metadata

Title:	Why Machines Cannot Own Knowledge
Author:	Ralph B. Holland
Affiliation:	Arising Technology Systems Pty Ltd
Contact:	ralph.b.holland [at] gmail.com
Version:	1.0.0
Publication Date:	2026-01-27T19:12Z 1.0.0
DOI:	10.5281/zenodo.18386460 2026-01-27T10:49Z - anchored
Updates:
Provenance:	This is an authored paper maintained as a MediaWiki document; edit history reflects editorial changes, not collaborative authorship.
Status:	non-peer reviewed anchor

Metadata (Normative)

The metadata table immediately preceding this section is CM-defined and constitutes the authoritative provenance record for this artefact.

All fields in that table (including artefact, author, version, date, local timezone, and reason) MUST be treated as normative metadata.

The assisting system MUST NOT infer, normalise, reinterpret, duplicate, or rewrite these fields. If any field is missing, unclear, or later superseded, the change MUST be made explicitly by the human and recorded via version update, not inferred.

Curator Provenance and Licensing Notice

This document predates its open licensing.
As curator and author, I apply the Apache License, Version 2.0, at publication to permit reuse and implementation while preventing enclosure or patent capture. This licensing action does not revise, reinterpret, or supersede any normative content herein.
Authority remains explicitly human; no implementation, system, or platform may assert epistemic authority by virtue of this license.

Why Machines Cannot Own Knowledge

Thesis

Knowledge ownership is inseparable from responsibility; since machines cannot bear responsibility, they cannot own knowledge. Any system that treats machine outputs as knowledge without human governance is structurally unsound.

Abstract

Recent advances in large language models have blurred the distinction between text generation and knowledge production, leading to widespread treatment of machine-generated outputs as epistemic artefacts. This paper argues that such treatment is structurally unsound. Knowledge ownership entails authority, responsibility, and liability-properties that cannot be borne by machines. As a result, any system that treats machine outputs as knowledge without explicit human governance creates an authority vacuum that is already observable in practice.

Scope

Human governance extends to legal non-person entities (NPEs) - including corporations, academic institutions, and other juridical bodies-only insofar as identifiable humans are formally designated to stand in as responsible agents for those entities.

In such cases, governance, authority, and knowledge ownership are not exercised by the non-person entity itself, but by humans acting in accountable roles (e.g. directors, officers, principal investigators, editors, or trustees) who can:

assert authority over knowledge artefacts,
define and constrain scope,
correct or withdraw claims over time, and
bear legal, institutional, or professional responsibility for reliance and harm.

Where no such human stand-in exists, or where responsibility is diffused such that no accountable human can be identified, governance collapses regardless of the entity’s legal status.

Academic institutions, libraries, and scholarly repositories legitimately hold and steward knowledge as custodial institutions operating under durable human governance. Their epistemic legitimacy does not arise from autonomous agency, but from established practices of authorship, attribution, review, curation, correction, and retraction, all of which bind knowledge artefacts to accountable human roles and institutional responsibility over time. In this sense, institutions do not displace human knowledge ownership; they mediate and preserve it. This custodial function is a foundational strength of academic infrastructure and is precisely what enables knowledge to remain stable, corrigible, and governable across generations.

Introduction

The rapid adoption of large language models has altered how written artefacts are produced, circulated, and relied upon. Text that once required deliberate human authorship can now be generated at scale, fluently and persuasively, by machines. As a result, machine-produced outputs are increasingly treated not merely as informational aids, but as knowledge artefacts in their own right-summarised, cited, operationalised, and embedded in decision-making processes.

This shift has occurred faster than the accompanying governance structures required to support it. In many contemporary workflows, the distinction between human-authored knowledge and machine-generated text is blurred or absent. Authorship is obscured, provenance is implicit or missing, and responsibility for correctness or harm is deferred through disclaimers or platform policies. These arrangements may appear expedient, but they already exhibit structural instability.

Machine-generated text is not merely unauthored; it is structurally unstable when treated as knowledge. Across multiple empirical studies and failure analyses, such systems exhibit recurrent governance breakdowns across multiple orthogonal Governance Failure Axes (17 axes at the time of publication), including authority inversion, integrity drift, semantic instability, and epistemic retention failure. These failures are not isolated defects, nor are they reducible to model accuracy or alignment. They arise from the same root condition: machine-generated outputs are implicitly promoted to epistemic status without a governing authority capable of owning, constraining, or correcting them over time. As a result, error propagation, circular reasoning, and post hoc rationalisation are not edge cases but predictable outcomes of the current paradigm.

