Speakers

Training data are increasingly treated as a core component of openness in AI, yet systematic evidence on how dataset sharing shapes innovation in creator-driven communities remains scarce. We study this question on a large online platform where users develop and share fine-tuned text-to-image models and, in some cases, publicly release the datasets used for training. Using panel data and a staggered difference-in-differences design around prominent dataset-sharing events, we find that sharing increases both the quantity of new models created and their subsequent performance, measured by user engagement and ratings. We then examine how creators use shared datasets, focusing on two sources of heterogeneity. First, effects are stronger when shared datasets are less redundant with a creator’s existing content space, consistent with complementary reuse. Second, experienced creators drive much of the production increase and are more likely to expand the novelty and breadth of their outputs after sharing, whereas less experienced creators exhibit larger gains in model performance. Consistent with a barrier-lowering mechanism, dataset sharing is also associated with more new creator entries. Despite these heterogeneous benefits at the individual level, models on the platform grow more similar to prior community outputs in aggregate, indicating that openness can produce homogenization even when some creators expand their creative scope. These findings contribute to research on digital platforms and open innovation by providing empirical evidence on how releasing a foundational AI resource shapes both the level and the nature of innovation in creator ecosystems. They also clarify a governance tradeoff that platform designers face between broadening participation and sustaining creative differentiation.

Xi Wu is an Assistant Professor in the Department of Information Systems and Operations Management at Tilburg School of Economics and Management. She received her Ph.D. in Business Administration (Concentration: Management Information Systems) in June 2022 from Fox School of Business, Temple University. Her primary research interest is in information economics, focusing on how technologies influence firms' value capture and creation. Her work centers on digital platforms, open-source software, and intellectual property in the digital age, using econometrics, analytical modeling, and machine learning.

The Information Systems field is increasingly leveraging digital trace data—multi-modal records of digital activity—in research. Yet, papers leveraging this data collection mode are surprisingly secretive about their practices, thereby raising ethical, legal, and methodological challenges. Our systematic review of 176 papers from leading IS journals (MISQ, ISR, JAIS, JMIS) reveals that ethical concerns and reporting standards for scraping practices are often overly brief, undermining the rigor and replicability of research. While calls for transparent data collection methods persist, guidance remains scarce, risking questionable research practices and hindering theory development. Adjacent to these concerns, digital trace data has tremendous value and can also help us to investigate interesting phenomena, or help us in tough review processes. Based on my recent publication, I will share our experiences in the review process, in which we switched partially to trace data, and how some of the challenges above have affected our choices.

Nicolai Fabian is an Assistant Professor of Information Systems (IS) at the University of Groningen. His research deals with technological change and the implications of digital technologies - specifically digital transformation. He is interested in how emerging digital technologies (e.g., AI, cloud, Big Data) change organizations and what are the implications of such change. He works mostly with a quantitative approach and especially relies on computational methods. For example, he uses web scraping but also machine learning to gather and analyze large datasets. Moreover, he is skilled in econometric techniques but also configurational methods such as fsQCA. His work has been published in the European Journal of Information Systems, the Journal of Business Research, the Communications of the Association for Information Systems and is highly cited. He also presented his work at major conferences such as the AOM, ICIS, or ECIS. He also published and co-edited a free-access managerial eBook on digital transformation. Furthermore, he was a guest editor for a special issue on knowledge engineering in the Journal of Business Research. He is currently co-editing a book on AI and sustainability for Edward Elgar Publishing.

The societal, technical, and organizational risks posed by artificial intelligence (AI) increasingly motivate organizations to invest in governance efforts to promote responsible AI development and use. Over three years, we conducted an in-depth field study in an organization to investigate the on-the-ground practices of AI governance. Beyond creating new roles (e.g., governance officer) and governance bodies (e.g., ethical council), the organization saw an AI development platform as a key part of governing AI. Our findings revealed a frequent misalignment between the platform’s intended use and its actual adoption in practice. Instead of relying on the platform, organizational actors often relied on simpler shadow tools to perform AI governance activities. Establishing, encoding, and enforcing rules around responsible AI unfolded through a series of discursive practices including frequent model review meetings, community workshops, and self-regulating standards. These practices socialized values and norms around AI and helped bridge the gap between aspiration-driven policy and day-to-day practice. Our findings highlight the role of discourse in governing through technological tools as means for overcoming limitations of bureaucratic controls.

Tamara Thuis is assistant professor AI Ethics and Governance at VU Amsterdam, KIN Center for Digital Innovation. She holds a PhD from the Erasmus University Rotterdam (2025). During her PhD, she was a visiting scholar at NYU. Beyond her academic work, Tamara co-founded EDSA (Ethical Data Science Association) with Joris Krijger, a foundation that connects academia, public institutions, and industry to discuss real-world AI ethical dilemmas. Her research sits at the intersection of technology, organization studies, and philosophy. She uses qualitative, ethnographic methods to explore how organizations translate ethical values into AI practices, govern the responsible AI development and use, and navigate evolving regulatory landscapes.