AI and ethics最新文献

{"title":"Exploring latent heterogeneity in AI responsibility: a person-centered analysis of ethical dispositions","authors":"Yanyi Chen,&nbsp;Samad Zare","doi":"10.1007/s43681-026-01363-1","DOIUrl":"10.1007/s43681-026-01363-1","url":null,"abstract":"<div><p>The increasing presence of artificial intelligence (AI) in higher education has intensified questions about what constitutes responsible use. Although institutional guidelines often presume a shared understanding of ethical AI conduct, the extent to which individuals differ in their orientations toward AI responsibility remains insufficiently examined. This study investigates such variation by examining how students conceptualize and enact responsible AI behavior. Using data from a large sample of higher education students, a person-centered analytical approach was applied to identify patterned differences in ethical dispositions toward AI. Three distinct responsibility profiles were identified, namely High Responsibility, Moderate Responsibility, and Low Responsibility, reflecting consistent, situational, and limited commitments to responsible AI use. These profiles differed systematically in students’ value orientations, self-beliefs, and experiences of academic strain, indicating that ethical engagement with AI reflects meaningful psychological heterogeneity rather than a uniform user perspective. Instead of framing lower responsibility as simple non-compliance, the findings suggest differential ethical readiness and vulnerability across profiles, highlighting the limitations of generic AI ethics policies. The study contributes empirical grounding to ongoing debates in digital ethics and AI governance by demonstrating the need for differentiated, context-aware approaches to supporting ethical AI engagement in educational settings.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148987026","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Taxonomies for connecting technical components of AI-based fraud detection systems and fairness approaches","authors":"Sebastiaan Berendsen,&nbsp;Emma Beauxis-Aussalet","doi":"10.1007/s43681-026-01375-x","DOIUrl":"10.1007/s43681-026-01375-x","url":null,"abstract":"<div><p>Our study investigates the inter-dependencies between the technical components of AI-based fraud detection systems and fairness approaches. Here, fairness approaches refer to any method or process to address, specify, measure or mitigate unfairness in AI systems. Despite extensive research on both AI-based fraud detection systems and AI fairness, these domains are often treated separately, leading to a fragmented understanding of both domains. This fragmentation leads to inconsistent terminologies and concepts, and fails to identify the inter-dependencies between technical components of fraud detection systems and fairness approaches. These are the research gaps we address in this study. Through a literature review of 78 papers, we synthesize existing concepts and terminologies into two structured taxonomies that allow to document these inter-dependencies. One taxonomy describes the technical components of AI fraud detection systems and another describes fairness approaches. A joint evaluation of the taxonomies reveals how technical components are impacting fairness, and how fairness considerations should govern the design and implementation of these technical components. Our findings show that four technical components (regarding class imbalance, overlapping class distributions, data quality, and adaptive learning) are intertwined with fairness approaches in fraud detection. The proposed taxonomies, and the inter-dependencies we identify, support AI practitioners and decision-makers in designing fair fraud detection systems. Additionally, the taxonomies provide structure to research investigating how technical components and technical decisions impact fairness and, vice versa, investigating how fairness approaches impact the technical design of fraud detection systems.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01375-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The pitfalls of AI detection in academic writing: bias, false positives, and the need for inclusive assessment","authors":"Victor Angelier","doi":"10.1007/s43681-026-01376-w","DOIUrl":"10.1007/s43681-026-01376-w","url":null,"abstract":"<div><p>AI-text detectors are increasingly used to police authorship in student assessment and scholarly publishing. This paper argues that they are the wrong instrument for that task. Its design combines a critical synthesis of empirical, information-theoretic, and policy evidence with an original documented exploratory multi-tool audit. The review synthesises empirical evaluations, information-theoretic results, and institutional policy decisions to develop a conceptual account of what detectors can directly observe: statistical or learned properties of submitted text, not provenance itself. Predictability-related properties such as low perplexity, reduced variation in local predictability, and formulaic density provide one mechanism through which both additional-language writing and highly conventional academic prose may become vulnerable to machine-authorship classification. While a growing number of universities across several national higher-education systems have disabled or restricted AI detection in high-stakes assessment, and a New York court has annulled a misconduct finding that rested on a detector score, detector reliance persists in