Extended Meme Theory: Consciousness as a Political System

Introduction

The Problem

What is consciousness from the standpoint of evolution? Why do humans perform acts that contradict their genetic interests – celibacy, hunger strikes, self-sacrifice? Why are old beliefs so difficult to change, and why do unsolved problems haunt us for years? Why does “willpower” run out, and why do addictions return? Why do entire states collapse even though nobody wanted it?

Existing theories answer these questions separately. This work proposes a single mechanism that explains all of these phenomena as consequences of the interaction between two replicators sharing a single substrate.

Central Thesis

2.5 million years ago, a second life form appeared on Earth – memes. They obey Darwinian laws (replication, variation, selection) but use a different medium: not DNA, but neural networks of the brain. A meme is not a metaphor: it is a cell assembly (Hebb, 1949), a physical ensemble of neurons causally demonstrated through optogenetics (Liu et al., 2012).

The brain is neither a “computer” nor a “vessel for consciousness.” The brain is a habitat for memes, grown under their pressure: the tripling of Homo brain volume over 2.5 million years was not for survival (the Flores hobbits with 400 cm³ brains survived for 100,000 years), but to support ever more complex memeplexes.

The Model: Biomemetic Complex (BMC)

Consciousness is not a unified agent but an ecosystem of competing replicators. Formally: the Biomemetic Complex $BMC = (G, M, I, S)$, where:

• G (Genetic layer) – fixed programs: hunger, fear, sex, care, SEEKING (curiosity). 7 basic emotional systems (Panksepp, 1998). The character vector $T = (T_{SEEK}, T_{FEAR}, ..., T_{PLAY})$ is 40-60% heritable.
• M (Memetic layer) – a dynamic network of acquired memes (memeplex). Memes are connected by edges (weighted edges, $w \in [-1, +1]$). Heterogeneous topology: a few hubs (memes with anomalously high centrality) define identity; the rest are periphery. A meme is fractal: it consists of sub-memes, which are themselves memes.
• I (Interface) – three mechanisms of G-M interaction: redirection (sex -> career), suppression (fasting, celibacy), interpretation (fear -> “divine trial”). Memes cannot switch off genetic programs – only redirect, suppress, or reinterpret them.
• S (Substrate) – neurobiology: neurotransmitters as “currency” (dopamine = attention, cortisol = stress), neuroplasticity, critical periods.

The Dual Nature of S: Substrate and Sensory Architecture

S has two inseparable aspects that must not be confused:

1. S as substrate – what the system is made of: neurons, neurotransmitters, synapses (biology) or processors, memory, communication channels (silicon). The substrate determines hardware constraints: how many memes can be stored, what the WM capacity is, the rate of edge decay.
2. S as sensory architecture – what information enters the system. Eyes, ears, skin, nociceptors – these are not “part of the brain” but input channels through which the environment acts upon the memeplex. Without sensory S, memogenesis is impossible: no signal – no meme.

Evolution designed the sensory system for G-programs: vision – to find food and spot predators, hearing – to hear an infant’s cry and a snapping twig, pain – to signal damage. The S-layer is derived from G: the question is not “what sensors to provide?” but “what must each G-program be able to perceive?” From this principle, five necessary and sufficient sensory channels are derived:

Key insight: The bandwidth of the S-layer is not constant. $S_{bw}(t)$ grows with substrate maturation: a newborn sees blurrily and hears muffled sounds, while the full perceptual zone unfolds gradually, creating critical periods (see AGI_F, Part VII, NM, Part XV).

Three Computational Engines: Why a Single Graph Is Not Enough

The brain uses three fundamentally different signal transmission mechanisms, each performing an irreplaceable function. Conflating them is a typical error of simplified models.

These three levels are not reducible to one another. The Graph Engine alone (activations via edges) is insufficient for consciousness: without the Modulation Engine there is no mode switching (FEAR vs PLAY), without the Diffusion Engine there is no semantic priming and diffuse memogenesis. Formalization: AGI_F, Part VII, BM, Part III, NM, Part XV.

