arXiv paper proposes commons governance as third AI institutional model

A new arXiv paper from Eduardo C. Garrido-Merchán argues that AI governance is too often framed as a binary choice between market-driven private ownership and top-down state regulation. The paper proposes a third path: commons-governed artificial intelligence, where communities self-organize to steward the resources that make AI possible—data, compute, models, knowledge, and even energy. Drawing on Elinor Ostrom’s Nobel-winning design principles for common-pool resources, the author builds a two-dimensional taxonomy: one axis covers the AI stack layers held in common, the other covers governance functions like boundary rules, conflict resolution, and collective-choice arrangements. The taxonomy is populated with ten recurrent institutional archetypes drawn from real-world examples such as data cooperatives, federated learning consortia, public compute initiatives, and open-weight model collaborations.

The paper also surfaces four critical tensions that constrain the commons approach: openwashing (labeling proprietary systems as open), the compute bottleneck (concentrated ownership of GPUs and cloud infrastructure), free-riding on community contributions, and the unavoidable trade-off between scale and sustainability. Crucially, the author elevates energy and computational sustainability from an externality to a first-class commons governance problem, arguing that the carbon footprint of training and inference must be managed collectively. The work closes with a research agenda for a polycentric AI commons, urging the field to move beyond theoretical frameworks and into practical institutional design. For professionals building or governing AI systems, this paper offers a concrete vocabulary and toolkit to design community-owned alternatives to Big Tech’s walled gardens.