FMT vs. Integrated Information Theory (IIT)¶

IIT is FMT's most serious competitor -- mathematically rigorous and philosophically ambitious -- but its panpsychist commitments, the unresolved Combination Problem, and the computational intractability of Phi represent structural costs that FMT avoids entirely.

Integrated Information Theory (Tononi, 2004; Albantakis et al., 2023) and the Four-Model Theory are the two frameworks in contemporary consciousness science that attempt to address the Hard Problem head-on rather than bracketing it. This makes their comparison especially instructive: both aim at the same target but arrive via fundamentally different routes.

IIT's Genuine Strengths¶

IIT brings three substantial contributions to consciousness science that deserve recognition.

Mathematical rigor. IIT is the most mathematically developed consciousness theory. Its axioms, postulates, and the Phi formalism provide a precision that most theories -- including FMT in its current form -- lack. The formal apparatus makes IIT's claims testable in principle, even where computation is infeasible in practice.

Qualia space. IIT's treatment of experiential structure through a multidimensional qualia space is arguably its greatest achievement. The idea that the quality of an experience corresponds to the shape of the cause-effect structure captures something deep about why red feels different from blue. FMT's account of experiential structure through the Explicit World Model is less formally developed.

The exclusion postulate. IIT provides a principled answer to the Boundary Problem through the exclusion postulate: the system with maximum Phi defines the boundary of consciousness. This is elegant, even if computationally unrealizable for biological systems.

IIT's Structural Problems¶

Three difficulties are not incidental to IIT but follow necessarily from its core commitments.

Panpsychism. IIT's axiom-based identification of consciousness with integrated information (Phi) entails that any system with non-zero Phi has some degree of experience. This includes thermostats, logic gates, and simple feedback circuits. IIT's proponents accept this consequence; most neuroscientists find it a reductio ad absurdum. FMT "requires no fundamental experiential property and no panpsychist micro-experience, and it cannot empirically exclude either." On FMT consciousness emerges weakly at a specific computational level, in a system that meets the sufficient conditions: the four model kinds under closure, free compute (the open-ended Class 4 regime, whose neural signature is near-criticality), embodiment, and at least one pain-pleasure axis.

The Combination Problem. If fundamental entities have micro-experience, how do these micro-experiences combine into the macro-experience of a human mind? This is not a gap in IIT's development -- it is a structural consequence of panpsychism itself (Chalmers, 2016). FMT requires no micro-experience, so the Combination Problem does not arise for it as a difficulty: consciousness emerges weakly from the running self-simulation, just as a spreadsheet sum emerges from transistors that contain no sum. FMT cannot exclude micro-experience either. "If panpsychism holds, one further ingredient is needed, a property of physics not yet known, and closure is what forms a subject from it; the theory would then answer the combination problem."

Phi is computationally intractable. Calculating Phi for a realistic neural system is not merely difficult but computationally intractable (Aaronson, 2014). This means IIT's predictions are often untestable in practice. FMT's predictions -- psychedelics alleviating anosognosia, controllable ego dissolution content, DID alter-switch neural signatures, lucid dream onset as criticality crossing -- require no intractable computation.

The Divergence Point¶

FMT adopts IIT's central diagnosis: consciousness is irreducible, integrated, and causally structured, not mere information processing. The divergence lies in what that irreducible structure is and where each theory locates phenomenality. IIT identifies consciousness with intrinsic causal power: a system is conscious to the degree that it has Phi. FMT identifies consciousness with a specific process -- ongoing self-simulation across four model kinds in the Class 4 regime. For IIT, consciousness is everywhere Phi is non-zero. For FMT, consciousness is nowhere without the four-model architecture and that regime, regardless of how much information is integrated.

The unfolding argument (Doerig et al., 2019) holds that any recurrent network can be unfolded into a feedforward network with identical input-output mappings. Its target is any theory on which two systems agreeing in all input-output behavior can differ in consciousness -- IIT, whose feedforward twin has zero Phi, and FMT alike, since self-referential closure is a property of the architecture rather than of the mapping. FMT's reply is that unfolding externalizes the loop: no stage of the unfolded chain models the stages that follow it, so closure is gone. And the two systems agree only in the limit of unbounded resources. Unfolding trades a reused loop for replicated stages; in the theory's model-system results on one connectome family, freedom from any closed path costs 71.5-74.0% of the connectome, against 12.0-14.0% for merely relocating closure. Where a resource bound applies, the unfolded chain and the closed one stop agreeing on behavior. On an unbounded substrate the theory has no experiment to offer.

Figure¶

graph TB
    subgraph IIT_PATH["IIT's Route"]
        direction TB
        I1["Axioms<br/>(intrinsic existence, composition,<br/>information, integration, exclusion)"]
        I2["Phi (Φ)<br/>integrated information"]
        I3["Consciousness =<br/>non-zero Φ"]
        I4["Panpsychism +<br/>Combination Problem"]
        I1 --> I2 --> I3 --> I4
    end

    subgraph FMT_PATH["FMT's Route"]
        direction TB
        F1["Self-Simulation<br/>(four-model architecture)"]
        F2["Class 4 regime<br/>(near-criticality is<br/>its neural signature)"]
        F3["Virtual Qualia<br/>(computational-level properties)"]
        F4["Weak Emergence<br/>no micro-experience required"]
        F1 --> F3
        F2 --> F3
        F3 --> F4
    end

    HP["Hard Problem"] --- I3
    HP --- F3

    style IIT_PATH fill:#1a1a2e,stroke:#a4243b,color:#fff
    style FMT_PATH fill:#1a1a2e,stroke:#2d6a4f,color:#fff
    style I4 fill:#6a1b2a,stroke:#a4243b,color:#fff
    style F4 fill:#2d6a4f,stroke:#40916c,color:#fff
    style HP fill:#2d1b69,stroke:#9b59b6,color:#fff

Both theories engage the Hard Problem, through fundamentally different routes. IIT's path leads through panpsychism to the Combination Problem. FMT's path through virtual qualia arrives at weak emergence, which requires no micro-experience, so the Combination Problem does not arise for it as a difficulty.

Key Takeaway¶

IIT and FMT are the field's two most philosophically ambitious theories. IIT pays for its mathematical elegance with panpsychism, the Combination Problem, and computational intractability. FMT pays for its architectural specificity with limited formal mathematical development -- a gap, not a structural flaw, since formalization can follow.