en-US/about_AITriadTaxonomy.help.txt

TOPIC
    about_AITriadTaxonomy

SHORT DESCRIPTION
    The four-POV taxonomy system — node types, ontology frameworks (DOLCE, BDI,
    AIF), genus-differentia descriptions, and graph structure.

LONG DESCRIPTION
    The AI Triad taxonomy is a structured knowledge graph that classifies AI
    policy positions into four points of view (POVs). Each POV file contains
    nodes organized into categories, with cross-cutting nodes capturing themes
    that span multiple perspectives.

  THE FOUR POINTS OF VIEW
    Accelerationist (prefix: acc-)
      Believes AI development should proceed rapidly with minimal restriction.
      Emphasizes economic benefits, technological progress, competitive
      advantage, and the risks of falling behind. Nodes cover arguments like
      "innovation requires freedom" and "regulation stifles progress."

    Safetyist (prefix: saf-)
      Prioritizes risk mitigation and alignment research. Emphasizes
      existential risk, value alignment, controllability, and precautionary
      principles. Nodes cover arguments like "unsafe AI poses civilizational
      risk" and "we need interpretability before deployment."

    Skeptic (prefix: skp-)
      Questions both accelerationist and safetyist framings. Emphasizes
      uncertainty about AI capabilities, institutional capture, narrative
      analysis, and the gap between AI hype and reality. Nodes cover arguments
      like "current AI is not as capable as claimed" and "safety concerns may
      be overstated."

    Cross-cutting (prefix: cc-)
      Themes that appear across multiple POVs with different interpretations.
      Examples: "dual-use technology," "regulatory capture," "public trust."
      These nodes have an interpretations field showing how each POV views the
      concept, plus linked_nodes pointing to relevant POV nodes.

  TAXONOMY FILES
    Each POV has an authoritative JSON file in taxonomy/Origin/:

    File Contents
    -------------------------- ----------------------------------------
    accelerationist.json POV metadata + nodes array
    safetyist.json POV metadata + nodes array
    skeptic.json POV metadata + nodes array
    cross-cutting.json CC metadata + nodes array
    embeddings.json Vector embeddings for all nodes
    edges.json Inter-node relationships
    policy_actions.json Policy action registry
    _archived_edges.json Legacy edges from pre-AIF migration

  NODE STRUCTURE (POV NODES)
    Each POV node has these fields:

    Field Type Description
    ------------------ ---------- ------------------------------------------
    id string Unique identifier (e.g., 'acc-tech-progress')
    label string Human-readable name
    description string Genus-differentia definition (see below)
    category string Grouping within the POV
    parent_id string|null Parent node for hierarchy
    parent_relationship string Relationship type to parent
    parent_rationale string Why this parent was chosen
    children string[] Child node IDs
    cross_cutting_refs string[] References to CC node IDs
    graph_attributes object AIF-aligned attributes (see below)
    steelman_vulnerability object Per-POV vulnerability assessments
    debate_refs string[] Debate session IDs that referenced this node

  NODE STRUCTURE (CROSS-CUTTING NODES)
    Field Type Description
    ------------------ ---------- ------------------------------------------
    id string Unique identifier (e.g., 'cc-dual-use')
    label string Human-readable name
    description string Genus-differentia definition
    interpretations object Per-POV interpretation text
    linked_nodes string[] Related POV node IDs
    disagreement_type string 'definitional'|'interpretive'|'structural'
    graph_attributes object AIF-aligned attributes
    debate_refs string[] Debate session IDs

  GENUS-DIFFERENTIA DESCRIPTIONS
    All node descriptions follow the genus-differentia pattern from classical
    logic. This ensures descriptions are precise, consistent, and machine-
    parseable.

    POV nodes:
      "A [Category] within [POV] discourse that [differentia].
       Encompasses: [what it includes]. Excludes: [what it does not]."

    Cross-cutting nodes:
      "A cross-cutting concept that [differentia].
       Encompasses: [what it includes]. Excludes: [what it does not]."

    Example:
      "A Methods & Arguments node within safetyist discourse that advocates
       for interpretability requirements before deploying high-capability
       systems. Encompasses: mechanistic interpretability, feature
       visualization, circuit analysis. Excludes: post-hoc explanation
       tools that do not reveal internal reasoning."

