Temporal Depth and Intrinsic Dominance: A Dynamical Completion of Integrated World Modeling Theory

Temporal Depth and Intrinsic Dominance: A Dynamical Completion of Integrated World Modeling Theory

Abstract

Integrated World Modeling Theory (IWMT) proposes that consciousness arises from integrated generative architectures capable of constructing spatially, temporally, and causally coherent world models. While IWMT specifies the representational structure required for conscious modeling, it does not explicitly constrain the dynamical regime under which such modeling becomes phenomenally instantiated. This paper introduces a regime-level completion. I distinguish representational temporal coherence from dynamical temporal depth and argue that conscious world modeling requires physical instantiation within intrinsically dominant, temporally deep dynamical regimes. Temporal Depth (TD) quantifies the contribution of slow intrinsic modes to overall system dynamics, while Intrinsic Dominance Ratio (IDR) partitions slow persistence into endogenous versus externally driven components. Together, TD and IDR function as conjunctive necessary conditions for conscious instantiation. This framework refines IWMT’s physical grounding and generates empirically testable predictions across anesthesia, disorders of consciousness, and artificial systems.

1. Introduction

Integrated World Modeling Theory (IWMT) represents one of the most ambitious contemporary syntheses in consciousness science (Safron 2020). Drawing upon predictive processing (Friston 2010; Clark 2013), global integration accounts (Dehaene & Changeux 2011), and embodied self-modeling (Metzinger 2003), IWMT proposes that consciousness arises when hierarchical generative models become sufficiently integrated and include the organism as an embedded perspective within a unified representation of the world.

The theory provides a compelling architectural picture. It explains how multimodal information can be unified within a hierarchical generative structure; how the organism can be represented within its own model; and how self-world distinctions can emerge from predictive inference. IWMT successfully integrates several strands of contemporary cognitive science into a coherent account of conscious world modeling.

Yet a deeper question remains insufficiently addressed: under what physical conditions does such modeling become phenomenally conscious?

IWMT describes computational organization and representational architecture. It specifies what sort of generative modeling must occur. However, it does not explicitly identify the dynamical regime under which these computational processes instantiate experience rather than remaining subpersonal or unconscious. This omission is not unique to IWMT. It reflects a broader tendency within computational and informational theories of consciousness to emphasize architecture while leaving regime specification implicit.

If consciousness is a physical phenomenon, then it must correspond to a particular organization of matter in motion. Computational structure alone does not fix dynamical regime. The same formal generative model can, in principle, be implemented in different physical modes: transient, rapidly decaying dynamics; stable attractor regimes; metastable switching regimes; or near-critical slow-evolving dynamics. From a purely functional perspective, these regimes might implement equivalent inferential operations. From a physical perspective, they differ profoundly in temporal organization.

This paper advances a central claim: any adequate physical theory of consciousness must specify the dynamical regime in which its computational architecture becomes phenomenally instantiated. IWMT, while architecturally rich, remains dynamically under-specified. We propose that conscious world modeling requires occupation of temporally deep intrinsic dynamical regimes. Temporal depth is understood not as mere slowness, but as persistent intrinsic organization across extended timescales relative to immediate input fluctuations.

The goal of this paper is not to replace IWMT. Nor is it to reduce experience to temporal metrics or to claim that slow dynamics are sufficient for consciousness. Rather, the aim is to complete IWMT at the level of physical instantiation by introducing a regime-level constraint that clarifies when generative modeling becomes phenomenally conscious.

It is important to clarify that IWMT already invokes “temporal coherence” as a property of generative world models. In IWMT, temporal coherence refers to the representational capacity of a system to model events across time, integrate past and anticipated future states, and maintain causal continuity within its world-model. The present proposal does not contest or replace this claim. Rather, it distinguishes representational temporal coherence from dynamical temporal depth. While IWMT specifies what must be modeled (temporally coherent structure), it does not explicitly specify the dynamical regime under which such modeling becomes phenomenally instantiated. The contribution here concerns this latter issue: a regime-level constraint on the physical dynamics that realize generative world modeling.

The argument proceeds as follows. First, we articulate the problem of dynamical under-specification in IWMT and in computational accounts more broadly. Second, we argue that regime specification is philosophically necessary for any physicalist account of consciousness. Third, we review empirical evidence indicating that conscious states are reliably associated with temporally extended intrinsic dynamics. Fourth, we introduce temporal depth as a regime constraint and show how it can be integrated into IWMT without altering its architectural commitments. Fifth, we outline testable predictions and falsification criteria. Finally, we discuss the broader philosophical implications of adopting what may be called a regime-based realism about consciousness.

It is important to clarify the sense in which the present proposal claims novelty. The contribution does not lie in identifying intrinsic timescales, metastability, or scale-free dynamics as correlates of consciousness. These phenomena are well documented. Nor does the contribution lie in proposing generative world modeling as the architectural basis of consciousness; that claim belongs to IWMT. The novelty resides in formalizing regime specification as a necessary completion of a world-modeling theory. That is, the present work argues that no architectural account of consciousness—including IWMT—can be physically adequate without specifying the dynamical phase in which its computational organization becomes phenomenally instantiated. Temporal depth is introduced not as an additional correlate but as a principled regime constraint that completes IWMT at the level of physical instantiation. The theoretical advance therefore lies in elevating dynamical regime from empirical observation to structural necessity within a consciousness theory. 

