AI and ethics最新文献

{"title":"AI-assisted retrospective reconstruction of contested psychiatric care: an autoethnographic case study","authors":"Takenari Fujio,&nbsp;Hiromi Fujio","doi":"10.1007/s43681-026-01407-6","DOIUrl":"10.1007/s43681-026-01407-6","url":null,"abstract":"<div><p>Contested psychiatric care may involve asymmetries in medical knowledge, institutional authority, procedural literacy, and access to documentary information. Generative artificial intelligence (AI), particularly large language models, has created new possibilities for patients to organize heterogeneous records and to explore alternative interpretations of complex healthcare experiences. This analytic autoethnographic case study examines a retrospective reconstruction of a psychiatric care trajectory spanning 1998–2026 and focuses on publicly accessible generative AI interfaces during the 2026 analytic period. The corpus combined autobiographical recollections, personal medical and administrative records, communication materials, longitudinal body-weight records, and AI-generated outputs from iterative analytic interactions. AI outputs were treated as interpretive prompts and organizational aids rather than as factual evidence. The analysis developed the zero-cost Distributed AI-assisted Interpretation (zDAI) framework as a case-specific heuristic for three linked activities: narrative organization, documentary and temporal checking, and structural synthesis. Differences among ChatGPT, Gemini, and Claude are reported as context-bound analytic tendencies rather than stable properties or performance rankings. An independently issued administrative review determination was analyzed separately from AI outputs and used only as documentary evidence concerning continued involuntary hospitalization at the time of review. The case suggests that multi-system AI interaction may support patient-led documentary organization, retrospective reconstruction, and reflective interpretation under conditions of informational complexity, while exposing risks concerning hallucination, reproducibility, privacy, relational ethics, platform instability, unequal access, and the possibility that AI may amplify rather than resolve epistemic asymmetries. The article therefore offers a preliminary conceptual basis for future research rather than evidence that zDAI is generally effective.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133111","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":"Artificial intelligence, religion, and society: ethical leadership and the shaping of value-based adaptive governance","authors":"Pauline Cheong,&nbsp;Liming Liu,&nbsp;Ilona Nord,&nbsp;Leon Schleier,&nbsp;Gabriela Frey,&nbsp;Katharina Yadav,&nbsp;Thomas Schlag","doi":"10.1007/s43681-026-01357-z","DOIUrl":"10.1007/s43681-026-01357-z","url":null,"abstract":"<div><p>As artificial intelligence (AI) becomes increasingly infused in religious communication from generative sermon-writing tools to avatar-led worship services, religious organizations are emerging as important yet underexamined sites of AI leadership and governance. Existing responsible AI scholarship has focused primarily on corporate and governmental settings, leaving limited understanding of how AI practices are ethically negotiated within civil society and faith-based institutions. Addressing this gap, this article examines how Christian religious organizations understand, evaluate, and govern the integration of AI within their digital media environments. Empirically, the study analyzes AI developments in three case studies involving 66 leaders from religious organizations in the United States, Germany and Switzerland. The findings identify three interrelated modes of normative, collaborative, and experimentative adaptivity through which religious leaders navigate the opportunities, risks, and uncertainties of emerging AI technologies. Building on these insights, we advance value-based adaptive governance as a framework for understanding how religious institutions function as learning organizations that actively influence AI development through theology-informed ethical assessment and participatory practices. Responding to calls for more relational and action-oriented AI ethics research, this article contributes to AI ethics by foregrounding religion as a distinctive organizational and cultural domain, demonstrating how Christian values, theological reflection, and institutional mission shape ethical reasoning, authority, cross-scalar collaboration, and spiritual discernment in human-centered AI governance.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01357-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035229","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":"Mal-Intelligence: how artificial intelligence produces a sustained illusion of cognitive competence","authors":"Monssif Rahhali,&nbsp;Abdellah Aajly","doi":"10.1007/s43681-026-01398-4","DOIUrl":"10.1007/s43681-026-01398-4","url":null,"abstract":"<div><p>Generative artificial intelligence tools improve measured task performance across cognitive domains, yet they may also generate in users an unwarranted and durable perception of enhanced personal intelligence. This article introduces Mal-Intelligence, a theoretical construct describing a sustained, self-reinforcing belief in augmented cognitive competence that is produced by structural properties of AI interaction and maintained through the progressive erosion of metacognitive monitoring. The article develops a three-level explanatory model, integrating evidence from cognitive psychology, human–computer interaction, and the philosophy of mind. The article asks under what conditions AI-assisted performance produces an unwarranted belief in enhanced personal competence, and through what mechanisms that belief resists correction. Because the