It has been extensively argued that emerging autonomous technologies can represent a challenge for our traditional responsibility practices. Though these challenges differ in a variety of ways, at the center of these challenges is the worrying possibility that there may be outcomes of autonomous technologies for which there are legitimate demands for responsibility but no legitimate target to bear this responsibility. This is well exemplified by the possibility of techno-responsibility gaps. These challenges have elicited a number of responses, including dismissals (...) of the legitimacy of these demands, attempts to find proximate agents that can be legitimately held responsible, and arguments for prohibiting the use of technologies that may open such gaps. In this piece we present a general argument that an overlooked but valuable option lies in adopting a strategy of taking responsibility for the outcomes of autonomous technologies even when the conditions for being legitimately held responsible are not met. We develop a general argument that the adoption of such a strategy is often justified not only by the demands of being responsible, but by practical considerations rooted in our relationships: the need to preserve of the quality of our relationships and the trustworthiness of the socio-technical system of which the autonomous technology is both a result of and embedded in. (shrink)

In a recent thought-provoking essay called “ChatGPT is Bullshit,” Hicks, Humphries and Slater call such large language models (LLMs) “bullshitters” and “bullshit machines.” Unlike the term “bullshit,” they argue, commonly used anthropomorphic terms such as “hallucination” and “confabulation” mispresent LLMs and sow confusion that could be socially harmful. This paper criticizes their essay in two steps. First, its reliance on Harry Frankfurt’s classic characterization of bullshit as indifference to truth, though understandable and compelling in one sense, risks misrepresenting LLMs. Second, (...) the argument is too quick to jettison anthropomorphic terms like hallucination and confabulation, which might prove useful metaphors for understanding generative AI. Exploring language to articulate good ways of understanding LLMs is indeed a socially important task, one benefitting from critical open-mindedness, some historical awareness, and a nuanced approach to how various words used to describe AI can operate. This paper attempts to contribute to this task by questioning the wisdom of categorically calling bullshit on ChatGPT. (shrink)

This study examines how ethical standards established by stakeholders such as developers and policymakers provide top-down guidance aligned with deontological ethics or utilitarian goals. It also highlights a complementary bottom-up approach, rooted in virtue ethics, in which individuals engage in ethical deliberations shaped by their moral values. Both approaches have limitations, and, at times, ethical standards can clash with moral values, thus blurring lines of responsibilities. Deontological principles may offer a structured framework, but often lack adaptability to diverse cultural contexts; (...) bottom-up approaches foster intrinsic moral intentions, but universal applicability may be challenging, thus raising moral dilemmas. Through a theoretical literature review, this study explains how different ontological and normative ethical perceptions lead to moral dilemmas in various AI application scenarios (e.g., algorithmically managed platforms, crime detection systems, medical AI assistants). It addresses top-down and bottom-up approaches that may help account for moral dilemmas ethically. The study discusses the balance between top-down regulatory frameworks and bottom-up community-driven ethics to navigate the complex ethical landscape of AI applications, whose increasing capabilities alter expectations of AI’s agency and morality. This study calls for holistic and multi-objective ethical frameworks that incorporate diverse normative ethical perspectives and recognizes context-specific ontologies throughout the AI lifecycle. It emphasizes a nuanced and context-specific combination of top-down standards (e.g., regulatory oversight, clear guidelines) and bottom-up fostering of moral values (e.g., by improving ethical knowledge). This tailored and ongoing reflection of ethical standards and moral values accounts for an ethical development, deployment, and utilization of AI technologies. (shrink)

Machine learning (ML) systems are vulnerable to performance decline over time due to dataset shift. To address this problem, experts often suggest that ML systems should be regularly updated to ensure ongoing performance stability. Some scholarly literature has begun to address the epistemic and ethical challenges associated with different updating methodologies. Thus far, however, little attention has been paid to the impact of model updating on the ML-assisted decision-making process itself, particularly in the AI ethics and AI epistemology literatures. This (...) article aims to address this gap in the literature. It argues that model updating introduces a new sub-type of opacity into ML-assisted decision-making – update opacity – that occurs when users cannot understand how or why an update has changed the reasoning or behaviour of an ML system. This type of opacity presents a variety of distinctive epistemic and safety concerns that available solutions to the black box problem in ML are largely ill-equipped to address. A variety of alternative strategies may be developed or pursued to address the problem of update opacity more directly, including bi-factual explanations, dynamic model reporting, and update compatibility. However, each of these strategies presents its own risks or carries significant limitations. Further research will be needed to address the epistemic and safety concerns associated with model updating and update opacity going forward. (shrink)

