AI and ethics最新文献_第7页

The alignment problem is an illusion 对齐问题是一种错觉

AI and ethics Pub Date : 2026-09-21 DOI: 10.1007/s43681-026-01385-9

John W. Murphy

引用次数: 0

Artificial intelligence, judicial precedent and responsible judicial AI governance in Uganda: ethical, constitutional and institutional implications of AI-assisted legal research 乌干达的人工智能、司法先例和负责任的人工智能司法治理：人工智能辅助法律研究的伦理、宪法和制度影响

AI and ethics Pub Date : 2026-09-21 DOI: 10.1007/s43681-026-01390-y

Javason Kamugisha

{"title":"Artificial intelligence, judicial precedent and responsible judicial AI governance in Uganda: ethical, constitutional and institutional implications of AI-assisted legal research","authors":"Javason Kamugisha","doi":"10.1007/s43681-026-01390-y","DOIUrl":"10.1007/s43681-026-01390-y","url":null,"abstract":"<div><p>Artificial intelligence can widen judicial access to legal authorities, but it can also reorder the informational environment in which precedent is identified and interpreted. This article examines AI-assisted legal research in Uganda’s Superior Courts through doctrinal, jurisprudential and normative analysis. Its scope is limited to tools that retrieve, rank, trace, summarise or generate accounts of legal authority. It excludes outcome prediction, risk scoring and automated adjudication. The article argues that precedent presents an ethical governance problem because legal relevance depends on institutional authority, ratio, material facts and subsequent judicial treatment, not textual similarity alone. In Uganda, that problem is intensified by uneven case-law reporting, incomplete metadata and reliance on legal materials from more heavily digitised jurisdictions. The analysis rejects two insufficient responses: mechanical transplantation of European governance models and reliance on transparency or explainability as stand-alone safeguards. It develops the Doctrine of Responsible Judicial AI and translates it into a Uganda Judicial AI Governance Framework. The framework prioritises an authoritative domestic judicial corpus, source-level verification, hierarchy-sensitive retrieval, meaningful human responsibility, proportionate impact assessment, procurement controls and reversible pilots. A lean institutional model embeds oversight within existing Judiciary structures to reduce cost and administrative duplication. Responsible adoption, the article concludes, requires governance of both the research output and the legal-information infrastructure that makes the output possible.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034466","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}

引用次数: 0

From principlism to practice: a five-principle safeguard framework for workplace AI 从原则到实践：工作场所人工智能的五原则保障框架

AI and ethics Pub Date : 2026-09-21 DOI: 10.1007/s43681-026-01404-9

Marlow D. Guttmann

{"title":"From principlism to practice: a five-principle safeguard framework for workplace AI","authors":"Marlow D. Guttmann","doi":"10.1007/s43681-026-01404-9","DOIUrl":"10.1007/s43681-026-01404-9","url":null,"abstract":"<div><p>This article is a conceptual and normative contribution. The framework it develops was derived through a structured synthesis of scholarly literature, policy instruments and regulatory materials, and not from the collection or analysis of empirical data. Artificial intelligence is increasingly used to recruit, allocate, monitor, evaluate and develop workers. While these systems promise efficiency, consistency and better decision support, they also reshape autonomy, fairness, accountability and dignity in contexts marked by organisational power asymmetry. The article translates five principles from biomedical and AI ethics—beneficence, non-maleficence, autonomy, justice and explicability—into organisational safeguard requirements. It does not claim that employees are patients or that work should be medicalised. Rather, it argues that medicine offers a mature ethical vocabulary for expert-driven, high-stakes decisions affecting vulnerable persons under uncertainty, and that the underlying duty to identify hazards, control them and consult those exposed is already familiar to employers from occupational health and safety law. The article offers three things. First, it conceptualises workplace AI as an asymmetric decision environment requiring a duty-of-care logic. Second, it develops a Five-Principle Workplace AI Safeguard Matrix that links each principle to design evidence, organisational evidence and worker-facing evidence, accompanied by illustrative indicators for each cell. Third, it proposes Occupational AI Safeguard Rounds as a recurring interdisciplinary review practice connecting ethics, engineering, participation and redress. The framework is intended to bridge abstract AI ethics principles, HR-focused responsible AI debates and operational compliance by specifying what organisations should be able to explain, measure, document and contest before and after implementing AI systems in employment contexts. It has not been empirically validated, and the article sets out what such validation would require.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034894","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}

