Upholding Epistemic Agency: A Brouwerian Assertibility Constraint for Responsible AI
Pith reviewed 2026-05-15 16:50 UTC · model grok-4.3
The pith
AI systems in high-stakes settings must return Undetermined unless they can supply a publicly inspectable certificate of entitlement to their claims.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is that responsible AI requires an assertibility constraint inspired by Brouwer, where assertion or denial is allowed only with a certificate of entitlement that is publicly inspectable and contestable; otherwise the output must be Undetermined. This creates a semantics where the statuses indicate entitlement to categorical speech, not the truth of the claim itself. The constraint is implemented by gating outputs based on internal witnesses like bounds or margins, ensuring that Undetermined is required when no such witness exists. A design lemma establishes that complete binary interfaces with sound certificates already fully decide the predicate, making the third status a
What carries the argument
The Brouwer-inspired assertibility constraint, which mandates a publicly inspectable certificate of entitlement for any assertion or denial in high-stakes domains, serving as a boundary object between internal model processes and public justification.
If this is right
- AI interfaces adopt a three-status output: Asserted, Denied, or Undetermined, based on available certificates.
- Certificates provide a stable yet revisable time-indexed profile of entitlement.
- Outputs become answerable to challengeable warrants instead of confidence scores alone.
- Any sound binary decision interface necessarily resolves to a determined status only when a forcing witness is present.
Where Pith is reading between the lines
- This approach could be tested in medical diagnosis systems by checking if requiring certificates reduces overconfident diagnoses that later prove wrong.
- Extending the constraint to legal AI tools might force abstentions in ambiguous precedent cases, preserving room for human interpretation.
- The certificate mechanism suggests a way to integrate AI outputs into democratic deliberation by making them subject to the same standards of evidence as human claims.
Load-bearing premise
That it is technically and socially possible to produce and maintain publicly inspectable certificates for the outputs of deployed generative models without undermining their utility.
What would settle it
Deploy a generative model with the proposed gating in a high-stakes domain like medical advice and observe whether users can successfully contest incorrect assertions using the provided certificates, or whether the system frequently defaults to Undetermined in ways that reduce its practical value.
read the original abstract
Generative AI can convert uncertainty into hypersuasive, authoritative-seeming verdicts, displacing the justificatory work on which democratic epistemic agency depends. As a corrective, I propose a Brouwer-inspired assertibility constraint for responsible AI: in high-stakes domains, systems may assert or deny claims only if they can provide a publicly inspectable and contestable certificate of entitlement; otherwise they must return Undetermined. This constraint yields a three-status interface semantics (Asserted, Denied, Undetermined) in which statuses mark entitlement to categorical speech rather than truth values of the underlying world-claim. The semantics cleanly separates internal entitlement from public standing while connecting them via the certificate as a boundary object. It also produces a time-indexed entitlement profile that is stable under numerical refinement yet revisable as the public record changes. I operationalize the constraint through decision-layer gating of threshold and argmax outputs, using internal witnesses (e.g., sound bounds or separation margins) and an output contract with reason-coded abstentions. A design lemma shows that any total, certificate-sound binary interface already decides the deployed predicate on its declared scope, so Undetermined is not a tunable reject option but a mandatory status whenever no adequate forcing witness is available. By making outputs answerable to challengeable warrants rather than confidence alone, the paper aims to preserve epistemic agency against the hypersuasive pull of automated speech in public justification.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a Brouwer-inspired assertibility constraint for responsible AI: in high-stakes domains, generative systems may assert or deny claims only if they supply a publicly inspectable and contestable certificate of entitlement; otherwise they must return Undetermined. This yields a three-status interface semantics (Asserted, Denied, Undetermined) that separates internal entitlement from public standing via the certificate as boundary object, operationalized through decision-layer gating of threshold/argmax outputs using internal witnesses such as sound bounds, together with a design lemma establishing that Undetermined is mandatory rather than tunable whenever no adequate forcing witness exists.
Significance. If the central mechanism can be realized, the framework supplies a normative design principle that prevents hypersuasive categorical outputs from displacing justificatory labor in democratic deliberation. The time-indexed entitlement profile and certificate-based boundary object offer a concrete way to make AI speech answerable to challenge rather than confidence alone, extending intuitionistic assertibility ideas to deployed systems.
major comments (2)
- [Operationalization section] Operationalization section: the gating of threshold and argmax outputs is defined in terms of internal witnesses (sound bounds or separation margins) that are required to produce the certificate, yet the manuscript supplies no derivation, construction, or existence proof showing how such witnesses can be obtained from typical generative models whose outputs are probabilistic and non-constructive. This gap renders the certificate unable to function as a Brouwerian entitlement warrant.