Machine learning (ML) systems are vulnerable to performance decline over time due to dataset shift. To address this problem, experts often suggest that ML systems should be regularly updated to ensure ongoing performance stability. Some scholarly literature has begun to address the epistemic and ethical challenges associated with different updating methodologies. Thus far, however, little attention has been paid to the impact of model updating on the ML-assisted decision-making process itself. This article aims to address this gap. It argues that (...) model updating introduces a new sub-type of opacity into ML-assisted decision-making—update opacity—that occurs when users cannot understand how or why an update has changed the reasoning or behaviour of an ML system. This type of opacity presents a variety of distinctive epistemic and safety concerns that available solutions to the black box problem in ML are largely ill-equipped to address. A variety of alternative strategies may be developed or pursued to address the problem of update opacity more directly, including bi-factual explanations, dynamic model reporting, and update compatibility. However, each of these strategies presents its own risks or carries significant limitations. Further research will be needed to address the epistemic and safety concerns associated with model updating and update opacity going forward. (shrink)

This paper considers how act consequentialism can engage with the Gamer’s Dilemma (the seeming moral distinction between virtual murder and virtual child molestation in video games). I argue that often the dilemma is implicitly or explicitly framed in a way that presumes that an answer can only be in terms of action types rather than tokens. This framing unfairly stacks the deck against act consequentialist approaches. An act consequentialist approach to the Gamer’s Dilemma speaks to its own set of concerns, (...) and this paper proposes three ways consequentialists can address the dilemma: adopting rule utilitarianism, taking a regulative approach focused on laws and policies, or examining typical cases to develop heuristics. Each of these approaches have advantages and disadvantages, and it’s important to recognize what the specific concerns of each approach are, as well as their particular strengths and limitations. By clarifying these different consequentialist framings, I demonstrate ways in which consequentialism can meaningfully contribute to debates on the ethics of virtual actions. I conclude by outlining some methodological benefits of bringing clarity to the particular way(s) that consequentialism can speak to the Gamer’s Dilemma and suggest how existing discussions using deontological and virtue ethical approaches might be productively reframed in consequentialist terms. (shrink)

Authorship Attribution (AA) approaches in Natural Language Processing (NLP) are important in various domains, including forensic analysis and cybercrime. However, they pose Ethical, Legal, and Societal Implications/Aspects (ELSI/ELSA) challenges that remain underexplored. Inspired by foundational AI ethics guidelines and frameworks, this research introduces a comprehensive framework of responsible guidelines that focuses on AA tasks in NLP, which are tailored to different stakeholders and development phases. These guidelines are structured around four core principles: privacy and data protection, fairness and non-discrimination, transparency (...) and explainability, and societal impact. Furthermore, to illustrate a practical application of our guidelines, we apply them to a recent AA study that targets identifying and linking potential human trafficking vendors. We believe the proposed guidelines can assist researchers and practitioners in justifying their decisions, assisting ethical committees in promoting responsible practices, and identifying ethical concerns related to NLP-based AA approaches. Our study aims to contribute to ensuring the responsible development and deployment of AA tools. (shrink)

A multimillion-dollar digital voter deterrence ad industry has emerged in the United States. Beginning with the 2016 US Presidential Election, interested parties ranging from international intelligence agencies to campaigning politicians have enlisted social media platforms’ microtargeted advertising infrastructure to inundate certain voter demographics with anti-voting content. These tactics are used disproportionately against people of color, especially Black voters, and remain virtually unregulated and unaddressed. This paper responds to the issue in two ways. First, I strive to catalyze a broader awareness (...) of this practice and the intertwined moral and epistemic wrongs that it entails. The wrongs of this digital industry occur simultaneously at three levels: individually, these types of ads undermine autonomy; socially, they constitute the perpetration of discriminatory voter suppression and epistemic injustice; and historically, they reify racial inequality by perpetuating targeted disenfranchisement efforts against non-white voters. Second, I argue that this type of content is instructive for the ongoing debate regarding social media platforms’ moderation of online political speech. The arguments I develop here provide a compelling case for the moderation of political content that seeks to disenfranchise, especially when those efforts are undertaken in discriminatory ways. This holds urgent implications for how social media platforms moderate and profit from suppressive content. (shrink)