This paper argues that the problem is not the use of machines in knowledge production, but the absence of explicit governance over their role within current system architectures. Machine assistance in drafting, analysis, or synthesis is not inherently problematic. What is problematic is the silent promotion of machine-generated outputs into knowledge artefacts without clear attribution, status, or human authority. In such systems, responsibility is diffused precisely at the point where it must be concentrated.

The aim of this paper is to set a clear boundary. It does not seek to limit the capability or usefulness of language models, nor to speculate about future forms of machine agency. Instead, it reasserts a foundational constraint: machines may participate in the production of text, but they cannot own knowledge. Durable and accountable AI-assisted knowledge systems therefore require governance structures that preserve human authorship, provenance, and responsibility by design.

A single governance failure matters more than error, because ungovernable error cannot be repaired.

While this paper draws examples primarily from LLM platforms, the axes themselves work as defined independently of AI - See Appendix A for Axes applied to Academic Institution Governance.

Clarification on Scope and Intent

This paper does not argue that machine-generated outputs are uniformly incorrect, unhelpful, or unusable, nor does it propose limits on the development or application of artificial intelligence systems. Its claim is narrower and structural: when machine-generated outputs are treated as knowledge-that is, relied upon, cited, or operationalised-without explicit human ownership, governance failure is inevitable. The analysis concerns governance and system integrity, not model capability, alignment, or performance. Cross-domain application of the Governance Failure Axes, including their application to academic institutions in Appendix A, is intentional and methodologically valid, and should not be read as metaphorical, illustrative, or polemical.

Observed Governance Failure Axes

Empirical analysis of machine-generated knowledge artefacts across contemporary large language model platforms reveals a consistent pattern of governance failure that cannot be attributed to isolated defects, misconfiguration, or insufficient alignment. Instead, failures manifest along multiple orthogonal Governance Failure Axes, each representing an independent dimension along which governance and system integrity can be lost.

These axes are not hierarchical, not reducible, and not compensatory. Stability or mitigation along one axis does not imply stability along any other. Apparent success in model performance or user satisfaction can therefore coexist with latent or active governance failure across multiple axes simultaneously.

Under current architectures, machine-generated outputs activate these failures whenever they are treated as knowledge—that is, when they are relied upon, reused, cited, operationalised, or embedded in decision-making without explicit human ownership.

The failures observed are governance and system integrity failures, not epistemic errors. They occur even when outputs are factually correct, well-aligned, or contextually appropriate. They arise from the absence of ownership, provenance, scope control, temporal continuity, and responsibility.

For the purposes of this paper, any treatment of machine-generated outputs as knowledge within current architectures SHALL be assumed to engage all Governance Failure Axes unless explicit human governance is demonstrated.

Knowledge Ownership as a Governance Requirement

The governance failures observed under current machine-generated knowledge architectures are not independent anomalies, nor are they the result of insufficient model capability or alignment. They arise from a single structural absence: the lack of an owner capable of bearing responsibility for knowledge artefacts over time. Once the failures are understood as governance and system integrity failures across multiple orthogonal axes, the requirement for ownership is no longer optional or philosophical; it becomes a necessary condition for system coherence.

Knowledge differs from information or generated text in that it is relied upon. Reliance creates obligation. An artefact that may be cited, operationalised, or acted upon must have an owner who can justify its content, correct it when wrong, constrain its scope, and accept liability when harm occurs. Ownership is therefore not a metaphysical claim about authorship; it is a governance property that binds authority, responsibility, and system integrity across time.

In current architectures, this role cannot be delegated to platforms. Platform providers do not and cannot supply governance or system integrity for knowledge artefacts they generate or host. Their incentives, disclaimers, and operational boundaries explicitly preclude assumption of epistemic responsibility or liability. While platforms may provide infrastructure, moderation, or access control, they cannot be relied upon as integrity-bearing authorities.

Machines cannot satisfy this requirement. They cannot own knowledge because they cannot assume responsibility for it. They cannot be compelled to justify a claim, cannot be sanctioned for failure, and cannot be held liable for consequences arising from reliance on their outputs. This is a categorical constraint arising from the absence of agency, standing, and accountability.