three under-examined arenas: individual grading practice at scale, scholarly journal screening, and institutions without formal policy revision. The paper adds a documented exploratory multi-tool audit in which a single human-authored manuscript, submitted to five commercial detectors on the same day, received materially different classifications spanning from “0% human” to “Human Generated”, demonstrating substantial cross-tool inconsistency and limited independent verifiability of the underlying classifications. A paired exploratory comparison further found that extensive machine rewriting by a commercial “humanizer” materially degraded the manuscript while producing only a small change in the reported detection score. The paper distinguishes genuine false positives from definitional boundary cases involving AI-assisted translation and polishing, engages the counter-argument that AI-polished prose may itself flatten linguistic diversity in academic English, and proposes a concrete process-oriented assessment model based on declared starting points, versioned drafting, process artefacts, and targeted oral verification. The model predates generative AI in its component practices but acquires renewed relevance because final-text classification cannot establish authorship, understanding, or misconduct on its own.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01376-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148987016","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"How ethics operationalization breaks down in practice: expert perspectives from healthcare AI","authors":"Riana Minocher,&nbsp;Isabel Velarde,&nbsp;Valerie Buerger,&nbsp;Gabriele Pluktaite,&nbsp;Vince I. Madai","doi":"10.1007/s43681-026-01381-z","DOIUrl":"10.1007/s43681-026-01381-z","url":null,"abstract":"<div><p>Efforts to promote “Trustworthy AI” have focused on operationalizing high-level ethical principles through tools, standards, and governance mechanisms. Yet a persistent gap remains between ethical guidance and day-to-day development practice. We conducted semi-structured interviews with ten experts in AI ethics, healthcare AI, and socio-technical research to examine how ethical principles are operationalized in real-world AI development. Participants described several ways in which efforts to operationalize ethical principles break down in practice. While existing operationalization approaches were viewed as useful, participants discussed conceptual, epistemic, institutional, and socio-technical conditions that constrain their effectiveness. Our findings suggest that the operationalization gap cannot be understood independently of the broader contexts in which AI systems are designed, developed, and governed.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01381-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148987019","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Workplace AI ethics through a multi-stakeholder lens: a qualitative inquiry into responsibility, trust, and the limits of principles","authors":"Andrej Miklosik,&nbsp;Tanja Rosenbaum","doi":"10.1007/s43681-026-01344-4","DOIUrl":"10.1007/s43681-026-01344-4","url":null,"abstract":"<div><p>This study examines how a heterogeneous group of workplace stakeholders interprets the ethical risks, responsibilities, and opportunities associated with the adoption of artificial intelligence (AI) at work, and considers how their lived accounts align with—and depart from—the principle-based frameworks that currently dominate AI ethics policy. Seven semi-structured interviews were conducted in October and November 2024 with stakeholders spanning seven distinct workplace roles (ethicist, AI expert, entrepreneur, social scientist, politician, employee, artist). Interviews were conducted in German, translated, transcribed, member-checked, and coded in MaxQDA through an abductive, three-stage coding procedure (open, axial, and selective coding), in which an a priori topic structure derived from the interview guide was placed in dialogue with data-driven codes. Coded themes were subsequently mapped against the five-principle AI ethics framework that has emerged in international policy discourse (transparency, justice and fairness, non-maleficence, responsibility, privacy). Six recurring concerns structured the data: responsibility and control; impact on employment and work; societal power structures; cultural and creative dimensions; trust and transparency; and ethics and values. Across roles, participants converged on the irreducibility of human responsibility and on the demand for verifiable transparency, but diverged sharply on questions of cultural homogenisation, the locus of accountability between developers and deployers, and the appropriate regulatory response. The pattern is consistent with what the literature describes as the gap between principle and practice in AI ethics: high-level principles enjoy broad assent, but their operational interpretation is contested role by role. The study contributes an exploratory, multi-stakeholder qualitative account of workplace AI ethics in which seven deliberately heterogeneous roles speak to a common set of themes. Findings are positioned as hypothesis-generating for subsequent, properly powered work, and connect the workplace-AI-ethics literature to the principle-based debate within AI governance scholarship. For managers, the data point to the need for verifiable transparency mechanisms (audits, logs, explanations) and to a redistribution of accountability across developer, deployer, and operator. For regulators, the findings illustrate why the high-risk classification of workplace AI under the EU AI Act is welcomed by frontline stakeholders but also why its meaningful implementation requires going beyond principles toward enforceable practice.