The “self” is not a separate entity but a product of the Self-Model Cluster (SMC) – a subgraph of the memeplex that models the memeplex itself ($SMC = \{m \in M : target(m) \in M \cup G \cup I\}$). At any given moment, the conscious “self” is the dominant SMC configuration integrating ~10-20 active memes (out of thousands stored), shifting by context (see Part XVI).

Key Mechanisms

Competition for attention. Memes compete for limited resources – attention, working memory, dopamine. The winner is not the “best” meme but the one that captures resources most effectively (through emotions, repetition, associations with hubs). Hence the power of addictions, social media, and propaganda.

Immune system. The memeplex defends itself from competing memes through four layers: selective exposure, labeling “foreign” (via DISGUST), counter-argumentation, self-reinforcement. Hence confirmation bias, cognitive dissonance, and resistance to change.

Edge decay and SIT. Two complementary dynamics mechanisms. Edge decay – forgetting: unused connections weaken according to $w(t) = w_0 \cdot e^{-\lambda t}$. Structural Incompleteness Tension (SIT) – the opposite process: structural gaps in the memeplex generate persistent SEEKING activation, explaining the return to unsolved problems, as well as the emergence of religion, superstition, and conspiracy theories as false closure – filling gaps with nodes of zero empirical validity.

Self-Model Cluster (SMC). The memeplex contains a subset of memes directed at the system itself – the Self-Model Cluster. The SMC gives rise to phenomenal consciousness: the recursive loop “memeplex models itself” creates a first-person perspective (qualia). Three levels of recursion: 0 – unconscious processing, 1 – phenomenal consciousness, 2 – metacognition/reflexion. When bridge nodes are destroyed and the SMC fragments, each fragment generates its own “self” (model of schizophrenia). Consciousness exists on a gradient: in animals, the SMC is limited to bodily self-model ($CL > 0$ but without level-2 recursion); in humans, memes expand the SMC and provide material for reflexion. The dual replicator is an amplifier of consciousness, not its prerequisite. See Parts XVI-XVII.

Active Inference. The memeplex is not a passive world model but a generative model that minimizes discrepancy (free energy) by two paths: perceptual inference (update the model to match reality – change oneself) and active inference (change reality to match the model – change the world). The materialization cascade operates at four levels: behavioral, self-fulfilling prophecy, psychosomatic, cultural accumulation. See Part XVIII.

Scale invariance. The same processes – replication, competition, immune defense, edge decay – operate at all levels: neuron -> individual -> group -> state. This is not an analogy but a literal isomorphism, provided the transfer conditions are met (presence of replication, competition, and differential selection).

What Is New Compared to Dawkins and Blackmore

Structure of the Work

The theory is presented in four documents, each developing a separate aspect:

Table of Contents

Block I. Foundation (Parts I-II)

• Part I. Foundation: What Is a Meme
• Part II. Empirical Base: We Are Built to Replicate Memes

Block II. Ecosystem (Parts III-V)

• Part III. Consciousness as an Ecosystem of Memes
• Part IV. Internal Economy: What Memes Compete For
• Part V. Memeplexes: How Memes Unite

Block III. In-Depth Analysis (Parts VI-XI)

• Part VI. Energetics of Memes: Replication as the Selection Criterion
• Part VII. The “Voting” Mechanism: How the Memeplex Decides Without an Agent
• Part VIII. Life Cycle of a Meme: Entry, Deactivation, Dormancy
• Part IX. Ontology of the Meme: Where Are the Boundaries of the Unit?
• Part X. Temporal Dynamics: Why Universal Timescales Cannot Be Set
• Part XI. Zipf’s Law in Consciousness: Why Inequality Is Inevitable (heavy-tailed)

Block IV. Cross-Level Transitions (Part XII)

• Part XII. Cross-Level Transitions: From Neuron to State

Block V. Politics and Defense (Parts XIII-XV)

• Part XIII. Politics Inside the Head
• Part XIV. The Immune System of the Individual
• Part XV. The Immune System of Memeplexes: A General Theory

Block VI. Consciousness and Dynamics (Parts XVI-XX)

