Paradigms in Change? Generative AI as an Epistemic Actor in Empirical Social and Educational Research

Edited by Tamara Ehmann¹ and Friedrich Wolf¹

¹ Goethe-Universität Frankfurt a. Main, Germany

Please submit your abstract until 15 November 2026 at https://www.medienpaed.com/about/submissions. Please also find the author guidelines there.

Call for Papers as PDF

Generative AI systems (large language models such as ChatGPT, Claude, and Gemini) mark a turning point in empirical social and educational research, as they do not merely serve data processing purposes but also intervene in analytical and interpretive research activities, among other things. In doing so, they open up novel methodological approaches and research practices. At the same time, they fundamentally call established methodological boundaries into question by potentially shifting, permeabilizing, or even dissolving the paradigmatic dividing lines between qualitative and quantitative research logics, thereby going beyond established methodological integrations such as those found in mixed-methods research. In these established paradigms, methodological choices are not merely epistemic; they also determine what counts as an object of research and as a legitimate result (Seyferth-Zapf 2023). With AI acting both as a research method and as a subject of media education, the central question becomes which assumptions the technology itself implicitly sets, and how they can be reflected upon.

Digital technologies have long shaped the research process as established and, in part, indispensable research tools, influencing it at the level of both research methods and methodology and thereby also raising epistemological questions. In the past, the implementation of analysis software such as MAXQDA or ATLAS.ti in qualitative research, as well as SPSS, R, or Mplus in quantitative research, has led to methodological standardization, changed analytical practices, and new possibilities for data analysis (cf. Bollen et al. 2022; Kuckartz and Rädiker 2024; Mayring 2025). A fundamental ambivalence becomes apparent here: while tools such as SPSS tend to promote methodological standardization, the spread of R, for example – together with its open-source community – simultaneously creates spaces for methodological reflexivity and epistemic openness. This tension intensifies with the use of LLMs and touches on the fundamental epistemic question of who, within the research process, makes, justifies, and assumes responsibility for analytical and interpretive judgments. While (sub-)disciplines such as the digital humanities or media studies have, for a longer period of time, engaged with the handling of and influence of digital technologies on epistemology and methodology (cf. König 2026; and see, among others, AI meets Qualitative Methods – AIQM¹), the current discourse on the use of LLMs in social and educational research remains strongly focused on the associated risks and research-ethical concerns (cf. Nguyen and Welch 2026). Systematic analyses of productive fields of application, methodological implications, and concrete conditions for successful use are still largely lacking at present, or are being discussed primarily on the basis of conceptual working papers (cf. Aufenanger 2025; Rosenbohm 2025; Wannemacher et al. 2025).

This special issue responds to the gaps currently emerging in the discourse within empirical social and educational research by offering a methodologically differentiated stocktaking of the field. It adopts an explicitly practice-oriented perspective on research and seeks, in particular, to present and critically discuss promising use cases and conditions for the successful application of generative AI, while also reflecting on their epistemological implications. The issue further aims to engage with the fundamental question of how generative AI, as an epistemic actor, reconfigures the boundaries between human intuition and algorithmic pattern recognition. In doing so, it seeks to contribute to the professionalisation of AI-supported research and to foster the development of a community of practice within empirical educational research.

The special issue is structured, first, along the methodological paradigms of quantitative (1), qualitative (2), and mixed-methods research (3), as each entails distinct epistemic logics and methodological requirements. Second, it addresses the implications of these developments for research methods education in higher education (4).

1. Within quantitative research, one key question concerns the extent to which AI-supported code generation might make complex statistical procedures accessible to a broader range of researchers and thereby enhance the quality of quantitative inquiry. Generative AI may, for instance, support the identification of syntax errors, the generation of comparable analysis code, the construction of questionnaires and psychometric instruments, and the documentation of methodological decisions. New analytical possibilities may also emerge where AI itself is employed as an analytical tool, for example in the automated classification of large text-based datasets or in the identification of patterns in complex relationships. In addition, new forms of graphical representation of empirical data may become conceivable. At the same time, such developments require critical scrutiny. Lowering methodological access thresholds may also foster a form of apparent expertise in which the availability of analytical procedures is no longer matched by the competence to apply them in a methodologically reflective manner. From a methodological perspective, it is therefore necessary to consider the extent to which the automation of analytical steps affects objectivity and replicability, particularly when the underlying models themselves remain epistemically opaque.