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"What is AI ethics, and what is it good for? Clarifying debates on AI ethics","authors":"Astrid Schomäcker,&nbsp;Timo Speith,&nbsp;Luca Deck,&nbsp;Thorsten Helfer,&nbsp;Hanwei Zhang,&nbsp;Lena Kästner","doi":"10.1007/s43681-026-01326-6","DOIUrl":"10.1007/s43681-026-01326-6","url":null,"abstract":"<div><p>The rapid integration of artificial intelligence (AI) into various aspects of society has raised pressing ethical concerns, from privacy and fairness to autonomy and safety. In response, the field of AI ethics has grown significantly, with a particular focus on developing guidelines to support the responsible development and use of AI. However, this trend has faced criticism, with concerns about “ethics washing” and skepticism regarding the overall value of AI ethics. In this paper, we argue that such criticisms often rely on an incomplete understanding of AI ethics. To address this, we propose an understanding of AI ethics that distinguishes between diverse <i>instances</i> of AI ethics (viz., research, education, practice, and governance instruments) and the <i>purposes</i> they serve (viz., knowledge, communication, and application). By disentangling these elements, we can illuminate different ways of doing AI ethics and provide a structured way to assess their value. Our arguments highlight that critiques of specific instances and purposes of AI ethics do not justify dismissing AI ethics altogether. We conclude by demonstrating how our proposed distinctions can reconcile existing defenses of AI ethics and guide future discourse, ensuring that the field remains both robust and relevant in addressing the ethical challenges posed by AI.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01326-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986674","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The synthetic moral agent (SMA) framework: navigating the ontological void and the crisis of stewardship in the age of generative AI","authors":"Blaine Garfolo","doi":"10.1007/s43681-026-01377-9","DOIUrl":"10.1007/s43681-026-01377-9","url":null,"abstract":"<div><p>When an algorithm denies parole, flags a mortgage application, or informs a medical triage decision, who answers for that judgment? Contemporary AI systems increasingly occupy the space between tool and actor, generating consequential outputs while governance frameworks struggle to locate responsibility for them. This paper develops the Synthetic Moral Agent (SMA) framework through two arguments that are mutually reinforcing but independently sufficient, rather than a single chain in which the second depends on the first. The first argument is philosophical: no currently available evidence establishes that artificial systems possess the phenomenal consciousness, unified self-model, or normative understanding that intrinsic moral agency would require — a condition this paper terms the Ontological Void. The second is a governance argument, grounded independently in control over deployment, creation of risk, knowledge asymmetry, capacity to insure and intervene, and economic benefit: institutional uncertainty about machine consciousness must never function as a mechanism for diluting human responsibility. Together, the two arguments converge on Strict Stewardship, under which increasing operational autonomy proportionally elevates the obligations of designers, deployers, and institutional stewards, supporting strict enterprise liability at deployment and chain-based allocation of responsibility upstream. Read separately, each supplies its own warrant for the same conclusion: a reader who declines the philosophical argument still confronts an institutional structure that concentrates risk, benefit, and knowledge in a single class of actors, itself sufficient grounds for anchoring liability there. Algorithmic complexity must never become an institutional shield against responsibility, regardless of how convincingly a system appears to understand.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986454","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Calculated inefficiency: sustaining human supervisory competence as an ethical design objective for hybrid human-AI systems","authors":"Mohamed El Louadi,&nbsp;Maher Kallel","doi":"10.1007/s43681-026-01341-7","DOIUrl":"10.1007/s43681-026-01341-7","url":null,"abstract":"<div><p>The ethics of human oversight over artificial intelligence systems has been articulated at length as a normative requirement, most notably in Article 14 of the EU AI Act and in the philosophical literature on meaningful human control. Far less attention has been paid to the organisational question that determines whether such oversight is substantive or nominal: how the human competence on which oversight depends is sustained over time, in environments where the practice opportunities that historically reproduced it have been absorbed by the machine. This paper develops the construct of calculated inefficiency, defined as the structural property of a hybrid information system in which human interventions in automatable processes are maintained at levels above what pure efficiency optimisation would select, regardless