philosophical commitments on the criteria of moral agency and the overarching purpose of criminal law. (shrink)

Computational Reliabilism (CR) has emerged as a promising framework for assessing the trustworthiness of AI systems, particularly in domains where complete transparency is infeasible. However, the rise of sophisticated adversarial attacks poses a significant challenge to CR’s key reliability indicators. This paper critically examines the robustness of CR in the face of evolving adversarial threats, revealing the limitations of verification and validation methods, robustness analysis, implementation history, and expert knowledge when confronted with malicious actors. Our analysis suggests that CR, in (...) its current form, is inadequate to address the dynamic nature of adversarial attacks. We argue that while CR’s core principles remain valuable, the framework must be extended to incorporate adversarial resilience, adaptive reliability criteria, and context-specific reliability thresholds. By embracing these modifications, CR can evolve to provide a more comprehensive and resilient approach to assessing AI reliability in an increasingly adversarial landscape. (shrink)

Digital platforms allow wildlife enthusiasts to share information with larger audiences than ever before. However, by heightening awareness and human interaction, they can threaten the well-being of non-human animals (henceforth “animals”). One example of this is increased stress to Snowy Owls (Bubo scandiacus) from close observation by large numbers of people made possible by accurate, timely, and widely-distributed location information. In this paper, we examine the ethics of animal privacy, wildlife observation, and information sharing on digital platforms, using Snowy Owls (...) as a case study. We begin by reviewing Snowy Owl migration and population statistics, trends in information sharing about and observation of Snowy Owls, and resulting harms to Snowy Owls. We then argue that animals have morally weighty privacy interests that ground human obligations to protect their privacy. These obligations are strengthened by additional concerns about species protection and ecological conservation. Our arguments support an array of privacy protections, including obligations on the part of wildlife observers, researchers, and other groups. However, because the catalyst for this paper is novel information technologies, we make a focused set of recommendations applicable to digital platforms. Specifically, we argue that platforms ought to restrict public information sharing when disclosure is likely to endanger individual animals, species, or ecosystems. (shrink)

This essay provides a framework for characterizing changes to social phenomena that accompany the digitalization of society. It begins by discharging two preliminary tasks: presenting the social ontology used in the analysis—a version of practice theory—and surveying extant general accounts of sociodigital phenomena to give readers a sense of the accounts on offer and to indicate through contrast how broad my account is. The starting point for the essay’s own account is the great number and variety of sociodigital phenomena and (...) changes today. The essay proposes to base a general account of this profusion on the notion of form, where the form of something is the features it exhibits in key dimensions of variation that characterize a population to which the something belongs. It is claimed that the four central dimensions of variation for social, and thus sociodigital, phenomena are (1) constitution, (2) interactions and associations, (3) spaces and times, and (4) meaning and significance. Following discussion of this claim, sketches are offered of the forms of four pre-digital social phenomena and four sociodigital ones. A final section offers general observations about how sociodigital phenomena are different in these four dimensions from their pre-digital counterparts. (shrink)

Artificial intelligent (AI) systems used in medicine are often very reliable and accurate, but at the price of their being increasingly opaque. This raises the question whether a system’s opacity undermines the ability of medical doctors to acquire knowledge on the basis of its outputs. We investigate this question by focusing on a case in which a patient’s risk of recurring breast cancer is predicted by an opaque AI system. We argue that, given the system’s opacity, as well as the (...) possibility of malfunctioning AI systems, practitioners’ inability to check the correctness of their outputs, and the high stakes of such cases, the knowledge of medical practitioners is indeed undermined. They are lucky to form true beliefs based on the AI systems’ outputs, and knowledge is incompatible with luck. We supplement this claim with a specific version of the safety condition on knowledge, Safety*. We argue that, relative to the perspective of the medical doctor in our example case, his relevant beliefs could easily be false, and this despite his evidence that the AI system functions reliably. Assuming that Safety* is necessary for knowledge, the practitioner therefore doesn’t know. We address three objections to our proposal before turning to practical suggestions for improving the epistemic situation of medical doctors. (shrink)