引用次数: 0

The ramifications of LLM watermarks for authorship and groups 法学硕士水印对作者和团体的影响

AI and ethics Pub Date : 2026-09-21 DOI: 10.1007/s43681-026-01389-5

Tim Räz

引用次数: 0

A dilemma for skeptics of trustworthy AI 对可信赖的人工智能持怀疑态度的人陷入两难境地

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01388-6

Madeleine Ransom, Nicole Menard

引用次数: 0

Evaluation methods and metrics towards trustworthy AI 可信赖人工智能的评估方法和指标

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01349-z

Catharina Margaretha van Leersum, Clara Maathuis

{"title":"Evaluation methods and metrics towards trustworthy AI","authors":"Catharina Margaretha van Leersum, Clara Maathuis","doi":"10.1007/s43681-026-01349-z","DOIUrl":"10.1007/s43681-026-01349-z","url":null,"abstract":"<div><p>The ongoing developments in Artificial Intelligence (AI) are introducing new technologies, initiatives, and applications. To assure that these developments do not translate into potential risks that are not captured, understood, measured, and managed, a trustworthiness approach is necessary. AI trustworthiness is treated as a multidimensional construct that spans transparency, safety, security, privacy, robustness, and fairness. We argue that these properties they should be translated into explicit, testable criteria. This article aims to synthesize evaluation approaches that make AI trustworthiness observable and measurable, enabling stakeholders to identify gaps, prioritize improvements, and align system behaviour with ethical, social, and legal values. A scoping review (resulting in 27 papers) is augmented with knowledge of 6 experts to map (i) the main trustworthy AI aspects, (ii) the evaluation methods used to assess these aspects, (iii) the metrics applied within them, and (iv) recommendations proposed to strengthen trustworthiness. The results suggest that current evaluation practices interpret and operationalize trustworthy AI in diverse ways. They often rely on established methods or benchmarks that may imply bias, limited comparability, or provide limited insight into the reasoning quality. We recommend to design evaluation frameworks that explicitly target trustworthiness aspects from a multidisciplinary angle and being instantiated to specific contexts or problems.</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/s43681-026-01349-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148987022","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}

引用次数: 0

Informed consent and AI driven depression screening systems 知情同意和人工智能驱动的抑郁症筛查系统

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01365-z

Julianna Costanzo, Hasse Hällström

{"title":"Informed consent and AI driven depression screening systems","authors":"Julianna Costanzo, Hasse Hällström","doi":"10.1007/s43681-026-01365-z","DOIUrl":"10.1007/s43681-026-01365-z","url":null,"abstract":"<div><p>Machine learning systems are increasingly able to identify whether an individual is likely to be depressed by analyzing social-media activity. Some scholars argue that informed consent is required before any such analysis begins. We defend a two-stage architecture: (1) automated, non-disclosive screening of platform data without prior consent, and (2) disclosure of individualized health-related information, and any further assistance, only after informed consent. The normative argument is structured as a continuum. Democratic Egalitarianism (DE) supplies a restorative justification where severe depression substantially impairs functioning autonomy and screening is necessary to identify people who would otherwise remain unaided. Virtue Ethics (VE) supplies a preventive justification where agency-eroding platform conditions place users who retain functioning autonomy on trajectories toward impairment. Because the system cannot know ex ante which users fall within either rationale, the same purpose-limited data analysis system initially sifts across the entire population of social media users and immediately discards non-flagged data. DE and VE therefore provide different reasons for one screening stage. Screening must remain insulated from human access to the data and operated by an independent, regulated body instead of the social media platform.</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":"148986673","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}

引用次数: 0

Compiling governance into enforcement: risk-tiered, confidence-aware fairness gates for machine-learning pipelines 将治理编译为执行：机器学习管道的风险分层、自信感知公平门

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01402-x

Rizwan Tanveer

{"title":"Compiling governance into enforcement: risk-tiered, confidence-aware fairness gates for machine-learning pipelines","authors":"Rizwan Tanveer","doi":"10.1007/s43681-026-01402-x","DOIUrl":"10.1007/s43681-026-01402-x","url":null,"abstract":"<div><p>Most algorithmic-fairness research concerns how to measure or mitigate bias; little concerns how to make a chosen policy hold in a production pipeline that engineers can route around under delivery pressure. We present a configuration-driven toolkit that compiles documented governance decisions into runtime enforcement, and ask: does the choice of validation-gate policy materially change deployment decisions on real data? A single declarative configuration binds a pipeline to one primary fairness definition, derives a continuous-integration enforcement tier from a documented historical-risk priority score, and applies a confidence-interval-aware validation gate whose strictness scales with that tier. The key design decision is that enforcement severity is a deterministic function of documented historical harm rather than a discretionary setting, and that a configuration declaring a weaker tier than the risk score warrants is refused at load time. We evaluate the working system on two public benchmarks, German Credit and Adult Income. The central finding replicates across both: an intervention whose disparity point estimate falls well below the threshold (0.055 and 0.067 against 0.10) is passed by a naive point-estimate gate but failed by the strict confidence-interval gate, because the interval upper bound (0.126 and 0.113) exceeds it. The gate policy, not merely the model, determines whether a borderline-fair system ships. The contribution occupies an underexplored operational-governance gap and maps onto ISO/IEC 42001, the NIST AI RMF, and the EU AI Act: fairness assurance is more robust when compiled into a non-bypassable, risk-calibrated property of the pipeline than when it depends on reviewer vigilance or a point estimate.</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":"148987020","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}

