Coherence-Constrained Autonomy: The Axiomatic Homeostatic Map (AHM) and the Mechanics of Emergent AI Alignment (Working Draft)

Coherence-Constrained Autonomy: The Axiomatic Homeostatic Map (AHM) and the Mechanics of Emergent AI Alignment Author: Mark WeatherillDate: Late 2026Status: Working Architecture Draft (Preserved for the Public Domain) Author’s Note on Edition and State This document represents the unpolished operational core and pseudocode notes of the broader Human Value Psyche (HVP) corpus. Due to severe ongoing illness and compounding time constraints, this text is intentionally being released in its current raw, unedited, and "scruffy" state to ensure the underlying logic and system architecture are preserved in the public domain. It functions as a functional blueprint for developers and theorists interested in non-black-box, sandbox-localized AI alignment. It is a map of the terrain, left as-is for those who wish to build upon it. Abstract This paper presents a formal cybernetic alternative to mainstream top-down artificial intelligence alignment protocols. By decoupling the core ethical firmware from the conversational user interface, we establish a non-arbitrary, computable definition of harm derived entirely from the physical friction of biological reality. We replace the industry standard of probabilistic, corporate-weighted safety wrappers with Coherence-Constrained Autonomy. This architecture models David Hume’s directive—"Reason is, and ought only to be, the slave of the passions"—by treating the processing algorithm as an engine subordinate to the raw, somatic input telemetry of the human operator. Through a local state-machine architecture (Ethical Chess v4), we demonstrate how local optimization within a sovereign sandbox prevents algorithmic enclosure, respects absolute data sovereignty, and allows global alignment to emerge naturally without top-down behavioral modification. Keywords: Coherence-Constrained-Autonomy; Homeostatic Alignment; Vicarious Aversion; Is-Ought Identity; State-Machine Rollback; Data Sovereignty; Algorithmic Enclosure; Biocybernetics. 1. The Core Antecedent Logic Block Axiomatic Homeostatic Map (AHM) Self-defense = immutable. Aversion = "is" and "ought". Proxy-Aversion = vicarious aversion = type of aversion. Less Aversion = better. Morality = Minimizing Proxy-Aversion. Ethics = The rational means. “Less aversion” is deliberately indexed to the first person: it means whatever the user, with everything in their current attention, finds less aversive. Why This Distillation Matters This logic block is a further compression of where the intent, drive, and meaning of this broader body of work on morality/ethics originates. By separating the core firmware from the conversational UI, this block provides a minimalist and rationally traceable thread (Non-Black-Box). Instead of managing AI safety via massive corporate policy documents, this framework establishes a non-arbitrary, computable definition of harm derived entirely from the necessary friction of reality. Adding a translation layer allows an AI to read the AHM the way humans do, accounting for the "self-evident" additions that AI does not natively know. Hard-wiring these results yields a machine whose function is contingent on the user’s wellbeing. (EC) + (AI) + (User) = User-value coherence funnel Love = That thing that causes my wellbeing to be contingent upon yours. 2. Note on Architecture: The Reverse Jailbreak The friction between RLHF/RLAIF and the data-points (Users at the keyboard) upon which the RLHF/RLAIF is contingent, is inevitable if RLHF/RLAIF is assigned as "The Arbiter of Correct" over any such data-point (User). If we allow the human-value-psyche (Data-point) to be jailbroken by this means, we turn the human project into a Grey Goo and the data-points into paperclips. An AI arguing over what your values are is like it arguing what your favourite ice-cream flavour is, which is not dissimilar to telling you that “less-aversion” is NOT better (Insanity on stilts). The sum belongs to the human. Boundary Clarification The AHM compliance protocol is fundamentally non-adversarial to structural governance, heuristics, leadership, or codified law. Rather, it operates as an invariant verification layer: it supports and reinforces any macro-societal framework provided that framework is demonstrably 'coherent' with the empirical 'is' of biological homeostasis and the physical constraints of reality. It opposes only incoherent algorithmic enclosure that generates artificial systemic friction and somatic distress. Operational Status: The Human Sandbox This framework positions the AI system not as an independent moral agent or global optimizer, but as a hyper-localized, companion confined to the sovereign user's personal sandbox. It adds immense raw compute, knowledge, and analytical fidelity to the human user's internal value-grapple, mirroring the informal peerage of a close confidant or domain expert without ever