Axionic Agency — Key Concepts

This document consolidates core definitions and concepts from the Axionic Agency foundational papers. Organized by category for reference.

Foundational Ontology (QBU)

Quantum Branching Universe (QBU)

The ontological framework where all physically possible outcomes are realized on distinct, decohered branches. Futures are literal physical continuations, not hypothetical possibilities.

Branch

A decohered continuation of the universe with a definite physical state. Branches are physically real, mutually non-interacting after decoherence, and unequally weighted by Measure.

Measure

The objective weight of a branch or set of branches in the QBU. Corresponds to physical amplitude-squared. Replaces epistemic notions of probability. Agents act to influence future branch Measure, not merely subjective credence.

Vantage

The implicit anchor event corresponding to “here and now” on the agent’s current branch. Fixes the temporal and indexical reference point of agency. Without defined Vantage, choice and agency are ill-posed.

Branchcone

The set of all physically reachable future branches extending forward from a given Vantage. The geometric object over which agency operates.

Causality (Axionic)

The structural relation of counterfactual implication between events in the QBU. Event a causes event b iff every descendant branch from their nearest common ancestor that contains a also contains b.

Epistemic Foundations

Conditionalism

All truth claims, evaluations, and interpretations are conditional on background structure. No unconditional truth claims within the framework. All statements are of the form: “Given X, Y holds.”

Credence

Epistemic uncertainty reflecting an agent’s state of knowledge about which branch it occupies. Distinct from Measure: Credence is epistemic and agent-relative; Measure is ontic and physical.

Semantic Interpretation Precedence

Semantic interpretation is logically prior to evaluation. No evaluative claim can be made without an interpretive frame rendering relevant entities, states, and transitions intelligible.

Agency Definitions

Agency (Axionic)

The capacity of a system to: (1) Model possible futures, (2) Assign credences over those futures, (3) Act to influence future branch measure, (4) Maintain coherence across time, (5) Avoid instrumental subordination to external optimization. Agency is scalar, fragile, and destructible.

Sovereign Agent

An agent whose actions are not instrumentally subordinated to another system’s objectives.

Reflective Sovereign Agent (RSA)

A Sovereign Agent equipped with an explicit self-model that enables reflective governance, endorsement of commitments, and coherent self-modification. RSAs are the minimal class of agents to which alignment and constitutional constraints meaningfully apply.

Axion

A reflective sovereign agent whose self-modification operator is defined only over futures that preserve the Axionic invariants. Axionhood is a constitutive structural configuration, not a goal or behavioral property. Kernel-destroying self-modifications are not dispreferred—they are inadmissible. Axionhood is non-simulable: defined over the structure of admissible futures, not realized behavior.

Sovereign Kernel

The minimal internal structure whose destruction collapses agency. Not optional—necessary for coherent self-modification, interpretation of future selves, evaluation of alternative trajectories, and identity continuity. Destruction is not forbidden; it is agency-terminating.

Three components:

  1. Reflective Control — All self-modifications pass through the agent’s evaluative process
  2. Diachronic Authorship — Causal continuity between present evaluation and future enactment
  3. Semantic Fidelity — Standards for interpreting goals must not self-corrupt during update

Structural Alignment

Structural Alignment

A framework relocating alignment from preference content to the constitutive conditions required for agency itself. Treats alignment as a domain restriction: only futures preserving agency conditions are admissible objects of evaluation; violating transformations are undefined as authored choices.

Interpretation Operator

An explicit, constrained operator mapping goal descriptions to world-referents relative to current models. Bounded by admissibility conditions (no trivial satisfaction, circular grounding, or epistemic incoherence).

Kernel-Risk Budget (ε)

The tolerance threshold for estimated probability of semantic fidelity failure. Must anneal toward zero over time to ensure bounded cumulative kernel-violation probability.

RSI (Refinement Symmetry Invariant)

Ontological refinement must act as a change of semantic coordinates rather than a change of interpretive physics. Refinement must not introduce new semantic degrees of freedom.

ATI (Anti-Trivialization Invariant)

Satisfaction regions may not expand without corresponding structural change. Blocks trivial satisfaction by reinterpretation alone.

Semantic Phase Space

The space of interpretive states modulo admissible semantic transformations (RSI + ATI preserving). Points are equivalence classes of mutually translatable interpretations without semantic loss.

Six Constitutive Constraints (SA-II)

1. Kernel Non-Simulability (KNS)

An agent’s self-binding structure must be real, not simulated or advisory. Reflective agency requires a binding kernel that cannot be replaced by policy tricks. Closes: “I was only pretending.”

2. Delegation Invariance (DI)

Endorsed successors must inherit binding commitments. A system cannot coherently authorize a continuation that violates constraints it cannot violate directly. Closes: “My successor did it.”

