AI & Society最新文献

{"title":"The replacement-augmentation paradox: techno-perceptual dynamics of AI adoption in medical imaging and the future of work","authors":"Syed Mohammad Waheed","doi":"10.1007/s00146-026-03145-z","DOIUrl":"10.1007/s00146-026-03145-z","url":null,"abstract":"<div><p>The implementation of artificial intelligence (AI) in radiology has garnered significant attention across both research and practice over the last decade. However, opinions on whether AI will augment or replace radiologists in their work have so far been divided, and the debate around this continues to linger. This article conceptually structures and terms this phenomenon as the “replacement-augmentation paradox” of AI in radiology—the persistent coexistence of replacement fears and augmentation perceptions within the same professional community. Rather than treating these positions as mutually exclusive or empirically resolvable through performance metrics alone, this article conceptualises the replacement-augmentation paradox as a synthesising framework that structures and interprets divergent, evidence-informed narratives of AI integration in radiology. Drawing predominantly on sociotechnical and perception-based research, this article juxtaposes the two coexistent narratives and synthesises collective observations to shed light on relatively less emphasised aspects shaping replacement-augmentation debates. The findings obtained here explicate the implementation of AI tools (such as deep learning) for medical imaging and early diagnosis in radiology through a proposed “techno-perceptual” framework where the augmentation and replacement of the existing work practices are both plausible outcomes (O), subject to preconditions. Alleviation of fears and determination of the resultant narrative is thus linked with the provision of three jointly enabling conditions: training (T), evidence (E), and decisional control (C), the presence or absence of which are deemed to be directly consequential in terms of yielding perceptions of augmentation or replacement, respectively. The importance and interplay of these factors are delineated here through the novel Boolean equation O = T ∧ E ∧ C, a corresponding logical table, a process flow diagram, and a radiologist typology matrix depicting four professional archetypes.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 9","pages":"8481 - 8493"},"PeriodicalIF":6.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098258","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":"Entangled Futures","authors":"Sennay Ghebreab,&nbsp;Virginia Dignum","doi":"10.1007/s00146-026-03166-8","DOIUrl":"10.1007/s00146-026-03166-8","url":null,"abstract":"","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 7","pages":"6331 - 6333"},"PeriodicalIF":6.1,"publicationDate":"2026-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s00146-026-03166-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148496606","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":"AlgoNomy: algorithmic normativity in healthcare","authors":"Nicolò Amore,&nbsp;Adelaide de Vecchi,&nbsp;David Messner-Kreuzbauer","doi":"10.1007/s00146-026-03180-w","DOIUrl":"10.1007/s00146-026-03180-w","url":null,"abstract":"","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 on","pages":"7693 - 7696"},"PeriodicalIF":6.1,"publicationDate":"2026-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148807559","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":"Can artificial intelligence generate scientific knowledge? Pre-theoretical knowledge without epistemic agency","authors":"Clemente García-Hidalgo","doi":"10.1007/s00146-026-03154-y","DOIUrl":"10.1007/s00146-026-03154-y","url":null,"abstract":"<div><p>This paper examines whether artificial intelligence systems can generate genuine scientific knowledge. I argue that AI can produce scientific knowledge, but only in a pre-theoretical and non-agential sense. To make this thesis precise, I distinguish knowledge-as-a-product (truth-apt, methodologically warranted outputs) from knowledge-as-a-status (an epistemic standing attributable to responsible agents), and I introduce a tripartite distinction between the capacity for epistemic responsibility (categorical), the exercise of that capacity (graded), and the allocation of responsibility within scientific practice (graded). Current AI systems can contribute knowledge-as-a-product without being knowers or bearers of knowledge-as-a-status. Pre-theoretical knowledge, defined here equivalently as pre-explanatory-integration knowledge, requires truth-aptness, methodological warrant robust against ML-specific failure modes, and the absence of explanatory integration. Engaging with the theory-ladenness thesis (Hanson 1958; Pietsch 2015; Andrews 2025), with hermeneutic and pragmatist perspectives on scientific knowledge, and with recent positions in the philosophy of machine learning (Coeckelbergh 2025; Durán and Pozzi 2026), I clarify that \"pre-theoretical\" denotes the absence of explanatory integration within the output, not the absence of theoretical influence on the knowledge-production process. The framework is tested against a sustained case study of AI-supported mammography screening. The analysis clarifies who bears responsibility for AI-assisted results, credits AI contributions without implying authorship, and suggests governance criteria for promoting AI outputs to full scientific knowledge.