of whether that maintenance was deliberately architected or emerged from other organisational sources. Whether such arrangements actually sustain or improve supervisory competence is treated as a separate empirical question. We distinguish three types of calculated inefficiency (architected, institutional, residual), formalise a four-dimensional hybrid-resilience framework (verifiability, distributed competence, decision traceability, systemic robustness) as a threshold weakest-link specification for hybrid resilience, and develop eight theoretical propositions, organised under seven proposition labels (P1, P2, P3a, P3b, P4, P5, P6, P7), that operationalise the construct at empirically testable scope. We engage three counter-positions, including the simulation-based competence reserve and the specialised recovery unit, and we specify the political-economy mechanisms (regulatory pre-commitment, insurance linkage, audit-trail requirements) through which the construct can be protected from short-term efficiency pressures. The contribution is to translate the ethical imperative of meaningful human control into an architectural and institutional design object that organisations can adopt, regulators can audit, and researchers can measure.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986455","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"“Unprecedented Injustice”: network determinism in the Dutch digital welfare state","authors":"Diletta Huyskes","doi":"10.1007/s43681-026-01350-6","DOIUrl":"10.1007/s43681-026-01350-6","url":null,"abstract":"<div><p>Automated decision-making systems in the welfare state have produced discriminatory outcomes at scale, as critical scholarship has extensively documented. This article examines the mechanisms through which such outcomes are produced by tracing the cultural, institutional, and technical determinants embedded in their design and use. Drawing on qualitative research conducted in the Netherlands between 2023 and 2024, it combines semi-structured interviews, ethnographic observation, and documentary analysis across three cases: SyRI, the childcare benefits scandal (Toeslagenaffaire), and Rotterdam’s welfare fraud detection system. Rather than asking whether automated systems discriminate, the article asks how discriminatory social classifications become translated into algorithmic risk and acquire the authority of seemingly objective decisions. The analysis identifies four recurring mechanisms: manual pre-selection of groups for scrutiny, the use of records of past suspicion as training data, proxy substitution for sensitive attributes, and the cumulative, intersectional layering of risk indicators. To conceptualize these dynamics, the article develops the notion of network determinism: the practical closure produced when administrative institutions, databases, policy objectives, statistical categories, organizational routines, and computational systems repeatedly reinforce one another, turning contingent classifications into effectively deterministic outcomes for those subjected to them. The findings show that discrimination cannot be attributed to algorithms alone, but emerges from chains of human and institutional choices that distribute both decision-making and responsibility. The Dutch cases therefore demonstrate why technical or legal corrections are insufficient without addressing the socio-cultural assumptions that automated systems are built to operationalize.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01350-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986497","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Will AI save ethics—and make ethicists obsolete?","authors":"Max Tretter","doi":"10.1007/s43681-026-01384-w","DOIUrl":"10.1007/s43681-026-01384-w","url":null,"abstract":"<div><p>In 1982, Stephen Toulmin famously argued that medicine “saved the life of ethics.” This article asks whether AI is now doing something similar—and finds that the answer splits. Working from a broadly Toulminian conception of ethics as a discipline that responds to societal demand for normative orientation by providing analyses and reasoned recommendations, the article develops its argument in two stages. The first is diagnostic. AI is undeniably <i>revitalizing</i> ethics: it generates unprecedented demand for normative orientation and brings ethical questions to the center of public and academic debate. Yet AI appears to be on a trajectory toward becoming increasingly capable of supplying the very outputs for which ethicists have traditionally been consulted. If it can provide these outputs faster, more cheaply, and, in some respects, better, demand for ethics and demand for ethicists may increasingly come apart: ethics as a discipline may flourish even as ethicists as a profession face the prospect of becoming obsolete. In light of this diagnosis, the second stage asks how ethicists might respond. Four strategies are reconstructed: (1) defense and counterattack, (2) adaptive specialization, (3) ethics as an activity, and (4) cooperative transformation. Each strategy is critically assessed, culminating in a reflective appraisal of their respective strengths and limitations. The article concludes by arguing that ethicists must fundamentally reconsider the nature of their expertise and the role they can still claim for themselves in an age increasingly shaped by AI.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01384-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}