Current Generative Artificial Intelligence models have become incredibly close to the human level of linguistic and artistic excellence, defying our conception of artworks as uniquely human products, resulting in an authorship problem, i.e., whether artificial agents can be regarded as genuine authors of their products. This paper provides a definition of institutional authorship to evaluate this possibility, using John Searle’s Speech Act Theory and Theory of Institutions. To apply the definition, we focus on artistic cases, assuming the institutional theory of (...) art. According to the institutional authorship definition proposed, authorship is the product of an agent’s declarative acts in the context of relevant institutions granting the agent’s deontic powers for performing such acts. These acts depend on sincerity conditions encompassing intentional states, which bind the product with an agent. We conclude that artificial agents cannot satisfy the institutional definition of authorship due to their inability to be institutional agents, meaning they cannot be authors. (shrink)

This article defends the claim that diversions, which are actions that cause distraction, are a unique way to modify someone’s behavior and that they are morally salient. While the focus of this article is dedicated to understanding the moral features of attention and diversion, it is crucial to keep in mind that the moral evaluation of these concepts is most pressing within a technological society deeply intertwined with an attention economy. We are inundated with distraction technologies, which are technologies whose (...) function partly or wholly depends on capturing the attention of its targets. Distraction technologies are widely used to capture the attention of billions of people. Once we come to treat diversions as unique moral actions, it will be clear that the most serious violations to our right to attention can be committed by those who control these technologies. (shrink)

This paper offers a political analysis of the platform and sharing economy—an economic model in which digital platforms facilitate social and economic interactions. Its two central models, mainstream and cooperative platforms, offer similar applications and services (e.g., home-sharing, food delivery), but fundamentally differ in their ownership and governance structures, economic models, and technical designs. Building on literature from the politics of technology (PoT), the paper develops an approach for the political analysis of platform technologies, combining central components from the works (...) of Winner, Feenberg, and Pfaffenberger. This approach is then applied to analyze the platform and sharing economy, highlighting the political significance of platform technologies. The analysis reveals three key insights. First, when incorporated into particular social arrangements, digital platforms become means for shaping social realities rather than mere tools for specific uses. Second, mainstream platforms perpetuate capitalist conditions in the digital sphere and therefore necessitate platform capitalism to function, whereas cooperative platforms resist and undermine it. Third, the dynamics between the platform models embody a struggle over the question of the good life in the digital economy. Additionally, the paper uncovers a philosophical weakness in Winner’s definition of “inherently political technologies” that warrants further attention in PoT literature. (shrink)

A broad set of public policy debates concern the limits of humanity’s control over nature. Attitudes towards such topics are not well explained by the standard 2-dimensional political model favored by political scientists of i) a left/right economic spectrum and ii) a liberal/authoritarian social spectrum. I pose a new, orthogonal, political spectrum to fill the void. It is a spectrum of value held for, on the one hand, nature, and on the other, technological progress. This harks back to the 18th (...) Century Enlightenment and Romanticism, but manifests again and again, including in recent debates about human enhancement biotechnology. (shrink)

In anticipation of the market introduction of highly and fully automated vehicles, regulations for their behavior in public road traffic are emerging in various countries and regions. Yet, as we show using the example of EU and German regulations, these rules are both incomplete and exceptionally vague. In this paper we introduce three traffic scenarios highlighting conflicting ethical, legal, and utility-related claims, and perform a legal analysis with regards to the expected behavior of AVs in these scenarios. We show that (...) the existing regulatory framework disregards the realities of algorithmic decision-making in automated vehicles, such as the incomplete and imprecise perception of their environment and the probabilistic nature of their predictions. Importantly, the current regulations are written in abstract language addressing human interpreters rather than the precise logical-numerical computer code. We argue that the required interpretation and translation of the abstract legal language into the logical-numerical domain is so ambiguous that the regulations as they stand fail to guide or limit automated vehicle behavior in any meaningful way. This comes with significant ethical implications, as the interpretation and translation is unavoidable and, if not provided by regulatory bodies, will have to be performed by manufacturers. We argue that ethical decisions with significant impact on public safety must not be delegated to private companies, and thus, regulatory frameworks need significant improvements. (shrink)

This paper explores the limitations of artificial intelligence (AI) in fulfilling the obligations inherent in close personal relationships, particularly in the context of loneliness. While AI technologies may offer some of the goods that we associate with close personal relationships, they lack the capacity for genuine commitment and individualized care that characterize human interactions. The finitude of human existence—our cognitive, emotional, and temporal limitations— and our capacity to make judgments concerning distinct kinds of value imbues human relationships with significance that (...) AI cannot replicate. (shrink)