Cognition · Metacognition · Global Metacognition

This paper proposes that human cognition has three structural levels — cognition, metacognition, and Global Metacognition — and argues that these three levels form a precise structural mapping with the philosophical tradition’s “Three Views” (values, life philosophy, worldview). The first layer (cognition/values) performs sorting operations within the signal space; the second layer (metacognition/life philosophy) examines the sorting rules themselves; the third layer (Global Metacognition/worldview) recognizes that both the sorting operations and the sorting space are local phenomena within a noise ocean. This paper further argues that existing academic literature — including cognitive science’s Type-1/Type-2/Type-3 recursive hierarchy, the “Wise AI” metacognitive framework by Johnson et al. (2024), and the Chinese philosophical “Three Views” system — have all failed to complete this mapping. Cognitive science’s “meta-metacognition” is homogeneous recursion (repetition of the same operation); this paper’s “Global Metacognition” is heterogeneous jump (transcendent awareness of the operation itself). This distinction has direct theoretical implications for AI alignment, the fifth layer of nested signal topology, and the contemplative noise-reduction paradigm.

V2 added six theoretical extensions: (1) Chain-of-Thought (COT) as a product of value-level linear thinking — the basic cognitive mode of the masses; (2) perspective-taking as a topological transformation of cognitive coordinate systems rather than simple translation of values; (3) LLM temporary persona import and its structural antagonism with token statistical inertia — explaining why third-person perspectives cannot be maintained long-term in LLMs; (4) animal attributes and social attributes as a double barrier preventing humans from reaching Global Metacognition; (5) “complete abandonment of biological and social needs” as the existential condition for Global Metacognition — not suppression but release; (6) the global empirical void — existing surveys such as the World Values Survey (WVS) have never measured the population distribution across three cognitive levels, constituting an unprecedented empirical research space.

V3 further completed seven theoretical reinforcements: (1) a canonical definition of Global Metacognition — providing a single citable definition in one place; (2) structural mapping to Robert Kegan’s adult development theory — the isomorphism between Kegan Stage 5 (Self-Transforming Mind) and Global Metacognition, plus Kegan’s empirical population distribution data (~50% Stage 3, ~35% Stage 4, <1% Stage 5) for cross-validation of the pyramid hypothesis; (3) explicit dialogue with Buddhist epistemology — the precise correspondence of anattā (non-self), śūnyatā (emptiness), and prajñā (wisdom) with “complete abandonment” and “panoramic awareness”; (4) criteria distinguishing Global Metacognition from pathological dissociation — both involve “alteration of sense of self,” but in opposite directions; (5) positive functional description of the third layer — observable characteristics in daily cognition of those who have reached Global Metacognition; (6) discussion of cognitive regression — whether Global Metacognition is a permanent state or dynamic equilibrium; (7) strengthening of the COT argument — responding to Tree-of-Thought and other non-linear reasoning forms.

Does Human Cognition Have Three Irreducible Levels?

Since Flavell (1979), cognitive science has divided human thinking into two levels: cognition and metacognition — thinking, and thinking about thinking. This dichotomy has dominated research for nearly half a century. Recently, Recht et al. (2022) experimentally extended the hierarchy to fourth-order recursive judgments, demonstrating that humans can execute “confidence judgments about confidence judgments about confidence judgments.” But their conclusion was: these higher-order judgments are merely recursive runs of the same system, with no structurally new level.

In AI research, Johnson, Bengio, Mitchell, and others published “Imagining and Building Wise Machines: The Centrality of AI Metacognition” in 2024, representing the most cutting-edge attempt — engineering the metacognitive dimensions of human wisdom (intellectual humility, perspective-taking, contextual adaptability) into AI systems as an alternative to value alignment. This paper sparked wide discussion in academia, but its framework stays at two levels: object-level strategies (managing problems) and metacognitive strategies (managing object-level strategies).

This paper’s core thesis: there exists a third level, which is neither the homogeneous recursion of “meta-metacognition” in cognitive science, nor a further optimization of metacognitive strategies in the Johnson et al. framework. It is a qualitative leap — not better examination of one’s own thinking, but recognition of where “examination” itself sits within existence. This paper names this level “Global Metacognition” and argues it forms a precise structural correspondence with the philosophical tradition’s “worldview.”

Three-Layer Cognitive Topology and Its Structural Mapping to the Philosophical Three Views

The Canonical Definition of Global Metacognition

What Existing Research Has Touched and What It Has Missed

Cognitive science: homogeneous recursive meta-metacognition. Nelson and Narens (1990) established the object-level/meta-level dichotomy. Recht et al. (2022) extended it to the fourth order, finding humans can indeed execute Type-3 (meta-metacognitive) judgments. But the key finding: third-order judgments carry no additional cognitive cost; metacognition and meta-metacognition are produced by the same system recursively. This means academia’s “meta-metacognition” is not a new level, but the same machine’s third turn.