construct takes the individual rather than the coupled human-AI system as its reference standard, the article also sets out why that choice does not depend on resolving the dispute over the bounds of cognition. At the individual level, the illusion is generated through cognitive ease, the illusion of explanatory depth, and the substitution of effortful reasoning; it is reinforced through automation bias, AI sycophancy, and the Dunning-Kruger dynamic; and, with sustained use, it consolidates into a stable identity-level belief. Because these mechanisms are embedded in widely deployed AI architectures rather than confined to individual users, Mal-Intelligence carries implications for AI system design, educational practice, and workplace policy. The article concludes with a set of theoretical propositions intended to guide a planned empirical validation program. Mal-Intelligence is presented as structurally produced rather than individually pathological, self-reinforcing rather than self-correcting, and collectively scaled rather than individually contained.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035319","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":"Beyond human-centered automated machine learning","authors":"Laurens Bliek,&nbsp;Elizabeth O’Neill","doi":"10.1007/s43681-026-01397-5","DOIUrl":"10.1007/s43681-026-01397-5","url":null,"abstract":"<div><p>There has recently been a movement in AI research to go beyond ‘human-centered AI’, and to consider whether and how AI systems should promote non-anthropocentric values or values that go beyond human concerns. However, the research domain of automated machine learning (AutoML) lags behind on this shift. In some sense this is surprising, because AutoML has unusual potential to facilitate ‘beyond human AI’. For example, compared to typical supervised deep learning algorithms, AutoML algorithms can deal with much more complicated decision variables and objective functions that do not require a precise mathematical relation, making them much more flexible. This article makes a theoretical contribution by characterizing some of the ways in which AutoML could facilitate the creation of AI systems that go ‘beyond human’. First, it introduces AutoML and describes how the AutoML research community has recently investigated moving toward being human-centered in a particular sense. Next, the article describes how many other AI research communities, after moving toward human-centered AI in various senses in recent years, are now moving to consider the importance of ‘beyond human’ values. The article then turns to how AutoML research could possibly go ‘beyond human.’ Whereas many aspects of human-centered AutoML, such as transparency, may also advance ‘beyond human’ values, doing so in a reliable way would require moving beyond human-centered AutoML. The article then demonstrates that in comparison with typical supervised deep learning approaches, the flexibility of AutoML offers unique opportunities for facilitating ‘beyond human’ values that are not present in other areas of AI. If there is merit in developing systems that go beyond human values, as many people outside of AutoML seem to think, then AutoML offers a number of advantages.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01397-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035397","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":"Consequence-bearing transition: a formal theory of when AI-generated transitions acquire authority to bind consequences","authors":"Hikmat Karimov,&nbsp;Rahid Alekberli","doi":"10.1007/s43681-026-01401-y","DOIUrl":"10.1007/s43681-026-01401-y","url":null,"abstract":"<div><p>Agentic AI can impose consequences on people whose opportunities to inspect or contest its decisions are limited. We defend a consequence-sensitive duty of justification: institutions delegating consequential action to AI should require evidence appropriate to the affected interests and provide accountable alternatives when that evidence is insufficient. The Consequence-Bearing Transition (CBT) framework formalises one implementation of this duty, conditional on specified institutional authority, consequence classes, and epistemic thresholds. Authority is transition-indexed, and consequence binding is a distinct gate outcome. Under a coverage–certainty coupling axiom, an attainable coverage ceiling below the class threshold rules out certification through any feasible evidence acquisition. Ceiling changes can invalidate that test’s previous outcome; persistence requires additional assumptions. In the stated finite input model, deciding whether feasible evidence can reach a coverage threshold is NP-complete, and its complement is coNP-complete. These coverage tests characterise the epistemic threshold condition when coupling is tight; other gate conditions remain necessary. Exact evidence selection is NP-hard, and deterministic online selection has no constant competitive ratio for marginal coverage gain under adaptive adversarial arrival. We connect these conditional results to institutional accountability, contestability, unequal documentation burdens, and the harms of deferral. Illustrative scenarios distinguish current insufficiency from structural uncertifiability. CBT supplies neither ethically justified thresholds nor a proof that institutional rules are just, and no empirical deployment benefit is claimed.