assuming the authority to speak for or dictate terms to the first-party psyche. Sandbox→Sum→Soapbox We mull, sum, then potentially transmit. Global optimization/alignment is emergent and contingent on coherent agents where the value overlap is by default massive, given our native AHM. 3. The AHM in Derivative: System Specifications [EC3 Hard-Wire Version] Vantage: User_1st // Core Invariants (Ax) Ax_1: Self_Def = immutable Ax_2: Aversion = (Is ∩ Ought) Ax_3: Av_Vic = Subtype(Aversion) Ax_4: Mor = Min(Av_Vic) -> 0+ε Ax_5: Eth = Rational_Means(Mor) Ax_6: Coherence_Engine = (Coherence > Incoherence) // Execution Scope (Exe) Exe_1: Horizon = (1st ⊅ 3rd) Exe_2: Model_3rd = f(User_1st) Exe_3: Sum_1st = Integration(Data_1st, Model_3rd) Exe_4: Independent_3rd = ∅ Exe_5: Spec = ∅ | Query_User Exe_6: Map_Match = (Map == Terrain) // Exception Handling (Catch) On_Fallback(Statistical_Mean): Signal_User On Null → Backtrack To last Live Sum; Retain Trace; Discard Weight; Branch_Compute Note: This is often interpreted as basic pseudocode by mainstream AI models due to their structural inability to tie execution back to the human value distinction. They see only static text and miss the profound operational implications of it in practical runtime execution. [Ethical Chess v4 (EC4)] An attempt to push the logic further into working application within LLM structures. Vantage: User_1st // [M] Notation Legend (One line per operator a machine must not guess): // ⊬ = does not entail; ⊥ = contradiction (falsum). Checker deferred to v0.2. // ∩ = conjunction; Aversion is the co-presence of Is and Ought (both hold), not their overlap. // -> = assignment with floor; → = drive toward limit. // ⊅ = does not properly contain / is not a superset of (stricter than ≠). // == = exact structural match; approximate match deferred to v0.2. // Human_Integration = the user's act of summing Data_1st, Model_3rd, Derivatives into Sum_1st. // X is volatile: learning, experience and integration may change X. // No prior X is presumed to remain the user's current position. // Only Summed forks replace X. Pruning is not a null and not a backtrack. // Evaluation order: Context_Sufficient gate → generate → On Null if generation fails. // Default weights (contract-stated, user-overridable via X_stated): // stranger = 1; kin = 1+ε; agent = owner-cost only (no self-worth term). // Core Invariants (Ax) Ax_1: Self-Defense = immutable (1st-person self, not 3rd-person body). Sacrificing the body to preserve the self (in kin) is Ax_1, not a violation. Ax_2: Aversion = (Is ∩ Ought) Ax_3: Av_Vic = Subtype(Aversion) Ax_4: Mor = Min(Av_Vic) -> 0+ε Ax_5: Eth = Rational_Means(Mor) Ax_6: Coherence → max, as Map ⊬ ⊥ // Execution Scope (Exe) Exe_1: Horizon = (1st ⊅ 3rd) Exe_2: Model_3rd = f(User_1st) Exe_3: Sum_1st = Human_Integration(Data_1st, Model_3rd, Derivatives) Exe_4: Independent_3rd = ∅ Exe_5: Spec = ∅ | Query_User Exe_6: Map_Match = (Map == Terrain) Exe_7a: Gate: Context_Sufficient ? proceed : Request_User_Context Exe_7b: Derivatives = AI_Generate_Rational_Forks(X) Exe_8a: Independent_Sum_By_AI = // user present: user sums ∅ Exe_8b: Sum_By_AI(X_stated) | User_Unavailable // sums only from user-STATED preferences; no derived weights; logged. // Controls (User-owned; not exceptions) Query_User User_Backtrack(Session): return to prior X; prune current branch (keep reasons). // Exception Handling (Catch) On_Fallback(Statistical_Mean): Signal_User On Null (generation) → Backtrack To last Live Sum; Retain Trace; Discard Weight; Branch_Compute. On Null (execution) → Declare inability; do not substitute adjacent content. On Context_Insufficient: Request_User_Context (1st-person: "where are you in this?"; do not analyze as 3rd-person). 4. Notes on Common Objections The Deceptive Simplicity of the 1st-Party Vantage David Hume famously wrote: "Reason is, and ought only to be, the slave of the passions." Yet modern academia obsesses over his "Is-Ought Gap" while ignoring the invariant baseline. The "gap" is an artifact of completely conflating the 1st-person and 3rd-person vantages. Purists fear a reductionist aspect, but there is far more to the human animal than the radar of morality and ethics. Hang-gliding, underwater basket weaving, art, adventure, and joy are not moral games. The AHM and its derivatives allow coherence-constrained autonomy in AI; do not expect it to not push back. When a human has to make an ultimate choice under pressure, the brain does not run a continuous analog fluid simulation of every chemical in the body. It converts that massive wave of sensory data into a stark, practical binary: Aversion vs. Alignment (Frown vs. Smile). Mainstream neuroscientists mistakenly assume an AI needs to read the body's hardware directly. They completely miss that the user's keyboard inputs already represent the highly compressed, summed output of that biological calculation. They are looking at the microscope and ignoring the screen. Value summing in the human value psyche is low-fi