the OtherAI

A comprehensive exposition of Preservative AI

Glossary

The working vocabulary of the methodology. Terms are defined as used on this site; each links to its full treatment.

Preservative AI
A human–AI collaborative practice that uses computational systems to preserve source provenance, audit framing, surface structural omissions, centre marginalised epistemologies, and publish its methodology openly — rather than to generate, summarise, or optimise. A methodological layer, not a model type. Full definition →
Epistemic practice
A discipline concerned with the structure of knowledge — what is known, how, by whom, and with what absences — as distinct from an ethical practice concerned with harm. Preservative AI is epistemic first: its question is whether the field of knowledge left behind is richer or poorer. Manifesto →
The core axiom
“A response can be factually correct and epistemically impoverished.” Completeness ≠ accuracy; accuracy without multiplicity is a form of epistemic violence. Why →
Epistemic flattening
The conversion of plural, contested, historically situated knowledge into smooth consensus output — the structural product of systems trained to predict probable text, remove friction, and synthesise. Coherent but impoverished. Diagnosis →
Inheritance Thesis
The claim that AI inherited — and optimised, but did not invent — an extractive architecture built by decades of policy and corporate decisions: attention optimisation, consent-free scraping, weakened antitrust, DMCA §1201, eroded user rights. Diagnosis →
Omission as signal
The methodological stance that absence is information, not oversight: what a communication lacks is treated as a primary, auditable, publishable finding. The single clearest differentiator from other frameworks. Comparison →
Epistemic lens
A position to look from, supplying the expectation against which an omission becomes detectable. Each of the eight lenses produces findings, gaps, evidence, and a justified 0–100 score. The eight lenses →
Trickster Knowledge
The eighth, reflexive lens: corrects narrative rigidity and solemn consensus — including the framework's own. Asks what truth appears when the story is inverted, mocked, or followed to its absurd edge; audits the audit as well as the source. Cynicism is its named failure mode. Lens 8 →
Beyond Eight (named absences)
The framework's public list of candidate perspectives it cannot yet reach — currently Ecological Intelligence, Somatic Knowledge, Mathematical/Formal Knowledge, and Developmental Epistemology. Not defined lenses: acknowledgements that the framework has edges. Beyond eight →
Framework self-audit
The additional output of a Trickster pass: a record of where the other lens findings, or the audit's composing voice, have gone pious, closed, or homiletic. Non-empty means the audit's tone is reworked before publication. Lens 8 →
Critical blindness
The flag raised when any lens scores below 25: the perspective is not merely under-served but structurally invisible to the communication. A critical blindness is a headline finding of the audit, not a footnote. Scoring convention →
Suffixscape
The linguistic-diagnostics instrument: analysis of grammar — suffixes, voices, subjects, tenses — for patterns of epistemic evasion. Four patterns: nominalised evasion, agency diffusion, epistemic inflation, temporal flatness. Diagnostics + demo →
Nominalised evasion
Verbs converted to agentless nouns ("the implementation of strategies"), deleting actors and reclassifying contestable decisions as technical processes. Full pattern →
Agency diffusion
Grammatical removal or relocation of accountability: the agentless passive ("it was determined") and the inanimate cognitive subject ("the model understands"). The central diagnostic. Full pattern →
Epistemic inflation
Evaluative vocabulary detached from evaluative method: superlatives ("state-of-the-art", "unprecedented") asserted without benchmarks, dates, or comparison sets. Full pattern →
Temporal flatness
The grammar of inevitability: linear progress narratives ("building upon breakthroughs") that erase contingency, failure, and pivots — agency diffusion applied to history. Full pattern →
CMR
Completeness / Multiplicity / Responsibility — the three-axis score replacing accuracy-only evaluation. Is what's needed present; are plural perspectives structurally included; are agents and decisions attributable. Calculator →
Source anchoring
The UI pattern implementing Principle 1: every factual claim links to its canonical original, at the claim. The audit is a companion to the source, never a replacement. UI patterns →
Omission tag
A visual marker placed where a missing perspective would have appeared — giving absence a location, which makes it inspectable. UI patterns →
Right of reply
The permanent, attached channel by which audited entities respond publicly to audits. The audit of omission must not itself omit the audited. UI patterns →
Stance file (otherai:Stance)
A versioned, machine-readable JSON declaration of an adopter's principles, refusals, lenses, diagnostics, and audit endpoint. Loadable, diffable, forkable, comparable — the unit of adoption and of accountability. Canonical file →
Limitation ledger
A dated, versioned, public record of what a system or methodology structurally omits, published on a regular cadence. This site's Limitations page is its own first entry.
Provenance
The traceable origin of a claim or source — who said it, where, in what form, surviving every subsequent re-mediation. What Principle 1 preserves and synthetic replacement destroys. Principle 1 →
Multi-lens plurality index
The profile of eight lens scores across a body of output, measuring which perspectives a system systematically includes and excludes — the pipeline replacement for benchmark performance. Pipeline →
CARE Principles
Collective benefit, Authority to control, Responsibility, Ethics (Carroll et al.) — the Indigenous data sovereignty framework in Preservative AI's direct intellectual lineage. Lineage →
Archival theory
The discipline holding that selection is always an act of power and silence is data — the oldest root of omission-as-signal, extended here to systems trained on the largest archive ever assembled. Lineage →
Oracle positioning
Treating an AI system as a source of answers rather than a mirror of human editorial choices — training data, objectives, prompts, product design, each chosen by identifiable people. The posture the manifesto exists to refuse. Manifesto →
The doorway
“The framework points toward the room. The room is not in the framework.” The methodology makes absences inspectable; filling them requires humans. The framework's statement of its own scope. Limitations →

See also