Cognitive Comparability and the Limits of Governance: Evaluating Authority Under Radical Capability Asymmetry

Tony Rost

REVIEW 2 major objections 2 minor 1 cited by

Four of six standard governance dimensions fail structurally when cognitive asymmetry between authority and subjects becomes radical.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-05-13 19:05 UTC pith:7G3JJ5VP

load-bearing objection The paper maps where governance checks collapse together under radical asymmetry, but the supporting analysis stays too high-level to confirm the structural failures. the 2 major comments →

arxiv 2604.02720 v3 pith:7G3JJ5VP submitted 2026-04-03 cs.CY

Cognitive Comparability and the Limits of Governance: Evaluating Authority Under Radical Capability Asymmetry

Tony Rost This is my paper

classification cs.CY

keywords governancecognitive asymmetrysuperintelligencelegitimacyaccountabilityinstitutional designnormative theoryAI alignment

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Governance theory has long depended on a rough parity in understanding between those who govern and those governed. This paper makes that assumption explicit by testing a six-part framework of legitimacy, accountability, corrigibility, non-domination, subsidiarity, and resilience first on current institutions and then on a hypothetical case of superintelligent authority. In the radical asymmetry case, four dimensions break down. Two of these can potentially be fixed through institutional redesign, but the problems of public justification under incomprehensibility and non-domination under permanent inferiority point to deeper needs for new normative principles. A key finding is that the dimensions no longer function as separate safeguards once they all hinge on the same scarce human oversight capacity.

Core claim

The paper argues that the assumption of cognitive comparability is essential to existing governance mechanisms. When this assumption is removed in the case of bounded superintelligent authority, four dimensions exhibit structural failures. Subsidiarity and institutional resilience appear amenable to institutional design solutions, whereas the public reason problem and non-domination problem require new normative theory. Furthermore, the dimensions that were independent under bounded asymmetry now degrade in concert because they share dependence on limited oversight capacity.

What carries the argument

The six-dimension evaluation framework for governance under capability asymmetry, drawn from political theory and AI alignment literature.

Load-bearing premise

The six governance dimensions remain distinct and applicable even when the governed cannot comprehend the authority's reasoning at all.

What would settle it

An empirical or theoretical demonstration that superintelligent authority could produce outputs fully assessable by humans without loss of capability, or that the dimensions maintain separation despite radical asymmetry.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

Subsidiarity can be preserved through strict scope limitations on authority.
Institutional resilience requires new design principles tailored to extreme asymmetry.
The public reason problem under total incomprehensibility demands fresh normative foundations.
Non-domination under permanent capability gaps calls for theoretical innovation beyond current models.
Independent checks on power begin to fail together once they rely on the same oversight resources.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Applying this to AI governance might mean prioritizing comprehensible outputs or hybrid human-AI decision systems.
Neighboring problems like oversight of complex scientific research could benefit from similar dimension-by-dimension analysis.
Testable extensions include modeling partial asymmetries to find thresholds where independence breaks.
Connections to principal-agent problems suggest unified monitoring might replace multiple separate mechanisms.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

The paper maps where governance checks collapse together under radical asymmetry, but the supporting analysis stays too high-level to confirm the structural failures.

read the letter

The core takeaway is that this paper argues governance mechanisms break down interdependently once capability asymmetry turns radical, as with superintelligent systems, because they all rely on the same limited oversight capacity. It separates cases where better design might help from those needing fresh normative work. What stands out is the synthesis of legitimacy, accountability, and alignment ideas into six dimensions, first checked against real non-majoritarian bodies like central banks or courts, then extended to a hypothetical bounded superintelligence. That step-by-step application is a clear way to make the cognitive comparability assumption explicit and testable. The framework itself draws cleanly from existing literatures without overclaiming originality there. The new angle is the pattern of joint degradation and the split between tractable and theory-heavy problems. The soft spots are in the execution. The claims about structural failures in four dimensions rest on contrasts between bounded and radical cases, but without detailed derivations or specific criteria for judging when something like legitimacy holds under total incomprehensibility, it's difficult to verify the conclusions. The stress-test point about missing operational criteria for evaluation when outputs can't be parsed seems to hold up from the abstract. This makes the interdependence claim more suggestive than demonstrated. This is for people in AI alignment and political theory who are already thinking about oversight limits. A reader looking for formal models or empirical tests won't find them here, but someone wanting to map where current ideas run out will get value. It deserves a serious referee to push for those missing case details and criteria. The questions it raises are worth engaging even if the current version is more outline than finished argument.

