In cultural heritage projects and artistic documentation, motion capture has emerged as a key archival strategy that is promised to be a “next stage” solution in preserving and accessing the past. However, motion capture is not an objective recording; it transposes a technological bodily imaginary onto the bodies whose movements it documents. This essay is situated at the intersection of current critical discourses on archives, dance and AI, bringing domain-specific knowledge to reimagine biased algorithmic systems. Although there is substantial risk for representational harms in how current AI motion models are used to render dancing bodies as data, recent projects show that retaining the entanglement of expert practitioners can refine data processing. We argue that incorporating the specificity of dance-based knowledge can support more meaningful historical research practices, in particular when understanding how bodies are themselves also archives. This contributes to identifying and countering the harms that arise from the mismatch between what automated motion extraction systems purport to accomplish and what they actually represent. The article outlines questions and guidelines that reimagine motion data through a visceral approach for an era of AI.

This article examines the collecting that occurred after the 2018 referendum to repeal the Eighth Amendment to the Irish Constitution, using the collections database as a lens for understanding how this recent period of Irish history is currently being narrativized. The Eighth Amendment prohibited abortion in almost all circumstances by equating the life of a pregnant person to that of the unborn child; its repeal was a result of four decades of grassroots, feminist campaigning. The collections now being preserved depict this activism, and are made up of campaign documents, photographs, and first-person stories archived from Facebook. Through a close analysis of the database of the Digital Repository of Ireland, where most of these materials are held, this article argues for an interpretation of the database as a political infrastructure, which refracts existing tensions surrounding the future of Irish feminisms and the activist archive. It also examines the database’s politics of visibility in relation to the shame and silence that defined women’s position in Ireland prior to the referendum and advances a theorisation of the archival database as a historiographical technology, which plays an active role in the production of Irish identity in the wake of the Eighth.

Foundation models are many things and encompass several modalities; they use text, images, sound, and more recently, action or inference units. But all of these forms share one thing in common: the (massive) scale. The “large” in large language models has been well studied by scholars in critical data, AI and archive studies, with several experts pointing at how these models are environmentally harmful, technically opaque and corporationally monopolistic primarily because of their scale. This piece discusses questions of technical and cultural scale – in the material, archival and procedural senses – within the contemporary technical and discursive landscape. At stake here is the role of critical and design studies within academic, artistic and para-academic worlds. It suggests that instead of corporate chatbots that aspire to pass the Turing test through multipurpose, encyclopedic service, we may be better served by playing with local models and reaching for small-scale AI development. This epistemological shift, in fact, may also provide some creative and critical potential that more effectively gets at the strangeness of machine learning systems while consciously and carefully handling the scalar environmental and social impacts of big AI.

This paper investigates a specific culture of interdisciplinarity that has gained traction at the intersection of applied AI and ethics. To address social and ethical harms of AI applications, scholars have suggested importing norms, methodologies and governance frameworks from established disciplines such as the social sciences or medicine. I show how this importation presupposes and endorses a framing of applied AI as a domain separate from established disciplines. Yet, such separation is what initially allows AI practitioners to operate outside those disciplinary norms that have evolved to prevent harms now associated with AI applications. Conversely, if AI applications were understood as situated firmly within these disciplines, practitioners would already be accountable to their norms and standards. Paradoxically, this culture of interdisciplinarity might thus reinforce a problematic disciplinary isolation of applied AI underlying the very ethical issues it seeks to mitigate – fighting symptoms while playing into their cause. In response, I outline three paths forward.