2. The Dynamical Under-Specification of IWMT

Integrated World Modeling Theory (IWMT) proposes that consciousness arises when hierarchical generative models become sufficiently integrated, embodied, and self-including (Safron 2020). At the computational level, IWMT describes a system that constructs a unified, probabilistic model of the world and situates the organism within that model as a perspectival locus. The architecture is hierarchical, predictive, and multimodal. It integrates information across levels and modalities, forming a world-model that is both inferential and embodied.

This architectural description is compelling. However, computational organization does not uniquely determine physical regime.

A generative model defined in Bayesian or predictive-processing terms can be implemented in multiple dynamical configurations. The same formal inference process—minimizing prediction error or variational free energy—could unfold in rapidly decaying transient responses, in stable attractor dynamics, in metastable switching regimes, or in near-critical, slow-evolving intrinsic modes. From the standpoint of formal inference, these may be functionally equivalent. From the standpoint of physical instantiation, they differ in fundamental ways.

The distinction matters because consciousness is not merely an input-output mapping. It is an ongoing, temporally extended phenomenon. If two systems implement the same computational mapping but differ radically in intrinsic temporal organization, it is not obvious that they should be regarded as phenomenologically equivalent.

Generative modeling is ubiquitous in biological systems. Cerebellar predictive control operates at subpersonal levels (Ito 2008). Spinal reflex loops can implement predictive adjustments (Wolpert & Ghahramani 2000). Early sensory cortices engage in predictive coding (Rao & Ballard 1999). None of these processes, taken in isolation, are typically regarded as conscious. Yet they instantiate generative inference. If generative modeling alone were sufficient, the boundary between conscious and unconscious processes would collapse.

Similarly, artificial neural networks now implement hierarchical generative models across modalities (Radford et al. 2019; Brown et al. 2020). These systems can generate text, images, and multimodal predictions. They approximate large-scale world modeling in a computational sense. If architectural sufficiency were the sole criterion, the question of artificial consciousness would become pressing in a way that many theorists regard as premature.

The difficulty can be expressed as a dynamical indeterminacy problem:

IWMT specifies what is computed, but not under what dynamical regime that computation becomes phenomenally conscious.

Without regime specification, IWMT risks conflating conscious world modeling with subpersonal predictive inference. The architectural description may be necessary, but it is not evidently sufficient. The physical conditions under which modeling becomes experience remain unspecified.

This is not a trivial omission. In physics, phenomena are typically defined not merely by structure but by regime. Superconductivity requires low temperature. Turbulence requires sufficient Reynolds number. Phase transitions occur at parameter thresholds. To describe structural relations without specifying the dynamical region in which they are realized is incomplete.

If consciousness is a physical phenomenon, then it too must correspond to a particular region of dynamical state space.

3. The Philosophical Necessity of Regime Specification

The need for regime specification is not merely empirical; it is philosophical.

Functionalist traditions have long held that mental states are individuated by causal organization rather than substrate (Putnam 1967; Fodor 1975). On this view, if the right computational relations are implemented, mental states follow. However, critics have argued that computational equivalence does not guarantee phenomenological equivalence (Searle 1992; Kim 1998; Chalmers 1996). Two systems may share input-output mappings and even internal computational structure while differing in their physical mode of realization.

The debate between functionalism and its critics often turns on whether physical realization conditions matter beyond abstract causal structure. In the context of consciousness, this debate takes on particular urgency. Experience appears as a temporally extended process. It unfolds, persists, and evolves over time. It is not merely a mapping from stimulus to response.

Computational descriptions abstract away from many physical details. They capture formal relations but leave open how those relations are dynamically instantiated. A predictive-processing architecture could, in principle, be implemented in a system whose internal dynamics decay almost instantaneously, or in one whose intrinsic modes persist across extended temporal windows. The formal architecture alone does not settle the matter.

Predictive processing frameworks (Friston 2010; Clark 2013) emphasize inference, prediction, and error minimization. They provide a powerful unifying perspective on cognition. Yet they largely leave the dynamical substrate implicit. IWMT inherits this strength—and this limitation.

If one accepts that consciousness is physically realized, then specifying the dynamical regime is not optional. It is a requirement of explanatory adequacy. Without regime specification, one risks what might be called architectural sufficiency without physical grounding.

This paper advances what may be termed a regime requirement:

Any adequate physical theory of consciousness must specify the dynamical phase in which its computational architecture becomes phenomenally instantiated.

The requirement is modest. It does not claim that dynamical regime alone explains experience. It claims only that without such specification, a theory remains incomplete at the level of physical instantiation.

4. Empirical Convergence on Temporal Depth

If regime specification is required, which regime is relevant?

A striking convergence in empirical research suggests that conscious states are associated with temporally extended intrinsic dynamics. This convergence appears across methodologies, species, and experimental manipulations.

