The Four-Model Theory of Consciousness: A Simulation-Based Framework Unifying the Hard Problem, Binding, and Altered States

Abstract

The science of consciousness remains in a pre-paradigm state, with no theory simultaneously satisfying the eight core requirements a complete theory must meet: the Hard Problem, the Explanatory Gap, the Boundary Problem, the Structure of Experience, Unity and Binding, Combination and Emergence, the Causal Role, and the Meta-Problem. This paper presents the Four-Model Theory, in which consciousness is constituted by real-time self-simulation across four nested models arranged along two axes - scope (world vs. self) and mode (implicit vs. explicit). The implicit models (Implicit World Model, Implicit Self Model) are substrate-level, learned, and non-conscious. The explicit models (Explicit World Model, Explicit Self Model) are virtual, transient, and phenomenal - they are the simulation in which experience occurs. The theory’s central claim is that qualia are constitutive properties of the computational level - digital constructs that exist at the level of the running computation but are incoherent at the substrate level, just as a spreadsheet cell’s value is incoherent at the transistor level. This dissolves the Hard Problem by revealing a category error - a level confusion that seeks phenomenal properties at the substrate level where they categorically do not exist. Self-referential closure explains why this specific computational process has experience when a weather simulation does not: the system’s model includes a model of itself, collapsing the inside/outside distinction and making experience constitutive rather than additional. Combined with a criticality requirement (the substrate must operate at the edge of chaos), the theory derives diverse phenomena from five principles: criticality, virtual qualia, a redirectable Explicit Self Model, variable implicit-explicit permeability, and virtual model forking. These principles unify psychedelic phenomenology, anesthetic mechanisms, dream states, split-brain phenomena, dissociative identity disorder, and animal consciousness. A systematic comparison shows the theory addresses all eight requirements. Unusually for a consciousness theory, the framework has substantial empirical grounding: five claims that follow from its core axioms - established in 2015 - have since been independently confirmed by research groups with no connection to the theory, including the anesthetic-criticality convergence (Casali et al., 2013; Hengen and Shew, 2025; Algom and Shriki, 2026), sleep-dependent criticality restoration (Bhatt et al., 2024), sleep onset as bifurcation (Li et al., 2025), and split-brain holographic degradation (Pinto et al., 2017). Four novel predictions remain untested - including that psychedelics should alleviate anosognosia and that ego dissolution content is controllable via sensory input - predictions no competing theory generates. Changelog v12 v12 changes (2026-07-06): independent adversarial-review pass, Hard-Problem rewrite, qualia-structure prediction (substantial revision; the core architecture is unchanged). Hard Problem (§3.4.3) rewritten: conditional dissolution — the level-confusion holds on a constitutive reading of self-referential closure the theory motivates but does not derive (rated partial, ◐); the closure→phenomenality link is made empirically approachable via spontaneous first-person report under a training control, with an explicit confabulation caveat (necessary but not sufficient). IIT engagement (§7.2): honest-convergence paragraph — grant IIT's diagnosis (experience as an irreducible, integrated, causally-structured whole) while rejecting substrate-boundness; consciousness located in substrate-neutral self-referential updating at criticality, not intrinsic Φ. Barrett & Mediano (2019) Φ-undefined-for-non-Markovian critique added with IIT's reply conceded; exclusion postulate stated correctly. New Prediction 5 (§8.6): qualia structure is cross-individually shareable (unsupervised Gromov–Wasserstein alignment of similarity structures; Kawakita et al. 2025) while absolute per-brain encoding is not — a falsifiable dissociation; new qualia-structure paragraph (§3.4.2) anchored to the phenomenal-concept strategy. Substrate independence: olfaction (thalamus-bypassing) anchor (Li & Gottfried 2010, Sela 2009); Seth biological-naturalism engagement; Bayne et al. 2024 cross-system testing; Milinkovic & Aru biological-computationalism engaged. Corrections & honesty: Casarotto (2016) PCI benchmark comparison corrected; 40/20 Hz reframed as a posit; No-Free-Lunch replaced by multiple-realizability; §1 automaton-identity aligned (consciousness = the self-simulation running on the Class-4 automaton, not the automaton itself); Passos-Ferreira & Chalmers 2026 infant-consciousness field-context; citation-breadth 2024–26. References: new verified citations added (Barrett & Mediano 2019, Kawakita 2025, Kriegeskorte 2008, Bayne 2024, Storm 2024, olfaction set); all