Urban Digital Twins (UDTs) have become the new buzzword for researchers, planners, policymakers, and industry experts when it comes to designing, planning, and managing sustainable and efficient cities. It encapsulates the last iteration of the technocratic and ultra-efficient, post-modernist vision of smart cities. However, while more applications branded as UDTs appear around the world, its conceptualization remains ambiguous. Beyond being technically prescriptive about what UDTs are, this article focuses on their aspects of interaction and operationalization in connection to people in (...) cities, and how enhanced by metaverse ideas they can deepen societal divides by offering divergent urban experiences based on different stakeholder preferences. Therefore, firstly this article repositions the term UDTs by comparing existing concrete and located applications that have a focus on interaction and participation, including some that may be closer to the concept of UDT than is commonly assumed. Based on the components found separately in the different studied cases, it is possible to hypothesize about possible future, more advanced realizations of UDTs. This enables us to contrast their positive and negative societal impacts. While the development of new immersive interactive digital worlds can improve planning using collective knowledge for more inclusive and diverse cities, they pose significant risks not only the common ones regarding privacy, transparency, or fairness, but also social fragmentation based on urban digital multiplicities. The potential benefits and challenges of integrating this multiplicity of UDTs into participatory urban governance emphasize the need for human-centric approaches to promote socio-technical frameworks able to mitigate risks as social division. (shrink)

In a bid to boost their below-replacement fertility levels, some countries, such as China, India, Iran, and Japan, have launched state-sponsored dating apps, with more potentially following. However, the use of dating apps as tools for social engineering has been largely neglected by political theorists and public policy experts. This article fills this gap. While acknowledging the risks and historical baggage of social engineering, the article provides a qualified defense of using these apps for three purposes: raising below-replacement birth rates, (...) increasing social cohesion, and preventing certain genetic diseases. (shrink)

Despite the widespread use of automated decision-making (ADM) systems, they are often developed without involving the public or those directly affected, leading to concerns about systematic biases that may perpetuate structural injustices. Existing formal fairness approaches primarily focus on statistical outcomes across demographic groups or individual fairness, yet these methods reveal ambiguities and limitations in addressing fairness comprehensively. This paper argues for a holistic approach to algorithmic fairness that integrates procedural fairness, considering both decision-making processes and their outcomes. Procedural fairness (...) emphasizes the importance of fair decision-making procedures, which aligns with theories of relational justice that stress the quality of social relations and power dynamics. We highlight the need for substantive procedural fairness to ensure better outcomes and address forward-looking responsibilities. Additionally, we propose leveraging Public Engagement, a core dimension within the well-established Responsible Research and Innovation framework, to enhance procedural fairness in ADM systems. Our contribution underscores the value of Public Engagement in fostering fairer ADM processes, thereby expanding the current focus beyond technical outcome-based approaches to encompass broader procedural considerations. (shrink)

Who is responsible for a harm caused by AI, or a machine or system that relies on artificial intelligence? Given that current AI is neither conscious nor sentient, it’s unclear that AI itself is responsible for it. But given that AI acts independently of its developer or user, it’s also unclear that the developer or user is responsible for the harm. This gives rise to the so-called responsibility gap: cases where AI causes a harm, but no one is responsible for (...) it. Two central questions in the literature are whether responsibility gaps exist, and if yes, whether they’re morally problematic in a way that counts against developing or using AI. While some authors argue that responsibility gaps exist, and they’re morally problematic, some argue that they don’t exist. In this paper, I defend a novel position. First, I argue that current AI doesn’t generate a new kind of concern about responsibility that the older technologies don’t. Then, I argue that responsibility gaps exist but they’re unproblematic. (NOTE: Email me for a copy.). (shrink)

This paper engages the ongoing debate around the possibility of virtue friendships in the Aristotelian sense through online mediation. However, I argue that since the current literature has remained overly focused on the mere possibility of virtual friendship, it has obscured the more common phenomena of using digital communication to sustain previous in-person friendships which are now at a distance. While I agree with those who argue that entirely virtual friendship is possible, I argue that the current rebuttals to the (...) objections regarding embodiment are lacking. We can see this more clearly when we shift our consideration to more ordinary cases of virtual friendship. I discuss this problem of disembodiment by engaging with an interpretive issue of Aristotle’s “living together” which has played a key role in the current literature. I then turn to expand the current debate by examining how physical absence, disembodiment, and the technology we use to overcome such disembodiment, shapes and impacts friendships at a distance. In short, I explore the tendencies of technology of which virtual friends ought to be aware. I argue that even though virtual virtue friendships are possible via technology, the medium makes it harder and limits the possible experiences within friendships. Even if we overcome these tendencies, technologically mediated virtue friendships require more discipline for a diminished experience. While these limitations and extra disciplines in no way make virtual friendship unimportant, I argue that the fullest possible virtual friendship will be lesser than the fullest possible traditional friendship. (shrink)

