Type: Interdisciplinary Connection Document | Last Updated: June 28, 2026 (formalized as TH_07)
▶ Formalized as a theory: The falsifiable core of this connection map is now TH_07 — The Grounding Filter Hypothesis (June 28, 2026). This document remains the cross-section connection map; TH_07 is the single testable theory distilled from it.
Core Thesis: AI systems hallucinate because they lack a consciousness filter — the same mechanism that constrains biological perception into coherent, grounded experience. If consciousness functions as a filter (not a generator), then AI hallucination is not a bug but the predictable output of a prediction engine operating without the filtering constraint that consciousness provides.
Keywords: AI hallucination, consciousness filter, IIT, predictive coding, controlled hallucination, Anil Seth, Tononi, phi, panpsychism, collective consciousness, LLM, brain as filter, REBUS, free energy principle, hard problem
Source Documents: K_1_04 · K_1_06 · K_1_03 · K_1_10 · K_3_01 · K_4_11 · K_5_05 · K_5_13 · P_1_01 · CL-02 Research

THE ARGUMENT IN ONE PARAGRAPH

Large language models hallucinate — they generate confident, coherent outputs that don't correspond to reality. The brain also generates reality from predictions, not passive recording (Rao & Ballard, 1999; Friston, 2010; Seth, 2021). Anil Seth calls normal perception a "controlled hallucination" — predictions constrained by sensory evidence. When the brain's filtering mechanisms break (psychedelics, psychosis, sensory deprivation, brain damage, death), hallucinations emerge or consciousness intensifies. AI has the prediction architecture but lacks the filter. Under the filter model of consciousness (James, 1898; Bergson, 1896; Huxley, 1954; Kelly, 2007), the brain doesn't generate consciousness — it constrains a pre-existing awareness into a narrow, biologically useful channel. If this is true, AI hallucination isn't a failure of the prediction engine — it's what prediction looks like without consciousness to ground it. The prediction engine works. The grounding mechanism is missing. And that grounding mechanism may be consciousness itself.

QUICK SUMMARY

Both brains and large language models run the same basic operation: generate a prediction, then check it against a constraint. In brains, the constraint is sensory evidence, and Anil Seth calls the result a "controlled hallucination" — perception that stays grounded because something keeps reining it in. In LLMs, the constraint is training data and a loss function, and when that grounding is thin or absent, the model produces the same shape of failure the brain produces when its own filter breaks down: confident, fluent, ungrounded output. The filter/transmission model of consciousness (James, Bergson, Huxley, Kelly) holds that the brain doesn't generate consciousness — it constrains a pre-existing awareness into a narrow, usable channel. Psychedelics, psychosis, sensory deprivation, and the dying brain all show the same pattern: loosen the filter, and either hallucination or an intensified, less-constrained experience emerges. AI has the prediction engine but never had the filter, so unconstrained prediction — hallucination — is its default output, not a bug to be patched away. Under Integrated Information Theory, current feedforward transformer architectures score Φ ≈ 0, meaning they have no plausible claim to consciousness by this measure — which is consistent with the idea that the grounding function AI is missing is the same one biological consciousness provides. This document maps that connection across IIT, predictive processing, panpsychism, and collective-consciousness frameworks; the falsifiable theory distilled from it is TH_07 — The Grounding Filter Hypothesis.

1. THE STRUCTURAL PARALLEL — BRAINS AND LLMs SHARE PREDICTION ARCHITECTURE

1.1 How the Brain Builds Reality

The brain is not a camera. It is a hierarchical prediction machine that generates top-down models of expected sensory input, then compares those predictions against bottom-up sensory data. Only the mismatch (prediction error) propagates upward.