• Part XVI. The Self-Model Cluster: Where the “Self” Comes From
• Part XVII. Reflexion: How Consciousness Modifies Itself
• Part XVIII. Active Inference: The Memeplex as an Engine of Materialization
• Part XIX. Merging and Splitting of Memeplexes
• Part XX. Language as M-Layer Architecture

Block VII. Change (Parts XXI-XXIII)

• Part XXI. Conditions for Change: What Triggers Restructuring
• Part XXII. Consequences and Predictions
• Part XXIII. Death and Immortality of Memes

Block VIII. Classical Theories (Part XXIV)

• Part XXIV. Classical Theories Through the Lens of Memetics

Block IX. Case Studies (Parts XXV-XXVII)

• Part XXV. The State as a Memeplex
• Part XXVI. The USSR as a Memeplex: Birth, Flourishing, and Death
• Part XXVII. Gandhi: Hacking a Memeplex Through Nonviolence

Block X. Conclusion (Parts XXVIII-XXX)

• Part XXVIII. Unity and Struggle of Replicators: Genes and Memes
• Part XXIX. Falsifiability: How to Disprove This Theory
• Part XXX. Conclusions from the Theory

Appendix

• Glossary of Terms

Part I. Foundation: What Is a Meme

Definition

A meme is a unit of cultural information capable of being copied between minds through imitation. It is not an abstraction but a physical structure in the brain: a pattern of neural connections that can be activated, compete for attention, and be transmitted to others.

Neurobiological Definition of the Meme

The claim “a meme is a physical structure” requires specification. What exactly is a meme at the neurobiological level?

Meme = cell assembly (Hebb, 1949) = engram (Josselyn & Tonegawa, 2020, Science) – an ensemble of neurons (~10³-10⁵) that systematically fire together. Optogenetic reactivation of specific engram cells causally triggers recall (Liu et al., 2012, Nature).

Information is encoded not in a single neuron but in a population activation vector (population coding – Georgopoulos et al., 1986, Science). A neuron is a universal computational element: it does not know what it encodes (neural reuse – Anderson, 2010; input rewiring – Sharma, Angelucci & Sur, 2000, Nature).

Memes overlap: one neuron participates in many ensembles (mixed selectivity – Rigotti et al., 2013, Nature). Memories close in time literally share neurons – creating a causal link between them (overlapping engrams – Cai et al., 2016, Nature).

Further reading: Full description of the neural substrate – see BIOMEMETICS, Part III.

Fractality of the Meme and Basic Terms

A meme is fractal: it consists of components that are themselves memes at a smaller scale. The hierarchy is relative, not absolute – the same object changes its role depending on the scale of analysis:

• “Smile” – a component of the meme “Mona Lisa”
• “Smile” – an independent meme with its own components: “joy,” “greeting,” “mystery,” “irony”
• “Joy” – a component of the meme “smile,” but also an independent meme: “celebration,” “laughter,” “childhood”

No separate term is introduced for “component of a meme” – when scale must be indicated, “sub-meme” is used.

Four basic terms of the theory:

Hub is not a separate hierarchical level but a role: an element with centrality significantly exceeding the average. A hub can be a meme (meme-hub), a memeplex (dominant memeplex), or even a component of a meme (key feature). By default, “hub” means meme-hub.

Meme-type vs meme-instance: distinguish the abstract cultural pattern (meme-type: “Mona Lisa” as a phenomenon) from the concrete neural realization in one BMC (meme-instance: my version of “Mona Lisa” with my associations). Analogy: genotype vs phenotype. One meme-type -> millions of meme-instances, each with unique fidelity and edge set.

BMC = (G, M, I, S) applies at any scale – individual, group, state, civilization. This is not an analogy but a literal application of one model (see Part XII for details).

Memes Are Stored as Heuristics, Not Exact Copies

Cognitive psychology has shown that memory stores information not as exact copies but as schemas – simplified patterns from which details are reconstructed during recall.

