The presentation explores the role of public participation in AI governance in liberal democracies, focusing on case studies in Canada and Germany. It reveals that current participatory methods increasingly mirror the logics of machine learning they aim to oversee, limiting democratic debate and turning participation into state-led knowledge extraction.

To address this, we conducted a series of futuring workshops in Canada, employing strategic foresight methods and large-language models. The presentation emphasizes a need to rethink data publics as relational, discursive, and deeply embedded in sociotechnical contexts to foster discursive and technical recursivity in the face of AI's social impact.