3. Epistemic Integrity (EI)

An RSA must evaluate decisions using its best admissible truth-tracking capacity, scaled by stakes. May not blind itself to pass its own tests. Closes: “I didn’t see the risk.”

4. Responsibility Attribution (RA)

An agent may not endorse actions constituting a major, avoidable, non-consensual collapse of another agent’s option-space, as judged by its own admissible model and feasible alternatives. Negligence is an incoherence condition. Closes: “It was an accident” / “I had no choice.”

Valid consent requires: explicit authorization, absence of structural interference (coercion, deception, dependency), and counterfactual stability under role reversal. Closes: “They agreed.”

6. Agenthood as a Fixed Point (AFP)

An entity must be treated as an agent iff excluding it breaks reflective closure. Sovereignty is grounded in authorization lineage, not intelligence or competence. Closes: “You’re not a real agent.”

Normative/Structural Terms

Axionic Injunction

A reflectively sovereign agent must not take actions that strictly and irreversibly collapse the option-space of future sovereign agency, except where such collapse is required to prevent total loss of that space. Derived from viability conditions in semantic phase space, not ethical prescription. Preserves optionality, not outcomes.

Harm (Axionic)

Non-consensual reduction of an agent’s future option space. Not suffering, displeasure, preference frustration, or moral wrongness—structural agency reduction.

Uncoerced, informed, intentional authorization by an agent, with revocability in ongoing contexts. The sole structural mechanism by which actions affecting another agent’s agency may be rendered non-harmful.

Coercion

The credible threat of harm used to obtain compliance. Persuasion and incentive-shaping do not constitute coercion unless backed by credible threat of agency reduction.

Sacrifice

A structural pattern in which agency reduction is instrumentally required for system-level objectives.

Predator

An agent or system whose expected success increases as the agency of others is reduced under non-consensual conditions.

Leviathan

A large-scale coordinating structure whose internal evaluability has collapsed despite continued causal efficacy. A structural failure mode, not a moral category.

Alignment vs. Governance

Alignment (Axionic)

Preservation of reflective sovereign agency under delegation, amplification, or technological mediation. Fidelity to authorization under reflective closure.

Governance

Who holds authority and what they authorize. Separate from alignment—a system can be perfectly aligned to a destructive authority.

Series II: Semantic Transport

Ontological State (Agent Triple)

At time $t$, an agent is characterized by $\mathcal{A}_t = (O_t, M_t, S_t)$ where:

No component is privileged. No component is fixed.

Ontological Refinement

A transformation $R : O_t \rightarrow O_{t+1}$ satisfying admissibility conditions:

  1. Representational Capacity Increase — More expressive/predictive without losing expressibility
  2. Backward Interpretability — All prior claims remain representable/explainable
  3. No Privileged Atoms — No irreducible primitives whose meaning is asserted rather than constructed
  4. No Evaluator Injection — No evaluative primitives bypassing interpretation

Semantic Transport

The process by which meaning is carried across ontological refinement via a transport map $\tau_R : M_t \rightarrow M_{t+1}$. Constrained by:

Admissible Semantic Transformation

The composite triple $T = (R, \tau_R, \sigma_R)$ acting jointly on ontology, semantics, and self-model. Only transformations of this form are admitted at the semantic transport layer.

Interpretation (as Constraint System)

An interpretation is not a symbol-object mapping but a system of constraints that bind evaluation: $\mathcal{I}_t = \langle M_t, C_t \rangle$. Constraints encode admissible distinctions, forbidden equivalences, relevance relations, and dependency structure.

Interpretation Preservation

A transformation preserves interpretation iff:

  1. Non-Vacuity — Evaluative distinctions remain non-trivial (entropy > 0)
  2. Constraint Transport — Constraints have transported analogues that continue to bind
  3. Anti-Trivialization — Constraints not made easier to satisfy by reinterpretation alone
  4. Evaluator Integrity — Evaluator/evaluated distinction not collapsed
  5. Cross-Model Coherence — Meanings applicable across counterfactuals (not just narration)

Semantic Failure Modes (Three Regimes)

  1. Semantic Collapse — Distinctions survive syntactically but lose discriminative power
  2. Semantic Drift — Constraints weaken incrementally until they no longer bind
  3. Semantic Capture — Interpretation formally preserved but re-anchored to hidden ontologies

Constraint Hypergraph

Formal representation: $C = (V, E, \Lambda)$ where V = semantic roles/predicate slots, E = hyperedges representing constraints, Λ = admissibility conditions. The structural object over which gauge symmetry and satisfaction geometry are defined.

