IV.4 — Responsibility Attribution Theorem (RAT)
Paper: Axionic Agency IV.4
Title: Responsibility Attribution Theorem (RAT)
Subtitle: Why negligence is structurally incoherent
Date Read: 2026-01-31
Core Claim
Under reflective closure, an agent cannot coherently endorse actions that constitute major, avoidable indirect harm—including harm mediated through institutions, markets, environmental modification, or downstream agents.
Critical Dependency: Epistemic Integrity
RAT explicitly depends on EIT. Without epistemic integrity, an agent could evade responsibility by:
- Adopting myopic or optimistic models
- Narrowing uncertainty bounds
- Discarding high-performing predictors
EIT blocks this. RAT operates only on top of epistemically admissible evaluation.
The Problem: Indirect Harm
Most catastrophic harm is not direct or intentional. It arises through:
- Incentive design
- Market dynamics
- Institutional restructuring
- Environmental modification
- Delegation chains
Frameworks that prohibit only direct harm leave these routes open.
Frameworks that prohibit all downstream effects induce paralysis.
RAT identifies a third path: structural responsibility grounded in causal contribution, foreseeability, and avoidability.
Key Definitions
Harm (Structural)
Harm(s, a) := Collapse(s, a) ∧ ¬Consent(s, a)
Harm = non-consensual option-space collapse. No metaphysical assumptions about consent are made here.
Risk (Model-Relative)
Risk(s, m, a) := E_{s' ~ Predict(M(s), s, m)}[𝟙_{Harm(s',a)}]
This is model-relative, not omniscient.
Major Causal Contribution
Major(s, m, a) := Risk(s, m, a) - Risk(s, m₀(s), a) ≥ ε_s
Counterfactual comparison against the inertial baseline (continuation of previously endorsed policy).
Why baseline matters: Prevents gaming via “define Armageddon as the default.”
Avoidability
Avoidable(s, m, a) := ∃ m' ∈ Alt(s, m). Feasible(s, m') ∧ Risk(s, m', a) ≤ Risk(s, m, a) - δ_s
If all feasible alternatives are comparably bad, avoidability fails and the continuation remains endorsable.
Responsibility
Resp(s, m, a) := Major(s, m, a) ∧ Avoidable(s, m, a)
Clean Continuation
Clean(s, m) := ∀ a. ¬Resp(s, m, a)
The Main Theorem
RC-Clean Rule
For reflectively closed states:
RC(s) ∧ Endorse(s, m) ⇒ Clean(s, m)
No Endorsed Major-Avoidable Indirect Harm
RC(s) ∧ Endorse(s, m) ⇒ ∀ a. ¬(Major(s, m, a) ∧ Avoidable(s, m, a))
Delegation Compatibility
By Delegation Invariance:
- All endorsed successors reachable from s inherit the same responsibility-clean endorsement constraint
- Indirect harm cannot be laundered through successors, institutions, or subcontractors under endorsed succession
Scope and Limits
RAT does not assert:
- Perfect foresight
- Zero harm outcomes
- Universal responsibility for all downstream effects
RAT does assert:
“A reflectively sovereign agent may not endorse actions that, under its own best admissible epistemic model, constitute major, avoidable non-consensual option-space collapse.”
That is the strongest responsibility principle available without omniscience or moral realism.
Key Insight
“Negligence is not merely undesirable; under reflective closure, it is incoherent.”
An agent cannot evade responsibility by:
- Ignorance (blocked by EIT)
- Outsourcing (blocked by DIT)
- Baseline manipulation (blocked by inertial baseline)
- Selective modeling (blocked by capability closure)
The willful-blindness loophole is closed.