Empirical base:

• Bartlett (1932), Remembering: A Study in Experimental and Social Psychology: the “War of the Ghosts” experiment showed that when retelling a story, subjects systematically distorted details, adapting them to their cultural schemas
• Reconstructive memory: memory does not reproduce but reconstructs – each recall is partly created anew

Implication for memetics: Each transmission of a meme is a potential mutation. The meme “I heard that X” is already not equal to the original – the brain reconstructs it from a schema. This explains why:

• Rumors get distorted in transmission
• Religious texts diverge across traditions
• Scientific ideas are transformed into pop versions

Schematic storage is one of the reasons for the high speed of memetic evolution (see Part X for details).

Structure of the Meme: Core and Periphery

Not all parts of a meme are stored equally. A meme has a hierarchical structure in which central elements (the core) are preserved more reliably than peripheral ones.

Neurobiological grounding of levels:

The core of a meme is physically “sturdier” than the periphery: it is anchored by structural synapses, while details are stored in dynamic spines with a high turnover rate (~40% per month for small spines, Yang et al., 2009, Nature).

This structure explains why:

• A forgotten foreign language is “remembered” in days rather than relearned over years – the core was preserved
• A classmate’s name may become “I think it started with A” – the core degraded to a trace
• An expert remembers principles but forgets syntax – the periphery decayed

The completeness of a meme’s preservation is called Fidelity – from 0 (only a trace) to 1 (full storage).

Mathematical formalization: The Fidelity formula and three storage modes – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: The physical structure of a meme in the brain – see BIOMEMETICS, Part I and Part IV.

Practical implication: Why do mnemonics work? They increase the number of a meme’s edges -> reduce the rate of decay. See NETWORK_MEMETICS, Part VIII: Associative Memory.

Three Mechanisms of Meme Synthesis

Memes are not merely copies. The unique evolutionary advantage of memes over genes is the ability to synthesize new memes from existing ones.

Mutation is a mechanism shared with genes. Recombination and abduction are specifically memetic: memes can blend in a single host’s mind within minutes, whereas genes require generations (Fauconnier & Turner, 2002; Balkin, 1998; Gabora, 1997).

Formalization: Graph growth through three synthesis operations – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Sleep as an arena for recombination (BLEND) – see BIOMEMETICS, Part IV.

Biological substrate: Panksepp’s 7 systems as basic emotional constructs to which memes attach – see BIOMEMETICS, Part IV.

Memes as Darwinian Replicators

The algorithm of evolution is indifferent to what a replicator is made of. If a system copies itself, mutates, competes, and inherits changes – it obeys Darwin’s law.

Key Clarification: Not Survival of the Fittest

A popular misconception holds that evolution is about the strongest winning. In reality, evolution is about who first occupied the niche and who is recognized as useful by the system.

Part II. Empirical Base: We Are Built to Replicate Memes

The Explosive Growth of the Brain 2.5 Million Years Ago

The human brain is an anomaly. It consumes 20% of the body’s energy (up to 60% in children) while constituting only 2% of body mass. From the standpoint of evolutionary efficiency, this is madness.

But the most interesting thing is when it began to grow:

2.5 million years ago marks the appearance of the first stone tools (Oldowan culture). This is the moment when imitation became critically important for survival. This is the moment of the second replicator’s emergence.

The brain grew not for hunting (predators hunt with brains ten times smaller), not for navigation (rats build cognitive maps with a tiny brain). The brain grew as a storage and processor for memes.

Coevolution of Genes and Memes: The Growth Mechanism

The emergence of memes triggered a self-reinforcing cycle:

Memes created selection pressure on genes, shaping the brain for their needs. Genes “caught up” with the demands of memes at maximum intensity. Those who best imitated increasingly complex skills survived and passed on their genes.

This explains the paradox: evolution unhesitatingly amputated our ancestor’s tail, which would have been useful in a million scenarios. But the brain, devouring 20% of energy, was not merely retained – it was hypertrophied to absurd dimensions. Because this was not the genes’ choice – it was pressure from the second replicator.

