Harminder Pal Monga (Indian Institute of Technology, Patna) has posted Epistemic Extraction in AI Governance: Institutional Design and Participation in the Global South on SSRN. Here is the abstract:
Artificial intelligence governance increasingly conditions participation in digital administrative and regulatory systems on adoption of externally defined epistemic categories. Participation is therefore not neutral inclusion but conditional recognition. This article conceptualizes this dynamic as the Guru Dakshina Problem: a recurrent institutional mechanism through which access to governance infrastructures depends on epistemic translation, producing exclusion where translation fails and dependency where it succeeds through epistemic surrender. Drawing on the Eklavya narrative as an interpretive heuristic, the article examines how governance systems reorder authority over what counts as valid knowledge before participation in decision-making becomes possible.
The article argues that AI governance systems regulate not only technological deployment but also the conditions under which knowledge becomes administratively intelligible. Procurement standards, certification systems, interoperability requirements, and compliance architectures determine which epistemic forms are recognized within governance processes and which are rendered administratively illegible. Exclusion is therefore shifted upstream from adjudication to institutional design.
Drawing on comparative institutional analysis of the European Union Artificial Intelligence Act, India’s Digital Personal Data Protection Act, and the African Union AI Policy Framework, the article demonstrates that governance outcomes are shaped less by abstract ethical commitments than by regulatory design, specifically the interaction between compliance conditions, enforcement mechanisms, sovereignty structures, and epistemic recognition. Across all three frameworks, participation continues to depend upon prior ontological standardization.
To address this gap, the article advances Epistemic Parity as an evaluative standard and institutional design principle for democratic AI governance. Epistemic Parity is defined as the structural condition in which diverse knowledge systems carry equal institutional weight before administrative decisions are made. Building on this principle, the article develops the Jambudvipa AI (JAI) Framework as a regional governance architecture grounded in South Asian institutional conditions and existing BIMSTEC-related coordination mechanisms.
At the center of the framework is the Zero-Feature Protocol, a governance and procurement standard that shifts regulatory assessment away from data standardization toward process interoperability. Unlike conventional AI architectures that require local data categories to be translated into externally defined schemas before administrative processing can occur, the Zero-Feature Protocol enables interoperability without requiring ontological surrender. Local knowledge systems and community-defined ontologies remain institutionally valid rather than being normalized before governance participation becomes possible.
The article argues that epistemic extraction is not an accidental by-product of AI deployment but a structural feature of contemporary regulatory architecture reproduced through procurement law, standards governance, certification systems, and interoperability requirements. It contributes to governance and development scholarship by demonstrating how certification regimes, procurement systems, and interoperability standards allocate authority over what counts as valid knowledge within AI governance systems.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum