Theory Map: Where the Major Theories Sit in FMT¶

Most theories of consciousness describe one part of the architecture the Four-Model Theory specifies. FMT keeps most of their mechanisms and rejects specific claims about what is sufficient for experience, whether phenomenal properties are real, and whether the right computation is enough without biology.

The other comparative pages set FMT against one theory at a time, or score theories against the eight requirements. This page places eighteen theories and variants on a single picture of FMT's architecture. Each theory is broken into separate claims, because one theory can be right about a mechanism and wrong about what that mechanism is sufficient for. Each claim gets one of four verdicts:

Absorbed — FMT contains the claim as one of its own components or as an implementation of one.
Compatible — FMT does not contain the claim and nothing in FMT conflicts with it.
Neutral — FMT takes no position.
Contradicts — the claim and FMT cannot both be true.

The Reference Architecture¶

FMT's architecture has two kinds of model along two axes: implicit models stored in the substrate's connectivity (IWM, ISM) and explicit models generated from them as running processes (EWM, ESM). The self-model is nested inside the world-model. The explicit models read from and write back to the implicit ones along a low-rank channel, whose occupancy is working memory. The theory's third principle is an identity: phenomenology "is that level's modeling closed on itself", the self-model and the world-model each entering the other's update rule. That is self-referential closure. Underneath everything runs free modeling, the open-ended (Class 4) computation the system recruits for its own modeling; criticality is the signature free modeling leaves in some brains, not a requirement.

These components give each theory a place to sit:

the wired band (implicit models, connectivity);
the channel between the bands (workspace, working memory, permeability);
the running band (explicit models);
the closure loop;
the ladder of recursive self-modeling depth (graduated consciousness);
the opacity between self-model and substrate (meta-problem);
the free-modeling field beneath it all;
substrate independence at the outside edge.

Figure¶

flowchart BT
    subgraph WIRED["Wired band: implicit models (stored in connectivity)"]
        IWM["IWM<br/><i>implicit world model</i>"]
        ISM["ISM<br/><i>implicit self model</i>"]
    end
    CH["Channel / working memory<br/><i>permeability</i>"]
    subgraph RUNNING["Running band: explicit models (virtual qualia live here)"]
        EWM["EWM<br/><i>explicit world model</i>"]
        ESM["ESM<br/><i>explicit self model</i>"]
        LAD["Ladder<br/><i>simply / doubly / triply extended</i>"]
    end
    FM["Free modeling<br/><i>criticality = its signature in some brains</i>"]
    SI["Substrate independence"]

    FM --> WIRED
    IWM <-->|read / write| CH
    ISM <-->|read / write| CH
    CH <--> EWM
    CH <--> ESM
    ESM -->|closure| EWM
    EWM -->|closure| ESM
    ESM --- LAD
    ESM -. "opaque (meta-problem)" .-> ISM

    GNW["GNW, CTM<br/><i>keeps: workspace, ignition</i><br/><i>rejects: broadcast constitutive</i>"] --> CH
    PP["PP / active inference<br/><i>keeps: inference machinery</i><br/><i>rejects: PP sufficient</i>"] --> WIRED
    REBUS["REBUS<br/><i>keeps: permeability change</i>"] --> CH
    DIT["DIT<br/><i>keeps: candidate cortical realizer</i>"] --> CH
    RPT["RPT<br/><i>keeps: recurrence</i><br/><i>rejects: local recurrence sufficient</i>"] --> EWM
    PRM["Reality monitoring<br/><i>keeps: source distinction</i><br/><i>rejects: monitor output constitutive</i>"] --> RUNNING
    IIT["IIT<br/><i>keeps: integration, recurrence</i><br/><i>rejects: identity with substrate</i>"] --> RUNNING
    BL["Beautiful Loop, ICT<br/><i>keeps: closure</i><br/><i>ICT rejected: no self-model needed</i>"] --> EWM
    HOT["HOT<br/><i>keeps: higher-order self-modeling</i><br/><i>rejects: per-state transitivity</i>"] --> LAD
    AST["AST, Metzinger, beast machine<br/><i>keeps: self-model mechanisms</i><br/><i>rejects: deflationism, phenomenality with no self-model, life as a requirement</i>"] --> ESM
    IWMT["IWMT<br/><i>compatible: world + self modeling</i>"] --> RUNNING
    ILL["Illusionism<br/><i>rejects: real phenomenal properties</i>"] --> ESM
    BN["Searle, Seth (biological naturalism)<br/><i>rejects: computation suffices</i>"] --> SI
    ENA["Enactivism<br/><i>compatible: autonomy</i><br/><i>rejects: no internal models</i>"] --> FM

