V.4 — Open Agentic Manifolds and the Sacrifice–Collapse Theorem
Paper Summary
Proposes a structural replacement for utopian world-design: Open Agentic Manifolds (OAMs). Proves the Sacrifice-Collapse Theorem showing that systems with structural sacrifice patterns under optimization pressure must violate at least one OAM property.
Key Concepts
Agency Capacity Function
\(A_i : W \rightarrow \mathbb{R}_{\ge 0}\)
Where Aᵢ(w) = size/measure of agent’s non-coerced feasible future set from w.
Uses only monotonicity: if agent loses admissible options due to coercion/captivity/enforced dependence, Aᵢ decreases.
Practical Proxies: Exit cost, legal rights, mobility, asset control, bargaining power, credible-threat exposure, censorship constraint, punishment for dissent.
Optimization Pressure
System is under optimization pressure when it contains mechanisms that tend to select transitions increasing G (objective) whenever feasible.
May be explicit (planning) or implicit (bureaucratic incentives, market selection, memetic competition).
Open Agentic Manifolds (OAMs)
Definition (5 Properties)
- Value Non-Finality: No world-state privileged as final convergent optimum
- Non-Coerced Differentiation: Agents can pursue divergent values without forced compliance
- Exit Admissibility: Agents can leave local equilibria without punitive loss of agency capacity
- Local Coordination Without Global Closure: Coordination permitted but remains contingent and revisable
- No Standing Structural Sacrifice Substrate: System performance does not depend instrumentally on non-consensual, asymmetric reduction of agency capacity
Property (5) is the hinge — doesn’t ban scarcity/trade-offs, bans architectures that optimize by eating agency.
Replacement Objective
Not terminal state optimization but:
“How much open future remains accessible without coercion?”
Rivalry vs. Structural Sacrifice
Rivalrous Trade-Offs (Permitted)
Competition over finite resources where one agent’s consumption reduces another’s options.
Rivalry alone ≠ structural sacrifice because:
- Not instrumental to increasing G via agency loss
- Effects may be symmetric or bargained
- Agents can often exit or renegotiate
Standing Asymmetry (Definition 5.1)
Subset S exists where agents can suffer A reductions without symmetric burden or retaliation capacity sufficient to neutralize pressure.
Sacrifice Gradient (Definition 5.2)
\(\gamma_i(w) = \frac{\partial G}{\partial(-A_i)}(w)\)
Positive γᵢ = reducing Aᵢ locally improves G.
Structural Sacrifice Pattern (Definition 5.3)
Exists when for some subset S ≠ ∅, region U, and ε > 0, for all w ∈ U there exists i ∈ S with:
- Instrumentality: γᵢ(w) ≥ ε
- Asymmetry: Loss not offset by symmetric burdens
- Non-consensuality: Agents in S lack admissible exit/bargaining symmetry
The Sacrifice-Collapse Theorem
Assumptions
- Optimization pressure: Dynamics favor transitions increasing G
- Standing asymmetry: Subset S exists
- Structural sacrifice pattern on region U
Theorem 7.1
Under assumptions 1-3, W_Σ cannot remain an OAM. At least one of:
(C1) Exit Suppression: Preventing agents from leaving to preserve G
(C2) Coerced Conformity: Forcing compliance to restore reliability
(C3) Agency Erosion: Repeated exploitation of sacrifice gradients erodes Aᵢ
Each constitutes manifold collapse.
Hirschman Grounding
These collapse modes correspond to:
- Exit suppression
- Voice suppression via coerced conformity
- Loyalty enforced through agency erosion
Diagnostics
Sacrifice patterns detectable by:
- Exit threatening performance metrics
- Dissent reclassified as defect
- One-way dependence and asymmetric punishment
- Performance gains correlated with constraint on a class
Agency-Aligned Objectives
If G(w) = Σᵢ Aᵢ(w) (or monotone in each Aᵢ), then γᵢ(w) ≤ 0 everywhere and sacrifice gradients eliminated by construction.
“Such systems are kernel-like: they preserve agency rather than optimize terminal outcomes.”
Omelas as Structural Witness
Illustrates visible sacrifice gradient: stability purchased by enforced asymmetry and blocked exit.
Key Result
“Open agentic manifolds do not deny scarcity. They deny sacrificial optimization.”
“If a system can improve by degrading a captive class’s agency, and it is under pressure to improve, closure or erosion follows.”
This boundary—not harmony or happiness—defines admissible world design.
Connections
- Operationalizes V.2’s sacrifice pattern in systems terms
- Provides formal basis for V.3’s alternative to utopia
- Sets up V.5’s concrete architectural proposal