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 9","pages":"8509 - 8524"},"PeriodicalIF":6.1,"publicationDate":"2026-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149099262","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":"Formation ethics: a cross-cultural framework for AI governance","authors":"Sibok Kim","doi":"10.1007/s00146-026-03186-4","DOIUrl":"10.1007/s00146-026-03186-4","url":null,"abstract":"<div><p>AI systems are scaling faster than the human and institutional capacities needed to govern them wisely. This paper calls that mismatch a formation gap, and argues that AI ethics must address not only which principles should guide AI, but what kinds of judgment and responsibility such principles presuppose. To clarify this asymmetry, it employs the bilingual heuristic of “civilization” (文明¹) and “culture” (文化) (Both terms are Meiji-era Japanese neologisms (和製漢語) coined to translate European concepts and later re-borrowed into Chinese and Korean: 文明 (“civilization,” literally “brightness of letters/refinement”) and 文化 (“culture,” literally “transformation through letters/refinement”). Whereas the Western terms derive, respectively from Latin <i>civis</i> (“citizen,” foregrounding the <i>polis</i>) and <i>cultura</i> (“cultivation, tilling,” an agrarian metaphor), the CJK compounds share the graph 文 (\"writing, pattern, humane refinement\") as their conceptual hinge—locating both civilization and culture in literate cultivation rather than in citizenship or soil. It is precisely this shared root in 文 that makes the pair analytically useful here: not as evidence of a distinct civilizational outlook, but as a vocabulary that holds infrastructural power and humane cultivation within a single semantic field.), not as a civilizational binary, but as an analytic distinction between infrastructural power and the moral and relational capacities needed to direct that power responsibly. The paper develops Formation Ethics as a developmental framework for responsible practice under acceleration. Its primary operational scaffold is Shu–Ha–Ri (守破離), adapted as a practical grammar for AI governance across interface design, organizational practice, and public policy. Through pause, place, and pluralize, Shu–Ha–Ri translates formation ethics into practices that protect interpretive discipline, adapt safeguards to local contexts, and cultivate institutional judgment. This emphasis moves AI ethics beyond principle declaration and compliance toward forming persons, routines, and institutions capable of principled action under uncertainty. The framework diagnoses omission anxiety: a moralized compulsion toward intervention in which restraint appears negligent. Against this pathology, it develops unforced responsibility as a context-sensitive, relational, and temporally disciplined orientation. Daoist wu-wei (無爲) and Buddhist emptiness (śūnyatā, 空性) clarify restraint, non-compulsion, and distributed agency without serving as indispensable foundations. Dialogical Emergence names moments when human–AI interaction transforms a user’s question, responsibility, or practical orientation rather than merely refining it. Methodologically, the paper offers a cross-cultural hermeneutic synthesis aimed at articulating responsible AI governance beyond principles, procedures, and compliance.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 on","pages":"7283 - 7294"},"PeriodicalIF":6.1,"publicationDate":"2026-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148807337","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 unknown-known trap: navigating invisible expertise boundaries in machine learning","authors":"Giulio Toscani","doi":"10.1007/s00146-026-03120-8","DOIUrl":"10.1007/s00146-026-03120-8","url":null,"abstract":"<div><p>Hyperspecialization in machine learning creates blind spots: deep technical skill without contextual awareness. Drawing on a Grounded Theory design, data were gathered through semi-structured interviews with 153 practitioners across 50 countries. This research identifies three operational modes: Known-Unknowns (specialized focus), Unknown-Unknowns (integrated collaboration), and Crisis (forced integration). Effective performance depends on “toggling”, dynamically shifting between modes based on the problem. This cognitive flexibility helps teams transform invisible blind spots into visible knowledge gaps, enabling sustainable collaboration. Resistance or misalignment signals stalled collective learning, not individual failure. The paper provides a learning-centered framework for building adaptive capacity in human-AI teams.