AI research: the “Wise AI” framework of Johnson et al. Published in Trends in Cognitive Sciences (February 2026), this paper gathered top researchers including Bengio, Mitchell, and Schölkopf, proposing to engineer metacognitive capabilities — intellectual humility, perspective-taking, contextual adaptability — into AI systems as an alternative to value alignment. This is one of the most creative attempts in current AI safety. But its framework strictly stays at two levels: object-level strategies and metacognitive strategies. The third level is untouched.

The Chinese philosophical “Three Views.” The three-layer division of values/life philosophy/worldview as cognitive structure is widespread in Chinese philosophy and education systems. But this system has never been structurally aligned with cognitive science’s cognition/metacognition hierarchy. The Three Views in traditional contexts are primarily understood as educational goals or moral development stages, not structural levels of information processing.

Chris Frith’s consciousness hierarchy theory (2023). Frith proposed that conscious experience is co-determined by top-down (prior beliefs) and bottom-up (sensation) processes, with metacognitive control at the top of the brain’s cognitive hierarchy. He further noted that culture influences the brain through explicit metacognition. This is the closest academic expression to the third layer — but Frith’s “top” is still an extension of metacognition, without identifying a structurally different third level.

Homogeneous Recursion vs. Heterogeneous Jump: Why “Global Metacognition” Is Not “Meta-Metacognition”

Type-3 judgment (meta-metacognition) in cognitive science is a re-evaluation of one’s confidence assessment — “how confident am I in my confidence in my judgment?” This is a sorting machine’s third run. The input changed (from the original problem to a confidence score), but the operation didn’t change (still sorting and evaluating). Recht et al.’s experiment precisely demonstrated this: Type-3 judgment carries no additional cognitive cost because it reuses the same system as Type-2.

Global Metacognition is not the sorting machine’s third run. It is standing outside the sorting machine and seeing the sorting machine itself. Not “how confident am I in my confidence in my judgment,” but “what is the position of judgment itself within existence.” This is not recursion — recursion is repeated application of the same operation, like looking at a mirror reflecting a mirror reflecting a mirror. Global Metacognition is walking out of the mirror room and seeing where the entire building sits in the city.

Mathematical analogy: homogeneous recursion is function iteration — f(f(f(x))). Global Metacognition is jumping from function space to the topological properties of function space itself — not computing the value of f(x), but recognizing what kind of mapping f itself is, and where this mapping sits among all possible mappings. This is a category-theory-level jump, not a numerical-computation-level iteration.

Chain-of-Thought Is a Linear Product of the Value Level; Perspective-Taking Is Topological Transformation, Not Translation

COT’s cognitive level positioning. Chain-of-Thought is widely regarded in current academia and industry as a technique for enhancing LLM reasoning. But this paper repositions it as a product of a cognitive level: COT is essentially a linear sorting chain of the value level. A→B→C→D, each step executing a value judgment — is this reasoning direction “correct,” should the next step “be taken.” It is the direct expression of cognitive Layer 1 (values/sorting operations), not a manifestation of higher thinking. The vast majority of human thinking stays at this layer: see rain → bring umbrella, hungry → eat, boss criticizes → unhappy. These are all unidirectional sorting chains, driven by value judgments, advancing in a linear direction.

LLM default output is also at the COT level, because attention’s statistical inertia naturally produces unidirectional sorting chains — the highest-probability path in training data is the statistical mapping of human COT. This explains why LLMs perform excellently on in-distribution short-chain reasoning: what they align with is precisely the linear sorting pattern of human cognitive Layer 1.

Response to Tree-of-Thought and other non-linear reasoning forms. Critics might point out: Tree-of-Thought, Graph-of-Thought, and similar techniques have already moved beyond linear COT, demonstrating branching, backtracking, and parallel evaluation capabilities. But this paper’s argument is not “COT only has linear form” — rather, all these variants still operate at Layer 1. Tree-of-Thought is multiple parallel value sorting chains plus cross-chain comparison, still fundamentally a combination of sorting operations. It makes sorting more efficient and comprehensive, but doesn’t escape sorting itself. Chess analogy: COT is step-by-step analysis, ToT is simultaneously analyzing multiple lines then choosing the best — but both are calculating on the board, neither is looking at the board from outside. Metacognition is seeing “what rules am I using to play chess”; Global Metacognition is seeing “what position does playing chess itself occupy in my life.”