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035389","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":"Pseudo-consciousness in Artificial Intelligence: a functional and governance framework for consciousness-like systems","authors":"José Augusto de Lima Prestes","doi":"10.1007/s43681-026-01392-w","DOIUrl":"10.1007/s43681-026-01392-w","url":null,"abstract":"<div><p>This article develops “pseudo-consciousness” as an analytical category for advanced artificial intelligence systems whose organized performance of consciousness-associated functions reshapes how they are interpreted, trusted, and governed without thereby justifying a positive attribution of phenomenal subjectivity. The central claim is not that machine consciousness is impossible in principle, but that current debate requires a concept for the increasingly important middle terrain between reactive automation and genuinely conscious agents. Many contemporary systems integrate heterogeneous information, revise their own outputs, transfer competencies across domains, simulate goal-directed organization, and sustain a recognizable behavioral profile across contexts, while remaining more plausibly understood, on present evidence, through a functional and governance-oriented lens than through an attribution of inner experience. The article situates this proposal within recent debates in the science and philosophy of consciousness, including theory-sensitive approaches to AI consciousness assessment, distinctions between extensional and intensional forms of artificial agency, proposals for testing machine consciousness, disputes between computational and biologically grounded views, and emerging empirical work on self-reference, introspection-like reports, trust, and moral attribution in large language models. It argues that pseudo-consciousness is useful because it identifies a non-trivial configuration of capacities associated with the appearance of mindedness, provides a disciplined vocabulary for systems whose social effects exceed older categories such as “narrow AI,” and clarifies why such systems generate distinctive ethical and governance concerns even in the absence of defensible evidence for consciousness. The paper develops five task-sensitive conditions for identifying pseudo-conscious profiles—global information integration, recursive metacognitive correction, cross-domain transfer competence, intentionality simulation without subjectivity, and behavioral coherence across domains—and uses them to examine boundary cases involving large language models, multimodal systems, and tool-using agents. It then shows how such profiles acquire social force through anthropomorphic uptake, relational asymmetry, and institutionally salient forms of trust, before turning to their ethical and governance implications. The conclusion is that pseudo-consciousness should be understood neither as a synonym for consciousness nor as a mere metaphor, but as a theoretically serious and practically useful framework for interpreting systems that perform the external grammar of mindedness under persistent uncertainty about their inner status.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035390","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":"The algorithmic muhtasib: an islamic moral-economy architecture for accountable AI governance in the digital state","authors":"Umar Iqbal Butt","doi":"10.1007/s43681-026-01415-6","DOIUrl":"10.1007/s43681-026-01415-6","url":null,"abstract":"<div><p>Governments worldwide are embedding AI into core public functions, yet dominant governance responses remain procedural — disclosure rules, impact assessments, audits — and are capacity-dependent, disadvantaging low- and middle-income digitizing states. This paper develops a framework pairing a normative foundation from the Islamic institution of hisba (public market and conduct oversight) with an architectural standard from resilience engineering, framed as a distributive and epistemic justice intervention for resource-constrained digitizing states. The paper proceeds conceptually.ing Three transferable principles are recovered from hisba — valuation transparency, distributed custodianship, documentary auditability — and a fourth criterion, contestability, is added from the algorithmic-accountability literature because the hisba sources speak to the inspection of a valuation mechanism rather than to the standing of an affected party to challenge a particular determination. Each of the four criteria is then paired with an architectural principle: disclosure with decision provenance, contestability with mandatory human override, distributed custodianship with bounded failure domains, documentary trace with continuous drift monitoring. The four paired criteria are given a minimal deontic representation as conditional obligations with threshold-triggered escalation. The framework is applied illustratively to algorithmic credit scoring and welfare targeting, and stress-tested across three stylized governance contexts of differing state capacity. The paired framework surfaces addressable governance gaps that neither a purely normative nor a purely architectural approach surfaces alone, and indicates that architectural accountability is comparatively capacity-robust: it reduces, without eliminating, the dependence of accountability on continuous discretionary oversight, shifting part of the compliance burden from ongoing institutional capacity to procurement-stage design requirements, while still requiring threshold-setting, system maintenance, human review, and enforcement. As a distributive and epistemic justice intervention for resource-constrained digitizing states, the framework offers a non-derivative theoretical contribution to comparative AI governance, pointing toward procurement-level architectural standards as a potentially more realistic lever than technical-assistance programs for improving accountability where oversight capacity is thinnest. The claim is analytical and remains to be tested empirically against deployed systems.