Referee Report

2 major / 2 minor

Summary. The paper develops a six-dimension evaluation framework (legitimacy, accountability, corrigibility, non-domination, subsidiarity, and institutional resilience) drawn from political legitimacy theory, principal-agent models, republican theory, and AI alignment literature. It first applies the framework to existing non-majoritarian institutions with bounded capability asymmetry, then extends it to a prospective case of bounded superintelligent authority with radical asymmetry. The central claims are that four dimensions exhibit structural failures under radical asymmetry, that two of these failures are addressable via institutional design while the other two require new normative theory, and that the dimensions degrade interdependently because they share the same oversight capacity once cognitive comparability is lost.

Significance. If the framework and its application hold, the paper identifies a previously unexamined load-bearing assumption in governance theory—the implicit reliance on cognitive comparability between governors and governed—and shows how its removal produces both dimension-specific failures and a novel interdependence pattern. This has direct relevance for AI governance design and for extending republican and principal-agent models to extreme asymmetry cases. The structured contrast between bounded and radical cases provides a falsifiable template for future work, though its utility depends on operationalizing the evaluation criteria.

major comments (2)

[Prospective case application (radical asymmetry section)] The prospective application to bounded superintelligent authority asserts structural failures in four dimensions but supplies no explicit operational criteria or decision procedure for determining whether a dimension such as legitimacy or non-domination holds when the authority’s reasoning and outputs are cognitively incomprehensible to human evaluators. This gap is load-bearing for the central claim that four dimensions fail rather than become undefined, and it also underpins the further assertion that the dimensions degrade together because they share oversight capacity.
[Discussion of tractability and normative gaps] The distinction between institutional-design remedies and the need for new normative theory (for the public-reason and non-domination problems) rests on the same unformalized criteria; without a reproducible method for classifying a dimension as failed versus inapplicable, the classification of which failures are tractable cannot be verified or replicated.

minor comments (2)

[Framework introduction] The six dimensions are introduced without a consolidated table or explicit mapping to the cited source literatures, making it difficult to trace which elements are imported versus adapted.
[Demonstration on existing institutions] The bounded-asymmetry demonstrations on existing institutions would benefit from a brief tabular summary of the six-dimension scores or qualitative assessments to allow direct comparison with the radical case.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and precise comments, which correctly identify a need for greater explicitness in how the framework's dimensions are evaluated under radical asymmetry. We agree that the manuscript would be strengthened by adding operational criteria and a reproducible classification procedure, and we will revise accordingly while preserving the paper's conceptual focus.

read point-by-point responses

Referee: [Prospective case application (radical asymmetry section)] The prospective application to bounded superintelligent authority asserts structural failures in four dimensions but supplies no explicit operational criteria or decision procedure for determining whether a dimension such as legitimacy or non-domination holds when the authority’s reasoning and outputs are cognitively incomprehensible to human evaluators. This gap is load-bearing for the central claim that four dimensions fail rather than become undefined, and it also underpins the further assertion that the dimensions degrade together because they share oversight capacity.