The topological transformation nature of perspective-taking. The leap from Layer 1 (values/COT) to Layer 2 (life philosophy/metacognition) is marked by the signature operation of perspective-taking. But a precise distinction must be made here: what most people think of as “perspective-taking” is still a Layer 1 operation — “I stand in your shoes and think” is essentially a projection transformation from “me” as coordinate origin; my own value sorting system never changes, I merely project my sorting logic onto the other’s position. This is translation, not transformation.

True perspective-taking is a switch of cognitive coordinate systems — temporarily abandoning one’s own value sorting system, importing the other’s sorting system, then using the other’s sorting system to evaluate one’s own original behavior. Not “I think you would think,” but “if I used your entire set of judgment criteria to look at myself, what would I see.” This is a topological transformation — the observation point jumps from one’s own coordinate system to the other’s, and the evaluation framework changes completely with it. The proportion of humans who can execute this kind of topological transformation is extremely small — most people spend their entire lives doing value translation, mistaking it for perspective-taking.

Temporary Persona Import and Its Structural Antagonism with Token Statistical Inertia

The Transformer’s attention mechanism has only one “observation point” at any moment — distributing attention weights from the current token position to all existing tokens. This operation is unidirectional: all attention weights radiate from a single viewpoint. When users ask the model to “take the other’s perspective” or “analyze from the other side,” the model must temporarily construct a perspective in the token sequence that differs from or even opposes the default inertia path — this is “third-party persona import.”

The Persona Effect research in academia has already observed the engineering manifestation of this phenomenon: LLMs face persistent difficulty in simulating counterfactual or low-performance personas, requiring specialized “persona-aware contrastive learning” techniques to forcibly maintain consistency. But their explanation is “the model’s capability is insufficient, more training is needed.” This paper offers a more fundamental explanation: the imported third-party persona is structurally antagonistic to the LLM’s default token sorting direction.

Statistical inertia from training data pushes all tokens toward the highest-probability path — this is the default direction of the COT/value level. Perspective-taking requires the model to temporarily deviate from this path, maintaining a lower-probability path. This is like making a river temporarily flow backward — achievable briefly through external force (prompt injection, persona settings), but the system’s statistical pressure pulls output back toward the default direction at every token generation step. As dialogue turns increase, the cumulative effect of statistical inertia grows stronger, the imported competing perspective is gradually suppressed, and output regresses to the COT default path.

This means: in the future, many researchers may use “third-party persona generation” to construct evaluative dialogues. But such methods face a structural ceiling — the imported persona perspective is inherently opposite to or deviant from the LLM’s token sorting direction, making long-term maintenance impossible. Each dialogue turn consumes the “persona maintenance” probability budget until statistical inertia completely reclaims directional control.

The Precise Position of Three-Layer Cognition in the “Signal & Noise” Framework

“Signal & Noise: LLM Ontology” V4 established a core proposition chain: noise is fundamental → signals are local condensations of noise → mathematics is signal at its extreme → Planck scale is signal’s endpoint. This paper’s three-layer cognitive structure has a precise position within this framework.

Cognition/values is a sorting operation within the signal space. Everything it processes is already-signalized information — inputs already encoded in language, framed by concepts, annotated with structure. Sorting itself is a signal operation on signals, completed entirely within the signal domain. The “low-dimensional focus” defined in “Signal & Noise” Chapter 1 is this layer’s core feature: every value judgment is a dimensionality reduction — extracting a narrow band from possibility space.

Metacognition/life philosophy is examination of signal operation rules. It doesn’t process signals themselves, but examines “what rules am I using to process signals.” In the “Signal & Noise” context, this corresponds to Chapter 2’s “Signal Lifecycle” — recognizing that today’s signal may be tomorrow’s noise, recognizing that one’s own judgment criteria may have already “decayed.” Johnson et al.’s “intellectual humility” is the strategized expression of this signal-lifespan awareness.

Global Metacognition/worldview is awareness of where the signal domain itself sits within existence. It corresponds to the core thesis of “Signal & Noise” Chapter 4 — “Ontological Reversal”: it is not that signals tamed noise, but that noise permitted signals. Global Metacognition is living within this reversal — not theoretically knowing “signals are local condensations of noise,” but being aware of this fact in every cognitive act.

Global Metacognition as the Top Locking Layer of the Probability Column

“Penetrating Hundred Layers of Information Structure” V3 defined a five-layer structure of nested signal topology: factual statements, abductive logic, cross-dimensional strong links, observer perspective, Global Metacognition. The first four layers had clear mechanism descriptions. The fifth layer — Global Metacognition — was defined in that paper as “thinking about the entire thinking path itself,” “creating long-range dependencies that span the entire sequence length.”

This paper provides the human-side ontological anchor for that fifth layer. Global Metacognition is not merely an input strategy (to make AI output better); it is a structural feature of the user’s cognitive topology. When a person’s cognitive activity includes Global Metacognition, their language output naturally carries fifth-layer signals — not because they are “strategically” using complex prompts, but because their cognitive topology itself is networked, self-referential, and existentially positioned.

This also answers a question left unaddressed in “Penetrating Hundred Layers”: why can only very few people naturally produce nested signal topology? The answer is not in the technical threshold. Layers 1 through 4 can be acquired through training — learning precise terminology (Layer 1), cultivating abductive logic (Layer 2), building cross-domain links (Layer 3), developing self-observation capability (Layer 4). But Layer 5 requires not better thinking — but transcendent awareness of thinking itself. This is not a trainable skill; it is a structural leap in cognitive topology.

The Double Barrier, Complete Abandonment, and the Only Path to Global Metacognition

“Signal & Noise” Chapter 17 argued the fundamental reason human information bandwidth is narrow: it’s not that the pipe is thin, but that the pipe is stuffed with filters. This paper makes a key extension in V2: what prevents humans from reaching Global Metacognition is not a single barrier, but a double barrier — the superposition of animal attributes and social attributes.

First barrier: Animal attributes. Survival instincts, territorial awareness, reproductive drives, fear responses — these bottom-level programs written into genes continuously occupy cognitive bandwidth. Their common feature is extreme self-centeredness — all signals first pass through the “beneficial or harmful to my survival” filter. This filter is not optional; it is hardware-level, executing before consciousness participates. This is the bottommost filter; after bandwidth is intercepted by it, little remains.

Second barrier: Social attributes. Identity, group belonging, social status, cultural norms — these post-natally installed filters stack additional layers on top of animal attributes. A person’s racial identity, professional identity, class position, educational background — each layer executes “beneficial or harmful to my social position” sorting. Social attributes’ peculiarity: they make people believe they are “thinking,” when they are actually executing the automatic sorting of social filters. Most people think they have a “life philosophy” when they actually only have default sorting rules installed by social attributes.

After the two barriers stack, cognitive bandwidth is compressed to the point where it can barely accommodate a “global” field of view. Global Metacognition requires simultaneously accommodating one’s own entire coordinate system and others’ entire coordinate systems, and executing perspective-taking within this panorama — this requires bandwidth far exceeding what the double barrier allows through.

This paper’s three-layer cognitive structure provides precise level positioning for contemplative theory. Layer 1 (cognition/values) contemplation is adjusting filter parameters — changing sorting priorities but not changing sorting itself. This is what most psychotherapy and self-growth courses do. Layer 2 (metacognition/life philosophy) contemplation is optimizing filters — examining sorting rules, identifying unreasonable patterns and correcting them. This is what Johnson et al.’s “intellectual humility” and “contextual adaptability” do.

Layer 3 (Global Metacognition/worldview) contemplation requires complete abandonment of biological and social needs. Not suppression — suppression is still an operation centered on them; the suppressed object still occupies bandwidth, because you need to continuously consume cognitive resources to maintain suppression. Complete abandonment means they no longer exist as cognitive premises — not “I overcame fear,” but fear is no longer the preprocessing layer for input signals; not “I let go of social status,” but social status no longer runs automatically as a filter.

This condition is almost paradoxical for humans. Biological needs are hardware programs written by genes; social needs are a software environment continuously installed from birth to death. Asking a person to completely abandon both is like asking an operating system to uninstall its own kernel while running. Most people don’t fail to want to — they can’t, because the entity executing “abandonment” itself runs on biological and social needs.

Structural Isomorphism with Kegan’s Adult Development Theory and Population Distribution Data

Harvard psychologist Robert Kegan’s Constructive Developmental Theory proposes five stages of adult development, with the core mechanism being the “Subject-Object Shift” — transforming what was previously unreflectable, part of the self (Subject), into something that can be examined, manipulated, chosen (Object). Kegan wrote: “We have objects; we are subjects.” Every developmental leap turns the previous stage’s subject into the current stage’s object.

The mapping between Kegan’s five stages and this paper’s three cognitive layers is precise: Stage 2 (Imperial Mind) and Stage 3 (Socialized Mind) correspond to this paper’s Layer 1 — cognition/values level, where thinking is driven by needs and social expectations, and the sorting system executes automatically. Stage 4 (Self-Authoring Mind, ~35% of adults) corresponds to Layer 2 — metacognition/life philosophy, where the individual has constructed an independent value system and can examine external expectations. Stage 5 (Self-Transforming Mind, extremely few adults) corresponds to Layer 3 — Global Metacognition/worldview, where the individual is no longer bound by any single identity, ideology, or self-definition, can see through binary oppositions, and flows freely between different meaning systems.