- [Design Lemma] Design Lemma: the lemma asserts that any total, certificate-sound binary interface already decides the deployed predicate on its declared scope, thereby making Undetermined mandatory. However, the lemma's applicability rests on the unshown premise that certificate-sound interfaces exist for modern generative models; without a concrete reduction or example for probabilistic architectures, the lemma does not establish the claimed separation of entitlement from public standing.
minor comments (2)
- [Abstract] The abstract introduces a 'time-indexed entitlement profile that is stable under numerical refinement' but the manuscript does not supply an explicit definition or update rule for this profile, leaving its stability properties underspecified.
- The three-status semantics is presented as cleanly separating internal entitlement from public standing, but the manuscript does not address how the certificate remains contestable when the underlying model is updated or when the public record changes, which is central to the revisability claim.
Simulated Author's Rebuttal
We thank the referee for the constructive and precise comments. We address each major point below, clarifying the manuscript's scope as a normative proposal while indicating revisions to improve clarity and address the identified gaps.
read point-by-point responses
-
Referee: [Operationalization section] Operationalization section: the gating of threshold and argmax outputs is defined in terms of internal witnesses (sound bounds or separation margins) that are required to produce the certificate, yet the manuscript supplies no derivation, construction, or existence proof showing how such witnesses can be obtained from typical generative models whose outputs are probabilistic and non-constructive. This gap renders the certificate unable to function as a Brouwerian entitlement warrant.
Authors: We agree that the Operationalization section presents the gating mechanism at a conceptual level without supplying explicit derivations or constructions for obtaining internal witnesses from probabilistic generative models. The framework treats such witnesses as prerequisites that must be supplied by the underlying model (e.g., via interval bounds or separation margins) for the certificate to be issued. The manuscript does not claim to solve the technical problem of witness extraction for current architectures. In the revised version we will expand this section with a high-level discussion of candidate techniques drawn from uncertainty quantification literature, such as conformal prediction for producing sound intervals or Lipschitz-margin analysis in embedding spaces, while explicitly stating that full existence proofs remain model-dependent and outside the paper's normative remit. revision: yes
-
Referee: [Design Lemma] Design Lemma: the lemma asserts that any total, certificate-sound binary interface already decides the deployed predicate on its declared scope, thereby making Undetermined mandatory. However, the lemma's applicability rests on the unshown premise that certificate-sound interfaces exist for modern generative models; without a concrete reduction or example for probabilistic architectures, the lemma does not establish the claimed separation of entitlement from public standing.
Authors: The design lemma is a general logical claim about any interface that meets the certificate-soundness condition: if the interface is total and produces a valid certificate whenever it asserts or denies, then it has already decided the predicate on its declared scope, rendering Undetermined mandatory wherever no forcing witness exists. The lemma does not assert or require the existence of such interfaces for modern generative models; it shows what normative consequence follows once an interface satisfies the condition. The separation between internal entitlement (availability of a witness) and public standing (provision of a contestable certificate) is encoded directly in the three-status semantics. We will revise the lemma's presentation and surrounding discussion to make this scope explicit, add a short illustrative example using a simple threshold classifier with verifiable bounds, and note that concrete reductions for probabilistic generative models constitute an open implementation question. revision: yes
- Supplying a concrete reduction or existence proof for certificate-sound witnesses in large-scale probabilistic generative models, which would require substantial additional technical development beyond the conceptual and normative focus of the manuscript.
Circularity Check
No significant circularity; proposal is normative and self-contained
full rationale
The paper advances a Brouwer-inspired normative constraint on AI outputs in high-stakes domains, requiring certificates of entitlement before assertion or denial. No equations, fitted parameters, or derivations appear in the provided text. The three-status semantics and design lemma are presented as conceptual consequences of the assertibility rule rather than reductions to prior inputs or self-citations. Operationalization via internal witnesses is stated as a requirement without any self-referential definition or smuggling of an ansatz. The argument does not rely on load-bearing self-citations or uniqueness theorems from the same authors. The central claim remains a forward normative proposal whose feasibility is left as an external technical and social question, not resolved by construction within the paper itself.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Brouwerian principles of assertibility and constructive proof
invented entities (2)
-
Certificate of entitlement
no independent evidence
-
Three-status interface semantics (Asserted, Denied, Undetermined)
no independent evidence
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.