Linnaean taxonomy, which imposes hierarchical classifications based on morphological characteristics, has become deeply embedded in modern data architecture, from databases to metadata schemas to AI training datasets. With its hierarchical structure and rigid categorization, Linnaean taxonomy privileges one type of knowledge while marginalizing alternative taxonomies that offer more fluid, contextual, and relational understandings of the natural world. This paper examines how the legacy of Linnaean taxonomy continues to shape contemporary classification systems and artificial intelligence (AI). Indigenous knowledge systems, which include spiritual, cultural, and ecological dimensions, view entities not as isolated objects but as nodes in dynamic, interconnected networks. We draw from the French naturalist, Comte de Buffon, who, in line with Indigenous knowledge systems, viewed nature as continuous and contextual rather than discretely compartmentalized. The dominance of Linnaean-style classification in AI and data systems perpetuates colonial power dynamics and contributes to knowledge homogenization while losing Indigenous languages and classification systems crucial for addressing contemporary environmental challenges, particularly in agriculture and biodiversity conservation. In this Age of AI, we call for a holistic and ecological approach to archives. Therefore, we propose applying ‘rhizomatic hylomorphism,’ an ethnobiological, alternative classification that transcends hierarchical taxonomies to embrace multiplicity, relationality, and contextual meaning.

AI assistants are increasingly used for navigating and analysing the contents of major archives. Applying Retrieval Augmented Generation to existing large language models, these tools draw on indexes of the relevant archives to answer, in natural language, users’ questions. In addition to being powerful finding aids, archival AI assistants are also presented as being capable of providing useful, automated answers to questions about the past. This article argues that such tools and how they are marketed result in major conceptual disruptions and uncertainties, placing pressure on our understanding of a range of roles, forms of information and outputs involved in the production of historical knowledge. In particular, we argue that these tools may obscure well-established beliefs that “sources” and “archives” are not unmediated, clearly navigable or necessarily comprehensive, and that the processes by which these are used to write “history” are by no means straightforward or instantaneous. With the aim of mitigating these misunderstandings, the article makes suggestions for how deployers could more carefully frame and describe the intended use of archival AI assistants (especially for public users), to ensure that their benefits for accessibility are exploited while also avoiding misconceptions and safeguarding rigorous historical practice.

Data slicing is an inherent practice in machine learning (ML) evaluation, where different subsets of a dataset are used for training and evaluating ML systems. Drawing on ethnographic fieldwork conducted between September 2023 and February 2024 among the data scientists who develop ML-driven recommender systems for the British Broadcasting Corporation (BBC), this reflection piece highlights the important, yet often overlooked, ML practice of data slicing. Building on archival influences in Critical Dataset Studies and scholarship on critical curation, the paper proposes recasting data slicing as a curatorial practice. This shift makes visible the often highly tacit slicing practices undertaken by data scientists when working with ML datasets. Specifically, the paper traces three central considerations regarding data slicing: replicability, representativeness and generalisability. Using these as examples, the paper reflects on the implications of the selection, organisation, and choices of how to best represent the past to predict the future preferences of audiences. By engaging with data slices as curatorial constructs, we can better understand and intervene in their material politics by making visible how their orderings convey certain meanings more effectively than others. Through this approach, the paper expands on existing work within Critical Dataset Studies by identifying the political role of data slicing in the evaluation of ML systems.

This reflection explores the gap between explaining and understanding artificial intelligence’s (AI) outcomes, focusing on the example of AI-driven medical diagnostics. Explainable AI (XAI) is fundamentally rooted in machine learning and probabilistic methods, which is why its relationship to domain-specific contexts, such as medical diagnostics or policymaking, as well as to social concepts, such as trust or accountability, remains ambiguous. This creates a gap between what is technically considered as explaining and what, in turn, constitutes the understanding of a model’s predictions and its concrete, situated implications. In our contribution, we draw on the philosophy of science and its contextual notion of scientific understanding to highlight the need for explainable algorithmic outcomes that enable actors to both explain and intervene in creating, applying or interpreting knowledge. Our reflection highlights the potential tension between XAI as providing insights into an AI model’s technical operations and explainability as part of a situated understanding of how the model works within specific social contexts. We believe that bringing the concept of scientific understanding to empirical AI ethics can help us rethink explainability and contribute to more responsible ways of making sense of and interacting with black-box models, particularly in medical diagnostics.

