Safron’s Integrated World Modeling Theory

Neuroscientist Adam Safron proposes “Integrated World Modeling Theory” (IWMT) as a unifying model of phenomenal consciousness and conscious access, as explained across computational/functional, algorithmic, and implementational/mechanistic (supervenient) levels of analysis (Safron, 2020, 2022b). IWMT engages Integrated Information Theory (IIT), Global Neuronal Workspace Theory (GNWT), and the Free Energy Principle and Active Inference (FEP-AI) framework, but it is also compatible (and synergistic) with other models such as Recurrent Processing, Predictive Processing, Dynamic Core, Temporo-Spatial Sentience (Northoff et al., 2023), Higher Order Thought, and Attention Schema theories. IWMT shows how different perspectives on consciousness may describe different aspects of minds, understood abstractly/algorithmically as hybrid machine learning architectures, with multiple computational motifs jointly functioning as cybernetic controllers for embodied agents/organisms as they predictively model their environments to adaptively navigate the world (Safron, 2021a; Safron et al., 2022).

Phenomenal Consciousness as Coherent Modeling

IWMT argues that consciousness emerges from processes capable of generating integrated models of systems and worlds with coherence with respect to space (as organized locality), time (as proportional changes in space), and cause (as predictable/modellable regularities in these changes). These coherence-making properties are inspired by Kant’s preconditions for judgment, with spatiotemporal and causal coherence (and potentially integration into unified self-models [c.f. transcendental apperception]) also being necessary for there to be coherently modeled properties, and so any kind of experience whatsoever.

IWMT moves beyond other approaches in suggesting that consciousness is “what it feels like” to generate coherent system-world estimates with sufficient rapidity that they can both inform and be informed by action-perception cycles on the timescales over which they evolve as embodied agents interact with their environments.

In other words, IWMT views phenomenal consciousness as a computational object akin to a “deep fake” created by generative AI, but where the information generated is sensorium states for embodied organisms with various combinations of modal features (e.g., sight, touch, sound, interoception), rather than pixel arrays.

While IWMT currently does not have a clear position on the range of physical implementations capable of entailing such generative modeling, the theory attempts to identify both necessary and sufficient conditions for consciousness by identifying the coherence-making properties and computational principles required to generate subjectivity as a stream of experience, organized according to coherently structured perspectival reference frames (thus being compatible with views emphasizing projective geometry as a basis for visuospatial awareness).

By focusing on the generation of likely patterns of sense data that correspond to all the various aspects/qualities of embodied experience, IWMT may help answer the (often begged) question as to why processing would entail consciousness, rather than happening “in the dark.” Functionally speaking, phenomenal consciousness is proposed to constitute a kind of evolutionary adaptation and data structure that combines different sensory modalities into an iteratively estimated, unified field of experience. It is highly adaptive—in terms of more intelligent action selection and learning—for an agent to have a spatiotemporally and causally coherent model of self and world, organized according to egocentric perspectival reference frames, with organism-salient/relevant features being given greater attention based on experience.

Mechanisms and Theoretical Integration

IWMT identifies points of convergence between IIT and GNWT and attempts to reconcile conflicting claims regarding physical substrates of consciousness. Mechanistically, IWMT proposes that “self-organizing harmonic modes” (SOHMs) are synchronous complexes of neural activity that emerge as metastable attractors within brain networks, functioning as both dynamic cores of integrated information as well as workspaces. With respect to the inconclusive adversarial collaboration between IIT and GNWT, both theories are proposed to be correct in their claims about whether consciousness is realized by either a “posterior hot zone” (understood as SOHMs forming/evolving at alpha frequencies) or by a broader functional network involving the frontal lobes (as SOHMs potentially requiring synchronization at theta frequencies). In these ways, IIT and GNWT are suggested to be making differing and potentially complementary claims about the systemic properties required to generate different aspects of experience/cognition (i.e., phenomenal consciousness or conscious access).

Applications and Extensions

IWMT has been further developed to inform models of intentional goal-oriented behavior (as enacted imaginings [Safron, 2021a]), high-level control for robotic systems (as generalized navigation [Safron et al., 2022]), and to explain a diverse range of psychedelic phenomena (as kinds of waking dreams [Safron et al., 2025]).

While not yet precisely specifying the physical/computational substrates of consciousness (Fields et al., 2025; Safron, 2021b), IWMT makes significant inroads towards explaining how there could be “something that it feels like” to be a physical process, such that conscious states may be helpfully altered and potentially realized in artificial (body-)minds (Safron, 2022a).