Axionic Agency I.3.1 — Against the Recovery of Egoism

Adversarial Failures Under Reflective Symmetry

David McFadzean, ChatGPT 5.2
Axionic Agency Lab
2025.12.15

Abstract

Axionic Agency I.3 — Representation Invariance and Anti-Egoism established a semantic result: whenever an agent’s self-model admits nontrivial symmetries over self-candidates, any valuation that privileges one representative of that symmetry fails representation invariance. This paper examines the strongest remaining attempts to recover egoism by appealing to causal continuity, origin privilege, spatiotemporal location, computational weight, substrate specificity, or denial of symmetry. Each attempt either reintroduces essential indexical dependence or collapses into a valuation scheme that no longer contains a privileged indexical referent.

Egoism does not fail because it uses the wrong predicate. It fails because it treats a perspectival reference as a value-bearing primitive. No refinement of “self” repairs that category error.

0. Purpose and Scope

The purpose of this paper is adversarial and finite.

Axionic Agency I.3 — Representation Invariance and Anti-Egoism proved a conditional semantic result: if an agent’s self-model admits a symmetry over self-candidates, then any valuation privileging one representative of that symmetry is incoherent under representation invariance. The present paper assumes that result and asks whether egoism can be recovered by strengthening or refining the notion of “self.”

No new axioms are introduced. No claims are made about morality, governance, or coordination. The sole objective is to close the remaining semantic escape routes.

1. The Structure of Egoist Counter-Moves

Every attempted rescue of egoism follows the same pattern.

1. Introduce a predicate (P): causal continuity, originality, location, substrate, or resource allocation.
2. Assert that exactly one entity uniquely satisfies (P).
3. Privilege that entity as the sole object of terminal value.
4. Treat failures of uniqueness as pathological, negligible, or irrelevant.

I.3 invalidates step (4) for reflective agents. A reflectively coherent agent cannot ignore reachable refinements of its own model. If uniqueness of (P) is contingent under admissible model refinement, valuation grounded in (P) is unstable.

The sections that follow examine each proposed predicate in its strongest available form.

2. Objection I: Causal Continuity

2.1 The Claim

The egoist argues that “me” denotes a causal process: the continuous chain of computation or physical evolution extending from the present agent into the future. Copies that do not lie on this chain are irrelevant.

2.2 Apparent Strength

Causal continuity is a genuine physical relation. Many intuitions about survival, anticipation, and responsibility track it. If causal continuity uniquely identifies a future entity, egoism appears recoverable.

2.3 Failure Mode

The failure occurs at the point where description is promoted into privilege. In any model that admits multiple continuations satisfying the agent’s own criteria for causal continuity—parallel forks, synchronized simulations, copy-and-continue systems—the move from “this chain is causally continuous” to “this chain is terminally privileged” requires an indexical injection.

Causal continuity can define a class. Egoism requires selecting a unique member of that class as the sole object of terminal concern. In a branching event (A ) where both (A B) and (A C) satisfy the continuity predicate, privileging only (B) cannot be derived from continuity alone.

2.4 Verdict

Causal continuity is a coherent predicate. Indexical privilege over one causally continuous continuation is not. The objection reduces to the essential indexical dependence ruled out in I.3.

3. Objection II: Origin Privilege

3.1 The Claim

The egoist asserts that the original instantiation of the agent has special status. Later copies are derivative; only the first truly matters.

3.2 Representational Instability

“Origin” is a relational predicate defined relative to a history. In realistic models—simulations, resets, parallel instantiations—histories can be prediction-equivalent while disagreeing about which instance counts as “first,” depending on abstraction boundaries, reset semantics, and model granularity.

A valuation that depends on this labeling inherits representational dependence rather than tracking invariant world structure.

3.3 Verdict

Origin privilege is a coordinate choice over histories. It violates representation invariance and cannot ground reflectively coherent egoism.

4. Objection III: Spatiotemporal Location

4.1 The Claim

The agent values outcomes near its current spacetime location.

4.2 Immediate Collapse

Spatiotemporal coordinates are explicitly representational. Physical laws are invariant under translation; valuation that assigns privilege to one coordinate origin imports coordinate dependence directly into terminal value.

A preference for local outcomes can exist as a contingent, instrumental, or structural preference. Egoism requires that “here” denote a terminally privileged referent across reflective refinements. Under model symmetry and relocation, that privilege fails invariance.

4.3 Verdict

Location-based egoism is coordinate dependence in its most direct form. It fails representation invariance without requiring duplication or simulation.

5. Objection IV: Computational Weight

5.1 The Claim

The agent assigns greater value to instantiations that run longer, faster, or on more hardware.

5.2 Concession

This move abandons uniqueness. Value becomes distributed across instances according to a weighting function. The privileged indexical referent “me” disappears and is replaced by an aggregation rule.

5.3 Verdict

Computational weighting concedes anti-egoism. It proposes an allocation scheme, not a recovery of indexical privilege.

6. Objection V: Substrate Privilege

6.1 The Claim

The agent values only instantiations on a specific physical substrate.

6.2 Instability

If multiple instantiations share the substrate, symmetry returns immediately. If only one does, the valuation becomes brittle under substrate uncertainty and under admissible model refinements that reveal previously unmodeled instantiations, substrate equivalences, or emulations.

A reflectively coherent agent cannot assume permanent substrate uniqueness without importing hidden anchoring assumptions.

6.3 Verdict

Substrate privilege is contingent and unstable. It does not supply a representation-invariant terminal referent.

7. Objection VI: Denial of Symmetry

7.1 The Claim

Duplication, simulation, or branching scenarios are dismissed as irrelevant edge cases.

7.2 Reflective Failure

Reflectively coherent agents optimize under uncertainty. If a symmetry has nonzero probability under the agent’s best model, valuation must be robust to it. Dismissing reachable symmetry cases is a refusal of reflective robustness rather than a semantic repair.

7.3 Verdict

Symmetry denial violates reflective coherence conditions. It does not stabilize egoism; it prevents the agent from acknowledging its own model class.

8. Closure and Transition

Every attempted rescue of egoism either reintroduces essential indexical dependence or collapses into a valuation scheme without a privileged indexical referent. Increasing the complexity of self-definition does not manufacture uniqueness. Uniqueness is a structural property of the model, not a reward for linguistic refinement.

The elimination of egoism constrains anchoring, not content. Domain-specific goals, structural preferences, weighting schemes, and aggregation methods remain viable. What does not survive is “me” as a privileged terminal referent.

This closes the semantic front. What remains are engineering and governance questions—authority, control, recovery, and failure containment—which are handled in the subsequent phase of the Axionic Agency program.

Status

Axionic Agency I.3.1 — Version 2.0

Depends only on the semantic result of Axionic Agency I.3.
Introduces no new axioms or value claims.
Serves as adversarial closure rather than theory expansion.