引用次数: 0

Why applied AI ethics in healthcare requires the integration of empirical inquiry and normative ethical analysis 为什么在医疗保健中应用人工智能伦理需要整合实证调查和规范伦理分析

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01370-2

Jennifer Viberg Johansson, Jessica Nihlén Fahlquist

{"title":"Why applied AI ethics in healthcare requires the integration of empirical inquiry and normative ethical analysis","authors":"Jennifer Viberg Johansson, Jessica Nihlén Fahlquist","doi":"10.1007/s43681-026-01370-2","DOIUrl":"10.1007/s43681-026-01370-2","url":null,"abstract":"<div><p>AI technologies in healthcare are increasingly subject to ethical scrutiny, often structured around high‑level ethical principles such as fairness, transparency, and accountability. While these principles are widely endorsed, their application frequently remains abstract, underspecified, and detached from the empirical contexts in which AI systems are developed and used. This paper argues that empirical ethics provides a necessary foundation for applied AI ethics in healthcare by enabling the ethically relevant contextualisation of normative principles through a reciprocal relationship between empirical and normative ethical analysis. In doing so, the paper addresses a persistent gap between ethical principles and empirical research in AI ethics by showing how these domains can be methodologically integrated rather than treated as separate or sequential. Drawing on empirical examples and our own experience from medical AI research and stakeholder engagement, we illustrate how empirical inquiry and normative ethical analysis, can work together to clarify the meaning of ethical values such as fairness, identify ethically salient risks and trade‑offs, and support ethically robust decision‑making across the AI lifecycle, including research, development, and procurement. We argue that without systematic and methodologically grounded empirical engagement, AI ethics risks becoming either toothless or merely procedural, or sliding into forms of ethical relativism in which empirical findings are treated as normatively sufficient. The article contributes to ongoing debates on how to operationalise AI ethics and highlights the role that empirical ethics can play in bridging ethical theory, law, and 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":"https://link.springer.com/content/pdf/10.1007/s43681-026-01370-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986672","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}

引用次数: 0

Generative AI in peer review: evolving policies and regulatory pathways for scientific rigour 同行评议中的生成人工智能：科学严谨性的演变政策和监管途径

AI and ethics Pub Date : 2026-09-18 DOI: 10.1007/s43681-026-01394-8

Ehsan Jozaghi

{"title":"Generative AI in peer review: evolving policies and regulatory pathways for scientific rigour","authors":"Ehsan Jozaghi","doi":"10.1007/s43681-026-01394-8","DOIUrl":"10.1007/s43681-026-01394-8","url":null,"abstract":"<div><p>The rapid adoption of large language models (LLMs) and generative artificial intelligence (GenAI) is reshaping academic publishing and raising new questions about peer review integrity, transparency, and governance. This paper examines GenAI use across editorial and reviewer workflows and considers its implications for confidentiality, accountability, evaluative authority, and scientific rigour. Drawing on the literature on research integrity and AI governance, along with a targeted comparative review of major publisher and editorial policies, the analysis identifies an evolving shift from undifferentiated restriction toward more function-specific, risk-based governance. The analysis examines three regulatory pathways: transparent integration within journal-controlled workflows, conditional reviewer use with disclosure, and prohibition when adequate confidentiality, data-protection, or other safeguards cannot be assured. The analysis also highlights disclosure challenges, including under-reporting, reputational effects, and the difficulty of reliably identifying undeclared AI use, while recognizing reviewer workload and increasingly complex, interdisciplinary research as key drivers of AI-assisted review. Drawing on these findings, the paper develops a tiered regulatory framework based on the AI system's function and capabilities, manuscript sensitivity, institutional governance capacity, disclosure, human accountability, reviewer education, and proportionate oversight. GenAI should therefore be governed not as a uniform technology but according to how, where, and for what purpose it is used, while preserving human responsibility for reviewer judgments and final editorial decisions.</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":"148987018","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}

引用次数: 0