Data-Centric Ai and The Future of Research: Rethinking The Role of Rdm

Researchers today produce large volumes of heterogeneous data throughout the research lifecycle. With the rapid integration of artificial intelligence (AI), these data are no longer only research outputs but increasingly serve as direct inputs for AI systems involved in idea generation, planning, data analysis, literature reviews, and scholarly writing. While AI is often portrayed as a means of accelerating and optimising research, its growing dependence on researcher-produced data raises fundamental questions about the role of researchers and the place of Research Data Management (RDM). This presentation argues that AI does not reduce the role of researchers but fundamentally transforms it. As analytical and technical tasks become increasingly automated, researchers shift from direct execution toward orchestration. Their responsibility lies less in performing analyses manually and more in managing, contextualising, validating, and interpreting data and AI-generated outputs, while retaining full accountability for research quality, transparency, and meaning. From this perspective, RDM emerges not as a peripheral support activity but as a core scientific infrastructure enabling accountable AI-supported research. Empirical studies in AI development already show that data-related challenges—such as data quality, structure, provenance, and documentation—remain major bottlenecks for reliable and trustworthy AI systems. These challenges align with the principles of data-centric AI, which emphasise improving data rather than models as the primary driver of AI performance. They also resonate with established RDM frameworks, including the FAIR principles and their AI-oriented extensions (FAIR²), which promote readiness for machine reuse and automation. The presentation invites discussion on how AI reshapes the relationship between researchers, data, and RDM, focusing on researchers’ emerging role as data orchestrators, the expansion of RDM responsibilities, and the readiness of current RDM practices to support increasingly automated yet accountable research processes.