2. Within the qualitative paradigm, different potentials and research-practical considerations arise, which vary according to the respective analytical objectives. For approaches grounded in qualitative content analysis, the question is whether and to what extent generative AI can support the systematic development and application of categories. For more reconstructive approaches, by contrast, it is of particular interest how generative AI might be used to assist case-specific analysis. Generative AI may also function as a «critical friend» by generating alternative interpretive perspectives, supporting the systematic analysis of extensive text corpora, and critically accompanying coding processes (see, for example, Törnberg 2024; von Unger 2025; for a reconstructive approach, see Schäffer and Lieder 2022 or; for qualitative content analysis, see Schneijderberg, Wieczorek, and Steinhardt 2026). At the same time, these practices raise fundamental methodological and epistemological questions: How does algorithmic pattern recognition based on probabilistic systems relate to interpretive processes of understanding? What role might AI play in the development of theoretical concepts grounded in empirical material? Which implicit assumptions and cultural codings are embedded in AI models, and in what ways might these reproduce hegemonic interpretive patterns (Carius and Teixeira 2025)? Can AI-assisted analysis do justice to the interpretive depth of qualitative methodologies, or does the use of LLMs contribute to a gradual standardisation that undermines the epistemic potential of ‘understanding the other’? What privacy policy issues must be considered when using LLMs for qualitative research? It is crucial to recall that, within the qualitative paradigm, the researcher is understood as the primary instrument of knowledge production — shaped by positionality, prior understanding, and interpretive judgement. AI-supported analytical practices therefore affect not only questions of innovation, but also the methodological premise that understanding is inherently bound to subjectivity.

3. Mixed-methods research may acquire a qualitatively new dimension through the use of generative AI. The analysis of open-ended responses in standardised surveys, the integration of qualitative interviews with quantitative text analysis, and the triangulation of heterogeneous data formats may all be facilitated by AI-supported procedures. This, in turn, may enable more complex mixed-methods designs and more seamless forms of integration between qualitative and quantitative data strands. At the same time, the AI-supported integration of heterogeneous forms of data raises fundamental epistemological questions: To what extent can algorithmic procedures not only technically bridge the paradigmatic tensions between qualitative and quantitative logics of inquiry, but also render these tensions productive for the methodological advancement of mixed-methods designs? An initial exploration of this topic was already undertaken in the recently published Jahrbuch Medienpädagogik 15 (Fromme et al. 2020): «Erziehungs-wissenschaftliche und medienpädagogische Online-Forschung: Herausforderungen und Perspektiven». Furthermore, a call for papers for a special issue dedicated specifically to mixed-methods research was issued in April 2025. This call should therefore also be read as a continuation of those debates and as a more specific epistemological engagement with the themes raised in the special issue «Mixed Methods im Spannungsfeld. Weiterentwicklung digitaler Forschungsmethoden zwischen KI, NLP, Big Data und transdisziplinäre Forschung».

4. These developments have immediate implications for research methods in education at colleges and universities. If generative AI is to become an essential component of empirical social and educational research, the question on the one hand arises with renewed urgency as to which competencies students must still acquire in order to conduct methodologically rigorous research of their own and to engage critically with research literature (see, in this regard, Märzinger 2025, on academic writing practices). On the other hand, media pedagogical competence and critical AI literacy are called upon on the part of students; this requires a reflection on how working with AI systems is to be understood as part of the media socialization of future researchers (see Issue 57 of MedienPädagogik, focusing on measuring media literacy). At the same time, AI also offers didactic opportunities: for the simulation of research processes, for exploratory work with larger datasets, and for the teaching of advanced methods (in this regard see, Lieder and Schäffer 2023, for reconstructive approaches). Building on this, AI itself should be regarded as a subject of learning that is to be reflected upon in media pedagogical scenarios. The development of pedagogical concepts capable of harnessing these potentials constitutes a central task for the future of methods education.

Across all of the thematic fields outlined above, contributions to this special issue are expected to engage explicitly with the broader epistemological and philosophy-of-science questions that accompany these developments. Across all research paradigms, the research process is mediated by the researcher. The use of generative AI as a co-researcher reconfigures this constellation: interpretive and analytical capacities are distributed between human and technical actors, thereby raising questions of epistemic responsibility, as well as of the transparency and traceability of research processes. From a philosophy-of-science perspective, the power dimension of the systems themselves must also be critically examined. Generative AI models are proprietary, commercially developed systems whose training data, parametrisation, and epistemic assumptions largely remain opaque – a condition that fundamentally challenges the transparency and traceability required of scientific research processes. Contributions are therefore invited to reflect, among other things, on the following questions: How should methodological quality criteria be conceptualised when AI systems are integrated into processes of knowledge production? What new forms of documentation and reflexivity are required to ensure the transparency of methodological decisions? Of central importance is the philosophy-of-science question of what status may be attributed to knowledge generated with or through AI, and how methodological control can be maintained in the sense of a reflexive research practice.

This call invites contributions to a two-stage review process:

• Abstracts of proposed contributions should be no longer than 200 words (excluding references) and include five to six keywords. Abstracts will undergo an editorial peer review process. Submission deadline: 15 November 2026 via: https://www.medienpaed.com/about/submissions
• Contributors will be notified of the outcome of the abstract review no later than 31 January 2027. Please note that a positive decision on the abstract does not constitute a final acceptance of the manuscript for publication.
• Full manuscripts must be submitted by 31 March 2027.
• Subsequently, full manuscripts will be reviewed in a double open peer review process, involving the authors, until approximately 31 May 2027, and revised accordingly.
• The publication of the special issue is planned for autumn 2027.

• Articles submitted in English or German must be original contributions and first publications.
• Full scholarly articles should comprise approximately 40.000 characters (including spaces, excluding abstract and references).
• An abstract of 150–200 words should briefly summarise the main arguments and findings.
• Both the title and the abstract must be provided in German and English and submitted together with the article.
• Please follow the author guidelines available at: https://www.medienpaed.com/about/submissions#authorGuidelines
• All submitted manuscripts will undergo an open peer review process.
• In the interest of scientific transparency, we encourage all researchers to make their research data available at the time of submission (e.g. software, datasets, questionnaires used).

Abstracts and full manuscripts should be submitted via: https://www.medienpaed.com/about/submissions

Authors are kindly reminded that MedienPädagogik follows the highest ethical standards in academic publishing. In this regard, we encourage the responsible and transparent use of AI tools, as outlined in the journal’s official statement:

«MedienPädagogik is committed to maintaining the highest ethical standards in academic publishing. In line with the recommendations of the Committee on Publication Ethics (COPE), we expect authors, reviewers, and editors to use artificial intelligence (AI) tools responsibly and transparently.

• Authors shall clearly disclose the use of any generative AI tools in the writing, editing, or translation of their manuscripts. This includes AI assistance in data analysis, media generation, or literature reviews.
• AI tools cannot be credited as authors. Responsibility for the content remains solely with the human authors.
• Data protection, fairness, and the potential for bias must be considered when using AI throughout the research and publication process.
• Reviewers and editors may use AI tools to support but not replace human judgment. Editorial decisions will never be made solely by algorithms.»

• Tamara Ehmann, Goethe-Universität Frankfurt, Fachbereich Erziehungswissenschaften, Associated Lecturer für Methoden in den Erziehungswissenschaften
• Friedrich Wolf, Goethe-Universität Frankfurt, Fachbereich Erziehungswissenschaften, wissenschaftlicher Mitarbeiter am Fachbereich Erziehungswissenschaften sowie am Center for Critical Computational Studies

Feel free to contact both editors if you have any questions.

Aufenanger, Stefan. 2025. «Künstliche Intelligenz und Forschungsethik – eine pragmatische Sicht». Erziehungswissenschaft 36 (70): 83–88.

Bollen, Kenneth A., Zachary Fisher, Adam Lilly, Christopher Brehm, Lan Luo, Alejandro Martinez, und Ai Ye. 2022. «Fifty Years of Structural Equation Modeling: A History of Generalization, Unification, and Diffusion». Social Science Research 107: 102769. https://doi.org/10.1016/j.ssresearch.2022.102769.

Carius, Ana Carolina, und Alex Justen Teixeira, «Artificial Intelligence and content analysis: the large language models (LLMs) and the automatized categorization». AI & Society 40 (2025): 2405-2415: https://doi.org/10.1007/s00146-024-01988-y.

Ernst, Julian, Christian Seyferth-Zapf, Judith Martinez Moreno, und Klaus Rummler, Hrsg. Medienkompetenz messen. MedienPädagogik 57. Zürich: OAPublishing Collective. https://doi.org/10.21240/mpaed/57.X

Fromme, Johannes, Stafan Iske, Therese Leik, Steffi Rehfeld, Jasmin Bastian, Manuela Pietraß, und Klaus Rummler, Hrsg. Erziehungswissenschaftliche und medienpädagogigsche Online-Forschung: Herausforderungen und Perspektiven. Jahrbuch Medienpädagogik 15. Zürich: OAPublishing Collective. https://doi.org/10.21240/mpaed/jb15.X.

