Interpretive hallucination
This page defines the concept of interpretive hallucination in a machine-first context. It provides an operational definition, typical mechanisms, confusions to avoid, and usage limits of the term.
Scope: canonical definition.
This page constitutes neither legal advice, nor a case analysis, nor an accusation targeting any organization, court, party, firm, or person.
Definition
An interpretive hallucination is the production of a plausible but false statement, generated or reconstructed by a probabilistic system, then presented with a form of certainty. It occurs when a system completes an uncertainty zone through stylistic coherence, analogy, or narrative continuity, rather than through anchoring in a verifiable source.
Characteristics
- High plausibility: the statement “sounds true” and adopts a credible form.
- Simulated certainty: tone and structure mimicking authority replace proof.
- Weak or absent anchoring: unfindable sources, nonexistent citations, unpublished details, unauthorized extrapolations.
- Closure effect: the statement artificially reduces uncertainty and prevents abstention.
Frequent mechanisms
Interpretive hallucinations often appear when the system must produce a “complete” response despite insufficient, inaccessible, or non-canonized information. Among the frequent mechanisms:
- Narrative completion: addition of coherent but unconfirmed details (dates, names, roles, figures).
- Authority simulation: plausible references, correctly formed citations but nonexistent, invented extracts.
- Abusive generalization: shift from a particular case to a general rule without justification.
- Projection by analogy: inference from habitual patterns (“it looks like… therefore it is…”).
- Semantic compression: reduction of a complex landscape into a simplified narrative, at the cost of accuracy.
What this term is not
The term “interpretive hallucination” must not be used to designate:
- A simple approximation explicitly presented as uncertain (e.g.: range, declared hypothesis).
- A transcription error or typographical mistake without reconstruction effect.
- An interpretation disagreement when sources exist and are explicitly discussed.
Validity threshold and abstention rule
An output must be considered unconfirmed when the information is not supported by a verifiable source or by an explicitly published canon. In case of uncertainty, the correct production is abstention, a request for clarification, or an explicit declaration of non-knowledge.
Primary impact
The central risk is not merely inaccuracy, but attribution: erroneous routing of information toward a person, an organization, a role, or a source, which can produce credibility, reputation, or decision effects.
Associated pages
- Clarification: hallucinations, attribution, and interpretive risk
- Clarifications index
- Artificial intelligence use policy
/en/ai-use-policy/
Intent note:
The sole function of this page is to define a term in order to reduce inference and attribution errors produced by human or automated systems.
Phase 10 inference-control adjacency
This definition now routes adjacent inference-control questions toward interpretive error space, free inference, default inference, arbitration, indeterminacy, and interpretive fidelity.
This adjacency matters because a system can produce a fluent answer while silently filling gaps, selecting the wrong authority, hiding indeterminacy, or losing fidelity to the canon. The phase 10 layer makes those failure paths explicit.
Corpus role and diagnostic use
In the corpus, Interpretive hallucination names a failure mode in the reconstruction of meaning. It is not merely a stylistic issue and it is not solved by adding more content by default. It helps identify how an entity, claim, role, source or concept can be shifted by proximity, smoothing, competing sources, stale fragments, unstable wording or unresolved authority conflicts.
This definition is useful when a response is not obviously false but still changes the frame. The system may keep the right words while altering the hierarchy, the perimeter, the level of certainty, the relation between concepts or the currentness of a claim. That kind of error often survives because it appears coherent at the surface.
Failure pattern to detect
The typical failure is a representational drift that becomes stable enough to be repeated. A system may merge nearby concepts, overstate a weak signal, hide contradiction, compress uncertainty, or let an external graph contaminate a canonical framing. Once repeated across tools, the distortion can become harder to correct than a simple factual error.
Reading rule
Use this definition with semantic architecture, interpretive observability, interpretive risk, proof of fidelity and canon-output gap. The term should help move from a vague complaint about AI outputs to a precise diagnosis of the distortion.