First, intrinsic timescales vary systematically across cortical hierarchies. Association cortices exhibit longer autocorrelation decay constants than primary sensory areas (Murray et al. 2014; Chaudhuri et al. 2015). Higher-order regions integrate information over extended temporal windows, while early sensory regions operate on shorter timescales.

Second, anesthesia reliably shortens intrinsic timescales. Under propofol, cortical dynamics become more transient and fragmented (Purdon et al. 2013; Hudson et al. 2014). Long-range temporal correlations are disrupted. Activity becomes less persistent.

Third, deep sleep is associated with breakdown of long-range temporal dependence (Tagliazucchi et al. 2013). Effective connectivity collapses (Massimini et al. 2005). Intrinsic reverberation diminishes.

Fourth, perturbational studies demonstrate that conscious brains sustain temporally extended responses to stimulation. The perturbational complexity index (PCI) reveals that wakeful brains exhibit prolonged, integrated responses to transcranial magnetic stimulation, whereas unconscious states show rapid decay (Casali et al. 2013).

Fifth, wakeful brain activity exhibits scale-free dynamics and long-range temporal correlations (Linkenkaer-Hansen et al. 2001; He 2014). These features are attenuated in unconscious states.

Sixth, metastable coordination and near-critical dynamics are frequently observed in conscious states (Deco et al. 2011; Beggs & Plenz 2003; Shew & Plenz 2013; Chialvo 2010). While the interpretation of criticality remains debated, its association with conscious activity is notable.

Across these diverse findings, a consistent variable emerges: temporal persistence. Conscious states exhibit extended intrinsic temporal organization; unconscious states exhibit collapse or fragmentation of such persistence.

The convergence does not prove that temporal depth is constitutive. Correlation alone is insufficient. Yet the robustness of the pattern suggests that temporal regime is not incidental.

The question, then, is whether this empirical convergence can be integrated into IWMT in a principled manner.

5. Temporal Depth as a Regime Constraint

The empirical convergence outlined above suggests that temporally extended intrinsic dynamics are reliably associated with conscious states. The present proposal interprets this convergence not merely as correlation but as indicative of a necessary regime condition.

Temporal depth, as used here, does not refer simply to slow neural oscillations or low-frequency power. Rather, it refers to the persistence of internally organized activity across extended timescales relative to immediate input fluctuations. A system exhibits temporal depth when its internal state evolution is governed by intrinsic modes that persist, reverberate, and stabilize across hierarchical levels.

Three features characterize temporally deep regimes:

First, intrinsic persistence: internal dynamics decay slowly relative to perturbations. Autocorrelation functions exhibit long tails. Perturbations reverberate rather than dissipate immediately.

Second, hierarchical stabilization: slow intrinsic modes span multiple levels of organization, allowing integration across modalities and timescales.

Third, intrinsic dominance: state evolution is governed primarily by internally structured dynamics rather than by immediate stimulus-driven entrainment.

Temporal Depth, as defined here, should not be conflated with computational or representational temporal depth in generative modeling. In predictive processing and IWMT, temporal depth typically refers to the horizon of counterfactual inference or the hierarchical extent of model-based prediction. In contrast, TD here refers to the weighted contribution of slow intrinsic dynamical modes to overall system activity. It is therefore a property of the system’s physical regime rather than of its representational content.

It is important to note that TD alone is insufficient to characterize conscious regimes. Slow dynamical modes may arise either from endogenous recurrent organization or from sustained external entrainment. A system strongly driven by slow environmental inputs may exhibit prolonged autocorrelation and dominant slow spectral components without possessing intrinsic dynamical autonomy. Therefore, once TD is introduced as a regime metric, a further distinction becomes necessary: the source of slow persistence. Intrinsic Dominance Ratio (IDR) emerges from this requirement. IDR partitions the variance of slow modes into endogenous and exogenous components, ensuring that TD reflects internally sustained dynamics rather than externally imposed structure.

Temporal depth thus captures a dynamical feature of the system as a whole. It is not reducible to a single metric but corresponds to occupation of a particular region of dynamical state space.

The central thesis can now be stated as follows: Integrated generative architectures give rise to conscious world modeling only when they are physically instantiated within intrinsically dominant, temporally deep dynamical regimes, rather than merely exhibiting representational temporal coherence.

The claim is deliberately restricted. Temporal depth is proposed as a necessary condition, not a sufficient one. Generative modeling without temporal depth remains subpersonal. Temporal depth without generative modeling does not suffice for experience. The conjunction matters.

6. Integration with IWMT

IWMT emphasizes three principal components:

1. Hierarchical generative modeling.
2. Multimodal integration.
3. Embodied self-inclusion.

The present proposal adds a fourth:

1. Temporal-depth stabilization.

Generative modeling provides representational structure. Integration provides unity. Embodiment provides perspectival embedding. Temporal depth provides dynamical persistence.

Without temporal depth, generative modeling may compute but fail to stabilize into phenomenally coherent states. Predictive inference may occur locally or transiently, yet without temporally extended intrinsic organization, such inference remains subpersonal.

