Theorizing the Data-AI Nexus

Editors

• Cristina Alaimo, ESSEC Business School
• Lauren Waardenburg, ESSEC Business School
• Jonny Holmström, Umeå University
• Lior Zalmanson, Tel Aviv University
• Reza M. Baygi, VU Amsterdam

Description

We are living through a profound epistemic shift in the history of computation. Unlike symbolic approaches associated with GOFAI (Good Old-Fashioned AI) (Haugeland, 1989), contemporary AI systems learn from, operate through, and increasingly generate data at unprecedented scale. Advances in machine learning and deep learning have repositioned data - not rules - as the primary substrate of computational intelligence (Goodfellow et al., 2016; Pearl & Mackenzie, 2018). This transformation compels us to rethink the foundations of AI scholarship within the IS discipline and to develop new theoretical vocabularies to understand the data-AI nexus as a complex socio-technical phenomenon rather than a purely technical achievement.

There are several reasons why the data-AI nexus is particularly urgent to rethink at this historical moment. A substantial body of scholarship has shown that data are sociotechnical artefacts, shaped by institutional practices, historical trajectories, cognitive frames, and semiotic conventions (e.g., Alaimo & Kallinikos, 2022, 2024; Bowker & Star, 1999; Boyd & Crawford, 2012; Kitchin, 2014; Leonardi & Treem, 2020; Treem et al., 2023). As AI systems are built on data, learn from data, and reshape data through processes of classification, prediction, and generation, data are simultaneously increasingly crafted to meet the requirements of AI systems (Van den Broek et al., 2021). Training datasets are curated, cleaned, labelled, augmented, and optimized with particular model architectures in mind (Sundberg & Holmström, 2024). In this recursive dynamic, the boundary between ‘data’ and ‘AI’ becomes blurred (Waardenburg & Huysman, 2022), demanding renewed conceptual attention to how organisations’ social fabric alters in the wake of AI (Baygi & Huysman, 2025).

Second, experts and organisations primarily encounter and shape AI through data practices: generating inputs, curating datasets, fine-tuning models, evaluating outputs, and conducting quality control (Ashuri et al., 2026). To understand AI in organisational contexts, we must therefore unpack the socio-cognitive and -technical processes of data production, cleaning, transformation, and governance; that is, the data work embedded in everyday organisational routines (Bertelsen et al., 2024; Parsons et al., 2025; Von Krogh, 2012). These processes extend beyond individual tasks to encompass institutional histories of data: organisational memory, sedimented classification systems, collective sensemaking, and enduring cognitive frames. Such historically layered infrastructures shape both how AI systems are designed and how their outputs are interpreted.

Third, AI is no longer trained only on data understood as structured database entries. Today, AI systems rely on an expanding range of materials, digital and analogue alike, that are transformed into training input. For the first time, AI also generates data (Yoo et al., 2025). Synthetic data - text, images, code, and other artefacts generated by machines - are becoming a key input for model training, fine-tuning, and evaluation (Narayanan & Kapoor, 2025; Nikolenko, 2021). The growing use of synthetic data for training purposes may be motivated not only by performance considerations, but also by issues of privacy, scarcity, access, and governance. When synthetic data, that are disconnected from social practice, recursively feed new models, the challenge, however, becomes not only technical but institutional, as human actors defer to AI even where scepticism and independent judgment are needed for legitimizing AI-generated material (Janssen, 2025; Liel & Zalmanson, 2025).

For this special issue, we invite submissions that advance our understanding of the data-AI nexus by foregrounding a dialogue between existing research traditions on data and AI, which have often remained disconnected. The topics outlined below are illustrative rather than exhaustive, and we encourage submissions that adopt different perspectives to creatively extend, challenge, or reframe these themes.

For this special issue, we invite contributions that critically examine the evolving relationships between data, AI, and social practice. We are particularly interested in theoretically ambitious and/or empirically rich studies that move beyond purely technical accounts of AI to explore its socio-technical, organisational, and epistemic dimensions. We encourage submissions that interrogate how data shape AI systems, how AI systems reshape the data they consume, and what happens when data becomes increasingly detached from practices of representation and meaning making.

Across these themes, we especially welcome work that advances theoretical and methodological discussions of the data-AI nexus in IS research. This includes analyses of co-constitution, entanglement, flow, recursion, performativity, infrastructuring, and classification in the context of AI and data. In particular, we invite more explicit attention to how constructs such as ‘intelligence’, ‘data quality’, ‘ground truth’, ‘training’, ‘fine-tuning’, ‘synthetic data’, and ‘model performance’ are represented, theorized, and operationalized in IS scholarship. How might IS research reconceptualise intelligence as relational, temporal, and emergent, rather than as residing solely in algorithms or datasets?