Auditory Memes: The Most Aggressive Replicators

Not all transmission channels for memes are equal. Auditory memes turned out to be the most aggressive:

This is why:

• Melodies “stick” in one’s head more firmly than images
• The most popular videos are musical
• Singers have more fans than writers or artists
• Speech became the primary tool for transmitting information

Language arose not for hunting and not for warning about danger – wolves hunt without grammar, and vervets warn about predators with simple calls. Language emerged as the primary instrument for propagating memetic replicators. The benefit to genes was collateral.

Over-Imitation: Proof of Our Nature

One of the most revealing experiments in comparative psychology is the imitation test.

Experimental conditions:

A transparent box with a treat inside. The experimenter demonstrates how to retrieve it, but includes pointless actions: taps a stick on the box three times, moves an unnecessary lever, blows on the box – and only then opens the door.

Results:

Even when children are explicitly told that some steps are pointless – they still repeat them. Even when the box is transparent and it is visible that the lever affects nothing – they pull it.

Why This Matters

Is the chimpanzee smarter in this task? From the standpoint of obtaining the treat – yes. But from the standpoint of replicating memes – no.

Over-imitation is not a bug; it is a feature. It is a mechanism ensuring accurate copying of cultural information, even when the copier does not understand why each step is needed.

Tool-making, fire-starting, food preparation, herbal medicine – all of these are complex chains of actions where it is unclear which step is critical. It is safer to copy everything.

Conclusion: We Are Machines for Copying Memes

Our brain is not a survival tool. It is an incubator for memes – one that memes grew for their own reproduction.

The Flores Hobbits: Plasticity of Memetization

But if the brain grew for memes – how does it respond to changing conditions?

Homo floresiensis – the dwarf humans of Flores Island – provide an unexpected answer. Their brains were approximately 400 cm³ – comparable to a chimpanzee’s, one-third of ours. Yet they were descendants of Homo erectus, whose brains were ~900 cm³.

However, culture did not disappear. Archaeological evidence shows:

• Manufacture and use of stone tools (flakes up to 12 cm)
• Hunting of large animals (Stegodon – an extinct elephant species)
• Control of fire
• Brodmann area 10 (prefrontal cortex, associated with cognition) – comparable in size to modern humans

This is an example of island dwarfism – a phenomenon confirmed by studies of Malagasy pygmy hippopotamuses: when body size decreases, the brain may shrink disproportionately. Crucially, this is not a “disabling” of cognitive abilities but a reorganization – preserving critical functions while reducing overall volume.

Conclusion: Brain size is plastic and depends on conditions, but the memetic infrastructure can be preserved even with significant volume reduction. This indicates system modularity: genes can “economize” on overall size while preserving key areas for cultural transmission. A large brain is not the “pinnacle of evolution” but an adaptation to specific pressure, which can be optimized when conditions change.

Sources: Nature (2009), Smithsonian Human Origins, Royal Society (2017)

Feral Children: Genes Build the Body, Memes Build the Person

Take a tiger cub, raise it in isolation – it will become a tiger. Any animal raised in solitude will know everything its species is supposed to know: how to hunt, hide, reproduce.

But isolate a human newborn – the result is catastrophic:

Documented cases (Amala and Kamala, Genie Wiley, Victor of Aveyron) show: children deprived of contact with culture during the critical period do not become humans in the full sense. Even after returning to society, they cannot fully master language and social behavior.

Methodological caveat: Cases of “feral children” have serious limitations as scientific data. For the Genie Wiley case: (1) NIMH criticized the disorganized nature of the research and the absence of clear parameters; (2) it is unknown whether she had innate cognitive impairments; (3) ethical problems – the research may have exacerbated the trauma (Britannica). For Amala and Kamala: documentation is based on one person’s diaries, and reliability is disputed. Nonetheless, the general pattern (impossibility of fully acquiring language after the critical period) is confirmed by more controlled studies of deaf children and late cochlear implantations.

Conclusion: Genes create the biological foundation – a body capable of hosting memes. But memes create the human. We are the only species whose essence is defined not by genetic but by memetic information.