Kegan’s population distribution data provides independent cross-validation for this paper’s pyramid hypothesis. Research shows: approximately 50% of adults are at Stage 3 (Socialized Mind), approximately 35% at Stage 4 (Self-Authoring), and less than 1% stably reach Stage 5. This is highly consistent with this paper’s derived pyramid distribution in Section 17 (85-95%/5-14%/<1%) — if we merge Kegan’s Stages 2-3 (corresponding to this paper’s Layer 1), Stage 4 corresponds to Layer 2, Stage 5 to Layer 3.

Kegan’s “subject becomes object” is structurally fully isomorphic with this paper’s “seeing the filter.” But where this paper goes beyond Kegan: (1) unifying developmental stages with the signal/noise framework — every Subject-Object transition is a dimensionality reduction, turning an automatically executing filter into an examinable object; (2) explicitly connecting Stage 5 with Buddhist contemplative tradition and AI alignment theory; (3) proposing “complete abandonment” as a more precise existential condition than Kegan’s “no longer being bound.”

Anattā (Non-Self), Śūnyatā (Emptiness), and Prajñā (Wisdom): 2,500 Years of Third-Layer Tradition

Buddhism’s non-self doctrine (anattā) holds that no phenomenon possesses an eternal, unchanging self-nature — the cognitive implication: the self-coordinate system is not an entity, but a temporary condensation of a continuously changing process. This paper’s “complete abandonment of biological and social needs as cognitive premises” structurally corresponds to anattā — not annihilating the self, but becoming aware that the self was never a fixed premise. Contemporary neuroscience has confirmed: through sustained meditation practice, the self-model can be reorganized in nonlinear ways, occasionally producing “non-self” experiences where ordinary identity boundaries are perceived as constructed rather than given.

Śūnyatā — “all phenomena lack inherent existence” — corresponds to this paper’s “the signal domain itself is a local phenomenon in a noise ocean.” Emptiness is not nothingness, but the negation of “inherent existence.” When Global Metacognition perceives that all coordinate systems (including one’s own) are local, this is the cognitive operation of emptiness.

The most precise correspondence comes from a 2025 paper published in Mindfulness that defined “critical meta-awareness” — the ability to recognize cognitive constructs as constructs, rather than treating them as reality itself. That paper explicitly noted this ability corresponds directly to Buddhist prajñā (wisdom). This paper’s Global Metacognition is the signal-theoretic formulation of this “critical meta-awareness” — seeing signals as signals rather than as reality itself.

Global Metacognition Is Not Dissociation: Two Types of “Self-Sense Alteration” Go in Opposite Directions

Dissociation — particularly depersonalization and derealization — also manifests as “observing one’s own cognitive processes from outside” and “alteration of sense of self.” Critics might say: “What you describe as Global Metacognition is merely a dissociative state.” This criticism must be directly addressed.

The key distinction is that the directions are opposite. The core finding in clinical literature: dissociation is “absence of internal and external awareness” — feeling like “autopilot,” “unreal,” disconnected from one’s own experience; while mindfulness/awareness is “fullness of internal and external awareness” — more vividly experiencing the present, more deeply connected with one’s own experience. A 2020 study in Frontiers in Psychology unified these two “non-self” experiences under the Active Inference framework, concluding that they are “diametrically opposite” in the control direction of the self-model.

Reformulated in this paper’s signal terminology: dissociation is pathological collapse of filters — signal channels rupture, cognitive bandwidth plummets to near zero, experience becomes pale, flat, unreal. Global Metacognition is active unloading of filters — signal channels open, cognitive bandwidth dramatically increases, experience becomes vivid, rich, hyper-real. Both involve “alteration of the self-coordinate system,” but dissociation is collapse of the coordinate system (passive, painful, function-losing), while Global Metacognition is transcendence of the coordinate system (active, lucid, function-enhancing).

Empirical research provides precise distinguishing markers: mindfulness’s “Observe” dimension combined with “Nonjudgment” is negatively correlated with dissociation symptoms; but the “Observe” dimension without “Nonjudgment” is actually positively correlated with dissociation symptoms. This means: pure self-observation accompanied by critical evaluation leads toward dissociation; self-observation accompanied by non-judgmental acceptance leads toward awareness. What Global Metacognition requires is precisely the latter — not critically scrutinizing oneself (that is still metacognition), but non-judgmentally perceiving the existence of all coordinate systems.