Artificial intelligence (AI) has the potential to help solve global problems and be employed “for good.” One area of immense recent investment and interest is the financial technology (“fintech”) sector. Boasting its ability to provide financial services for the underbanked, various startups are developing apps that collect mobile phone data and use machine learning (ML) to provide credit scores – and subsequently, opportunities to access loans – to groups often left out of traditional banking. Based on 25 semi-structured interviews with corporate leaders, data scientists and investors at fintech companies developing and managing ML-based alternative lending apps in low- and middle-income countries, this study delves into the different ways fintechs conceptualize and define fairness, including from both a process (“fair” algorithmic design) and outcome (“fair” credit assessment) perspective. By engaging insights from industry actors, this research reveals the cultural logics, power dynamics and institutional incentives that underpin fairness narratives, including how these dynamics shape gendered outcomes in access to credit. Rather than challenging systemic exclusion, fintechs prioritize scalability and profit over equity, developing and deploying “gender blind” algorithms that perpetuate and legitimize financial disparities under the guise of neutrality and objectivity. Ultimately, fairness in ML-driven lending is shaped by institutional priorities and economic logics, where fairness narratives inadvertently obscure systemic injustices even as ethical innovation is claimed. This paper contributes to the growing field of empirical AI ethics by examining how fairness is defined, contested and operationalized in fintech-driven algorithmic lending and the ensuing implications, before offering a feminist alternative to fairness in ML that centers community voices and sets equity as a normative horizon.

Mainstream artificial intelligence (AI) ethics primarily takes vertical approaches: top-down, principle-driven and expert-led frameworks that often exclude the very publics affected by algorithmic systems. This paper explores whether citizen juries (CJs) can function as a horizontal alternative empirical ethics method that emphasizes participation, situated knowledge and epistemic justice. We conducted a citizen jury with Brussels residents to deliberate on the use of an AI system for electricity distribution during brownouts. The jury process combined guidance ethics, speculative fiction, and generative AI tools to scaffold inclusive deliberation. Using affordance theory, we analyze how the citizen jury method structurally mediates certain forms of participation and ethical reasoning. We evaluate its normative value through the lenses of epistemic justice and effective performativity. Our findings show that citizen juries can foster testimonial inclusion and support hermeneutical sense-making, offering a partial corrective to abstract, generic AI ethics. However, we also identify critical constraints: problem-posing was narrowed, alternative epistemologies discouraged, and the outputs risked reproducing the generic performativity forms of vertical ethics. While citizen juries hold promise for doing AI ethics otherwise, their transformative potential hinges on reflexively examining their own methodological affordances and institutional framing.

Mainstream artificial intelligence (AI) ethics, with its reliance on top-down, principle-driven frameworks, often fails to account for the situated realities of diverse communities affected by AI. Critics have therefore argued that AI Ethics increasingly serves corporate interests through practices of “ethics washing,” operating as a tool for public relations rather than a means of preventing harm or advancing the common good. In response, this paper adopts a Science and Technology Studies perspective to interrogate AI Ethics itself as a matter of concern. This perspective brings into view a core tension between vertical approaches (external, top-down, principle-based) and horizontal approaches (from-within, risk-mitigating, implementation-oriented). By tracing how these models have shaped the discourse, we show their relevance, as well as their limitations. To move beyond these limitations, we propose a third model that integrates empirical research on local lifeworlds with explicit value commitments. We outline this empirical approach through four shifts: first, a transition from top-down reasoning toward empirical grounding; second, a transition from abstract principles to situated orders of worth; third, a proposal to operate across scales by attending to assemblages of concern; and fourth, a redirection from risk mitigation toward the cultivation of technologies of hope.

