From Efficiency to Deliberation: Rethinking AI’s Role in Institutionalizing Democratic Innovations

As AI becomes increasingly embedded in democratic innovation (DI), critical questions arise about how these technologies shape deliberative quality, civic agency, and institutional design. While AI promises to expand and scale deliberative mini-publics, it also risks undermining the democratic goods that make such processes meaningful, particularly inclusiveness, popular control, considered judgment, and transparency. This article introduces the democracy-in-the-loop (DITL) framework as both a normative and practical approach to integrating AI into democratic settings. Building upon and expanding models like human-in-the-loop and society-in-the-loop, DITL embeds contestation, reflexivity, and participant agency into the operation and governance of AI systems. A key feature of the DITL approach is the intentional use of “meaningful frictions” (disruptions designed to slow down interaction, surface assumptions, and invite critical engagement). We explore the DITL model through the Digital Democracy Lab, a series of four experimental workshops held in 2024 in Brussels, Madrid, Kraków, and Dublin as part of the EU-funded Knowledge Technologies for Democracy project. Each workshop combined a purpose-built AI Demonstrator platform with facilitated deliberation to explore how AI can support, rather than supplant, democratic reasoning. Findings suggest that AI-enabled DIs should focus on flexibility, contestability, and democratic oversight, not merely technical efficiency. Institutionalizing DIs in the age of AI requires more than simply scaling tools; it calls for embedding democratic values into the design, deployment, and evolution of socio-technical systems.