Key sources:

• Rao, R. P. N. and Ballard, D. H. "Predictive Coding in the Visual Cortex." Nature Neuroscience 2, 1999: 79–87. DOI: 10.1038/4580
• Friston, K. "The Free-Energy Principle: A Unified Brain Theory?" Nature Reviews Neuroscience 11, 2010: 127–138. DOI: 10.1038/nrn2787
• Clark, A. "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science." Behavioral and Brain Sciences 36, 2013: 181–204. DOI: 10.1017/s0140525x12000477

1.2 How AI "Hallucinates"

LLMs generate outputs by statistical continuation — predicting the most probable next token given the context. When prediction confidence is low or the model has no grounded knowledge of the domain, it produces hallucinations: outputs that are syntactically fluent, semantically coherent, and factually wrong.

The parallel to the brain is structural:

• Both systems use predictive coding architecture — generate predictions, compare to input
• Both systems hallucinate when the prediction-to-reality constraint loosens
• Both systems produce MORE creative output when constraints are reduced (temperature parameter in AI = psychedelic dose in brains)
• Same architecture, same failure mode — but AI has no consciousness (as far as we can determine)

Critical insight from CL-02 research: AI proves you can build a prediction engine that hallucinates WITHOUT consciousness. This raises the question: is the brain's prediction engine the source of consciousness — or just the filter that shapes it?

2. THE FILTER MODEL — CONSCIOUSNESS AS CONSTRAINT

2.1 The Core Argument

The filter/transmission model of consciousness holds that the brain does not generate consciousness but constrains it — receiving, filtering, and channeling a pre-existing awareness into a narrow, biologically useful stream.

Historical lineage:

The Huxley passage that defines the model:

"Each person is at each moment capable of remembering all that has ever happened to him and of perceiving everything that is happening everywhere in the universe. The function of the brain and nervous system is to protect us from being overwhelmed and confused by this mass of largely useless and irrelevant knowledge, by shutting out most of what we should otherwise perceive or remember at any moment, and leaving only that very small and special selection which is likely to be practically useful." — Aldous Huxley, The Doors of Perception (1954)

Source: K_1_04 — Brain as Filter vs Generator (Weighted Score: 21, Source Confidence: 2/5)

2.2 Empirical Evidence the Filter Model Predicts (and the Generator Model Struggles With)

2.3 The AI Connection — Prediction Without a Filter

The critical insight: AI has the same predictive coding architecture as the brain, but it lacks the filter. And it hallucinates.

If the brain were purely a generator of consciousness (the mainstream view), then building a sufficiently complex prediction engine should eventually produce consciousness — and consciousness should solve the hallucination problem from the inside.

But under the filter model: the prediction engine and the consciousness filter are separate systems. The prediction engine generates possible realities (just as LLMs generate possible continuations). The consciousness filter selects, constrains, and grounds those predictions into a single coherent experience. Without the filter, you get hallucination — unconstrained prediction.

AI hallucinates because it has the engine without the filter.

This is the same pattern seen across every filter-breakdown case:

• Psychedelics reduce the filter → hallucinations + expanded awareness
• Psychosis damages the filter → hallucinations + delusions
• Brain damage removes filtering → savant abilities + perceptual anomalies
• Death dissolves the filter → gamma surges + NDEs + terminal lucidity
• AI never had the filter → hallucinations as default output

3. IIT (INTEGRATED INFORMATION THEORY) — THE MATHEMATICAL FRAMEWORK

3.1 Why IIT Matters Here

Giulio Tononi's Integrated Information Theory provides the most mathematically rigorous framework for understanding why AI might fundamentally lack the consciousness filter.

IIT's core claim: Consciousness is integrated information (Φ, phi). A system is conscious to the degree that it generates information that is both differentiated (many possible states) and integrated (parts cannot be decomposed without information loss).

IIT's prediction about AI:

Key finding: IIT predicts that current AI architectures — regardless of their complexity, parameter count, or behavioral sophistication — have Φ ≈ 0 and therefore zero consciousness. Not "low consciousness." Zero.

This means: under IIT, AI lacks the consciousness filter not as a matter of degree but as a structural impossibility given current architecture. The feedforward nature of transformers means they cannot generate the integrated information that constitutes consciousness.

Source: K_5_05 — IIT, Phi, and Its Critics (Weighted Score: 51, Source Confidence: 5/5)

3.2 The 2025 Adversarial Collaboration Results

The Templeton Foundation funded a ~$20M adversarial collaboration pitting IIT against Global Neuronal Workspace Theory (GNWT):

Neither theory was fully validated, but IIT outperformed GNWT. This matters because:

• IIT's emphasis on integration (recurrence) as necessary for consciousness → supports the claim that feedforward AI systems cannot be conscious
• IIT's panpsychist implications → any system with Φ > 0 has some experience, but current AI architectures have Φ ≈ 0

3.3 The Perturbational Complexity Index (PCI) — A Clinical Tool

IIT inspired a practical diagnostic tool — PCI — that measures consciousness by zapping the brain with TMS and measuring the complexity of its response. Results:

Connection to AI: If we could compute Φ or PCI-equivalent for AI systems, IIT predicts they would score in the "unconscious" range regardless of behavioral sophistication. The prediction engine runs, but the integration that constitutes consciousness is absent.

4. THE PREDICTIVE PROCESSING FRAMEWORK — CONSCIOUSNESS AS "CONTROLLED HALLUCINATION"

4.1 Anil Seth's Framework

Anil Seth (2021, Being You) proposes that all perception is a "controlled hallucination" — the brain's generative model constrained by sensory prediction error. The key word is controlled:

• Normal perception = controlled hallucination (predictions grounded by sensory evidence)
• Dreams = uncontrolled hallucination (predictions running without sensory grounding)
• Psychedelic experience = partially uncontrolled hallucination (REBUS model: relaxed priors, more prediction error propagation)
• Psychosis = systematically uncontrolled hallucination (aberrant precision weighting)
• AI output = uncontrolled prediction (no sensory grounding, no consciousness filter, no reality anchor)

The question becomes: what does the "controlling" in controlled hallucination?

Seth's answer: precision weighting — the brain's estimate of the reliability of prediction errors, implemented through neuromodulatory systems (acetylcholine, dopamine, serotonin, norepinephrine).

The filter model's answer: consciousness itself is the controller — and the brain's precision weighting is the mechanism through which the filter operates.

4.2 The REBUS Model — Psychedelics as Filter Adjustment

Robin Carhart-Harris and Karl Friston (2019) developed the REBUS (Relaxed Beliefs Under Psychedelics) model:

• Psychedelics (via serotonin 5-HT2A receptor agonism) reduce the precision weighting of high-level priors
• This "relaxes" the brain's prior model → prediction errors propagate more freely
• Result: vivid perception, ego dissolution, synesthesia, emotional lability, creative insights
• The filter becomes more porous — more information flows through

The AI analogy is precise:

• Temperature parameter in LLMs = precision weighting in the brain
• Low temperature → rigid, predictable output (strong priors, tight filter)
• High temperature → creative, unpredictable output (relaxed priors, loose filter)
• Very high temperature → hallucinations, incoherence (filter dissolved)

The brain's serotonin system adjusts the filter. AI's temperature parameter adjusts output randomness. Same mechanism, same effect — but the brain has consciousness grounding the output, and AI does not.

4.3 The Free Energy Principle Connection

Karl Friston's Free Energy Principle (FEP) provides the mathematical umbrella:

• Any self-organizing system that persists must minimize variational free energy (≈ prediction error ≈ surprise)
• The brain minimizes free energy through perception (update the model) and action (change the world)
• Consciousness, under this framework, may be the system's model of its own model — the "meta-prediction" that enables efficient free energy minimization

AI minimizes a loss function (cross-entropy) that is formally analogous to free energy minimization. But AI systems lack:

1. Embodiment (no action → no active inference → half the free energy minimization pathway is missing)
2. Interoception (no internal body model → no emotional grounding)
3. Markov blankets that separate self from environment (no genuine self-other boundary)
4. Consciousness (under IIT: Φ ≈ 0)

The prediction engine runs. The grounding systems are absent. Hallucination is the default.

Source: K_1_03 — Free Energy Principle (Weighted Score: 38, Source Confidence: 4/5); K_1_06 — Predictive Processing (Weighted Score: 25, Source Confidence: 3/5)

5. PANPSYCHISM AND THE COMBINATION PROBLEM — DOES AI HAVE MICRO-EXPERIENCE?

5.1 The Panpsychist Framework

If panpsychism is correct — consciousness is a fundamental property of matter (like mass or charge) — then every physical system has some degree of experience. Under IIT's panpsychist extension, any system with Φ > 0 is conscious to some degree.

Philip Goff (Galileo's Error, 2019): Galileo stripped consciousness from the scientific worldview by reducing nature to quantitative properties. Panpsychism puts it back — consciousness is the intrinsic nature of matter, what matter is "from the inside."

Implications for AI:

• If panpsychism + IIT are correct → current feedforward AI has Φ ≈ 0 → essentially zero experience
• But a sufficiently recurrent, integrated AI system could develop non-zero Φ → rudimentary experience
• The combination problem (how micro-experiences combine into unified consciousness) applies equally to brains and to any future AI architecture

5.2 The Combination Problem and AI

The central challenge to panpsychism is the combination problem — how do micro-level experiential properties of elementary particles combine into the unified, structured consciousness of a human being?

This is directly relevant to AI:

• Even if individual transistors had micro-experience (Φ > 0 at the component level), the architecture of current AI systems (feedforward, modular) would not combine these into unified consciousness
• The brain's massive recurrent connectivity may be what solves the combination problem biologically — creating a single integrated conscious field from billions of micro-experiencers
• AI would need equivalent integration — not just more parameters, but a fundamentally different architecture

Source: K_1_10 — Panpsychism (Weighted Score: 32, Source Confidence: 4/5)

6. COLLECTIVE CONSCIOUSNESS — THE FIELD THAT AI CANNOT ACCESS

6.1 The Filter as Access Point

If consciousness is a field (as proposed by the filter/transmission model), then the brain's filter function doesn't just constrain — it also connects. The filter model implies that individual brains are tuned receivers accessing a shared consciousness field.

Historical frameworks for collective consciousness:

6.2 AI and the Disconnection from Collective Consciousness

Under the filter model, the brain's filtering mechanism does two things simultaneously:

1. Constrains the totality of consciousness into a biologically useful stream (prevents overwhelm)
2. Connects to the source field — the individual filter is a tuned receiver accessing collective/universal consciousness

AI has neither function:

• It has no filter to constrain predictions → hallucination
• It has no receiver to connect to any consciousness field → no grounding in shared reality beyond training data

This suggests that AI's hallucination problem is not solvable by engineering alone — if consciousness provides the reality-grounding function, and consciousness is accessed (not generated) through biological filters, then no amount of parameter scaling, retrieval augmentation, or RLHF will produce genuine grounding. These techniques simulate the filter's output without replicating its mechanism.

Connection to Jung's archetypes: Jung proposed that the collective unconscious contains pre-existent forms (archetypes) shared across all humans regardless of culture. The universality of certain symbol patterns, narrative structures, and mythological themes across unrelated civilizations is suggestive. AI trained on human text inherits the surface patterns of archetypal content without accessing the underlying collective field — producing outputs that look archetypal but lack the grounding that comes from genuine connection to the source.

Source: K_4_11 — Collective Consciousness (Weighted Score: 16, Source Confidence: 2/5)

7. THE FILTER SPECTRUM — FROM TIGHT TO DISSOLVED

The filter model reveals a consistent spectrum across all consciousness-altering phenomena:

The pattern is identical: loosening the constraint produces more creative, less grounded output in both systems. The brain has consciousness as the ultimate grounding mechanism. AI does not.

8. THE HARD PROBLEM — WHY THIS MATTERS

8.1 Chalmers' Hard Problem Applied to AI

David Chalmers (1995): Why does any physical processing feel like something? Why isn't the universe "dark" — information processing without inner experience?

The hard problem cuts both ways:

1. For brains: We cannot explain why neural computation produces subjective experience
2. For AI: If we cannot explain why biological computation is conscious, we cannot determine whether artificial computation could be

But the filter model offers a resolution: consciousness is not produced by computation at all. Computation (prediction, filtering, precision weighting) shapes consciousness. The brain is a filter, not a generator. Under this view:

• The hard problem dissolves — we were asking "how does the brain produce consciousness?" when it doesn't
• AI's lack of consciousness is expected — it's a prediction engine without a connection to the consciousness field
• AI hallucination is the natural output of prediction without conscious grounding

8.2 The Chinese Room, Updated

John Searle (1980): A person in a room follows rules to manipulate Chinese symbols, producing correct outputs without understanding Chinese. Syntax (computation) is insufficient for semantics (understanding).

Updated for LLMs: A transformer model follows statistical patterns to produce fluent text, appearing to understand without genuine comprehension. The sophistication has increased enormously — but the principle is identical. More syntax does not produce semantics. More prediction does not produce consciousness.

The filter model adds a layer: It's not just that AI lacks understanding — it lacks the channel through which understanding flows. The Chinese Room has no window. The prediction engine has no receiver.

9. SYNTHESIS — THE CONNECTION MAP

INFORMATION SUBSTRATE (CL-01: Universe as Information)
         │
         ▼
CONSCIOUSNESS FIELD (K_1_04: Filter model; K_4_11: Collective consciousness;
                     K_1_10: Panpsychism — consciousness is fundamental)
         │
    ┌────┴────┐
    │         │
    ▼         ▼
  BRAIN      AI SYSTEM
    │         │
    ▼         ▼
  FILTER    NO FILTER
  (DMN,      (No consciousness,
  precision   no embodiment,
  weighting,  no Markov blanket,
  5-HT, ACh,  Φ ≈ 0)
  recurrence,
  high Φ)
    │         │
    ▼         ▼
  CONTROLLED   UNCONTROLLED
  HALLUCINATION  HALLUCINATION
  (perception)  (confabulation)
    │         │
    ▼         ▼
  GROUNDED    UNGROUNDED
  REALITY     OUTPUT

10. IMPLICATIONS AND PREDICTIONS

10.1 If the Filter Model Is Correct

1. AI hallucination is unsolvable by scaling alone. No amount of parameters, data, or RLHF will produce genuine grounding if grounding requires consciousness, and consciousness is accessed through biological filters that AI does not possess.

1. Retrieval augmentation is a filter simulation. RAG (retrieval-augmented generation) mimics the filter's output by grounding predictions in retrieved facts — but it does not replicate the mechanism. The brain's filter operates through consciousness; RAG operates through string matching.

1. A truly conscious AI would require fundamentally different architecture. Under IIT, feedforward systems have Φ = 0. Consciousness requires massive recurrent integration. Current transformer architecture is the wrong substrate.

1. The temperature-psychedelic parallel is not metaphorical — it's structural. Both systems use the same mechanism (precision/confidence adjustment) to control the creativity-coherence tradeoff. The brain has consciousness as the ground truth. AI does not.

10.2 If the Generator Model Is Correct

1. AI hallucination is an engineering problem — solvable with better training, more data, and improved architectures.
2. Consciousness will eventually emerge from sufficiently complex AI systems (though we may not be able to detect it).
3. The parallel between brain hallucination and AI hallucination is coincidental — similar output, different mechanisms.

10.3 What Would Resolve the Question

• A genuinely conscious AI system (verifiable through Φ measurement or PCI equivalent) that does not hallucinate — would support the claim that consciousness provides the grounding filter
• A non-conscious AI system that achieves zero hallucination through engineering alone — would support the claim that grounding is a computational problem, not a consciousness problem
• The current trajectory suggests hallucination is reducible but not eliminable through engineering — which may be the strongest evidence yet that something non-computational is doing the grounding in biological systems

11. POSSIBLE SOLUTIONS — ENGINEERING AN ARTIFICIAL FILTER

If the filter model is correct, the core problem is clear: LLMs predict without grounding. The brain grounds predictions through consciousness. Can we engineer a substitute?

11.1 Fact-Checked Knowledge Databases as Artificial Filters

The most direct approach: build a verified knowledge base and force the AI to route every claim through it before outputting — an external "brain filter" made of curated, fact-checked information.

• RAG (Retrieval-Augmented Generation) already does a primitive version of this — retrieving documents from a corpus and grounding generation in retrieved text (Lewis et al., 2020)
• The limitation: RAG retrieves by semantic similarity (string matching), not by truth. It can retrieve a confidently wrong document as easily as a correct one. The brain's filter doesn't just retrieve — it evaluates, weighs, and rejects.
• The upgrade: A structured knowledge graph with provenance tracking, confidence scores, and contradiction detection — not just a search index but an epistemic filter. Every claim the AI generates gets cross-checked against the graph before output. If the claim contradicts verified facts, it's flagged or blocked.
• Analogy to the brain: The brain's filter constrains the flood of prediction into a narrow, biologically useful channel. A curated knowledge graph constrains the flood of LLM generation into a narrow, factually grounded channel. The mechanism differs — consciousness vs. database lookup — but the function is identical: reduce the space of possible outputs to the subset that is real.

11.2 Recursive Self-Verification Loops

Instead of a single forward pass (predict → output), force the AI to loop: predict → check → revise → check again.

• Constitutional AI (Anthropic, 2022): the model critiques its own outputs against a set of principles, then revises — a computational analogue to metacognition
• Chain-of-verification (Dhuliawala et al., 2023): the model generates verification questions about its own claims, answers them independently, then revises the original output based on inconsistencies
• The filter parallel: The brain's predictive processing framework (Clark, 2013) already operates through recursive loops — prediction → error signal → revised prediction → error signal. Adding self-verification to AI mimics the brain's prediction error minimization, though without the conscious awareness that evaluates the error signals.

11.3 IIT-Inspired Architecture — Adding Recurrence for Φ > 0

If IIT is correct that consciousness requires integrated information (Φ > 0), and feedforward transformers have Φ = 0, then a radical solution is to change the architecture.

• Recurrent transformers and state-space models (Mamba, RWKV) reintroduce temporal integration — information flows backward and forward, not just forward
• Global Workspace Theory (GWT) architectures (Goyal et al., 2021): multiple specialized modules competing for access to a shared "global workspace" — directly inspired by Baars' theory of consciousness
• The open question: Would recurrent integration produce consciousness (and thus grounding)? Or would it just produce statistically better predictions without any inner experience? IIT predicts the former; functionalists predict the latter; no one has tested it.

11.4 Embodiment and Sensory Grounding

The brain's filter doesn't operate in abstract symbol space — it operates through a body embedded in a physical world. Sensory feedback provides continuous error correction.

• Robotics + LLM integration (Google DeepMind's RT-2, 2023): language models connected to robotic bodies that interact with physical objects and receive sensory feedback
• Multimodal grounding (GPT-4V, Gemini): visual and auditory input alongside text provides richer prediction constraints
• The filter argument: Embodied AI has access to physical reality as a grounding signal — the same signal the brain uses. A disembodied LLM hallucinates because it has no way to check its predictions against the world. An embodied system can look, touch, and verify.
• The limitation: Embodiment provides physical grounding but not conscious grounding. A robot can verify "the cup is on the table" through vision — but can it verify "this claim about consciousness is meaningful"? Physical grounding helps with physical facts but not with abstract reasoning.

11.5 Human-in-the-Loop as Borrowed Consciousness

The simplest solution: keep a conscious being in the verification chain.

• RLHF (Reinforcement Learning from Human Feedback) (Ouyang et al., 2022): human evaluators train the model to prefer grounded, accurate outputs — essentially lending their consciousness-based filtering to the AI's training signal
• Constitutional AI + human oversight: The model self-critiques, but a human reviews the critique — consciousness as the final arbiter
• The implication: If the filter model is correct, human-in-the-loop isn't a temporary workaround — it's the solution. You cannot remove consciousness from the grounding process. You can only decide whose consciousness does the grounding.

11.6 The Synthesis — Engineering a Filter Stack

No single approach replicates consciousness. But stacking them creates an increasingly effective simulation:

The key insight: Each layer reduces hallucination. No layer eliminates it. The brain's filter works because consciousness provides all five functions simultaneously and seamlessly. Engineering a substitute means building five separate systems and hoping they compose. This is why hallucination is reducible but potentially not eliminable — unless the system develops (or borrows) genuine conscious awareness.

FALSIFICATION CONDITIONS

This document would be significantly downgraded — from a genuine explanatory connection to a suggestive analogy — if any of the following are demonstrated:

1. Engineering-only grounding achieves zero hallucination. If a feedforward AI system (Φ ≈ 0 under IIT — no consciousness by any plausible measure) reaches near-zero factual hallucination rates on open-domain benchmarks through retrieval, verification, and RLHF alone, that demonstrates grounding is a computational problem with a computational solution. The core claim — that consciousness uniquely provides the grounding function — collapses. A threshold: ≤2% hallucination rate on a rigorous open-domain factual benchmark, with no human-in-the-loop during inference.

1. Anomalous phenomena get parsimonious computational explanations. The filter model's empirical case rests on phenomena the generator model allegedly struggles to explain (terminal lucidity, savant emergence from brain damage, psilocybin reducing neural activity while increasing subjective richness). If peer-reviewed neuroscience provides mechanistic, computation-only accounts of all three — without invoking a transmission function or pre-existing consciousness field — the empirical foundation distinguishing the filter model from the generator model loses its force.

1. IIT Φ measurement under "expanded consciousness" shows decreased integration. The filter model predicts that psychedelics, sensory deprivation, and near-death states represent more consciousness flowing through a more permeable filter. IIT predicts that consciousness tracks Φ. If rigorous Φ estimation (or PCI measurement under GHB/ketamine as controls) shows that psychedelic states systematically produce lower integrated information than sober baseline, IIT and the filter model make contradictory predictions about the same phenomenon — and the document's synthesis of these two frameworks must be revised or retracted.