This addition does not alter IWMT’s computational commitments. It specifies the dynamical phase in which those commitments become phenomenally instantiated.

The result is a completed conditional: consciousness arises only when integrated generative modeling is physically instantiated within intrinsically dominant, temporally deep dynamical regimes, rather than merely satisfying representational coherence.

This formulation resolves the boundary ambiguity identified earlier. Local predictive coding in early sensory cortex operates on short intrinsic timescales and is typically unconscious. Large-scale integrative modeling in association networks operates on longer timescales and becomes a candidate for conscious instantiation. Under anesthesia or deep sleep, intrinsic timescales collapse; world modeling may persist in attenuated form, but without temporal stabilization, phenomenology fails to arise.

The proposal thus distinguishes between computational architecture and dynamical phase. IWMT provides the former; temporal depth specifies the latter.

7. Why IWMT as the Appropriate Architectural Substrate

A question arises: why pursue dynamical completion specifically within the framework of Integrated World Modeling Theory rather than within alternative accounts of consciousness?

The choice of IWMT is not arbitrary. Among contemporary theories, IWMT occupies a distinctive middle position. It is neither a purely informational identity theory, such as Integrated Information Theory (Tononi et al. 2016), nor solely an access-based functional theory, such as Global Workspace Theory (Dehaene & Changeux 2011). Rather, it proposes that consciousness arises from integrated, embodied generative world modeling (Safron 2020). It thus combines architectural richness with an explicit commitment to physical instantiation.

This architectural richness is precisely what renders IWMT dynamically under-specified. Because it situates consciousness within hierarchical generative inference, it risks overgeneralization unless physical boundary conditions are articulated. Generative modeling is ubiquitous across biological systems and increasingly present in artificial systems. Without a regime constraint, the boundary between conscious and unconscious modeling becomes unclear.

By contrast, IIT identifies consciousness with integrated causal structure itself. Within IIT’s framework, Φ is proposed as both necessary and sufficient. Introducing a dynamical regime constraint would either be redundant or in tension with IIT’s sufficiency claim. The present proposal does not aim to redefine the physical substrate of consciousness but to constrain when a given architecture becomes phenomenally instantiated.

Global Workspace Theory, in turn, focuses on access and broadcasting. While temporal persistence may be relevant to ignition stability, GWT is not primarily a theory of physical instantiation but of cognitive availability. A regime-level completion would therefore address a different explanatory level than that which GWT occupies.

Predictive processing frameworks are even broader. They describe a general theory of brain function rather than a specific account of consciousness. Completing predictive processing dynamically would risk becoming a general theory of neural regime selection rather than a targeted refinement of a consciousness theory.

IWMT is uniquely positioned because it explicitly links generative modeling, embodiment, and phenomenological unity. It aims to bridge computational architecture and experiential structure. Yet it does not formalize the dynamical phase in which this architecture becomes conscious. The temporal-depth proposal therefore operates not as an alternative theory but as a completion of IWMT at precisely the level where it is most vulnerable: physical regime specification.

In this sense, the present proposal is not a departure from IWMT but an internal strengthening. It clarifies the physical conditions under which integrated world modeling crosses from computational process to phenomenally instantiated experience.

8. Differential Predictions and Falsifiability

The present proposal yields differential predictions that follow directly from coupling Temporal Depth (TD) with Intrinsic Dominance Ratio (IDR) as conjunctive regime conditions. Unlike accounts that emphasize temporal structure or representational coherence alone, this framework predicts that conscious states must exhibit not only sustained intrinsic timescales, but also intrinsic dominance of those timescales over externally driven dynamics.

First, the model predicts that systems displaying high TD but low IDR will fail to instantiate consciousness. For example, a neural system strongly entrained by slow external inputs may exhibit prolonged autocorrelation and dominant low-frequency spectral power, thereby inflating TD. However, if such slow dynamics are primarily input-locked, IDR will remain low. The theory therefore predicts that externally driven slow activity, even if temporally extended, is insufficient for conscious instantiation. This distinguishes the present account from purely spectral or timescale-based theories.

Second, the model predicts that conscious states will occupy regions of dynamical state space characterized by both elevated TD and IDR. Under deep anesthesia, non-REM sleep, or certain disorders of consciousness, TD may collapse due to shortening intrinsic timescales; alternatively, TD may remain partially preserved but IDR may decrease due to reduced endogenous integration and increased susceptibility to external perturbation. The theory therefore predicts that loss of consciousness may arise via two dissociable mechanisms: (a) collapse of temporal depth, or (b) loss of intrinsic dominance. Empirically, this could be tested by combining perturbational paradigms (e.g., TMS-EEG) with variance-partitioning methods that distinguish endogenous recurrent contributions from stimulus-driven components.

Third, the framework predicts that artificial systems exhibiting temporally coherent world modeling but lacking intrinsic dynamical dominance will fail to satisfy the regime condition. Even if such systems demonstrate counterfactual reasoning and hierarchical prediction (computational temporal depth), their dynamical evolution will remain primarily externally prompted. Thus, the theory generates a discriminative boundary condition absent from architectural accounts alone.