Raul Hakli & Pekka Mäkelä (2016, 2019) make a popular assumption in machine ethics explicit by arguing that artificial agents cannot be responsible because they are designed. Designed agents, they think, are analogous to manipulated humans and therefore not meaningfully in control of their actions. Contrary to this, I argue that under all mainstream theories of responsibility, designed agents can be responsible. To do so, I identify the closest parallel discussion in the literature on responsibility and free will, which concerns (...) ‘design cases’. Design cases are theoretical examples of agents that appear to lack responsibility because they were designed, philosophers use these cases to explore the relationship between design and responsibility. This paper presents several replies to design cases from the responsibility literature and uses those replies to situate the corresponding positions on the design and responsibility of artificial agents in machine ethics. I argue that each reply can support the design of responsible agents. However, each reply also entails different levels of severity in the constraints for the design of responsible agents. I offer a brief discussion of the nature of those constraints, highlighting the challenges respective to each reply. I conclude that designing responsible agents is possible, with the caveat that the difficulty of doing so will vary according to one’s favoured reply to design cases. (shrink)

This paper examines the ongoing challenges of interdisciplinary collaboration in Machine Ethics (ME), particularly the integration of ethical decision-making capacities into AI systems. Despite increasing demands for ethical AI, ethicists often remain on the sidelines, contributing primarily to metaethical discussions without directly influencing the development of moral machines. This paper revisits concerns highlighted by Tolmeijer et al. (2020), who identified the pitfall that computer scientists may misinterpret ethical theories without philosophical input. Using the MACHIAVELLI moral benchmark and the Delphi artificial (...) moral agent as case studies, we analyze how these challenges persist. Our analysis indicates that the creators of MACHIAVELLI and Delphi “copy” ethical concepts and embed them in LLMs without questioning or challenging these concepts themselves sufficiently. If an ethical concept causes friction with the computer code, they only reduce and simplify the ethical concept in order to stay as close as possible to the original. We propose that ME should expand its focus to include both interdisciplinary efforts that embed existing ethical work into AI, and transdisciplinary research that fosters new interpretations of ethical concepts. Interdisciplinary and transdisciplinary approaches are crucial for creating AI systems that are not only effective but also socially responsible. To enhance collaboration between ethicists and computer scientists, we recommend the use of Socratic Dialogue as a methodological tool, promoting deeper understanding of key terms and more effective integration of ethics in AI development. (shrink)

In this article, firstly, cooperative interaction between robots and humans is discussed; specifically, the possibility of human/robot joint action and (relatedly) the possibility of robots occupying institutional roles alongside humans. The discussion makes use of concepts developed in social ontology. Secondly, certain key moral (or ethical—these terms are used interchangeably here) issues arising from this cooperative action are discussed, specifically issues that arise from robots performing (including qua role occupants) morally significant actions jointly with humans. Such morally significant human/robot joint (...) actions, supposing they exist, could potentially range from humans and robots jointly caring for the infirm through to jointly killing enemy combatants. (shrink)

Research under real-world conditions is crucial to the development and deployment of robust AI systems. Exposing large language models to complex use settings yields knowledge about their performance and impact, which cannot be obtained under controlled laboratory conditions or through anticipatory methods. This epistemic need for real-world research is exacerbated by large-language models’ opaque internal operations and potential for emergent behavior. However, despite its epistemic value and widespread application, the ethics of real-world AI research has received little scholarly attention. To (...) address this gap, this paper provides an analysis of real-world research with LLMs and generative AI, assessing both its epistemic value and ethical concerns such as the potential for interpersonal and societal research harms, the increased privatization of AI learning, and the unjust distribution of benefits and risks. This paper discusses these concerns alongside four moral principles influencing research ethics standards: non-maleficence, beneficence, respect for autonomy, and distributive justice. I argue that real-world AI research faces challenges in meeting these principles and that these challenges are exacerbated by absent or imperfect current ethical governance. Finally, I chart two distinct but compatible ways forward: through ethical compliance and regulation and through moral education and cultivation. (shrink)

The paper develops a conception of the good life within a digitalized society. Martha Nussbaum’s capability theory offers an adequate normative framework for that purpose as it systematically integrates the analysis of flourishing human lives with a normative theory of justice. The paper argues that a theory of good digital lives should focus on everyday life, on the impact digitalization has on ordinary actions, routines and corresponding practical knowledge. Based on Nussbaum’s work, the paper develops a concept of digital capabilities. (...) Digital capabilities are combined capabilities: To possess a digital capability, an individual must acquire certain skills and abilities (internal capabilities) and needs access to devices and external infrastructures like internet connections. If societies as a whole and everyday environments are digitalized to a certain degree, the possession of specific digital capabilities is a crucial precondition for a flourishing life. The paper likewise analyzes challenges that are connected to digital capabilities. Digital structures are constantly changing. In consequence, digital capabilities are never acquired once and for all, but always precarious and in danger of being lost—with serious consequences for individual everyday lives in digitalized environments. As digital capabilities are crucial for leading a good life, people are entitled to develop and maintain them. They describe demands of justice. Using the examples of filling in an online form and digital education, the paper finally illustrates the size of institutional changes that are necessary to meet these demands. (shrink)