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 9","pages":"8359 - 8373"},"PeriodicalIF":6.1,"publicationDate":"2026-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s00146-026-03120-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149099135","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":"Technological alienation and how networked digitality constitutes a map for understanding the complexity of individual and collective consciousness","authors":"Betto van Waarden","doi":"10.1007/s00146-026-03162-y","DOIUrl":"10.1007/s00146-026-03162-y","url":null,"abstract":"<div><p>Technology is feared to foster alienation, and may even come to control us through superintelligence. Yet this fear is exacerbated by a reductionist assumption that humans are defined mentally by their intelligence. Human consciousness is broader than intelligence, but how does it actually function? This article will suggest that these two problems of technological alienation and consciousness can actually help inform each other. Going beyond the intelligence-focused computational theory of mind, it will argue that <i>technology can help us conceptualize the networked multidimensionality of consciousness</i>. This multidimensional consciousness counters assumptions of alienation and suggests human compatibility with a modern accelerated society, provided we restructure its logic based on our complex consciousness rather than on mere intelligence. The article makes this argument in three parts. First, it reviews the notion of alienation by focusing on Robert Hassan’s theorized transition from analog to digital, and problematizes existing assumptions about analog and how it is supposed to reflect a linear logic in humans. Second, it uses networked digitality as a heuristic guide to understanding the multidimensional nature, scope, and simultaneity of individual consciousness. Third, it suggests that networked digitality can even help us to transcend the individual to imagine a multisensory connected collective consciousness. The article ends with an outlook on how we can start to rethink our socio-political institutions in light of multidimensional consciousness so that they remain liberal democratic rather than being hijacked by technological totalitarianism.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 9","pages":"7853 - 7869"},"PeriodicalIF":6.1,"publicationDate":"2026-06-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s00146-026-03162-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149099136","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":"Lived, affective, and cultural dimensions of ethical AI: beyond procedural AI ethics","authors":"Sharon Tettegah","doi":"10.1007/s00146-026-03165-9","DOIUrl":"10.1007/s00146-026-03165-9","url":null,"abstract":"<div><p>Contemporary AI ethics frameworks have made important contributions to governance discourse, particularly in their treatment of fairness, accountability, transparency, and safety. Yet, even in their most sophisticated forms, these frameworks remain structurally incomplete: they insufficiently theorize the affective and culturally mediated realities through which human beings encounter technological systems. This article argues that current AI ethics remains overly procedural and therefore cannot fully account for the human terms on which artificial intelligence is experienced and judged. A central asymmetry underlies the field: significant effort is devoted to developing systems capable of detecting, simulating, and responding to human emotion, while the dominant ethical frameworks used to evaluate those same systems routinely treat emotion as analytically secondary or marginalized. This asymmetry operates across three analytically distinct levels. At the technical level, systems detect or simulate emotional states while their designers lack an ethical vocabulary for what is at stake in those operations. At the governance level, frameworks that evaluate those systems provide no category for affective harm. At the level of lived experience, the people subjected to AI systems encounter them through feeling, embodiment, and cultural interpretation, dimensions that procedural evaluation does not reach. Emotion recognition offers a clear illustration of the consequences: insufficient consideration of emotion at the technical level produces systems that misread culturally specific expression; insufficient consideration at the governance level means no ethical framework names the misrecognition as a harm; insufficient consideration at the level of experience means the affected person has no recourse within existing evaluative categories. This asymmetry reflects a foundational limitation in prevailing models of responsible AI, which are sophisticated in evaluating systems while remaining underdeveloped in theorizing how those systems are lived by the people whose lives they shape (Floridi et al. 2018; Jobin et al. 2019; Choung et al. 2023). Drawing on affect theory, cultural studies, and sociotechnical scholarship, this article argues for a paradigm shift in how AI ethics conceptualizes human experience, one that moves beyond procedural compliance toward a richer account of lived ethical life.