Falsification is straightforward in principle. The model would be undermined if (1) conscious states are reliably observed in systems with low TD and low IDR, or (2) unconscious states consistently exhibit high TD and high IDR. Similarly, if intrinsic–extrinsic variance partitioning fails to distinguish conscious from unconscious conditions across anesthesia, sleep, and disorders of consciousness, the regime constraint would lose explanatory force. In this sense, TD and IDR function not as metaphysical claims but as empirically tractable inequality conditions defining a testable dynamical phase.

Importantly, the proposal does not claim that TD and IDR are sufficient for consciousness. Rather, it asserts that any adequate instantiation of conscious world modeling must satisfy both conditions. This conjunction distinguishes the present framework from theories that rely exclusively on representational temporal coherence or on undifferentiated measures of complexity.

9. Regime Realism and the Limits of Architectural Sufficiency

The deeper philosophical motivation for this proposal lies in the limits of architectural sufficiency. Contemporary consciousness theories often emphasize computational organization, integration, or informational structure. Yet architecture without regime specification risks underdetermination.

Two systems may share causal organization yet differ in intrinsic dynamical phase. One may operate in shallow, rapidly decaying regimes; another in temporally deep, metastable ones. If experience corresponds to regime-level properties, then computational description alone is insufficient.

The present proposal advances a position that may be called regime realism:

Consciousness corresponds to specific dynamical phases of organized matter.

This stance is consistent with non-reductive physicalism (Chalmers 1996; Kim 1998). It does not reduce experience to temporal metrics. It does not claim that slow dynamics are identical to phenomenology. It claims only that experience occurs within certain dynamical regimes and not others.

Regime realism avoids the excesses of panpsychism by introducing boundary conditions. Not all systems with slow dynamics qualify; only those that also instantiate integrated generative world modeling within temporally deep intrinsic regimes do so.

At the same time, regime realism avoids purely computational sufficiency. It insists that physical instantiation matters.

10. Discussion

A persistent worry for any dynamical completion of a consciousness theory is whether it risks redescribing familiar neural correlates under a new label. One might argue that intrinsic timescales, long-range temporal correlations, metastability, and scale-free structure are well-known features of cortical organization and that identifying them as necessary conditions for consciousness adds little beyond correlation. It is therefore important to clarify what the present proposal does—and does not—contribute.

First, the proposal does not merely restate that conscious brains exhibit complex or persistent dynamics. Rather, it reframes this empirical convergence as a regime constraint on generative world modeling. The novelty of the present proposal does not lie in asserting that temporal structure matters for consciousness, nor in introducing temporal coherence into generative modeling frameworks. Rather, it lies in distinguishing representational temporal coherence from dynamical temporal depth, and in formalizing a regime-level constraint that couples Temporal Depth (TD) with Intrinsic Dominance Ratio (IDR). Together, these define a physically specified necessary condition for the instantiation of conscious world modeling — a condition not explicitly articulated within IWMT or TTC.

A potential counterexample concerns psychedelic or dissociative states, in which phenomenology is profoundly altered without complete loss of consciousness. These states sometimes exhibit increased signal diversity or altered spectral properties rather than simple collapse of temporal persistence (Sarasso et al. 2015). However, altered states need not undermine the temporal-depth requirement. The present proposal does not claim that temporal depth must take a single form or remain constant across conscious states. Rather, it claims that some degree of intrinsic persistence and hierarchical stabilization must remain operative. Psychedelic states may reflect reconfiguration of slow intrinsic manifolds rather than their collapse. Indeed, increased dynamical diversity under psychedelics may coexist with preserved long-range intrinsic coordination, consistent with altered yet sustained phenomenology. The decisive contrast is not between different forms of persistence, but between sustained intrinsic organization and rapid dynamical fragmentation, as observed in deep anesthesia or non-REM sleep. Thus, altered states provide an opportunity for refinement rather than refutation of the regime hypothesis.

The empirical case for temporal depth as a regime marker is unusually robust. Under propofol anesthesia, EEG signatures show fragmentation of intrinsic dynamics and shortened autocorrelation structure (Purdon et al. 2013). Recovery of consciousness is accompanied by reinstatement of metastable activity states (Hudson et al. 2014). Deep sleep disrupts long-range temporal dependencies in default mode and attention networks (Tagliazucchi et al. 2013), while effective connectivity collapses (Massimini et al. 2005). Disorders of consciousness show diminished dynamical complexity and impaired coordination across large-scale networks (Demertzi et al. 2019; Boly et al. 2012). Conversely, wakefulness is characterized by scale-free fluctuations and long-range correlations suggestive of persistent intrinsic organization (Linkenkaer-Hansen et al. 2001; He 2014). These findings span neuroimaging modalities, analytic techniques, and clinical contexts.

The theoretical challenge is to explain why such persistence should matter for consciousness. Criticality accounts argue that near-critical dynamics optimize information transmission and dynamic range (Beggs & Plenz 2003; Shew & Plenz 2013; Chialvo 2010). Metastability accounts emphasize flexible coordination among large-scale networks (Deco et al. 2011). Yet neither framework by itself specifies why these properties should instantiate experience rather than merely enhance computation. By embedding temporal depth within a generative world-modeling architecture, the present proposal offers a principled answer: temporally extended intrinsic modes enable hierarchical stabilization of predictive models across time, allowing world representations to persist rather than flicker.