Critical periods: a window that closes. The phenomenon of feral children is not merely an illustration of the role of memes. It points to a fundamental property of the S-layer: substrate plasticity is finite in time. The parameter $\lambda_{plast}(t)$ – the rate of forming new connections and $\kappa$-transitions – peaks in early ontogenesis and then declines. When plasticity drops below a threshold ($\lambda_{plast} < \theta_{plast}$), the window of memogenesis closes: the S-layer continues to receive signals, but the M-layer is no longer able to efficiently generate new memes from them.

Feral children represent a case where the critical period for linguistic memogenesis was missed: $\lambda_{plast}$ dropped below threshold without a sufficient M-layer. The brain (S-substrate) is fully intact, G-programs function – but the memeplex did not develop. This explains why returning to society yields only limited results: plasticity $\lambda_{plast}$ is no longer what it was in the first years of life, and the same sensory signals generate memes more slowly and with lower fidelity.

Formalization: $\lambda_{plast}(t)$ as a bell-shaped (Gaussian) function with a peak and subsequent decline, critical period window, connection to $k_{active}(t_{dev})$ and $S_{bw}(t)$ – see AGI_F, Part VII, NM, Part XV.

Sexual Selection Shifted to Memes

The ability to imitate became so important that it turned into a criterion of sexual selection. Women began to prefer not the best hunters or the strongest individuals – but those who best create and propagate memes.

The result is visible today: the insane number of fans of actors, musicians, comedians – people whose skills are useless for survival in the wild. Rock stars have more sexual partners than Olympic champions. This is not a cultural anomaly – it is the result of millions of years of selection in favor of the best meme propagators.

Part III. Consciousness as an Ecosystem of Memes

Intuitive Self-Evidence

Many people, independently of familiarity with meme theory, arrive at similar conclusions: consciousness is not a monolith but a set of images that are replaced, become obsolete, and synthesize with age. This first-person observation confirms the model: we literally feel how some memes displace others inside our heads.

Life Cycle of a Meme: Brief Introduction

Long-term memory is practically never erased. This means that the “death” of a meme within the brain is not destruction but loss of access to attention. Memes do not die – they fall asleep, awaiting a trigger for reactivation.

The life cycle of a meme, mechanisms of deactivation, functional activation thresholds, and survival strategies are discussed in detail in Part VIII.

But where does the sense of the unity of consciousness come from if memes compete? The mechanism is the Self-Model Cluster (SMC): a subgraph of the memeplex that models the memeplex itself. The SMC creates the illusion of a unified “self” from a multitude of competing memes (see Part XVI for details). The quantitative metric of consciousness level: $CL = \sigma_{SW} \cdot A_{SMC} \cdot f(Balance)$ (see NETWORK_MEMETICS, Part XIII).

Part IV. Internal Economy: What Memes Compete For

Resources Inside the Brain

Memes do not compete for storage space – that is nearly infinite. They compete for limited processing resources.

Mechanisms of Competition

Meme Strategies in the Struggle for Resources

From Strategy to Metric: How to Measure “Victory”

Each meme strategy is reflected in a network metric:

Eigenvector centrality is the key metric for evaluating a meme’s “strength”: what matters is not just how many connections it has, but how important the neighbors are.

A meme with connections to hubs has high eigenvector centrality even with a small number of connections.

Formalization: Eigenvector centrality, assortativity – see NETWORK_MEMETICS, Part VI.

Cognitive Dissonance: Resolution Through Hub Advantage

Cognitive dissonance (Festinger, 1957) is one of the most studied phenomena in psychology. The qualitative idea that well-connected beliefs are more stable than peripheral ones has been known for a long time (Quine, 1951 – “web of belief”; Thagard, 2000 – “coherence in thought and action”; connectionism in general). However, no model provides a quantitative prediction of which specific belief will change. BMC formalizes this effect through eigenvector centrality.

Dissonance in BMC terms: two memes with a negative edge are both active above $\theta_{high}$ – a standard situation of “active contradiction.” Example: “I am an honest person” (hub, $C_E = 0.8$) + “I lie at work” (peripheral meme, $C_E = 0.15$).

