Upholding Epistemic Agency: A Brouwerian Assertibility Constraint for Responsible AI

This academic paper addresses the risk of generative AI converting uncertainty into authoritative yet potentially misleading verdicts, which undermines democratic epistemic agency. The author proposes a Brouwer-inspired assertibility constraint for high-stakes domains, requiring AI systems to provide publicly inspectable and contestable certificates of entitlement before asserting or denying claims. If such certification is unavailable, the system must return an 'Undetermined' status. This approach introduces a three-status interface semantics (Asserted, Denied, Undetermined) that distinguishes between internal entitlement and public standing. The framework operationalizes this constraint through decision-layer gating using internal witnesses like sound bounds, ensuring that 'Undetermined' is a mandatory status rather than a tunable rejection option. By prioritizing challengeable warrants over simple confidence scores, the proposed method aims to preserve human epistemic agency against the persuasive influence of automated speech in public justification contexts.