König, Mareike. 2026. «Fertig – vorerst. Unfertigkeit als epistemischer Wert in den digitalen Geisteswissenschaften». Zeitschrift für digitale Geisteswissenschaften 11. https://doi.org/10.17175/2026_002.

Kuckartz, Udo, und Stefan Rädiker. 2024. Qualitative Inhaltsanalyse: Methoden, Praxis und Umsetzung mit Software und künstlicher Intelligenz. Weinheim: Beltz Juventa.

Lieder, Fabio Roman, und Burkhard Schäffer. 2023. «Lehren und Lernen rekonstruktiver Forschungsmethoden mit genrativen Sprachmodellen in hybriden Forschungswerkstätten? Theoretische und empirische Befunde». Journal für Psychologie 31(2): 131-154: https://doi.org/10.30820/0942-2285-2023-2-131.

Mayring, Philipp. 2025. «Qualitative Inhaltsanalyse mit ChatGPT: Fallstricke, große Annäherungen und grobe Fehler. Ein Erfahrungsbericht». Forum Qualitative Sozialforschung 26 (1): https://doi.org/10.17169/fqs-26.1.4252.

Märzinger, Marina. 2025. «Begleitung des Denk- und Schreibprozesses durch KI-basierte Werkzeuge». Magazin erwachsenenbildung.at 55: 29–36. https://erwachsenenbildung.at/magazin/ausgabe-55.

Nguyen, Duc Cuong, und Catharina Welch. 2026. «Generative Artificial Intelligence in Qualitative Data Analysis: Analyzing–Or Just Chatting?». Organizational Research Methods 29 (1): 3–39.

Rosenbohm, Sophie. 2025. «Mit künstlicher Intelligenz forschen? Anwendungsmöglichkeiten und Herausforderungen in der arbeitssoziologischen Forschungspraxis». IAQ-Report 2025-09. https://doi.org/10.17185/duepublico/84315.

Schäffer, Burkhard und Fabio Roman Lieder. 2022. «Distributed interpretation – teaching reconstructive methods in the social sciences supported by artificial intelligence». Journal of Research on Technology in Education 55: 111-124.

Schneijderberg, Christian, Oliver Wieczorek und Isabell Steinhard, „The handbook of qualitative and quantitaiv conten analysis. Introduction to Classical, Digital, AI-supported, and Automated Data Analysis (2026): https://doi.org/10.4324/9781003496397.

Seyferth-Zapf, Christian. 2023. «Mixed Methods im Kontext gestaltungsorientierter Bildungsforschung: Beispiel einer Forschungsarbeit zur Förderung von Medienkritikfähigkeit bei Jugendlichen vor dem Hintergrund aktueller Propaganda». MedienPädagogik: Zeitschrift für Theorie und Praxis der Medienbildung (Jahrbuch Medienpädagogik 19): 195–228. https://doi.org/10.21240/mpaed/jb19/2023.03.08.X

Törnberg, Petter. 2025. «Large Language Models Outperform Expert Coders and Supervised Classifiers at Annotating Political Social Media Messages». Social Science Computer Review 34 (6): 1181–1195.

von Unger, Hella, Yves Jeanrenaud, und Thomas Raucheger. 2025. «Mit KI über KI qualitativ forschen». In Wie KI Studium und Lehre verändert: Anwendungsfelder, Use-Cases und Gelingensbedingungen, herausgegeben von Klaus Wannemacher, Elke Bosse, Maren Lübcke und Alena Kaemena, 29–31. Hochschulforum Digitalisierung, Arbeitspapier Nr. 87. https://hochschulforumdigitalisierung.de/wp-content/uploads/2025/04/HFD_AP_87_Wie_
KI_Studium_und_Lehre_veraendert_final.pdf.

Wannemacher, Klaus, Elke Bosse, Maren Lübcke, und Alena Kaemena, Hrsg. 2025. «Wie KI Studium und Lehre verändert: Anwendungsfelder, Use-Cases und Gelingensbedingungen». Hochschulforum Digitalisierung, Arbeitspapier Nr. 87. https://hochschulforumdigitalisierung.de/wp-content/uploads/2025/04/HFD_AP_87_Wie_KI_Studium_und_Lehre_veraendert_final.pdf.

1 https://www.ifib.de/blog/detail/ai-meets-qualitative-methods-ein-interdisziplinaeres-netzwerk-fuer-forschung-mit-ki