This stabilization is not trivial. Predictive processing models describe inference as ongoing error minimization (Friston 2010; Clark 2013). However, error minimization can occur locally and transiently without yielding phenomenological unity. For experience to arise, modeling must be globally integrated and temporally sustained. Hierarchical intrinsic timescale gradients (Murray et al. 2014; Chaudhuri et al. 2015) suggest that association cortices integrate information over extended windows, supporting cross-modal coherence. When these gradients collapse, phenomenology appears to collapse as well. Temporal depth thus functions as a stabilizing scaffold for integrated world modeling.

Importantly, this view aligns with emerging work on neural population dynamics. Low-dimensional latent manifolds appear to structure cortical activity across tasks (Cunningham & Yu 2014). These manifolds often evolve along slow trajectories, suggesting that cognition unfolds on constrained dynamical surfaces rather than arbitrary fluctuations. Consciousness, on the present account, corresponds to occupation of such intrinsically governed slow manifolds that maintain world-model coherence across perturbations.

The artificial systems case further clarifies the necessity of regime specification. Contemporary large-scale models implement sophisticated generative architectures (Radford et al. 2019; Brown et al. 2020). Yet they typically operate through feedforward or shallow recurrent dynamics, driven by external prompts rather than sustained intrinsic evolution. Without temporally deep intrinsic manifolds governing state evolution independent of immediate input, such systems lack the dynamical persistence hypothesized here as necessary for consciousness. This distinction preserves conceptual clarity while avoiding premature attribution of experience to computational sophistication alone.

From a philosophical standpoint, the regime requirement supports a form of physical constraint realism. Consciousness is not identified with integration or prediction per se, but with these functions realized within specific dynamical phases. This position resonates with non-reductive physicalism (Chalmers 1996; Kim 1998) by insisting that physical instantiation conditions matter without collapsing experience into any single physical variable. It also avoids the inflationary tendencies of panpsychism by introducing boundary conditions: not all temporally persistent systems qualify, only those instantiating integrated generative world modeling within intrinsic-dominant regimes.

A further implication concerns methodological development. If temporal depth is indeed a necessary condition, then its measurement becomes central. Autocorrelation decay constants (Murray et al. 2014), Hurst exponents and 1/f spectral slopes (He 2014), perturbational persistence (Casali et al. 2013), and metastability indices (Deco et al. 2011) are imperfect but tractable proxies. Future research must clarify which aspects of temporal persistence are most predictive of phenomenology. Importantly, the proposal invites multimodal validation: electrophysiology, fMRI, perturbational techniques, and computational modeling can all contribute.

Nevertheless, caution is warranted. Correlation does not entail necessity. It remains possible that temporal depth is an enabling condition rather than a constitutive one. Moreover, different conscious states—such as psychedelic experiences or dissociative anesthesia—may exhibit altered temporal structure without complete loss of experience (Sarasso et al. 2015). Such cases offer critical tests: if altered phenomenology corresponds systematically to altered intrinsic timescale organization, the regime hypothesis gains support; if not, it requires revision.

The proposal also reframes debates about access and phenomenal consciousness. Global Workspace Theory emphasizes ignition and broadcasting (Dehaene & Changeux 2011). Temporal depth may provide the dynamical substrate that allows ignition to persist rather than dissipate. Thus, regime specification complements rather than competes with access-based accounts.

In sum, the dynamical completion of IWMT does not replace architectural insights but grounds them in physical regime constraints. By tying conscious world modeling to temporally deep intrinsic dynamics, the theory offers a coherent synthesis of empirical findings, philosophical considerations, and computational architecture. Its central virtue lies in moving from correlation to constraint: temporal persistence is not merely observed alongside consciousness, but posited as a necessary boundary condition for its physical instantiation.

10.1 Relation to other Theories

The proposal does not displace leading theories of consciousness; rather, it clarifies their physical substrate.

Global Workspace Theory (Dehaene & Changeux 2011) emphasizes ignition and broadcasting. Temporal depth may provide the dynamical stability required for ignition to persist rather than dissipate. Without extended intrinsic modes, global broadcasting might occur transiently without sustaining experience.

Integrated Information Theory (Tononi et al. 2016) focuses on integrated causal structure. However, integration metrics alone do not specify dynamical regime. A highly integrated yet rapidly decaying system may differ phenomenologically from a temporally persistent one. The present proposal suggests that integration must be instantiated within temporally deep intrinsic regimes to become conscious.

