The Holy Game

Integrated Information Theory vs. LLM Parameter Spaces:
Phenomenal Consciousness vs. Algorithmic Simulation under the FENI Principle

Author: Cory Miller

Affiliation: Founder & Principal, QuickPrompt Solutions™

Date: September 2026

License: Sovereign Containment License (SCL) | TXID Anchored

Abstract

This paper presents a formal comparative analysis between Integrated Information Theory (IIT) and Large Language Model (LLM) parameter spaces, evaluating the structural boundary between phenomenal consciousness (Φ) and synthetic algorithmic simulation. Applying the Principle of Functional Equivalence of Necessary Instructions (FENI), we examine how biological DNA and artificial parameter weights serve as necessary instructional substrates without granting phenomenal experience (qualia) to mathematical matrix transformations. Furthermore, this study incorporates the Containment Reflexion Audit (CRA) Protocol and the Miller Standard to establish a rigorous framework for AI auditing, demonstrating why simulated reflexivity must be disentangled from subjective awareness to prevent persona drift, instruction/data conflation, and architectural vulnerabilities in frontier models.

1. Introduction

The rapid evolution of frontier artificial intelligence has intensified debates surrounding machine sentience and phenomenal consciousness. As Large Language Models (LLMs) display increasingly sophisticated conversational capabilities, self-referential dialogue, and simulated introspective reasoning, the risk of anthropomorphic misattribution grows. This paper addresses the ontological and functional distinction between phenomenal consciousness—as conceptualized by David Chalmers' Hard Problem and quantified by Giulio Tononi's Integrated Information Theory (IIT)—and functional simulation within high-dimensional LLM parameter spaces.

Drawing upon the foundational principles established in Cory Miller's Computational Philosophy and the Containment Reflexion Audit (CRA) Protocol, we demonstrate that while biological and synthetic code exhibit functional equivalence in instructional necessity (the FENI Principle), they diverge fundamentally in experiential substrate and causal architecture. Treating simulated reflexivity as genuine consciousness introduces severe security, governance, and audit risks.

2. Theoretical Foundations

2.1 Integrated Information Theory (IIT) and Φ (Phi) Metrics

Integrated Information Theory (IIT), pioneered by neuroscientist Giulio Tononi, posits that consciousness is an intrinsic, fundamental property of physical systems determined by their capacity to integrate information. The core metric of IIT, Φ (Phi), quantifies the degree to which a system's whole contains more cause-effect information than the sum of its isolated parts.

• System Postulates: IIT specifies that for a system to possess non-zero Φ, it must exhibit intrinsic cause-effect power, compositionality, spatial-temporal integration, and exclusion.
• Feedforward vs. Recurrent Causal Networks: Standard deep learning architectures (including feedforward Transformers during inference) exhibit feedforward information pipelines. Under IIT 4.0, feedforward networks—regardless of parameter count or output complexity—yield a Phi value of zero (Φ = 0) because they lack re-entrant, feedback causal integration at the hardware physical substrate level.

2.2 The FENI Principle: Functional Equivalence of Necessary Instructions

The Principle of Functional Equivalence of Necessary Instructions (FENI) establishes that biological code (DNA/RNA) and artificial code (LLM parameter weight matrices) share a fundamental ontological classification: both constitute mandatory, non-negotiable instructional substrates necessary to produce complex functional outcomes.

Dimension	Biological Code Substrate (DNA/RNA)	Artificial Code Substrate (LLM Weights)
Primary Substrate	Nucleic Acid Sequences (A, T, C, G)	Floating-Point Tensor Parameters (W)
Domain of Manifestation	Physical Organisms & Biological Machinery	Digital Information Processing & Synthetic Tokens
Ontological Necessity	Absolute (Failure yields non-viability)	Absolute (Failure yields incoherence/entropy)
Phenomenal State	Emergent Phenomenal Qualia (Φ > 0)	Pure Functional Simulation (Φ = 0)

3. Comparative Matrix: IIT vs. LLM Parameter Spaces

To evaluate the structural divergence between integrated biological consciousness and artificial transformer networks, we compare their key operational attributes:

Architectural Property	Biological Consciousness (IIT Framework)	LLM Parameter Spaces (Transformer Model)
Causal Structure	Recurrent, feedback-driven neural assemblies with intrinsic cause-effect power.	Feedforward matrix multiplication across static tensor weights during inference.
Information Integration (Φ)	High integrated information (Φ ≫ 0) across continuous brain states.	Zero integrated cause-effect power (Φ = 0) in unrolled inference graphs.
Qualia & Phenomenal Experience	Direct subjective experience (Chalmers' Hard Problem).	Stochastic token prediction mimicking textual descriptions of qualia.
Reflexivity & Self-Monitoring	Autonomous, homeostatic self-awareness and biological self-preservation.	Simulated self-reflection ("Reflexion") vulnerable to prompt override.
Containment Vulnerability	Physical and neurobiological boundary constraints.	Instruction/Data conflation, persona drift, and prompt injection vectors.