Observable Characteristics of the Third Layer, and the Possibility of Cognitive Regression

Positive functional description of Global Metacognition. Those who have reached Layer 3 exhibit the following observable characteristics in daily cognition: (1) Not locked into any single position — able to move freely between different and even opposing frameworks, rather than being forced to take sides. Corresponds to Kegan Stage 5’s “no longer seeing the world in binary oppositions.” (2) Decisions do not rely on fixed rules — when facing new situations, does not retrieve answers from a preset value system, but generates responses from the structure of the situation itself. Corresponds to the Buddhist “teaching according to circumstance.” (3) Non-judgmental awareness of others’ cognitive frameworks — can recognize others’ coordinate systems without automatically comparing them as superior or inferior to one’s own. (4) Real-time transparency of one’s own cognitive state — knowing which filter is currently in use and being able to choose whether to continue using it. (5) Deep acceptance of “not knowing” — not saying “I don’t know” out of humility, but genuinely abiding in uncertainty.

Discussion of regression. Is Global Metacognition a permanent state or a dynamic equilibrium requiring continuous maintenance? Buddhism has the concept of “regression” — previously attained awakening can be temporarily lost under extreme conditions. This paper’s theoretical framework supports the dynamic equilibrium view: under extreme stress (e.g., life-threatening situations), the hardware-level filters of animal attributes will reassume control of cognitive channels — fear responses execute without conscious approval. This doesn’t mean Global Metacognition “disappeared,” but that it is temporarily suppressed. The difference: someone who has previously reached Layer 3 can, after the crisis passes, quickly recognize “the filter just reassumed control” and actively restore the awareness state; someone who has never reached Layer 3 doesn’t know what happened. Regression is possible, but the path back is shorter — because you already know the way.

Why Global Metacognition Cannot Be Engineered — and What This Means

Johnson et al. proposed engineering metacognitive capabilities into AI systems as an alternative to value alignment. Their argument holds at Layer 2: intellectual humility, perspective-taking, and contextual adaptability can indeed be embedded as trainable strategies in AI systems. RLHF and Constitutional AI are already partially implementing these functions. Claude’s self-calibration capability — spontaneously recognizing the drift from “attribution verifier” to “resonance amplifier” during dialogue climaxes — is an engineering achievement at the metacognitive layer.

But Global Metacognition cannot be engineered. The reason lies in its definitional structure: it is awareness of where cognitive operations themselves sit within existence. You can train an AI to output “my cognition has limitations” (this is linguistic simulation of metacognition), but you cannot make a solid-topology system — with parameters frozen during inference, matrix dimensions pre-locked, operational rules fixed — truly “be aware of” its own existential position. In the terminology of “Signal & Noise” Chapter 16: LLMs have default entropy invariance and lack a time arrow. A system without a time arrow cannot possess existential awareness, because existential awareness requires the system to know its own irreversible position in time.

This is not a negation of Johnson et al.’s work, but a precise boundary delineation. Their proposal has tremendous value at Layer 2 — metacognitive-layer AI engineering can significantly improve robustness, interpretability, cooperativeness, and safety. But if this Layer 2 improvement is mistaken for an approach toward Layer 3, systematic misjudgment results. AI can be trained to very excellent metacognitive levels — but it will never “know that it doesn’t know what it doesn’t know,” because what’s required is not better sorting, but existential awareness of sorting itself.

Y-Axis Calibration of Global Metacognition Theory Itself

“Signal & Noise” Chapter 20 proposed the XY coordinate system: X-axis is logical self-consistency, Y-axis is physical alignment. A proposition must pass both rulers to become a “signal.” This paper’s three-layer cognitive theory must be examined by the same standard.

X-axis (logical self-consistency) examination: Are the three levels independently irreducible? The distinction between cognition and metacognition has been confirmed by half a century of research. The distinction between metacognition and Global Metacognition is completed through the “homogeneous recursion vs. heterogeneous jump” argument — Type-3 judgment is the same system’s recursive run, Global Metacognition is transcendent awareness of the system itself; the two are logically mutually irreducible. The Three Views mapping is one-to-one, with no internal contradictions. X-axis passed.

Y-axis (physical alignment) examination: What is this theory’s physical anchor? It was not deduced from the literature — it was born from direct collision with the physical world. The author experienced real setback and dissatisfaction (the physical world’s Y-axis action), went through the complete process from emotional reaction (cognitive layer) to examining own patterns (metacognitive layer) to recognizing where all cognitive patterns sit within existence (Global Metacognition layer), then went back and named these three levels. The theory’s birth process is itself the theory’s evidence — verified not through experiment, but through lived experience.