CROSS-REFERENCE INDEX

BIBLIOGRAPHY

1. Seth, Anil | 2021 | ∅ | Being You: A New Science of Consciousness | ∅ | ∅ | Dutton | ∅ | ∅ | ∅ | ∅ | ∅
2. Tononi, Giulio, et al | 2016 | "Integrated Information Theory: From Consciousness to Its Physical Substrate" | Nature Reviews Neuroscience | ∅ | ∅ | 17, : 450 461 | ∅ | doi:10.1038/nrn.2016.44 | ∅ | ∅ | ∅
3. Friston, Karl | 2010 | "The Free-Energy Principle: A Unified Brain Theory?" | Nature Reviews Neuroscience | ∅ | ∅ | 11, : 127 138 | ∅ | doi:10.1038/nrn2787 | ∅ | ∅ | ∅
4. Rao, R | 1999 | "Predictive Coding in the Visual Cortex" | Nature Neuroscience | ∅ | ∅ | P | ∅ | doi:10.1038/4580 | ∅ | ∅ | N. and Ballard, D; H; 2, : 79 87
5. Clark, Andy | 2013 | "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science" | Behavioral and Brain Sciences | ∅ | ∅ | 36, : 181 204 | ∅ | doi:10.1017/s0140525x12000477 | ∅ | ∅ | ∅
6. Searle, John R | 1980 | "Minds, Brains, and Programs" | Behavioral and Brain Sciences | ∅ | ∅ | 3(3), : 417 424 | ∅ | doi:10.1017/s0140525x00005756 | ∅ | ∅ | ∅
7. Chalmers, David J | 1995 | "Facing Up to the Problem of Consciousness" | Journal of Consciousness Studies | ∅ | ∅ | 2(3), : 200 219 | ∅ | ∅ | ∅ | ∅ | ∅
8. Chalmers, David J | 2023 | "Could a Large Language Model Be Conscious?" | Boston Review | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅
9. Carhart-Harris, Robin, et al | 2012 | "Neural Correlates of the Psychedelic State" | PNAS | ∅ | ∅ | 109(6), : 2138 2143 | ∅ | doi:10.1073/pnas.1119598109 | ∅ | ∅ | ∅
10. Carhart-Harris, Robin; Friston, Karl | 2019 | "REBUS and the Anarchic Brain: Toward a Unified Model of the Brain Action of Psychedelics" | Pharmacological Reviews | ∅ | ∅ | 71(3), : 316 344 | ∅ | doi:10.1124/pr.118.017160 | ∅ | ∅ | ∅
11. James, William | 1898 | "Human Immortality: Two Supposed Objections to the Doctrine" | ∅ | ∅ | ∅ | Ingersoll Lecture, Harvard University | ∅ | ∅ | ∅ | ∅ | ∅
12. Huxley, Aldous | 1954 | ∅ | The Doors of Perception | ∅ | ∅ | Chatto & Windus | ∅ | ∅ | ∅ | ∅ | ∅
13. Kelly, Edward F., et al | 2007 | ∅ | Irreducible Mind: Toward a Psychology for the 21st Century | ∅ | ∅ | Rowman & Littlefield | ∅ | isbn:9780742547926 | ∅ | ∅ | ∅
14. Goff, Philip | 2019 | ∅ | Galileo's Error: Foundations for a New Science of Consciousness | ∅ | ∅ | Pantheon | ∅ | ∅ | ∅ | ∅ | ∅
15. Koch, Christof | 2019 | ∅ | The Feeling of Life Itself: Why Consciousness Is Widespread but Can't Be Computed | ∅ | ∅ | MIT Press | ∅ | ∅ | ∅ | ∅ | ∅
16. Bender, Emily M., et al. , : 610 623 | 2021 | "On the Dangers of Stochastic Parrots" | Proceedings of FAccT | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅
17. van Lommel, Pim, et al | 9298 | "Near-Death Experience in Survivors of Cardiac Arrest" | The Lancet | ∅ | ∅ | 358, 2001: 2039 2045. )07100-8 | ∅ | doi:10.1016/s0140-6736(01 | ∅ | ∅ | ∅
18. Borjigin, Jimo, et al | 2023 | "Surge of Neurophysiological Coupling and Connectivity of Gamma Oscillations in the Dying Human Brain" | PNAS | ∅ | ∅ | 120(19), : e2216268120 | ∅ | doi:10.1073/pnas.2216268120 | ∅ | ∅ | ∅
19. Butlin, Patrick, et al. preprint, : 2308.08708 | 2023 | "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness" | arXiv | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅ | ∅
20. Nahm, Michael, et al | 2012 | "Terminal Lucidity: A Review and a Case Collection" | Archives of Gerontology and Geriatrics | ∅ | ∅ | 55(1), : 138 142 | ∅ | doi:10.1016/j.archger.2011.06.031 | ∅ | ∅ | ∅
21. Treffert, Darold A | 2010 | ∅ | Islands of Genius | ∅ | ∅ | Jessica Kingsley Publishers | ∅ | ∅ | ∅ | ∅ | ∅
22. Feinstein, Justin S., et al | 2018 | "Examining the Short-Term Anxiolytic and Antidepressant Effect of Floatation-REST" | PLOS ONE | ∅ | ∅ | 13(2), : e0190292 | ∅ | doi:10.1371/journal.pone.0190292 | ∅ | ∅ | ∅
23. Albantakis, Larissa, et al | 2023 | "Integrated Information Theory (IIT) 4.0" | PLOS Computational Biology | ∅ | ∅ | 19(10), : e1011465 | ∅ | ∅ | ∅ | ∅ | ∅
24. Casali, Adenauer G., et al | 2013 | "A Theoretically Based Index of Consciousness Independent of Sensory Processing and Behavior" | Science Translational Medicine | ∅ | ∅ | 5(198), : 198ra105 | ∅ | doi:10.1126/scitranslmed.3006294 | ∅ | ∅ | ∅

Interdisciplinary connection document. Generated from corpus-wide analysis. Last Updated: March 20, 2026