    style WIRED fill:#1a1a2e,stroke:#6c757d,color:#fff
    style RUNNING fill:#1a1a2e,stroke:#40916c,color:#fff
    style CH fill:#2d1b69,stroke:#9b59b6,color:#fff
    style FM fill:#1a1a2e,stroke:#9b59b6,color:#fff
    style SI fill:#1a1a2e,stroke:#9b59b6,color:#fff
    style GNW fill:#1b4332,stroke:#a4243b,color:#fff
    style PP fill:#1b4332,stroke:#a4243b,color:#fff
    style REBUS fill:#1b4332,stroke:#40916c,color:#fff
    style DIT fill:#1b4332,stroke:#40916c,color:#fff
    style RPT fill:#1b4332,stroke:#a4243b,color:#fff
    style PRM fill:#1b4332,stroke:#a4243b,color:#fff
    style IIT fill:#1b4332,stroke:#a4243b,color:#fff
    style BL fill:#1b4332,stroke:#a4243b,color:#fff
    style HOT fill:#1b4332,stroke:#a4243b,color:#fff
    style AST fill:#1b4332,stroke:#a4243b,color:#fff
    style IWMT fill:#1d3557,stroke:#457b9d,color:#fff
    style ILL fill:#6a1b2a,stroke:#a4243b,color:#fff
    style BN fill:#6a1b2a,stroke:#a4243b,color:#fff
    style ENA fill:#1d3557,stroke:#a4243b,color:#fff

Each theory is attached to the part of FMT its central claim is about. Green boxes: FMT keeps the theory's mechanism. Blue boxes: FMT can live with the theory's main claim. Red boxes: FMT rejects the theory's main claim. A red outline marks a specific claim FMT rejects, named in the box. The implicit models form one wired band and the explicit models one running band; the four labels mark poles of one modeling space, not four modules.

The Summary Table¶

Theory	Where it sits in FMT	FMT keeps	FMT rejects	Page
IIT (Tononi; Albantakis et al.)	Running band, closure, free modeling	Consciousness is irreducible, integrated, causally structured; recurrence	Identity with the substrate's cause-effect power; every Φ > 0 system conscious	vs-iit
GNW (Baars; Dehaene & Changeux)	Channel	Workspace = channel occupancy; capacity limit; ignition as marker	Broadcast as what makes content conscious	vs-gnw
CTM (Blum & Blum)	Channel; world model containing a self	World model with a self inside it	Broadcast constitutive; no recruitment condition	further theories
Predictive processing (Friston; Hohwy & Seth)	Wired band; generation of explicit models	Prediction-error machinery as the implicit layer's implementation	PP as sufficient for consciousness	vs-pp
Beast machine (Seth)	ISM → ESM	Interoceptive inference as one route to the self-model	Life as a possible requirement	self-modeling tradition
REBUS (Carhart-Harris & Friston)	Permeability	One implementation of global permeability change	—	vs-pp
Beautiful Loop (Laukkonen, Friston & Chandaria)	Closure	Self-evidencing loop = closure	—	—
HOT (Rosenthal; Lau & Rosenthal)	Ladder	Higher-order self-modeling as the extended rungs	A higher-order representation of the particular state, rather than closure of the whole system, as the condition of experience	vs-hot
Perceptual reality monitoring (Lau)	Source distinction	Reality monitoring as a function	Monitor output as what makes experience	further theories
RPT (Lamme)	Generation of the running models	Recurrence as the mechanism	Local sensory recurrence as sufficient	vs-rpt
AST (Graziano)	Self-model; attention into the channel	Attention schema as part of the self-model; meta-problem mechanism	That the experience depicted in claims of awareness does not exist ("small" illusionism)	vs-ast
Self-model theory (Metzinger)	Self-model	Self-model as constitutive; transparency; "selfless" minimal phenomenal experience read as no represented self	Phenomenality with no self-model at all	further theories
IWMT (Safron)	Running band; free modeling	World and self modeling; near-critical signature; formal mechanism	Graded demarcation (mild)	—
DIT (Aru, Suzuki & Larkum)	Step from wired to running band	Candidate cortical realizer	— (uniqueness: neutral)	—
ICT (Chang, Biehl, Yu & Kanai)	Closure	Closure as the primitive	Informational closure without a self-model as sufficient	further theories
Illusionism (Dennett; Frankish)	Substrate side; meta-problem	No phenomenal properties in the substrate	No phenomenal properties at any level	two-level ontology
Biological naturalism (Searle; Seth)	Substrate independence	Interoceptive self-regulation as one route	Searle: the right computation is never enough. Seth: consciousness may depend on being alive	further theories
Enactivism (Varela, Thompson & Rosch; Thompson)	Free-modeling recruitment	Autonomy and embodiment as conditions on recruitment	Experience without internal models; experience in every living system, if "mind" there includes it	further theories