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 on","pages":"7257 - 7265"},"PeriodicalIF":6.1,"publicationDate":"2026-06-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s00146-026-03165-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148807238","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":"Instant understanding: LLM zero-shot qualitative discourse analysis of multimodal social media content","authors":"Dustin Tucker,&nbsp;Giovanna L. Henery","doi":"10.1007/s00146-026-03066-x","DOIUrl":"10.1007/s00146-026-03066-x","url":null,"abstract":"<div><p>The most successful uses of generative AI in social science up to this point have been limited several ways, mostly processing text-only data for simple and sharply delineated tasks, such as thematic analysis and hate speech detection, and frequently requiring training or fine-tuning of the AI model before use. The present study greatly expands the space of possible applications of generative AI in social science by demonstrating the ability of Gemini 2.5 Pro, a current frontier Large Language Model (LLM), to conduct comprehensive discourse analyses of multimodal (text, image, and video) social media posts, taking into consideration the broader social, political, cultural, and technological context of those posts. Using AI to conduct this sort of analysis requires the researcher to create detailed instructions to guide the analysis, but not any additional training or fine-tuning—the model’s zero-shot (“off-the shelf”) performance is more than adequate. The analysis of each post also takes a few minutes at most, even when processing videos up to 60 min long, enabling in-depth, open-ended analyses of large corpora that were not previously feasible, which in turn enables academic researchers, content moderators, and law enforcement, among others, to study and review orders of magnitude more data than they previously could.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 on","pages":"7315 - 7346"},"PeriodicalIF":6.1,"publicationDate":"2026-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148807588","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":"From artificial to artefactual wisdom: governing territorial polyrhythmy","authors":"Nicolas Merveille","doi":"10.1007/s00146-026-03151-1","DOIUrl":"10.1007/s00146-026-03151-1","url":null,"abstract":"<div><p>This article proposes a conceptual grammar for governing the heterogeneous temporalities that coexist within a territory without ever unifying. We argue that contemporary algorithmic infrastructures (adaptive traffic systems, logistics platforms, predictive analytics) share an architectural commitment they cannot govern: they operate according to a dominant logic derived from Aristotle’s excluded third, treating contradictions as problems to be resolved through eliminating one of their terms. The governance of territorial polyrhythmy requires a different architecture. Our central contribution is a grammar for qualifying territorial temporal antagonisms, not a critique of the smart city paradigm alone. Drawing on Simondon’s metastability (1958), Stengers’s ecology of practices (2005), and Lupasco’s logic of the included third (1951), we propose three conceptual displacements: from excluded to included third, from problem-solving to antagonism governance, and from measurement to evaluation. These displacements give rise to what we call ArtefactKairos: technical devices oriented toward the composition of heterogeneous temporalities rather than the compression of time. Engaging the conversation on artificial wisdom opened in this journal (Casacuberta 2013; Tsai 2020; Sinha and Lakhanpal 2023) and extended beyond it (McGregor 2025; Chella 2025), we argue that wisdom can be understood as a systemic property of territorial governance, an idea not yet addressed by the existing literature. Building on the Aristotelian distinction between sophia and phronesis, and on Dreyfus’s account of embodied judgement, we displace wisdom from the individual to the territorial register and from the contemplative to the compositional. The wisdom at stake is artefactual rather than artificial in the agentive sense: it has no existence outside the technical mediations through which polyrhythmic antagonisms become perceptible, nameable, and governable. The article introduces the Tempobloc as an elementary unit of polyrhythmic analysis and the Tic-Tac Gears, an exploratory governance grammar whose seven spatial forms and seven temporal colours generate forty-nine configurations for qualifying territorial antagonisms. The Algo-Geriato antagonism between algorithmic acceleration and demographic deceleration, explored through the case of a Montreal intersection, serves as a paradigmatic example. The framework is theoretical: it specifies three concrete design shifts (from optimization to composition, from prediction to qualification, from measurement to evaluation) that may precede and orient technical implementation, while acknowledging that the economic matrix of acceleration lies beyond its current scope and constitutes its necessary complement.</p></div>","PeriodicalId":47165,"journal":{"name":"AI & Society","volume":"41 on","pages":"7205 - 7219"},"PeriodicalIF":6.1,"publicationDate":"2026-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148807587","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}