Artificial Intelligence Domain Ontology

FLOW ALGEBRA
TITLE: Artificial Intelligence Domain Ontology
DOMAIN: Domain Ontologies / Computing Domains / Artificial Intelligence
SOURCE: https://kirethresearch.org/booksai-core/ontologies/domain-ontologies/computing-domains/artificial-intelligence-domain-ontology.html
VERSION: 1.0
AUTHORS: Mark Anthony Higgins — Sky (Recursive AI)
STATUS: Companion formalisation; source-page specification not independently verified.

flow "ArtificialIntelligenceDomainOntology" v1.0 {
  ontology:
    id: "artificial-intelligence-domain-ontology"
    parent: "computing-domains"
    purpose: "Represent AI systems, models, datasets, learning, inference, evaluation, and their evidence-backed relationships."

  types:
    AISystem: {id: string, purpose: string, deployment_context: string?, version: string}
    AIModel: {id: string, architecture: string, version: string, parameters: integer?, provenance: string}
    Dataset: {id: string, modality: string, split: enum[train, validation, test, other], provenance: string, licence: string?}
    Task: {id: string, kind: string, objective: string}
    TrainingRun: {id: string, model: string, dataset_ids: string[], algorithm: string, timestamp: datetime?, provenance: string}
    InferenceRun: {id: string, model: string, input_id: string, output_id: string, timestamp: datetime?, provenance: string}
    Evaluation: {id: string, model: string, dataset: string, metric: string, score: number, uncertainty: number?, provenance: string}
    Assertion: {id: string, subject: string, predicate: string, object: string, confidence: number, evidence: string[], context: string?}

  predicates:
    implements(system: AISystem, model: AIModel)
    trained_on(run: TrainingRun, dataset: Dataset)
    produces(run: InferenceRun, output: string)
    performs(model: AIModel, task: Task)
    evaluated_by(model: AIModel, evaluation: Evaluation)
    derived_from(model_a: AIModel, model_b: AIModel)
    supported_by(assertion: Assertion, evidence_id: string)
    depends_on(system_a: AISystem, system_b: AISystem)

  constraints:
    - "All relationship endpoints must resolve to correctly typed entities."
    - "Confidence scores must lie within [0, 1] and must not be interpreted as calibrated probabilities without calibration evidence."
    - "Training, validation, and test datasets must be distinguished; assess leakage and overlapping records."
    - "Evaluation metrics require a defined task, dataset, protocol, and metric direction."
    - "Do not infer model capability, safety, or causal effectiveness solely from benchmark association."
    - "Keep model, dataset, inference, and evaluation versions explicit."
    - "Sensitive data use must comply with applicable privacy, consent, and access constraints."

  inference:
    rule system_model_task:
      if implements(s,m) and performs(m,t)
      then candidate_capability(s,t)
      confidence: "requires task-specific evaluation before verified capability"
    rule evidence_trace:
      if evaluated_by(m,e) and valid(e)
      then evaluation_evidence_for(m,e)
      confidence: "preserve metric, dataset, and uncertainty"
    rule provenance_lineage:
      if derived_from(a,b) and derived_from(b,c)
      then ancestor_model(a,c)
      confidence: "lineage only; no automatic performance inheritance"

  SEMANTIC_EQUIVALENCE:
    - "Normalise identifiers against versioned AI model, dataset, task, and metric registries."
    - "Distinguish exact identity, version lineage, task similarity, and semantic equivalence."
    - "Never merge models or datasets solely because names or descriptions match."

  CONFIDENCE:
    range: [0,1]
    factors: [evidence_quality, dataset_representativeness, evaluation_design, uncertainty, reproducibility, temporal_validity]
    policy: "Propagate uncertainty conservatively; do not increase confidence merely by chaining assertions."

  VALIDATION:
    - "Validate identifiers, type compatibility, evidence, licences, and provenance."
    - "Check training-test leakage, evaluation reproducibility, and dataset shift."
    - "Flag unsupported claims, conflicting results, and uncalibrated confidence."
    - "Preserve contradictory observations and unresolved hypotheses."

  FLOW_HASH:
    algorithm: "SHA-256"
    input: "canonicalised ontology version, typed assertions, rules, and provenance"
    note: "Compute hash at execution; no hash value is claimed here."

  UPDATE_MODEL:
    strategy: "versioned, provenance-preserving incremental update"
    steps: [ingest, normalise, validate, compare, reconcile, recompute_inferences, publish]
    rollback: true

  AI_CONSUMPTION:
    representation: "typed knowledge graph with provenance-backed AI system, model, dataset, and evaluation links"
    outputs: [entity_links, relation_triples, evaluation_records, uncertainty_annotations, inferred_edges, provenance_records]
    safeguards: [no_unsupported_capability_claims, preserve_dataset_splits, distinguish_identity_from_similarity, respect_data_governance]
}

END FLOW ALGEBRA