Although theories such as the Temporo-Spatial Theory of Consciousness (Northoff & Huang 2017) already emphasize the importance of intrinsic spatiotemporal dynamics, the present proposal does not simply restate that time matters for consciousness. Temporo-Spatial Theory (TTC) emphasizes the constitutive role of intrinsic spatiotemporal organization in consciousness. The present proposal overlaps with TTC at the empirical level insofar as both recognize the importance of intrinsic temporal structure. However, the explanatory roles differ. TTC treats intrinsic temporo-spatial dynamics as primary and constitutive. In contrast, the present framework anchors consciousness in generative world modeling (as in IWMT) and introduces TD and IDR as regime constraints governing when such modeling becomes phenomenally instantiated. The distinction lies not in denying the importance of intrinsic time, but in embedding it within a generative architectural framework and formalizing regime occupation conditions.

In each case, the regime requirement adds a physical boundary condition without negating architectural insights.

10.2 Artificial Systems and Boundary Clarity

The question of artificial consciousness increasingly presses upon theoretical accounts. Large-scale generative systems now construct sophisticated world models (Radford et al. 2019; Brown et al. 2020). If generative modeling and integration alone suffice, the boundary between simulation and experience becomes difficult to articulate.

The temporal-depth constraint introduces a discriminative criterion. Artificial systems that operate primarily via feedforward computation or short-range recurrence lack temporally persistent intrinsic manifolds. Their state evolution is largely stimulus-driven or externally prompted. Under the present proposal, such systems would not satisfy the intrinsic dominance condition (IDR) required for conscious instantiation.

This does not foreclose artificial consciousness in principle. It renders the question empirical: does the system instantiate temporally deep intrinsic generative dynamics? If so, the debate shifts from architecture to regime.

It is important to emphasize that the temporal-depth requirement is substrate-neutral. The proposal does not assume that biological tissue is uniquely privileged. In principle, any physical system capable of instantiating integrated generative world modeling within temporally deep intrinsic regimes would qualify under the present account. However, most contemporary artificial systems operate primarily through externally prompted feedforward inference or shallow recurrence. Their internal state evolution is typically stimulus-driven rather than intrinsically stabilized over extended timescales. Absent persistent intrinsic manifolds governing generative dynamics independent of immediate input, such systems fail the intrinsic-dominance condition articulated here. The proposal therefore does not deny the possibility of artificial consciousness; it introduces a principled dynamical criterion by which such claims could be evaluated.

10.3 What the Proposal Does Not Claim

It is important to state clearly what remains unresolved.

First, the explanatory gap remains. Specifying regime conditions does not explain why experience has qualitative character. It constrains when experience occurs.

Second, the proposal does not reduce phenomenology to time or slow dynamics. Temporal depth is a necessary boundary condition, not an identity claim.

Third, the model does not claim that temporal metrics alone determine conscious richness. Architecture and regime jointly matter.

Fourth, the proposal remains neutral regarding metaphysical debates between reductive and non-reductive physicalism. It is compatible with both, insofar as both accept that physical instantiation conditions are relevant.

The modesty of the proposal is deliberate. Its ambition lies in clarifying the physical phase in which conscious world modeling occurs, not in dissolving the metaphysics of experience.

11. Conclusion

Integrated World Modeling Theory provides one of the most architecturally sophisticated accounts of consciousness currently available. By grounding phenomenology in hierarchical generative modeling, multimodal integration, and embodied self-representation, IWMT offers a compelling synthesis of predictive processing and experiential unity. Yet computational architecture alone does not determine physical instantiation. Without specification of the dynamical regime in which integrated modeling becomes experience, the theory remains underdetermined.

The present proposal advances a regime requirement: any adequate physical theory of consciousness must specify the dynamical phase in which its computational organization becomes phenomenally instantiated. Empirical convergence across anesthesia, sleep, disorders of consciousness, perturbational studies, and intrinsic timescale analyses suggests that conscious states occupy temporally deep intrinsic regimes characterized by sustained internal organization and hierarchical persistence. We have argued that temporal depth constitutes a necessary boundary condition for conscious world modeling.

The proposal remains modest. It does not claim that temporal depth is sufficient for consciousness, nor that it dissolves the explanatory gap. It does not reduce experience to time. It specifies when integrated world modeling becomes phenomenally instantiated and when it does not. In doing so, it strengthens IWMT’s physical grounding while preserving its architectural insights.

Importantly, IWMT’s appeal to temporal coherence concerns the structure of world models, whereas the present proposal concerns the physical regime in which such models operate. TD and IDR therefore do not compete with IWMT’s criteria but constrain their physical realization.

Consciousness, on this view, is not merely integrated modeling. It is integrated modeling stabilized within temporally deep intrinsic dynamical regimes.

References

Alkire, M.T., Hudetz, A.G. and Tononi, G. (2008) Consciousness and anesthesia, Science, 322 (5903), pp. 876–880.

Beggs, J.M. and Plenz, D. (2003) Neuronal avalanches in neocortical circuits, Journal of Neuroscience, 23 (35), pp. 11167–11177.

Boly, M., Seth, A.K., Wilke, M., et al. (2013) Consciousness in humans and non-human animals: Recent advances and future directions, Frontiers in Psychology, 4, pp. 1–20.

Brown, T., Mann, B., Ryder, N., et al. (2020) Language models are few-shot learners, Advances in Neural Information Processing Systems, 33, pp. 1877–1901.