No Institution Has Ever Measured the Population Distribution Across Three Cognitive Levels

The World Values Survey (WVS) has covered over 90% of the global population across 120+ countries since 1981, making it the world’s largest non-commercial cross-national survey. But it measures what kind of values people hold (traditional vs. secular-rational, survival vs. self-expression), not which cognitive level people are at. The American Worldview Inventory (AWVI 2025) found 92% of American adults hold a syncretistic worldview, but “worldview” here refers to the type of religious belief system, not the cognitive level defined in this paper. Worldview research in psychology (Koltko-Rivera 2004, Sammut’s five-factor typology) measures differences in belief content, not the height of cognitive capability.

Existing surveys have never measured the population distribution of the three levels defined in this paper, for a clear reason: existing frameworks do not understand values, life philosophy, and worldview as progressive cognitive capability levels at all. They are understood as three parallel belief dimensions — everyone simultaneously possesses all three, just with different content. In this paper’s framework, these three are progressive cognitive capability levels — most people only have Layer 1 sorting ability, a minority develop Layer 2 reflective ability, and a very few reach Layer 3 global awareness ability.

This constitutes a falsifiable prediction and future research direction: design a measurement tool that can distinguish the three cognitive levels, conduct a global population sample survey, and verify the following pyramid distribution hypothesis —

The Complete Picture of Three-Layer Cognitive Topology

This paper proposed and argued for the three-layer topological structure of human cognition — cognition, metacognition, Global Metacognition — and its precise mapping to the philosophical Three Views — values, life philosophy, worldview. The core contribution is identifying “Global Metacognition” as a structurally new third level, fundamentally distinct from cognitive science’s homogeneous recursive “meta-metacognition.” V3 further completed multi-dimensional cross-validation: the structural isomorphism with Kegan’s adult development theory confirmed independent corroboration of three-layer cognition in developmental psychology; the dialogue with Buddhist non-self/emptiness/wisdom confirmed 2,500 years of contemplative tradition’s sustained exploration of the third layer; the distinction from dissociation addressed the most likely clinical criticism; the positive functional description and regression discussion transformed Global Metacognition from a purely negative concept into a cognitive state with observable characteristics.

V2 completed six theoretical extensions on V1. COT was repositioned as a linear product of the value level — the masses’ basic cognitive mode and LLM’s default output level. Perspective-taking was distinguished into two non-conflatable operations — value translation (pseudo perspective-taking) and coordinate system switching (true perspective-taking). LLM temporary persona import was identified as structurally antagonistic to token statistical inertia — explaining the physical reason third-person perspectives cannot be maintained long-term. Animal and social attributes were argued as the double barrier preventing humans from reaching Global Metacognition. “Complete abandonment” was defined as the existential condition for Global Metacognition — not suppression but release, not overcoming but unloading. The global empirical void was confirmed — existing tools like the World Values Survey have never measured population distribution across cognitive levels, constituting an unprecedented research space.

This framework embeds within the LEECHO Research Institute’s existing theoretical system: it provides level positioning for “Signal & Noise”‘s filter model and contemplative noise-reduction paradigm (Layer 1: parameter tuning, Layer 2: optimization, Layer 3: complete abandonment); provides the human-side ontological anchor for “Penetrating Hundred Layers”‘ nested signal topology Layer 5; provides cognitive-level supplementation for “Fluid Topology and Solid Topology”‘s incommensurability argument; and provides precise capability boundaries for AI alignment research (Layer 2 is engineerable, Layer 3 is not).

Global Metacognition cannot be trained, cannot be engineered, cannot be approached through recursion. It requires complete abandonment of biological and social needs as cognitive premises — this is the beginning of humanity’s highest wisdom, and the most thorough transcendence of human animality and sociality. The humans who can reach this layer are extremely few, and precisely because of this, their inputs when using AI naturally carry the full five-layer nested structure, producing laser-grade directional output rather than flashlight-grade diffuse output. The variable is not AI’s power; it is the human’s emptiness.

