Hierarchical Conscious Selves

Updated 12 November 2025

Hierarchical conscious selves are defined as nested layers of cognitive and neural processes, ranging from subliminal routines to meta-reflective awareness.
They integrate insights from neuroscience, evolutionary biology, and computational models to explain how information is broadcast and processed across self-levels.
These models inform both biological and potential machine consciousness, emphasizing structured representations that enable flexible, adaptive behavior.

Hierarchical levels of conscious selves refer to the organization of self-related cognitive processes, neural architectures, and computational models into discrete strata, each associated with different representational, functional, and phenomenological properties. Contemporary research integrates perspectives from neuroscientific network analysis, cognitive agent architectures, formal policy-based modeling, evolutionary building blocks, and introspective phenomenology to characterize these levels. These insights point to nested, interacting layers ranging from subliminal or reflexive routines to meta-conscious self-reflection, spanning both biological and potential machine consciousness.

1. Foundational Theories and Definitions

The concept of hierarchical selves arises in several traditions. In Global Workspace Theory (GWT), conscious access involves the wide broadcasting of select contents to distributed neural “hubs,” while non-conscious processing remains local and modular. Wiersma delineates three principal levels: subliminal/preconscious, conscious/cognitive, and meta-conscious/reflective selves, each defined by scope of broadcast, type of information processing, and reportability (Wiersma, 2017).

Other frameworks distinguish between phenomenal consciousness (the existence of subjective “what it’s like” experience) and access consciousness (contents are available for report and deliberation). Formalizations of self-models articulate first-order (phenomenal core) selves, second-order (social/other) selves, and third-order (meta-self-models), with increasing representational recursion (Bennett et al., 22 Sep 2024). Evolutionary and comparative approaches identify a ladder of subjective capacities, each corresponding to distinctive building blocks of experience and selfhood (Spencer, 9 May 2024).

Network neuroscience leverages structural properties such as k-shell cortical decomposition to map dense, integrative “nucleus” regions to higher-order conscious processing, with surrounding shells assigned to progressively lower levels of information integration (Lahav et al., 2018).

2. Levels of Self and Conscious Processing

2.1 Subliminal, Preconscious, and Conscious Selves

Subliminal (Subconscious) Self: Inputs remain local, fail to reach the global broadcast hubs associated with the GW, evoke negligible emotional intensity (E ≈ 0), and dissipate rapidly. Such processing supports transient priming but is not reportable and cannot anchor deliberate cognition (Wiersma, 2017).
Preconscious Self: Overlearned or routine stimuli are processed in short-term buffers, running multiple unobtrusive cycles below the threshold of reportability. These streams exhibit low emotional intensity (E ≈ 1) and cognitive effort (C ≈ 1), yielding moderate sustainability (S = E/C ≈ 1), facilitating automatic, expert performance (Wiersma, 2017).
Conscious (Cognitive) Self: Novel, salient, or emotionally charged stimuli cross the GW bottleneck and are broadcast via “rich-club” hubs. These inputs are highly sustainable (E ≈ 4, C ≈ 2, S ≈ 2), support deliberate reasoning, flexible integration, and rich phenomenal experience. They dominate cognition over seconds or longer, allowing for integrative, temporally extended thought (Wiersma, 2017).

2.2 Reflective and Meta-Conscious Levels

Meta-Conscious Self: This level supports awareness of being aware—explicit re-representations of conscious contents. Features include temporal dissociation (meta-conscious events follow conscious cycles), heightened cognitive effort, reduced emotional vividness, and low sustainability (high C, low E, thus S < 1). Meta-conscious episodes exert top-down regulatory control: they can interrupt ongoing routines, re-focus attention, and modulate future emotional appraisal (Wiersma, 2017).

3. Structural–Functional Hierarchies in Neural and Computational Architectures

Hierarchical organization in neural circuits and computational models underpins the emergence of layered conscious selfhood.

Cortical k-shell Decomposition: The human cortex, via k-shell analysis, reveals three nested hierarchies:
- Low hierarchy (k ≤ 15): Perceptual and associative regions (e.g., fusiform gyrus), supporting raw sensory processing.
- Middle hierarchy (k = 16–18): Multimodal integration and executive functions (e.g., Broca’s area, right dorsolateral PFC).
- High hierarchy/nucleus (k = 19): Default Mode Network (DMN), salience network, and integrative hubs (medial PFC, precuneus) implicated in core conscious processes (Lahav et al., 2018).

The nucleus exhibits high self-connectivity (72% ±1.6%), short average path length (L ≈ 2), and high clustering (C ≈ 0.42), matching conditions for a global workspace.

Dual-Hierarchy Information Processing: Each neuron acts within strongly connected delay-coupled subgraphs (SCDSGs), supporting non-binary multi-hypothesis decision making. Ensembles of SCDSGs create external hierarchies (E^k: objects → events → narratives → scenarios) and internal/mental hierarchies (M_total ≅ P(E_total)), the latter strictly richer due to power set cardinality (Grindrod, 2016). Interactions between layers allow for recursive construction and re-conjuring of subjective experience.
Agent-Based and Formal Policy Models: In agent architectures, three broad strata are defined—reactive, deliberative, and reflective/meta-management—each with a functional correlate in conscious self-organization. Hierarchical self-models as formal policies (first-order self, model of others/world, self-as-modeled-by-others) underpin access consciousness (Bennett et al., 22 Sep 2024). Transitions depend on natural selection for increasingly rapid, valence-driven weak-policy optimization.