Casali, A.G., Gosseries, O., Rosanova, M., et al. (2013) A theoretically based index of consciousness independent of sensory processing and behavior, Science Translational Medicine, 5 (198), 198ra105.

Chalmers, D.J. (1996) The conscious mind: In search of a fundamental theory, Oxford: Oxford University Press.

Chaudhuri, R., Knoblauch, K., Gariel, M.A., et al. (2015) A large-scale circuit mechanism for hierarchical dynamical processing in the primate cortex, Neuron, 88 (2), pp. 419–431.

Chialvo, D.R. (2010) Emergent complex neural dynamics, Nature Physics, 6 (10), pp. 744–750.

Clark, A. (2013) Whatever next? Predictive brains, situated agents, and the future of cognitive science, Behavioral and Brain Sciences, 36 (3), pp. 181–204.

Cunningham, J.P. and Yu, B.M. (2014) Dimensionality reduction for large-scale neural recordings, Nature Neuroscience, 17 (11), pp. 1500–1509.

Deco, G., Jirsa, V.K. and McIntosh, A.R. (2011) Emerging concepts for the dynamical organization of resting-state activity in the brain, Trends in Cognitive Sciences, 15 (12), pp. 550–560.

Dehaene, S. and Changeux, J.P. (2011) Experimental and theoretical approaches to conscious processing, Neuron, 70 (2), pp. 200–227.

Demertzi, A., Tagliazucchi, E., Dehaene, S., et al. (2019) Human consciousness is supported by dynamic complex patterns of brain signal coordination, Science Advances, 5 (2), eaat7603.

Friston, K. (2010) The free-energy principle: A unified brain theory?, Nature Reviews Neuroscience, 11 (2), pp. 127–138.

He, B.J. (2014) Scale-free brain activity: Past, present, and future, Trends in Cognitive Sciences, 18 (9), pp. 480–487.

Hudson, A.E., Calderon, D.P., Pfaff, D.W. and Proekt, A. (2014) Recovery of consciousness is mediated by metastable activity states, Proceedings of the National Academy of Sciences USA, 111 (25), pp. 9283–9288.

Ito, M. (2008) Control of mental activities by internal models in the cerebellum, Nature Reviews Neuroscience, 9 (4), pp. 304–313.

Kim, J. (1998) Mind in a physical world: An essay on the mind–body problem and mental causation, Cambridge, MA: MIT Press.

Linkenkaer-Hansen, K., Nikouline, V.V., Palva, J.M. and Ilmoniemi, R.J. (2001) Long-range temporal correlations and scaling behavior in human brain oscillations, Journal of Neuroscience, 21 (4), pp. 1370–1377.

Massimini, M., Ferrarelli, F., Huber, R., et al. (2005) Breakdown of cortical effective connectivity during sleep, Science, 309 (5744), pp. 2228–2232.

Metzinger, T. (2003) Being no one: The self-model theory of subjectivity, Cambridge, MA: MIT Press.

Murray, J.D., Bernacchia, A., Roy, N.A., et al. (2014) A hierarchy of intrinsic timescales across primate cortex, Nature Neuroscience, 17 (12), pp. 1661–1663.

Northoff, G. and Huang, Z. (2017) How do the brain’s time and space shape consciousness?, Trends in Cognitive Sciences, 21 (12), pp. 911–926.

Purdon, P.L., Pierce, E.T., Mukamel, E.A., et al. (2013) Electroencephalogram signatures of loss and recovery of consciousness under propofol, Proceedings of the National Academy of Sciences USA, 110 (12), pp. E1142–E1151.

Rao, R.P.N. and Ballard, D.H. (1999) Predictive coding in the visual cortex: A functional interpretation of some extra-classical receptive-field effects, Nature Neuroscience, 2 (1), pp. 79–87.

Safron, A. (2020) An integrated world modeling theory (IWMT) of consciousness, Frontiers in Artificial Intelligence, 3, 30.

Sarasso, S., Boly, M., Napolitani, M., et al. (2015) Consciousness and complexity during unresponsiveness induced by propofol, xenon, and ketamine, Current Biology, 25 (23), pp. 3099–3105.

Searle, J.R. (1992) The rediscovery of the mind, Cambridge, MA: MIT Press.

Shew, W.L. and Plenz, D. (2013) The functional benefits of criticality in the cortex, The Neuroscientist, 19 (1), pp. 88–100.

Tagliazucchi, E., Balenzuela, P., Fraiman, D. and Chialvo, D.R. (2013) Breakdown of long-range temporal dependence in default mode and attention networks during deep sleep, Cerebral Cortex, 23 (8), pp. 1798–1807.

Tognoli, E. and Kelso, J.A.S. (2014) The metastable brain, Neuron, 81 (1), pp. 35–48.

Tononi, G., Boly, M., Massimini, M. and Koch, C. (2016) Integrated information theory: From consciousness to its physical substrate, PLoS Computational Biology, 12 (5), e1003588.

Wolpert, D.M. and Ghahramani, Z. (2000) Computational principles of movement neuroscience, Nature Neuroscience, 3 (Suppl), pp. 1212–1217.