Is Openness in Decline? Data Sharing Between AI Commons, and Predatory Capture

Mayer, Katja; Skupien, Stefan; Knaus, Jochen

doi:10.5281/zenodo.17304840

Is Openness in Decline? Data Sharing Between AI Commons, and Predatory Capture

Mayer, Katja; Skupien, Stefan; Knaus, Jochen

Publisher: Zenodo
DOI: 10.5281/zenodo.17304840
Repository: zenodo.org
Language: en

Abstract

Open Science is meant to make research more transparent, collaborative, and equitable. But with the rise of machine learning and generative AI, new challenges around data sharing have emerged. AI development often depends on publicly shared or scraped data, yet the resulting models and infrastructures are typically closed and corporate-controlled. Thus, researchers increasingly express concern that open data are exploited in ways that strip context, reinforce bias, and concentrate power in proprietary AI systems (Birhane et al., 2022; Jernite et al. 2022, ; Widder et al., 2023; Zueger et al. 2023). In my own research, I have observed growing caution among researchers who once advocated openness. This is not just about lack of incentives but reflects deeper unease with how data circulates and is reused and often misused in AI-driven, commercialized environments. This moderated discussion session aims to learn from researchers and Open Science practitioners about how they experience these tensions. After a 10-minute introduction framing the issues, participants will discuss three guiding questions in three 15-minute rounds, sharing perspectives, dilemmas, and ideas. Contributions via live polls and an online board will help document collective insights and potential paths forward. Birhane, A., Kalluri, P., Card, D., Agnew, W., Dotan, R., & Bao, M. (2022). The values encoded in machine learning research. arXiv preprint arXiv:2106.15590. https://doi.org/10.48550/arXiv.2106.15590 Jernite, Y., Nguyen, H., Biderman, S., Rogers, A., Masoud, M., Danchev, V., Tan, S., Luccioni, A. S., Subramani, N., Johnson, I., Dupont, G., Dodge, J., Lo, K., Talat, Z., Radev, D., Gokaslan, A., Nikpoor, S., Henderson, P., Bommasani, R., & Mitchell, M. (2022). Data governance in the age of large-scale data-driven language technology. Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, 2206–2222. https://doi.org/10.1145/3531146.3534637 Widder, D. G., West, S., & Whittaker, M. (2023). Open (For Business): Big Tech, concentrated power, and the political economy of Open AI. SSRN Scholarly Paper 4543807. https://doi.org/10.2139/ssrn.4543807 Züger, T., Asghari, H. AI for the public. How public interest theory shifts the discourse on AI. AI & Soc 38, 815–828 (2023). https://doi.org/10.1007/s00146-022-01480-5

Full text

Is Openness in Decline? Data Sharing Between AI Commons, and Predatory Capture Katja Mayer (University of Vienna) Stefan Skupien (Berlin University Alliance) Jochen Knaus (Weizenbaum Institut) New “old” challenges in the “AI era” for Open Science? AI Context ●ML and GenAI rely on publicly shared or scraped knowledge ●But resulting models and infrastructures are closed and corporate-controlled AI Commons under Pressure ●Open layers of the AI stack (datasets, models, frameworks) depend on precarious, voluntary labour ●Few public or independent funders sustain these infrastructures ●Corporate actors dominate maintenance, direction, and access Predatory Capture ●Open data and code reused without context or credit ●Reinforces bias, opacity, and concentration of power in proprietary systems Researcher Perspectives ●Growing caution among long-time advocates of openness ●Concerns about circulation, misuse, and loss of control in commercial AI environments The 3 Questions for Today 1. Your experience: Is openness becoming harder to pursue in the age of AI? 2. What forms of recognition or institutional support would help? 3. How could new models of openness in the age of AI look like? This session: ●15-min introduction framing issues ●3 ×10-min discussion rounds on guiding questions ●Collective insights documented via live polls & online board → Documentation via collective pad, please add your name to the authors list zbw.to/osc25-pad03 Creating a space: “Yes, we’re open!?” (Berlin, 25.3.2025) - Event to bring together science, economy, civil society, and public service, and politics - 2 Panel discussion and 8 world-cafe-sessions - Aim was to explore experiences and needs to foster AI that includes and supports Open Science across domains Creating a space: “Yes, we’re open!?”: Key findings Results: Our shared vision is a socially accepted digital culture in which AI technologies are designed to be transparent, reproducible, sustainable, and oriented toward the common good. Key guiding principles are based on open science. Simon Brunel (Limo for Research) Creating a space: “Yes, we’re open!?”: Key findings Four key strategic areas: 1. Shared knowledge spaces and strategic networking, 2. Skills development and education, 3. Sustainable development through open digital infrastructures, 4. Governance: Combining responsibility and innovation. Discussion Paper: https://www.doi.org/10.34669/WI.DP/51 Policy Paper: https://www.doi.org./10.34669/WI.PP/15 Background II: Project Politics of Openness Open Data Practices in the Computational Social Sciences Funder: FWF Elise Richter Fellowship Duration: 2019–2025 PI: Dr. Katja Mayer Location: University of Vienna Aim: Investigate how openness is envisioned, negotiated, and enacted in the data practices of computational social science. Empirical Fields: ●Citizen science ●Data infrastructures ●Social media research and data science Methods: qualitative interviews, Group discussions, participatory observation Quotes: Open Science under new pressure in the GenAI era •We designed our repository for human access and scholarly reuse, not for bots pulling terabytes overnight. The scraping by AI developers is overwhelming our servers and our staff. (Infrastructure manager, research data repository, 2024) Infrastructures strained by large-scale scraping Machine-readability is supposed to make data more ‘FAIR,’ but in practice it often flattens them. Contextual metadata, methodological notes, even uncertainty - all get lost in translation to standardized formats (Social Scientist, 2023) Machine-readability improves interoperability but strips context and qualitative metadata •As soon as our open dataset is used for model training, it disappears into a black box. We don’t even know if our attribution stays attached, let alone how the data are interpreted (Open data practitioner in the social sciences, 2023) Once data enters proprietary AI pipelines: oversight and recognition often lost •AI research exposes how fragile that idea has become. The complexity of models, dependencies, and compute environments makes full reproducibility practically impossible —even for those who want to do it right. (Data Scientist, 2024) See also: Hosseini et al. 2025 Limits of openness for reproducibility: fragile in AI due to hyperparameters, environments, hidden dependencies Yes, openness is messy and sometimes misused, but closing off knowledge isn’t the answer. We need smarter forms of openness —ones that build trust, accountability, and real collaboration. (Data steward and Open Science advocate, 2023) Balancing Openness and Responsibility Mohammad Hosseini, Serge P. J. M. Horbach, Kristi Holmes, Tony Ross-Hellauer; Open Science at the generative AI turn: An exploratory analysis of challenges and opportunities. Quantitative Science Studies 2025; 6 22–45. doi: https://doi.org/10.1162/qss_a_00337 Question 1: Collecting Experiences In your own research or practice, have you noticed that openness is becoming harder to pursue - or even coming under attack - in the age of AI? For example, via infrastructural strain, loss of context in pursuing better machine-readability, lack of recognition for opening knowledge, challenges in governing data reuse, or the fragility of reproducibility in AI/ML workflows? Slido results Slido results Recap/Wrap + Outlook How will we use the outcomes of this session? ●Please add your names, notes and feedback to the shared Google document zbw.to/osc25-pad03 ●We will publish a short blog post summarizing today’s discussion ●The insights will inform our ongoing policy work on Open Science and AI ●We will invite you to take part in implementing some of the proposed activities ●To stay informed, please contact us at [email protected]