What the Pattern Shows¶

Mechanisms are kept. The global workspace becomes the channel and its occupancy. Recurrent processing becomes the dynamics that host the running models. Prediction-error minimization becomes the machinery of the implicit layer. Higher-order self-representation becomes the rungs of the ladder. Apical dendritic coupling in layer-5 neurons becomes a candidate cortical realizer of the step from stored to running models. The attention schema becomes part of the self-model. The Beautiful Loop and information closure theory both name closure.

Sufficiency claims are rejected. Each of these theories also says what is enough for experience: broadcast (GNW), local recurrence (RPT), a higher-order representation of the state (HOT), informational closure without a self-model (ICT), any integrated structure with Φ above zero (IIT). FMT's answer to all of them is the same: phenomenality is the self-model and world-model closed on each other, so none of these structures is enough without that closure.

Ontology and substrate claims are rejected. Illusionism denies that phenomenal properties exist, and AST that the experience our claims of awareness describe exists; on FMT they exist at the computational level as virtual qualia. Searle holds that running the right computation is never enough, and Seth that consciousness may depend on being alive. FMT holds that any physical system meeting its sufficient conditions is conscious: the four models under closure, free compute, embodiment, and at least one pain-pleasure axis. Strong enactivism denies that experience is made of internal models; on FMT experience is a running internal simulation.

Which Disagreements Can Be Tested¶

Each rejection comes with the experiment that would decide it, where one exists.

GNW: make content globally available without letting it into the self-model's update rule. No such manipulation exists yet in brains. Language models show the half that exists: a broadcast workspace without closure.
RPT: find a state with local sensory recurrence intact and self-model dynamics abolished. RPT predicts experience, FMT none. "Islands of awareness" are the candidate case, but they cannot report.
HOT: abolish metacognitive read-out while sparing first-order perception and the self-model loop. HOT predicts experience changes; FMT predicts it survives. HOT proponents need not accept that metacognitive measures track the higher-order representation their theory requires.
Reality monitoring: perturb the brain and track reality judgements and experience separately. FMT predicts source attribution fails first, as in dreaming.
IIT: no decider in brains yet. The unfolding argument targets both theories; FMT's reply is that a closed system and its unfolded feedforward twin stop agreeing on behavior once a resource bound applies, a difference measurable in silico (vs-iit).
ICT, IWMT: build matched artificial systems with and without a self-model in the loop. This separates the architectures in silico, though behavior does not settle experience.
Minimal phenomenal experience: record a measure of self-model activity during verified episodes. FMT predicts a minimal but non-zero signature; the strict selfless reading predicts none.
Biological naturalism, enactivism: decidable only with an artificial system that implements the full architecture, which does not yet exist.
AST and illusionism: no empirical test exists. Both sides explain the reports. The difference is ethical: on FMT, an artificial system that meets the sufficient conditions would have phenomenal states and be a moral patient for that reason.

Any experiment that tries to remove closure meets one complication. In a recurrent network, cutting one loop tends to leave two smaller closed systems. A manipulation has to show that it removed closure, not just that it opened a loop.

Key Takeaway¶

FMT contains most of these theories' mechanisms as components. Its disagreements with them concern what is enough for experience and whether experience is real, and several can be decided by experiments that can be described now.