4. Evolutionary Ladders and Building Blocks

Spencer identifies eight building blocks, each corresponding to a level of conscious self, leading to full human-style emotive consciousness (Spencer, 9 May 2024):

Level	Defining Capability	Example Species
Social Signaller	Intra-species communication	Honeybees, frogs, primates
Scripted Agent	Hard-wired scripts (responsive memory)	Migratory birds, insects
Learning Agent	Acquisition of new routines	Cats, rats, pigeons
Memory Optimizer	Unconscious engram replay (dreams)	Mice, birds, mammals
Social Observer	Sentinel and vicarious state detection	Wolves, horses, dogs
Self-Aware Entity	Mirror self-recognition	Chimpanzees, dolphins
Mind-Reader	Theory of Mind	Chimpanzees, ravens
Emotive Self	Full narrative, empathy, creativity	Homo sapiens

Each successive level introduces a distinct subjective capacity, with evolutionary transitions underpinned by selective pressures for learning, social coordination, and internal self-modeling. Flat (deficit) consciousness represents a loss of the highest level, with preservation of lower blocks.

5. Formal Criteria, Sustainability, and Model Implications

Engagement Criteria (Input-Based Entry):
- Stimulus familiarity: Overlearned patterns promote preconscious processing.
- Emotional intensity: Drives conscious access, with appraisal evaluating valence.
- Cognitive effort: Allocation of high-effort streams only if justified by emotional import (Wiersma, 2017).
Quantitative Sustainability:
- Sustainability $S = \frac{E}{C}$ , where E is emotional intensity (0–4), C is cognitive effort (0–4) (Wiersma, 2017).
- Subliminal streams: S ≈ 0, preconscious: S ≈ 1, conscious: S ≈ 2.
Recursion and Power of Mental Representation:
- Internal hierarchies always exceed external hierarchies in cardinality (|P(E_total)| > |E_total|), entailing that no finite automaton can learn the full space of subjective experience (Grindrod, 2016).
Self-Model Chains:
- First-order self (phenomenal), embedded recursively as second- and third-order models (agent models, self-modeled-by-other, meta-cognition), each requiring prior construction for the next (Bennett et al., 22 Sep 2024). This chain underpins the distinction between phenomenal and access consciousness.

6. Comparative, Phenomenological, and Machine Architecture Perspectives

Subliminal and Supraliminal Layers: Drawing on introspection, CogAff architecture, and Aurobindo’s taxonomy, layers beneath waking consciousness (submental, true subconscious, subliminal proper, superconscient) offer domain-specific processing inaccessible to report but influential on behavior and phenomenal intuitions (Kvassay, 2019). These layers can be mapped to functional architectures in artificial agents.
Thresholds and Integration Weights: Conscious states may be modeled as weighted integrals over proto-qualia generated at each depth, with thresholds θ_i and integration weights α_i modulating penetration and dominance of layers (Kvassay, 2019).
Implications for Machine Consciousness: Given a physical architecture supporting unrestricted functional roles, coexisting reflective and deliberative processes (including “proto-qualia”) could in principle instantiate multiple layers of conscious selves (Kvassay, 2019). However, absence of internal hierarchical recursion (as in finite automata) precludes open-ended subjective experience.

7. Theoretical and Practical Consequences

Hierarchical models of conscious selves support several key implications:

Cognitive Economy: Low-level processing conserves resources for routine tasks; conscious broadcasting enables flexible, integrative operations; reflective oversight provides adaptive self-regulation (Wiersma, 2017).
Evolution and Natural Selection: Selection favors architectures that accelerate generalization and complexity of self-models; deeper hierarchies correspond to increasing repertoire and flexibility of subjective experience (Bennett et al., 22 Sep 2024).
Neural Correlates of Consciousness: Cortical shell hierarchies correspond closely with functional neuroanatomy of default-mode and executive networks. High hierarchy/nucleus alignment with consciousness-related regions is statistically robust ( $r_{\rm observed}\approx 0.92$ ), and randomization tests reveal strong non-random overlap (Lahav et al., 2018).
Limits of Algorithmic Systems: No finite automaton, regardless of sophistication, can exhaust the full combinatorial richness of internal (mental) hierarchies; conscious experience is necessarily open-ended and “never in totality, not even in a lifetime” (Grindrod, 2016).
Computational Modeling: Architectures such as LIDA and Router instantiate multi-level routines and reflective layers, suggesting possible pathways for artificial implementation and further refinement of hierarchy-based consciousness accounts (Wiersma, 2017). Formal policy-based models specify precise criteria for transitions between levels and provide a foundation for further mathematical treatment (Bennett et al., 22 Sep 2024).

A comprehensive analysis of hierarchical conscious selves thus integrates neuroscientific, computational, evolutionary, and phenomenological dimensions, elucidating the multiple nested, interacting layers that collectively constitute conscious selfhood.