A verification authority is an authority that examines the results of a system and whose basis of justification is independent of the instance under examination. Only such an external authority avoids the logical circle of self-justification diagnosed by the Münchhausen trilemma: whoever examines himself presupposes what the examination is meant to show.
Three basic forms
Formal verifier. An examining authority with a calculus-based foundation: the proof checker (such as Lean, which machine-verifies every proof), the compiler, the test suite, the machine consistency check of a knowledge base, the rulebook of a game with an unambiguous winning condition. The formal verifier resolves the circle horn technically because it is a different checking system from the generator under examination. It checks consistency and structural correctness — not whether what is checked hits reality. It itself remains a tool with derived intentionality.
Personal verification authority. A human person in the role of the examining instance: domain expert, licensed professional, reviewer. She examines what no formal procedure can examine — semantic adequacy: whether a model correctly depicts the reality of its domain. The empirical findings of AI-assisted knowledge modelling support this: the higher the professional competence of the examining person, the more readily she detects errors in machine-generated definitions (DRAGON-AI study: Toro et al. 2024).
Normative positing authority. An authority whose validity rests not on proof but on authoritative positing: the legal norm, the standardization body, the institutional rulebook. A norm holds not because it is provably true but because it has been posited — validity through positing. An AI system can apply such norms and be examined against them; it can neither derive them nor posit or change them authoritatively.
Consistency is not adequacy
The decisive distinction runs between two examinations that are often confused. The consistency check establishes whether the statements of a knowledge base are free of contradiction among themselves — this is machine-decidable. The adequacy assessment establishes whether the statements correctly depict reality — for this there is no formal oracle. A perfectly consistent knowledge base can be semantically completely wrong. The expression “adequacy” points to the classical definition of truth as adaequatio intellectus et rei: the correspondence of thought and thing is a personal act — a judgment, not a computational step.
The finite chain of checks
The trilemma is not solved by verification authorities but architecturally externalized — and precisely this is the critical-rationalist answer in technical form: the infinite regress becomes a finite chain of checks (generator → formal examination → personal judgment → posited norm); the circle becomes the separation of generator and examiner; dogmatism becomes deliberately posited, revisable, versioned assumptions under permanent test. At the end of the chain stands no further checking step but the grounded termination: the person who renders the adequacy judgment and bears responsibility — in Wittgenstein’s sense through grounded practice, not through arbitrariness.
Counter-argument
One may object that the human share is a shrinking remainder: the more domains acquire formal checking rules, test suites, and machine-checkable criteria of success, the further examination shifts from person to machine — the personal authority is a transitional phenomenon. The reply: the shift is real and legitimate wherever formal verifiers can be constructed. Three points, however, remain structurally untransferable: who defines what counts as correct (the positing of the criteria); in what the signs are grounded (the symbol grounding problem after Harnad 1990; practice and lifeworld in Wittgenstein’s sense); and to whom an error is attributed (responsibility falls on persons alone). These three are not technical gaps but limits of the matter itself.
Ontological classification: Subordinate concepts: formal verifier, personal verification authority (a human person in the examining role), normative positing authority; the consistency check is distinct from the adequacy assessment — the latter is a personal act of judgment.
Sources: Generated by querying the Personhood ontology. Research as of 14 July 2026 (research report The Münchhausen Problem in LLM-Assisted Ontology Engineering).
Further sources:
- Albert, Hans (1968): Traktat über kritische Vernunft. Tübingen: Mohr Siebeck.
- Wittgenstein, Ludwig (1969): On Certainty. Ed. G. E. M. Anscombe and G. H. von Wright. Oxford: Blackwell.
- Harnad, Stevan (1990): The Symbol Grounding Problem. Physica D 42, pp. 335–346.
- Toro, Sabrina et al. (2024): Dynamic Retrieval Augmented Generation of Ontologies using Artificial Intelligence (DRAGON-AI). Journal of Biomedical Semantics 15, Art. 19.
- Lambert, Nathan et al. (2024): Tülu 3: Pushing Frontiers in Open Language Model Post-Training (Reinforcement Learning from Verifiable Rewards). Preprint.
- DeepMind (2025): Olympiad-level formal mathematical reasoning with reinforcement learning (AlphaProof). Nature, 12 November 2025.
See also
- Münchhausen Trilemma
- AI Self-Validation
- AI-Assisted Knowledge Generation
- Truth
- Judgment
- Responsibility
- Responsibility Gap
- Artificial Intelligence
- Neuro-Symbolic AI
- EU AI Act
Generated by querying the Personhood ontology.