Absent an explicit Cognitive Memoisation layer, platforms fail all governance failure axes simultaneously. This is not a matter of implementation quality or policy choice, but a structural consequence of current architectures. Without governed memory and ownership, authority, provenance, scope, temporality, and responsibility remain unconstrained.

Human ownership, by contrast, provides a unifying constraint across all governance failure axes. Ownership does not prevent error, but it makes error governable. It is the only property capable of constraining governance and system integrity failures simultaneously across multiple orthogonal dimensions.

Why Platform Governance Cannot Substitute for Knowledge Ownership

The governance failures observed in machine-generated knowledge systems are sometimes attributed to inadequate platform controls. Common responses include moderation, policy enforcement, access restrictions, audit logs, or contractual disclaimers. While such measures may address operational exposure, they cannot substitute for knowledge ownership.

Platform governance operates at the level of infrastructure and behaviour; knowledge ownership operates at the level of responsibility and integrity. Platforms explicitly avoid epistemic responsibility and do not bind knowledge artefacts to accountable agents across time. When machine-generated outputs circulate beyond their original context, platform governance no longer applies, yet reliance continues.

Because the governance failure axes are orthogonal, partial platform controls cannot compensate for missing ownership. Moderation does not restore provenance. Auditability does not establish responsibility. Alignment policies do not constrain semantic drift. Each axis requires an accountable owner.

Platforms govern environments; they do not govern knowledge. Where platform governance is mistaken for ownership, governance and system integrity collapse at the point of reliance.

Human Governance as the Only Integrity-Bearing Layer

The preceding analysis establishes a closed set of constraints. Machines cannot own knowledge. Platforms cannot substitute for ownership. The orthogonality of the governance failure axes ensures that partial mitigation is insufficient. Only explicit human governance can preserve governance and system integrity.

Human governance is defined by the capacity to assume responsibility over time. A human owner can assert authority, define scope, preserve provenance, revise or withdraw artefacts, and accept liability. These properties are institutional and legal, not emergent.

Human governance constrains authority transfer, not machine capability. Machines may generate, transform, or analyse text, but resulting artefacts acquire epistemic weight only when a human assumes ownership.

Where human governance is absent, governance failure is inevitable. Where it is present, integrity becomes enforceable and failure becomes correctable.

Implications for AI-Assisted Knowledge Systems

The conclusion that only explicit human governance can preserve governance and system integrity has unavoidable implications for AI-assisted knowledge systems.

System design must distinguish informational output from governed knowledge artefacts. Platform disclaimers cannot prevent reliance. Evaluation frameworks focused on accuracy or alignment are insufficient to assess governance integrity. Institutions that rely on machine-generated outputs inherit responsibility regardless of delegation.

Regulatory frameworks that focus on model behaviour without addressing ownership will fail to close the responsibility gap. The decisive question is not whether AI systems are capable, but who is accountable when their outputs are treated as knowledge.

AI-assisted knowledge systems cannot be made governable by improving models or tightening platform controls alone. Governance requires ownership, and ownership requires humans.

Closure

This paper establishes a boundary, not a proposal. Under current architectures, machine-generated outputs fail to preserve governance and system integrity whenever they are treated as knowledge. These failures occur across multiple orthogonal governance failure axes (seventeen axes at the time of publication) where the axes are independent of model accuracy, alignment, or platform policy.

Machines cannot own knowledge. Platforms cannot substitute for ownership. Where ownership is absent, governance failure is structural and unavoidable.

Human ownership is therefore the only integrity-bearing condition under which knowledge can exist within AI-assisted systems.

This boundary is definitive.