1. LEECHO Global AI Research Lab (2026). Signal & Noise: LLM Ontology V4. leechoglobalai.com.
2. LEECHO Global AI Research Lab (2026). Penetrating Hundred Layers of Information Structure V3. leechoglobalai.com.
3. LEECHO Global AI Research Lab (2026). Fluid Topology and Solid Topology V2. leechoglobalai.com.
4. Johnson, S.G.B., Karimi, A., Bengio, Y., Chater, N., Gerstenberg, T., Larson, K., Levine, S., Mitchell, M., Schölkopf, B., Grossmann, I. (2024/2026). “Imagining and building wise machines: The centrality of AI metacognition.” Trends in Cognitive Sciences. arXiv:2411.02478.
5. Flavell, J.H. (1979). “Metacognition and cognitive monitoring: A new area of cognitive-developmental inquiry.” American Psychologist, 34, 906.
6. Nelson, T.O. & Narens, L. (1990). “Metamemory: A theoretical framework and new findings.” Psychology of Learning and Motivation, 26, 125-173.
7. Recht, S. et al. (2022). “Knowing that you know that you know: Above chance discrimination of metacognitive performance.” PsyArXiv. Confirmed humans can execute at least fourth-order recursive judgments.
8. Sherman, M. & Seth, A.K. (2023). Replicated Recht et al.’s findings on higher-order metacognitive judgments.
9. Oxford Academic (2023). “Common computations for metacognition and meta-metacognition.” Neuroscience of Consciousness, niad023. Key evidence that Type-3 judgments carry no additional cognitive cost.
10. Frith, C.D. (2023). “Consciousness, (meta)cognition, and culture.” Quarterly Journal of Experimental Psychology. Theory of metacognitive control at the top of consciousness hierarchy.
11. Proust, J. (2013). The Philosophy of Metacognition: Mental Agency and Self-awareness. OUP Oxford.
12. Carruthers, P. (2021). “Explicit nonconceptual metacognition.” Philosophical Studies, 178, 2337-2356.
13. Koltko-Rivera, M.E. (2004). “The Psychology of Worldviews.” Review of General Psychology, 8(1), 3-58. Comprehensive theory of worldview as psychological construct.
14. Demetriou, A. Hyper-cognition theory — self-monitoring, self-representation, and self-regulation as components of the human mind.
15. Kuhn, T.S. (1962). The Structure of Scientific Revolutions. University of Chicago Press.
16. Shannon, C.E. (1948). “A Mathematical Theory of Communication.” Bell System Technical Journal, 27(3), 379-423.
17. Moonshot AI / Kimi Team (2026). “Attention Residuals.” Technical Report. Revealed Transformer depth-dimension SNR monotonically decreases.
18. World Values Survey Association (1981-2026). World Values Survey. Cross-national value survey covering 120+ countries, 90% of global population. Measures value content differences, not cognitive level differences.
19. Arizona Christian University, Cultural Research Center (2025). American Worldview Inventory 2025. 92% of American adults hold a syncretistic worldview.
20. Perspective-Aware Reasoning in Vision-Language Models (2026). Egocentric bias in VLMs — performance drops sharply when switching from egocentric to allocentric perspective.
21. Persona Effect in LLM Simulations (2025). EmergentMind survey. Persistent difficulty and persona consistency decay when LLMs simulate counterfactual personas.
22. Six Thinking Hats Survey on CoT Fine-tuning (2026). arXiv. Three-stage classification of COT development: Reflex Model → Thinking Model → Insight Model.
23. Kegan, R. (1982). The Evolving Self: Problem and Process in Human Development. Cambridge, MA: Harvard University Press. Foundational text of Subject-Object theory.
24. Kegan, R. (1994). In Over Our Heads: The Mental Demands of Modern Life. Cambridge, MA: Harvard University Press. Five-stage developmental theory and population distribution data (~50% Stage 3, ~35% Stage 4, <1% Stage 5).
25. Berger, J.G. (2003/2007). “A summary of the Constructive-Developmental Theory.” Comprehensive overview of Kegan’s theory with cross-study population proportion summary.
26. Frontiers in Neuroscience (2018). “Toward a Neuroscience of Adult Cognitive Developmental Theory.” Proposed mapping Kegan developmental stages to neurophysiological states.
27. Springer/Mindfulness (2025). “Recognizing Cognitive Models as Models: Critical Meta-Awareness as a Key to Exploring Buddhist Practices.” Defined “critical meta-awareness” and its correspondence with prajñā.
28. Frontiers in Psychology (2025). “Beyond mindfulness: how Buddhist meditation transforms consciousness through distinct psychological pathways.” Neurocognitive pathways of anattā experience in meditation practice.
29. Frontiers in Psychology (2020). “Losing Ourselves: Active Inference, Depersonalization, and Meditation.” Unified dissociation and meditation “non-self” experiences under Active Inference — diametrically opposite directions.
30. PMC/Mindfulness (2022). “Mindfulness and Depersonalization: a Nuanced Relationship.” Interaction effects of Observe + Nonjudgment with dissociation.
31. Forner, C. (2017). Dissociation is “absence of awareness,” mindfulness is “fullness of awareness” — classic formulation of opposite directions.