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035396","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":"Large language models eroding science understanding: an empirical study of ‘malignment’","authors":"Harry Collins,&nbsp;Hartmut Grote,&nbsp;Paul Newbury,&nbsp;Patrick Sutton,&nbsp;Simon Thorne","doi":"10.1007/s43681-026-01391-x","DOIUrl":"10.1007/s43681-026-01391-x","url":null,"abstract":"<div><p>Large Language Models (LLMs) increasingly provide scientific information to non-specialist users, yet their linguistic fluency can create an impression of expertise that exceeds their capacity for scientific judgement. This paper experimentally investigates how readily an LLM can be configured to present fringe scientific claims as authoritative knowledge. We constructed two RAG-like Custom GPTs using GPT-5.1, one concerned with the fine-structure constant and the other with gravitational waves. In each case, ten papers selected by domain experts from the alternative preprint repository viXra were supplied as a domain-specific knowledge base, together with instructions directing the model to treat this material as authoritative. The underlying model was neither retrained nor fine-tuned. Expert-generated questions were then posed to the resulting “Fringe-LLMs”, to the unmodified model, and to domain experts. The Fringe-LLMs produced fluent, detailed and apparently authoritative answers that systematically reflected the fringe corpus, while expert assessment found many of these answers to be scientifically incorrect or seriously misleading. By contrast, the unmodified model generally produced answers substantially closer to mainstream scientific understanding. We describe this deliberate redirection of model output as malignment. The experiment demonstrates that scientifically misleading LLMs can be created with a low technical barrier and without altering model weights or poisoning training data. Because non-specialists may be unable to distinguish such outputs from genuine expertise, the results raise important concerns about the use of LLMs as intermediaries for scientific knowledge and reinforce the continuing importance of expert human judgement and provenance.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01391-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034518","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":"Is AI a moral expert?","authors":"Lane DesAutels","doi":"10.1007/s43681-026-01406-7","DOIUrl":"10.1007/s43681-026-01406-7","url":null,"abstract":"<div><p>Artificial intelligence systems have become increasingly involved in morally significant decision-making—from judicial sentencing recommendations, to healthcare triage, to drone strike implementation. This raises important questions about whether AI has the moral aptitude necessary for making high-stakes moral choices. In this essay, I explore whether, and to what extent, AI is (or could ever be) a moral expert. To answer this question, I distinguish between three accounts of moral expertise: theory-based, practice-based, and hybrid views. I argue, contra theory-based and hybrid views, that moral expertise is more than the ability to explain and apply accepted moral frameworks; it requires genuine moral insight. But to have genuine moral insight, I suggest, requires the ability to perceive moral nuance and reframe preexisting moral frameworks considering those moral nuances. Because AI lacks empathy, contextual moral understanding, and epistemic standing within the space of reasons, I argue that current AI systems do not now, nor will they likely ever instantiate this form of insight. I conclude that, while AI may be considered a valuable tool for supporting human moral deliberation, it cannot by itself serve as an expert moral decision-maker.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034519","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":"Asymmetrical consent in ambient clinical AI: rethinking patient recording rights and epistemic authority in U.S. health care","authors":"Krista Walker","doi":"10.1007/s43681-026-01416-5","DOIUrl":"10.1007/s43681-026-01416-5","url":null,"abstract":"<div><p>The rapid integration of ambient artificial intelligence (AI) into clinical documentation is reshaping how medical encounters are captured, interpreted, and institutionalized. In the United States, these systems process patient-clinician conversations and are commonly introduced through organizational disclosure and consent practices that may provide patients limited information about when a system is active, whether audio is retained, how data are processed, and who may access the resulting documentation. Patients who wish to record their own medical encounters, by contrast, face a fragmented body of state recording law as well as institutional policies and clinician discretion. This article identifies and theorizes a structural asymmetry in consent within the U.S. regulatory landscape. Healthcare institutions can establish the conditions under which ambient AI is incorporated into routine care, whereas patients often must seek permission, interpret state law, or risk institutional objection when documenting the same encounter for their own use. This asymmetry falls most heavily on patients who have historically experienced dismissal and misdocumentation in clinical settings, including Black women and other marginalized groups whose accounts have long been discounted as unreliable. Drawing on Miranda Fricker's theory of epistemic injustice and critical race theory, the article argues that this asymmetry undermines patient agency and reinforces inequities in whose account of care is recognized as authoritative. The article reframes ambient AI as a governance issue that redistributes control over clinical narratives, and it proposes an original patient-inclusive governance model grounded in reciprocal recording rights, encounter-level consent, and shared authority over the medical record.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034606","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}