DOIs web-verified. v11 changes (2026-06-29): Criticality dimensions, causal roles, metacognition (theory extension and citation strengthening; the substantive architecture is unchanged). Criticality requirement (§3.7): introduced two orthogonal dimensions of criticality — extent (integration; the spatial fraction recruited into Class-4 dynamics) and complexity (differentiation; the richness of the Class-4 patterns computed) — framed as honest convergence with the integration–differentiation tradition (Tononi & Edelman 1998; IIT-Φ; Lempel-Ziv/PCI), claiming only the edge-of-chaos grounding and the seizure-discriminated definition of extent. Seizure framing (§3.7, §10.3): made route-independent — a generalized seizure abolishes consciousness by exiting the Class-4 regime via pathological hypersynchrony or supercritical runaway (Meisel et al. 2012); the invariant is loss of Class-4 dynamics, not a single branching-ratio value. Access vs. phenomenal (§3.4.6): added the FMT-native access mechanism — the transition from present to accessed content is explicit-model self-interaction / recursion depth (the §3.5 ladder), not a broadcast bottleneck. New §4.2.3 "The Two Causal Roles of the Self-Model": distinguishes the outward (slow, indirect) and inward (immediate, reality-decoupling) causal grips; the self-model contributes necessity, not freedom (non-epiphenomenal, yet with no independent causal power over the substrate); the reality-coupling lock-in is a second governor on the rare jointly-maximized states; content-steering is left as an open question (Soon et al. 2013; Koenig-Robert & Pearson 2019). Metabolic governor (§5.1): the finite metabolic budget identified as the first of two governors that make the jointly-maximized state rare. Open Questions (§9): the ESM/EWM double dissociation now cites supporting open-data metacognition evidence (Rouault et al. 2018, n=498; the Confidence Database, 2,752 participants, Rahnev et al. 2020; Fleming & Lau 2014) as consistent with the architecture, and reframes the decisive test as a selective causal dissociation — functional selectivity, not the focal excision of a localized module the theory does not posit. Philosophical sharpening (§3.4.5): separated the two limits on extracting a non-running description — the merely practical (a full static capture is unstorable and illegible the way source code is not) from the constitutive (a description is not an instance; experience is a property of the running process). References: 12 new citations added (all DOIs web-verified); the self-citation for the inner-dimensionality > sensory/motor-channel claim replaced with mainstream support (Stringer et al. 2019; Zheng & Meister 2025; Fiser et al. 2004; Raichle 2010). v10 changes (2026-06-10): Major exposition revision (sharpening of arguments; no change to the theory's substantive claims). Philosophical commitments (§4): named the metaphysical position as type-B physicalism / the phenomenal-concepts strategy (Loar 1997; Papineau 2002) combined with the ability hypothesis; reframed the explanatory gap as epistemic rather than ontological (supervenience holds — phenomenal facts are entailed by the physical description but cannot be had by description, only by instantiation); replaced the previously circular zombie rebuttal with the conceivable-but-impossible argument. Criticality requirement (§3.7): restated the Class 4 requirement as an explicit three-step argument — capability, free instantiation, evolutionary forcing — presented as a strongly-motivated principle rather than a theorem; clarified that merely universal but heteronomous systems (e.g. a laptop) are excluded because nothing drives them to criticality. Relation to Predictive Processing (§7.2): reframed PP as a theory of perception and inference rather than of consciousness (Hohwy & Seth 2020); FMT supplies the consciousness-making layer (self-model, self-referential closure, criticality) — complementary, not rival. References: 14 bibliographic corrections (including corrected first author Xu et al. 2024; corrected Chowdhury and Katlowitz titles and DOIs; COGITATE published title/DOI); 5 references added (Loar 1997, Papineau 2002, Langton 1990, Hohwy & Seth 2020, Xu et al. 2024). v9 changes (2026-06-08): Figures restored: Three figures (four-model architecture, real/virtual split, phenomenological content) added back to the paper. Open Question 7: Basal ganglia role in permeability gating — dopaminergic prediction-error mechanism as candidate for variable permeability, schizophrenia extension (Gershman 2019, Shepherd 2011, Benjamin & Kording 2023, Howes & Kapur 2009, Deperrois et al. 2022). Animal consciousness: FMT-operations-to-empirical mapping table (self-referential closure, variable permeability, criticality, graduated levels mapped to animal evidence). Barrett IIT critique: Phi shown mathematically undefined for non-Markovian systems (§7.2). Bach Machine Consciousness Hypothesis: Independent convergence on self-simulation (§7.3). fMRI 2×2 convergence: Fox 200