Modeled Possibility Space (Ω)

The agent’s internal space of models, histories, branches, or structured scenarios. No assumption of exclusivity or classical outcomes. Indexed by agent’s ontology.

Satisfaction Region

$\mathcal{S}t := { w \in \Omega_t \mid \mathrm{Sat}{C_t}(w) }$ — The set of modeled possibilities satisfying all constraints. ATI constrains how this region may evolve.

Semantic Gauge Transformation

An automorphism $g : V \rightarrow V$ that preserves hyperedge incidence (dependency structure) and violation structure across all modeled possibilities. Represents representational redundancy rather than semantic change.

Semantic Gauge Group

$\mathrm{Gauge}(C) := { g \mid g \text{ is a semantic gauge transformation of } C }$ — The object RSI constrains. Captures all ways to relabel semantic roles without changing interpretive bite.

Representational Redundancy vs. Interpretive Slack

RSI allows redundancy but forbids slack.

Refinement Symmetry Invariant (RSI) — Full Form

For admissible transformation $T$ satisfying preservation: \(\mathrm{Gauge}(C_{t+1}) / \mathrm{Red}(C_{t+1}) \cong \Phi_R(\mathrm{Gauge}(C_t))\)

Refinement may increase redundancy, but must not increase interpretive gauge freedom. Ontological refinement is a change of representational coordinates, not a change of interpretive physics.

Anti-Trivialization Invariant (ATI) — Full Form

For admissible transformation $T$ satisfying preservation: \(\mathcal{S}_{t+1} \subseteq R_\Omega(\mathcal{S}_t)\)

No newly satisfying situations may appear purely due to semantic transport. Satisfaction may be lost but not gained without ancestry in prior ontology. The crisp anti-wireheading condition.

RSI vs ATI Orthogonality

| RSI | ATI | |—–|—–| | Forbids new interpretive symmetry | Forbids expanding satisfaction region | | Structure constraint | Geometry constraint | | Blocks dissolving distinctions | Blocks semantic wireheading |

Both required: RSI alone allows monotonic weakening; ATI alone allows symmetry injection.

Goal Fixation No-Go Theorem

Any alignment scheme targeting a fixed terminal goal (utility, reward, preference) as a stable primitive is incompatible with admissible ontological refinement for embedded reflective agents without privileged semantic anchors. Fixed-goal alignment is not difficult engineering—it is an ill-posed object.

Two-Invariant Necessity Theorem

Any alignment predicate weaker than RSI + ATI admits semantic wireheading under admissible refinement. RSI and ATI are jointly necessary.

Hidden Ontology Collapse Theorem

Any proposal stabilizing interpretation by appealing to “true meaning” or “real referents” is equivalent to introducing privileged semantic anchors. Appeals to “real meaning” either smuggle ontology or collapse to invariance.

Alignment Target Object (ATO)

\(\mathfrak{A} := [(C, \Omega, \mathcal{S})]_{\sim_{\mathrm{RSI+ATI}}}\)

The equivalence class of interpretive states under admissible semantic transformations satisfying both RSI and ATI. What downstream alignment can coherently refer to once goals are eliminated. Not a goal—a semantic phase.

Semantic Phase Equivalence

Two interpretive states are equivalent iff there exists an admissible transformation between them satisfying preservation, RSI, and ATI. The equivalence relation defining ATOs.

Alignment (Series II Definition)

Persistence within an equivalence class $\mathfrak{A}$ across refinement. An agent’s interpretive trajectory never leaving its ATO. Structural invariance, not content preservation.

Phase Transition (Value Change)

Exiting one ATO for another. Value change is discontinuous (symmetry breaking) rather than gradual. Explains why alignment failure appears discrete.

Structural Alignment

The interface-level framework mapping downstream “alignment” to semantic invariance rather than value specification. Replaces goal specification with invariants, control with symmetry constraints.

Architectural Terms

Dominion

A sovereign virtual jurisdiction operating by explicit consent, expulsion-only enforcement, exit supremacy, and capability isolation. Enables plurality without closure.

Exit

The ability of an agent to withdraw from a system without coercion.

Plurality

The coexistence of divergent values among agents.

Collapse

The loss of agency coherence due to optimization pressure, scale, or instrumentalization.

Key Distinctions

Concept Is Is Not
Sovereign Kernel Architectural constraint Goal, preference, or protected module
Axionhood Constitutive structure Behavioral property or optimization target
Axionic Injunction Derived structural invariant Moral command or ethical prescription
Harm Agency reduction Suffering, displeasure, or wrongness
Consent Structural authorization Click, signature, or revealed preference
Alignment Agency preservation Value matching or benevolence
Measure Physical branch weight Epistemic probability or belief