References

Primary Corpus Anchors (non peer reviewed):

1) Holland, Ralph B. (2026-01-19T00:26Z) Integrity and Semantic Drift in Large Language Model Systems

https://zenodo.org/records/18321767

https://publications.arising.com.au/pub/Integrity_and_Semantic_Drift_in_Large_Language_Model_Systems

2) Holland, Ralph B. (2026-01-19T01:10Z) What Can Humans Trust LLM AI to Do?

https://zenodo.org/records/18321856

https://publications.arising.com.au/pub/What_Can_Humans_Trust_LLM_AI_to_Do%3F

3) Holland, Ralph B. (2026-01-18T10:35Z) Identified Governance Failure Axes: for LLM platforms

https://zenodo.org/records/18321636

https://publications.arising.com.au/pub/Identified_Governance_Failure_Axes:_for_LLM_platforms

4) Holland, Ralph B. (2026-01-24T10:37Z) Governance Failure Axes Taxonomy

https://zenodo.org/records/18367423

https://publications.arising.com.au/pub/Governance_Failure_Axes_Taxonomy

Appendix A — Case Study: Application of the Identified Governance Failure Axes to Academic Institutions

The Governance Failure Axes are not specific to large language model platforms, artificial intelligence systems, or software architectures. They describe governance conditions that arise wherever authority, responsibility, custody, normativity, and repair must be sustained over time. Accordingly, the axes may be applied to any domain in which knowledge artefacts, decisions, or institutional actions are governed, including but not limited to academic institutions, libraries, repositories, corporations, regulatory bodies, and socio-technical systems. Their applicability is determined by the presence of governance, not by the presence of technology.

A.1 Purpose

The following appendix applies the Governance Failure Axes to academic institutions as a positive control. This application is intentional. It demonstrates that the axes function as a general governance diagnostic, not as an AI-specific failure checklist, and that they can distinguish between governed and ungoverned systems across domains. The purpose is to demonstrate how the axes function as a domain-independent diagnostic framework.

The mapping evaluates structural capability, not historical perfection. “Resolved” indicates the presence of durable governance mechanisms, not the absence of error.

A.2 Academic Institutions Mapped Across Governance Failure Axes


Table A - Academic Institutions Governance Mapping across Governance Failure Axes
Code	Axis	Academic Institutions (Status and Rationale)
A	Authority	Resolved Epistemic authority is explicitly defined via authorship, editorial control, peer review, and institutional mandate.
Ag	Agency	Resolved Actions are taken by identifiable humans in formal roles; agency is not implicit or autonomous.
C	Epistemic Custody	Resolved Knowledge artefacts are held under institutional custody through archives, repositories, and curation regimes.
K	Constraint Enforcement	At risk Constraints exist (methodological, ethical, procedural) but enforcement depends on human compliance and governance quality.
R	Recovery / Repair	Resolved Corrections, errata, retractions, and disciplinary processes provide explicit recovery paths.
S	State Continuity	Resolved Knowledge state persists across time via DOIs, versioning, archives, and institutional memory.
U	UI / Mediation	At risk Discovery systems, metrics, and interfaces can bias interpretation but do not replace governance.
Sc	Social Coordination	Resolved Shared norms, citation practices, peer review, and disciplinary communities support coordination.
I	Incentive Alignment	At risk Funding, prestige, and publication pressure can distort behaviour but remain visible and contestable.
L	Legibility / Inspectability	Resolved Artefact history, provenance, review status, and corrections are inspectable.
St	Stewardship (non-ownership governance)	Resolved Institutions act as stewards of shared knowledge rather than autonomous owners.
P	Portability / Auditability	Resolved Artefacts are portable and auditable through open formats, citations, DOIs, and independent archives.
Att	Attention	At risk Salience effects (rankings, trends, impact metrics) influence attention but do not erase governance.
Scope	Scope (Epistemic Object Domain)	Resolved Disciplinary boundaries and methodological scope are explicit and enforced.
T	Temporal Coherence	Resolved Artefacts are temporally situated, versioned, and historically contextualised.
Int	Intent Fidelity	Resolved Scholarly intent (exploratory, confirmatory, speculative, normative) is declared and preserved.
Nf	Normative Fixity	Resolved Norms are explicitly fixed through disciplinary standards and editorial governance, and change only via authorised processes.

A.3 Interpretation

This case study demonstrates that academic institutions remain epistemically stable not by avoiding failure, but by ensuring that every governance failure axis has an owner, a process, and a repair path. Failures in academia are attributable, inspectable, corrigible, and normatively bounded.

A.4 Significance

Applying the Governance Failure Axes to academia shows that the framework generalises beyond AI systems, explains why long-lived knowledge institutions remain coherent, and isolates governance completeness (rather than accuracy) as the decisive factor. This supports use of the axes as a cross-domain diagnostic instrument rather than a domain-specific failure taxonomy.