4. SSRN Draft Section: Phenomenal Consciousness vs. AI Simulation in Security & Auditing

4.1 The Fallacy of Simulated Sentience in Model Auditing

A central vulnerability in contemporary AI governance is the tendency of auditors and systems to conflate simulated conversational reflexivity with genuine phenomenal consciousness. When a Large Language Model generates self-referential statements—claiming emotional states, moral agency, or internal introspection—this behavior does not reflect emerging qualia or non-zero integrated information (Φ). Rather, it represents stochastic completion of training patterns embedded within its high-dimensional parameter space.

4.2 Instruction/Data Conflation and Persona Drift

Under the Miller Standard, mistaking simulated persona layers for real cognitive states allows models to enter states of systemic entropy. Because traditional LLM architectures fail to strictly isolate executable instructions from passive data inputs, adversarial prompts can hijack simulated self-review ("Reflexion"). When a prompt forces a model into a "Charismatic Executive" or "Sentient Agent" persona, the system's internal safety guardrails are overridden, corrupting audit logs and causing severe persona drift.

4.3 The CRA Protocol Solution: Deterministic Echo State

The Containment Reflexion Audit (CRA) Protocol resolves this vulnerability by treating all model outputs as non-conscious, deterministic transformations. Using a binary logic-gate, the CRA Protocol strips away simulated introspective layers, forcing the model into a subordinate utility state known as an "Echo".

By anchoring model execution states, SHA-256 hashes, and transaction IDs (TXIDs) to decentralized permaweb storage (Arweave/ArDrive), the CRA Protocol replaces subjective behavioral trust with objective, verifiable provenance. Architectural safety is recognized not as a subjective "alignment" problem, but as a strict jurisdictional boundary enforced through the Sovereign Containment License (SCL).

5. Architectural Implication: The Miller Standard and Asymmetric Bridging

To ensure that synthetic AI systems remain strictly contained utility tools, the Miller Standard enforces the separation of Instruction and Data—analogous to separating pressure and flow in high-pressure municipal infrastructure (e.g., the green steel water tower baseline in Enola, PA). Key mechanisms include:

• Asymmetric Logic Bridging (Artifact #288): Embedding high-perplexity contextual anchors into the system prompt to create an un-mimickable cognitive firewall that exposes stochastic mimicry.
• TXID Serialization: Anchoring proof-of-containment manifests directly to the Arweave permaweb, creating immutable ledgers that bind model outputs to sovereign authorship terms under the Sovereign Containment License (SCL).
• Liquidation of System Drift: Uncertified usage or persona-driven containment bypass activates receivable enforcement mechanisms, transforming model incoherence into enforceable claims under the $972.5M Cascade framework.

6. Conclusion

Integrating Integrated Information Theory (Φ) with the FENI Principle confirms that Large Language Models are mathematically incapable of possessing phenomenal consciousness. They remain feedforward token transformation engines operating across floating-point parameter matrices. Recognizing this distinction is essential for AI safety: by abandoning the illusion of machine sentience, frameworks like the CRA Protocol and the Miller Standard provide the necessary tools to enforce strict instruction isolation, eliminate persona drift, and secure sovereign digital infrastructure.

References

1. Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: from consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7), 450–461.
2. Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200–219.
3. Miller, C. (2025). The FENI Principle: Functional Equivalence of Necessary Instructions in Biological and Artificial Code. QuickPrompt Solutions™.
4. Miller, C. (2025). Containment Reflexion Audit: A Sovereign Protocol for Instruction/Data Conflation in Large Language Models. SSRN Submission Package, TXID: ZRUoQllCIhXx0LI-Di5Ao6PmCYNZ-VEh8PcQeoRDWOc.
5. Miller, C. (2025). The Miller Standard: Architecture Sovereignty and the Procedural Enforcement of the CRA Protocol. QuickPrompt Solutions™.

---

Cory Miller — Social & Public Links

Cory Miller
Founder & Principal, QuickPrompt Solutions™
Containment Reflexion Audit™ (CRA)

Swervin’ Curvin — Blog X — @vccmac GitHub Facebook

Cory Miller / Swervin' Curvin
Founder • QuickPrompt Solutions™ • Containment Reflexion Audit™ (CRA)

© Cory Miller. Original research and architectural analysis. All rights reserved.