We particularly seek cross-level insights into how assumptions about data and intelligence are embedded in AI systems and how they (re)shape organisational practices and societal outcomes. This may include studies of data work in organisations (e.g., curation, labelling, cleaning, evaluation), model development practices, synthetic data pipelines, or the integration of generative AI into workflows.

We welcome methodological and conceptual contributions that help IS researchers to better study and theorize the data-AI nexus. Much empirical research treats data as given and AI as a technical artefact whose inner workings can remain in the background. We contend that the IS field is uniquely positioned not only to adopt methods from computer science or data science, but to inform and lead interdisciplinary debates on how data and AI should be studied as socio-technical systems. IS scholars can leverage trace data, computational methods, ethnographic methods, archival analysis, and experimental designs to examine how AI systems are built, trained, governed, and contested. We encourage methodological pluralism and reflexivity regarding the epistemic status of data used in our own research.

Finally, we also encourage contributions that link data, AI, and grand challenges. Addressing climate change, public health crises, inequality, digital governance, and geopolitical tensions increasingly relies on AI systems trained on vast and heterogeneous data infrastructures. Yet these infrastructures are historically layered, institutionally embedded, and politically contested. How can IS research theorize and design for responsible AI in contexts where data is incomplete, biased, synthetic, or recursively generated? What forms of governance, accountability, and stewardship are required when AI systems not only analyse the world but actively participate in constructing its data representations? In line with this, we particularly invite papers that examine AI and data in distinctive, novel, and perhaps non-commercial contexts, for example scientific, environmental, or planetary settings.

In bringing this special issue together, we aim to catalyse how IS scholars conceptualise and study the interdependencies between data, AI, and social practice. We seek contributions that treat the data-AI nexus not as a technical backdrop but as a core analytical problem that reshapes how knowledge is produced, validated, and institutionalized in digital societies. Although we welcome interdisciplinary studies, submissions should clearly foreground the IS artefact as central to the theoretical contribution. In line with the intellectual orientation of EJIS and other leading IS journals, we encourage papers that move beyond incremental extensions of existing models and instead develop original conceptual insights that will shape future debates on data, AI, and organizing.

Potential topics

• Data and AI co-constitution. We welcome submissions that explore the extent to which AI is co-constituted by data. Specifically, we invite authors to examine where and when ‘intelligence’ in AI emerges: in computational architectures, in datasets, or in their entanglement? How do choices about data selection, curation, labelling, and historical accumulation shape what AI systems can know and do? Contributions may theorize intelligence as a relational and emergent property arising from the interaction between models and data. They may critically interrogate the epistemological assumptions embedded in contemporary AI development practices.
• AI as a data-maker. We also encourage submissions that conceptualize AI as an active producer of data rather than merely a consumer. AI systems increasingly structure, generate, and transform the data they rely on, shaping its form, quality, and ontology. We welcome studies that investigate how classification systems, predictive models, and generative tools reorganize reality into particular categories and representations. We are interested in the rise of AI agents—autonomous or semi-autonomous entities that do not merely respond to prompts but interpret data to actively navigate digital and physical environments. How do these agentic systems transform the data-AI nexus by autonomously generating, filtering, and acting upon data without direct human intervention? What happens to data quality and provenance when the primary users and producers of data are agents rather than humans? Of particular interest are analyses of recursive dynamics in which synthetic data, untethered from original contexts or empirical ground truths, becomes the primary input for other AI systems. What are the epistemic, organisational, and societal implications of AI systems trained on AI-generated data?
• Data decoupled from practice. We invite research that examines the consequences of data production that is detached from institutionalized knowledge practices or situated practices of meaning-making. As data becomes increasingly abstracted, standardized, and mobilized across contexts, its links to the social practices and communities from which they originated may weaken. We encourage submissions that explore the social, political, and organisational implications of data that lacks clear connections to existing knowledge, truth claims, context, or lived experience. How do organisations govern and make decisions based on decontextualized data? What risks emerge when data-driven systems operate without strong ties to local knowledge, professional judgement, or collective sensemaking?
• The social life of the data-AI nexus. Finally, we encourage contributions that trace the institutional, historical, and cognitive trajectories of data as it is produced, curated, transformed, and mobilized. Data have a social life: they travel across systems, accumulate classifications, and become embedded in organisational routines and memories. We welcome studies that examine how organisational memory, institutional logics, and collective sensemaking shape both the datasets used to train AI systems and the interpretation of AI outputs. Moreover, this social life is increasingly populated by AI agents that function as new organizational actors. We welcome research that explores how these agents interact with human experts, how they participate in collective sensemaking, and how their autonomous data work—such as real-time evaluation or automated decision-making—reshapes the institutional memory of the firm. By situating AI within the broader social life of data, submissions can illuminate how historical trajectories and institutional dynamics influence contemporary AI practices.

Associate editors