This article discusses a case of how epistemic friction, discomfort and disagreement can become generative resources for interdisciplinary Responsible artificial intelligence (AI) research in health. Drawing on a four-year collaboration (2022–2026) on Responsible AI for health decision-making between AI Research, Medicine, Humanities and Law at a Dutch University and its University Medical Center, the article analyzes specific moments of “disconcertment” arising from interdisciplinary discussions on three central and interrelated concepts within the research group: (1) bias, (2) ground truth and (3) error. By analytically narrating the epistemic tensions produced by discipline-specific approaches to these concepts, the article shows how moments of disconcertment can create productive spaces for deepening interdisciplinary engagement. These moments may enable participating groups to address epistemological differences and reflect on taken-for-granted research relationships between their disciplines. Building on this insight, the article argues that moments of disconcertment can be mobilized as “little tools of friction” in interdisciplinary AI research more broadly. Such tools of friction support practices of “doing difference together” and may generate new possibilities and potentials for (inter)disciplinary thinking on AI, fostering new directions of doing interdisciplinary AI-based medicine and beyond.

This article examines the emergence of a new politics of vision shaped by the generative capacities of contemporary AI. It argues that generative technologies craft a distinct visual order, one that enables subtle yet devastating forms of harm through anticipation, projection, and the symbolic manipulation of the future by means of visual pre-enactment. I conceptualize this as speculative violence: a form of harm that does not depict bloodshed or explicit brutality but nonetheless embeds erasure and annihilation within the frictionless aesthetic of machine-generated futures. Speculative violence arises when generative AI is mobilized for its forward-oriented capacities in politically charged contexts, enabling images of possible futures to act upon the present; while not limited to war zones, this article focuses on the Israel–Palestine context for reasons of scope. It shows how, by giving visual form to what has not yet occurred, AI contributes to normalize the disappearance of populations and the destruction of places, stripping politically inflammatory ideas of their shock by rendering them aesthetically plausible. A striking instance is the “Riviera of the Middle East” video, circulated by Donald Trump in February 2025, whose synthetic fantasy of Gaza as a luxury resort visually anticipated – and helped normalize – the logic later formalized in policy proposals, such as the G.R.E.A.T. initiative and the 20-Point Peace Plan. More broadly, the article argues that generative AI functions as archival technology. Far from merely recalling past data, it generates speculative futures that accrue into a future archive: a corpus of synthetic images that shape political discourse and collective perception by visually scripting what should be imagined as likely, desirable or inevitable. In examining this archive, the article illuminates how generative AI transforms political imagination into an aestheticized form of world-making, with profound implications for contemporary regimes of power and violence.

This contribution argues that the practice of data provenance developed in critical research on artificial intelligence (AI), especially around publicly accessible data documentation, could inform a more transparent practice of provenance in archives. To set the ground for this cross-field exchange, I trace frameworks related to provenance, first from archival studies and then from AI/machine learning (ML). I ask how these frameworks overlap and diverge and how they have “flowed” or been assimilated into and adopted by the other field. I show how archival theories have been animated in documenting data in the context of AI and ML, including through documentation frameworks like Datasheets for Datasets, and then reverse that flow of knowledge, arguing that a practice of provenance developed in AI could forward some of the evolving goals of archival provenance, including those related to traceability and transparency.

This paper describes a prototype test harness that has been developed to evaluate the use of large language model retrieval-augmented generation systems for humanities and social science research, with an emphasis on collections held in the galleries, libraries, archives and museum sector. We propose the development of new modes of research software engineering oriented towards the values and principles of those communities, and rigorous modes of software testing grounded in industry best practices.