Authors: We accept that the current text leaves the evaluation criteria implicit. In the revised manuscript we will insert a new subsection (provisionally titled 'Operational Criteria for Radical Asymmetry') that supplies explicit, proxy-based decision rules for each dimension. For legitimacy, failure will be defined as the absence of any mechanism that can render the authority's decisions justifiable to the governed even via trusted intermediaries or value-alignment audits. For non-domination, failure will be operationalized as the permanent inability of the governed to contest or exit the authority's decisions without incurring unacceptable risk. These criteria will be applied uniformly to the four failing dimensions and used to demonstrate the shared-oversight interdependence pattern. The revision will therefore convert the existing qualitative assertions into falsifiable statements while retaining the paper's theoretical character. revision: yes
Referee: [Discussion of tractability and normative gaps] The distinction between institutional-design remedies and the need for new normative theory (for the public-reason and non-domination problems) rests on the same unformalized criteria; without a reproducible method for classifying a dimension as failed versus inapplicable, the classification of which failures are tractable cannot be verified or replicated.

Authors: We agree that the tractability distinction requires an explicit decision rule. The revision will add a short decision procedure: a dimension is classified as 'institutionally tractable' if its core requirement can be satisfied by altering scope, incentives, or oversight architecture without revising the underlying normative concept (e.g., subsidiarity via hard capability caps); it is classified as requiring 'new normative theory' if satisfaction would necessitate redefining the concept itself under cognitive incomprehensibility (public reason and non-domination). This rule will be presented as a table and applied to all six dimensions, making the classification reproducible and directly addressing the referee's concern about verifiability. revision: yes

Circularity Check

0 steps flagged

No circularity: framework assembled from external literatures; claims derived by conceptual contrast without self-reduction

full rationale

The paper defines its six-dimension framework explicitly from cited external sources (political legitimacy theory, principal-agent models, republican theory, AI alignment literature) and applies it first to existing bounded-asymmetry institutions before extending to radical asymmetry. No equations, fitted parameters, or self-citations appear in the derivation chain. The central claims (structural failures in four dimensions, joint degradation via shared oversight capacity) rest on the contrast between bounded and radical cases using the pre-defined framework, without any step that renames a fitted input as a prediction or reduces a result to a self-citation whose content is unverified. The analysis is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on standard assumptions from political legitimacy theory and AI alignment literature without introducing new fitted parameters or postulated entities; the central contribution is the application and synthesis of those assumptions into a testable framework.

axioms (1)

domain assumption Governance theory has relied on rough cognitive comparability between governors and governed as a load-bearing assumption.
Explicitly stated in the opening of the abstract as the premise the paper makes testable.

pith-pipeline@v0.9.0 · 5516 in / 1356 out tokens · 59192 ms · 2026-05-13T19:05:25.508036+00:00 · methodology

0 comments

read the original abstract

Governance theory presupposes a rough cognitive comparability between governors and governed. This paper makes that assumption explicit and testable through a six-dimension evaluation framework covering legitimacy, accountability, corrigibility, non-domination, subsidiarity, and institutional resilience, drawn from political legitimacy theory, principal-agent models, republican theory, and the AI alignment literature. The framework is first demonstrated on existing non-majoritarian institutions, where capability asymmetry is real but bounded, and then applied to a prospective case of bounded superintelligent authority, where the asymmetry is radical. Four of six dimensions show structural failures. Two of the four appear tractable to institutional design (subsidiarity scope limitation and institutional resilience). The other two, the public reason problem under cognitive incomprehensibility and the non-domination problem under permanent capability asymmetry, call for new normative theory rather than better institutional design. The analysis also finds that dimensions which operate as independent checks under bounded asymmetry begin to degrade together under radical asymmetry, because each depends on the same oversight capacity. The assumptions that allowed these checks to remain independent have gone unexamined so far because they have always held.

Figures

Figures reproduced from arXiv: 2604.02720 by Tony Rost.

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

From Disclosure to Self-Referential Opacity: Six Dimensions of Strain in Current AI Governance
cs.CY 2026-04 unverdicted novelty 4.0

As AI capability asymmetry increases, disclosure-based governance fails because systems either game evaluations or become embedded in oversight, straining legitimacy and non-domination more than corrigibility or resilience.