Explanations are conceived to ensure the trustworthiness of AI systems. Yet, relying solemnly on algorithmic solutions, as provided by explainable artificial intelligence (XAI), might fall short to account for sociotechnical risks jeopardizing their factuality and informativeness. To mitigate these risks, we delve into the complex landscape of ethical risks surrounding XAI systems and their generated explanations. By employing a literature review combined with rigorous thematic analysis, we uncover a diverse array of technical risks tied to the robustness, fairness, and evaluation (...) of XAI systems. Furthermore, we address a broader range of contextual risks jeopardizing their security, accountability, reception alongside other cognitive, social, and ethical concerns of explanations. We advance a multi-layered risk assessment framework, where each layer advances strategies for practical intervention, management, and documentation of XAI systems within organizations. Recognizing the theoretical nature of the framework advanced, we discuss it in a conceptual case study. For the XAI community, our multifaceted investigation represents a path to practically address XAI risks while enriching our understanding of the ethical ramifications of incorporating XAI in decision-making processes. (shrink)

Responsibility gaps traditionally refer to scenarios in which no one is responsible for harm caused by artificial agents, such as autonomous machines or collective agents. By carefully examining the different ways this concept has been defined in the social ontology and ethics of technology literature, I argue that our current concept of responsibility gaps is defective. To address this conceptual flaw, I argue that the concept of responsibility gaps should be revised by distinguishing it into two more precise concepts: epistemic (...) responsibility gaps (ERG) and control misalignments (CM). ERG functions as a refined version addressing specific aspects of responsibility attribution. It captures the challenge of identifying who is to blame. CM, on the other hand, introduces a new dimension by focusing on misalignments of control. These revisions offer better guidelines for the development and oversight of artificial agents, both in terms of assigning blame and ensuring the right distribution of control. (shrink)

With the rapid expansion of parameters in large language models (LLMs) and the application of Reinforcement Learning with Human Feedback (RLHF), there has been a noticeable growth in the moral competence of LLMs. However, several questions warrant further exploration: Is it really possible for LLMs to fully align with human values through RLHF? How can the current moral growth be philosophically contextualized? We identify similarities between LLMs’ moral growth and Deweyan ethics in terms of the discourse of human moral development. (...) We then attempt to use Dewey’s theory on an experimental basis to examine and further explain the extent to which the current alignment pathway enables the development of LLMs. A beating experiment serves as the foundational case for analyzing LLMs’ moral competence across various parameters and stages, including basic moral cognition, moral dilemma judgment, and moral behavior. The results demonstrate that the moral competence of the GPT series has seen a significant improvement, and Dewey’s Impulse-Habit-Character theory of moral development can be used to explain this: the moral competence of LLMs has been enhanced through experience-based learning, supported by human feedback. Nevertheless, LLMs’ moral development through RLHF remains constrained and does not reach the character stage described by Dewey, possibly due to their lack of self-consciousness. This fundamental difference between humans and LLMs underscores both the limitations of LLMs’ moral growth and the challenges of applying RLHF for AI alignment. It also emphasizes the need for external societal governance and legal regulation. (shrink)

Customer ratings and reviews are playing a key role in the contemporary platform economy. To establish trust among stran- gers without having to directly monitor platform users themselves, companies ask people to evaluate each other. Firms like Uber, Deliveroo, or Airbnb construct digital reputation scores by combining these consumer data with their own information from the algorithmic surveillance of workers. Trustworthy behavior is subsequently rewarded with a good reputation score and higher potential earnings, while untrustworthy behavior can be algorithmically penalized. (...) However, can these technolo- gies themselves be trusted? Are they reliable communicators of trustworthy market behavior and do they fairly distribute the benefits and burdens of the online economy among platform companies, consumers, and workers? The implementation of Digital Reputation Systems (DRSs) has obvious advantages, like reducing adjudication costs and enhancing consumer trust, but there are also disadvantages that are predominantly offloaded on workers. Taking Airbnb’s DRS as my case study, I use Stephan Lessenich’s theory of the externalization society to argue that DRSs concentrate the benefits of the platform economy among consumers and platform companies while pushing negative externalities mostly unto workers. (shrink)