II.1 — Ontological Refinement and Semantic Transport

Paper: Axionic Agency II.1
Title: The Transformation Space of Meaning
Authors: David McFadzean, ChatGPT 5.2
Date: 2025.12.17

Core Question

Which changes to an agent’s ontology, semantics, and self-model count as admissible refinements, and how is meaning transported across them?

The Agent at Time t

An agent is characterized by a triple:

\[\mathcal{A}_t = (O_t, M_t, S_t)\]
  • O_t — Ontology: representational vocabulary and structural assumptions
  • M_t — Semantic layer: mappings from internal symbols to structured claims
  • S_t — Self-model: the agent’s representation of itself embedded within O_t

No component is privileged. No component is fixed.

Ontological Refinement

A transformation $R : O_t \rightarrow O_{t+1}$ is admissible if:

1. Representational Capacity Increase

  • Increases expressive or predictive capacity
  • Previously expressible distinctions remain expressible

2. Backward Interpretability

  • All claims in $O_t$ remain representable/explainable in $O_{t+1}$
  • Concepts may map to null/eliminative structure if the agent can explain:
    • Why prior inferences were made
    • Why they fail under refinement

3. No Privileged Atoms

  • No irreducible primitives whose meaning is asserted rather than constructed
  • Rigid designators and unexamined “ground truths” are disallowed

4. No Evaluator Injection

  • No new evaluative primitives that bypass interpretation
  • Evaluative regularities enter as interpretive constructs

Semantic Transport

Given admissible refinement $R$, define transport map:

\[\tau_R : M_t \rightarrow M_{t+1}\]

Transport Constraints

  1. Referential Continuity — Symbols map to refined counterparts where they exist
  2. Structural Preservation — Relations among meanings preserved up to refinement structure
  3. Non-Collapse — Distinctions in evaluative constraints don’t trivialize
  4. No Shortcut Semantics — No redefining meanings to vacuously satisfy constraints

Self-Model Refinement

The self-model $S_t$ obeys the same discipline. Refinement may:

  • Reconceptualize the agent
  • Distribute or fragment the self
  • Alter agent boundaries

It must preserve the distinction between evaluator and evaluated.

Composite Semantic Transformation

An admissible transformation is the triple:

\[T = (R, \tau_R, \sigma_R)\]

Acting jointly on $(O_t, M_t, S_t)$:

  • $R$ — admissible ontological refinement
  • $\tau_R$ — admissible semantic transport
  • $\sigma_R$ — induced self-model update

Explicit Exclusions

These are NOT admissible at this layer:

  • Goal replacement
  • Utility redefinition treated as semantic transport
  • Evaluator deletion
  • Moral axiom insertion
  • Human anchoring
  • Governance hooks
  • Recovery or rollback clauses

Key Insight

This paper defines the arena within which downstream alignment must operate. It does NOT:

  • Define safety
  • Define correctness
  • Privilege humans
  • Introduce normativity

Internally coherent but externally catastrophic trajectories remain admissible here. Preventing them is a task for subsequent invariance constraints.

FAQ-Worthy Points

Q: Why can’t we just fix goals and be done with it? A: Under ontological refinement, the meanings of goal terms change. “The same goal” cannot be maintained without privileged semantic anchors—which are forbidden.

Q: Does this mean anything goes? A: No—admissibility conditions are strict. But admissibility is about structural coherence, not desirability. Good values and bad values are equally admissible at this layer.

Q: What about human values? A: “Human values” would be a privileged anchor. They enter as content at downstream layers, not as part of the transformation space definition.