As artificial intelligence (AI) systems rapidly develop in biomedical research, bioethical frameworks focused on human subjects research are often limited in their capacity to address ethical concerns related to data-driven and black-boxed AI algorithms. Calls to embed Ethical, Legal and Social Implications research into large research consortia focused on AI seek to address the challenges of applying models developed in the context of genomics research to AI projects. Foundational questions related to the conduct of such embedded ethics work include: what are the goals of this work, what counts as “ethics,” who qualifies as an ethics expert and what mechanisms, including organizational structures, can ensure success in achieving ethics goals alongside scientific goals by multidisciplinary research teams? We draw on empirical case studies (including semi-structured interviews and document analysis) of large, federally funded AI research projects in the US that required embedded ethics. Notable findings include the lack of consensus among key players regarding the definition of AI ethics and conflict over the scope of ethics work, which impedes the implementation of ethics-related goals in research practices. These conflicts highlight the differences in disciplinary cultures, contentious boundaries between technical and ethical expertise and the lack of power and impact of the ethics team. Our findings underscore that effective ethics integration depends on trust and respect for ethics scholarship through leadership and relationship-building. We argue that building long-lasting, equitable relationships requires institutional commitment through organizational frameworks and funding mechanisms that prioritize deliberation about ethical concerns, shared values and shared power, as well as co-developing ethics goals alongside science goals. A horizontal approach is needed to nurture joint relationships for fostering ethical and just biomedical AI research and development.

This paper describes and reflects on the imaginaries, knowledge, processes and methods of a new four-year research project called Performing AI (PAI): Governance, Agency and Action – An Interdisciplinary Inquiry. Funded by the Swiss National Science Foundation’s “collaborative and interdisciplinary research” program, the project assembles researchers from three Swiss universities (the Universities of Fribourg and Lausanne and the Zurich University of the Arts) and Japan (University of Tokyo’s General Systems Science department in the Graduate School of Arts and Sciences). PAI’s goal is to investigate “AI” from epistemic, ontological, aesthetic and ethical angles, neither taking for granted its “uncontroversial ‘thingness’”, nor assuming received disciplinary frames to fit the purpose. Instead, we address how AI is “performed” – that is, enacted and produced – across different discursive and material sites and contexts. PAI asks multiple questions: How is AI enacted in governmental policy? What does artistic practice do to AI, and what does AI do to artistic practice? How is AI reconfigured in interdisciplinary scientific domains such as artificial life (as opposed to computer science)? How is AI taken up, or reconfigured in the public sphere, including schools, museums and festivals focused on the intersections of art, technology and society?

This paper presents a brief history of archival provenance and its shifting role across different paradigms of archives, considering how this past work can provide insights into current trajectories of AI. It explores what archival provenance does, or can do, differently than current imaginations of data provenance in AI. It proposes three “lessons from archival provenance” that challenge the present focus of existing approaches in AI, in particular highlighting how narrowly this work structures and models evidence, authenticity and creatorship of data. These lessons can be used to identify new opportunities for applying concepts of archival provenance by presenting a roadmap for future work where AI research might align and converge with related work in archival theory. It concludes by asserting that grounding AI datasets in this perspective of archival provenance can realize a new paradigm of “data as archives,” serving to envision responsibilities to a range of data creators, determine needs for documentation of context and establish a crucial role for “data archivists.”

The Pacific island nation of Tuvalu, facing the threat of being the first country lost to climate change, has declared itself the world’s “First Digital Nation.” In the age of deep learning and generative artificial intelligence (AI), this article examines Tuvalu’s state-scale digital twin as a project of preservation and archiving – not a technical fix, but a governing paradigm built on pattern discovery and predictive efficiency. It asks what is preserved through this process, by whom and for whom. The article advances three claims. Firstly, Tuvalu’s digital twin can be read as a state archive recalibrated for the era of generative AI and climate crisis: a live, multimodal apparatus in which records do not merely describe but route attention and authority. Secondly, converting the nation into a digital twin renders Tuvalu an asset, commodifying its culture and infrastructure. This extends a broader pattern of “knowledge islandization,” enacted through archival violence and rooted in colonial practices. Thirdly, drawing on performance theory, the article reads both the modern state archive and the emergent digital twin as performative agents operating across networks of humans, apparatus and governance, actively refining the perceptions of citizenship, local knowledge and statehood.