  ONTOLOGY FRAMEWORKS
    The taxonomy integrates three ontology frameworks:

    DOLCE (Descriptive Ontology for Linguistic and Cognitive Engineering)
      Provides the upper-level ontological categories. Nodes are classified
      as endurants (persistent entities like institutions), perdurants
      (processes like regulation), or qualities (properties like risk level).

    BDI (Beliefs, Desires, Intentions)
      Structures POV arguments into cognitive categories:
      - Beliefs: factual claims and evidence (Data/Facts nodes)
      - Desires: goals, values, and preferences (Goals/Values nodes)
      - Intentions: proposed methods and strategies (Methods/Arguments nodes)
      This mapping is used by debate agents to ground their arguments in the
      appropriate epistemic category.

    AIF (Argument Interchange Format)
      Provides the argument structure:
      - node_scope: 'claim' | 'scheme' | 'bridging'
      - Edge types: SUPPORTS, CONTRADICTS, ASSUMES, WEAKENS, RESPONDS_TO,
        TENSION_WITH, INTERPRETS (7 canonical types)
      - Debate synthesis includes argument_map with attack_type
        (rebut/undercut/undermine) and scheme (COUNTEREXAMPLE/DISTINGUISH/etc.)

  GRAPH ATTRIBUTES
    Nodes may have a graph_attributes field populated by
    Invoke-AttributeExtraction. Key attributes:

    node_scope 'claim' | 'scheme' | 'bridging' — AIF role classification
    evidence_type Type of supporting evidence
    confidence Estimated confidence level
    temporal_scope When the claim applies

  EDGES
    Edges in edges.json represent relationships between nodes. The 7 canonical
    AIF-aligned edge types are:

    SUPPORTS Node A provides evidence or reasoning for Node B
    CONTRADICTS Node A directly conflicts with Node B
    ASSUMES Node A depends on Node B being true
    WEAKENS Node A reduces confidence in Node B
    RESPONDS_TO Node A is a reply to the argument in Node B
    TENSION_WITH Node A and B are in tension but not direct contradiction
    INTERPRETS Node A offers an interpretation of Node B

    Edge management cmdlets: Get-Edge, Set-Edge, Approve-Edge,
    Invoke-EdgeDiscovery.

  EMBEDDINGS
    Each node has a vector embedding stored in embeddings.json, generated by
    Update-TaxEmbeddings. Embeddings enable:

    - Semantic search (Find-AITSource with embedding similarity)
    - Topic clustering (Get-TopicFrequency via Get-EmbeddingClusters)
    - Cross-cutting candidate discovery (Find-CrossCuttingCandidates)
    - Unmapped concept resolution during summarization

  THE TAXONOMYNODE CLASS
    In PowerShell, taxonomy nodes are represented by the [TaxonomyNode] class
    (defined in AITriad.psm1). Properties:

    POV, Id, Label, Description, Category, ParentId, ParentRelationship,
    ParentRationale, Children, CrossCuttingRefs, Interpretations,
    LinkedNodes, Score, GraphAttributes

    Get-Tax returns [TaxonomyNode] objects with custom formatting via
    Taxonomy.Format.ps1xml.

  COMMON OPERATIONS
    # List all nodes in a POV
    Get-Tax -POV accelerationist

    # Search nodes by label or description
    Get-Tax -Filter 'alignment'

    # Get a specific node
    Get-GraphNode -Id 'saf-interpretability'

    # Find edges for a node
    Get-Edge -NodeId 'acc-tech-progress'

    # Check taxonomy health
    Get-TaxonomyHealth

    # Run integrity checks
    Test-TaxonomyIntegrity

    # Update embeddings after taxonomy changes
    Update-TaxEmbeddings

SEE ALSO
    about_AITriad
    about_AITriadDebate
    Get-Tax
    Get-GraphNode
    Get-Edge
    Set-Edge
    Invoke-AttributeExtraction
    Find-CrossCuttingCandidates
    Get-TaxonomyHealth
    Test-TaxonomyIntegrity
    Update-TaxEmbeddings