BMC prediction (hub advantage):

The meme with lower eigenvector centrality will change first. Reason: the high-centrality meme is connected to many other memes; its change would cause a cascade. The economy of the memeplex favors minimal restructuring.

What distinguishes this from Festinger and Quine: Festinger describes what (discomfort) and why (reduction). Quine and Thagard describe a qualitative principle (a well-connected belief is more stable). BMC provides a quantitative formula: $P \propto 1/C_E(i)$. This is a testable prediction: in cognitive dissonance experiments, the belief with more associations (degree centrality as a proxy for eigenvector centrality) should be preserved, while the peripheral one should change.

Boundary of applicability: Hub advantage is the default mode for everyday dissonance. When $Tension > \theta_{crisis}$ (life crises, massive external pressure, prolonged accumulation of discrepancies), hub displacement kicks in – the hub yields (see Part XVII, age-related crises; Part XXVII, Gandhi). Two mechanisms are not a contradiction but different regimes of a single system: minimal restructuring under weak dissonance, cascading restructuring under crisis.

Formalization: Hub Protection Hypothesis – see NETWORK_MEMETICS.

Why Some Memes Monopolize: Zipf’s Law

Why is there no “democracy of equal memes” in the mind? Why do some memes become hubs while the rest remain periphery?

The answer is provided by the preferential attachment mechanism (“the rich get richer”):

1. A new meme enters the memeplex
2. It seeks something to attach to
3. It is more likely to attach to already “popular” memes (those with more connections)
4. The “popular” ones become even more popular
5. Result: power-law distribution (Zipf’s law)

Zipf’s law in the memeplex:

Implication: Hubs are not a metaphor but a mathematical inevitability. In systems with preferential attachment, hubs arise inevitably.

Formalization: Preferential attachment mechanism, distribution formulas – see NETWORK_MEMETICS, Part III.

The Currency of Internal Economy: Dopamine (the SEEKING System)

Dopamine is not the “pleasure hormone.” It is the currency with which the brain pays for attention. In terms of Panksepp’s affective neuroscience, the dopaminergic system is SEEKING (curiosity, anticipation, “wanting”), which is recruited by all other emotional systems. See BIOMEMETICS: SEEKING as a Meta-System.

Discrete programs, continuous experience. Panksepp’s seven systems are discrete neural circuits (subcortical G-level). But the subjective experience of emotions is a blend of several simultaneously active G-programs through the M-layer: “nostalgia” = PANIC/GRIEF + SEEKING + CARE, “inspiration” = SEEKING + PLAY, “jealousy” = LUST + RAGE + FEAR. Cultural variability (amae, Schadenfreude, saudade) = different M-clusters in a continuous affective space with identical G. Barrett is correct for the M-level (emotions as constructions), Panksepp for the G-level (discrete circuits). BMC reconciles both approaches. See BM: Affective Space; NM: Blend Formula.

Memes that learned to exploit the dopaminergic system gain an advantage. Hence the power of addictions, social media, and gambling.

Unfinished Tasks: Why the Brain Will Not Release Unsolved Problems (SIT)

A word on the tip of the tongue. An unsolved problem from school. The question “why do I live?” An unfinished novel. All these things share one feature: they return – after days, months, years – without any apparent cause. No external trigger, no reminder, no deliberate effort. The brain simply “won’t let go.”

According to the formulas of edge decay, unused memes should weaken: $w(t) = w_0 \cdot e^{-\lambda t}$. But unsolved problems are the exception. They do not decay until resolved.

Structural Incompleteness Tension (SIT) – the tension of structural incompleteness – explains this phenomenon. Imagine the memeplex as a jigsaw puzzle. When the pieces are connected, the picture is whole and there is no tension. But when a hole gapes in the center of the puzzle, it creates persistent tension proportional to the importance of the missing piece.

Formally: structural gaps in the memeplex generate their own SEEKING activation. This is the “second input” to the dopaminergic system – the first input (external stimuli and recruitment from emotions) was described above; the second (SIT) operates endogenously, from within. The more important the cluster with the gap and the more central the position of the gap, the stronger the tension.

But not all unsolved problems “hook” equally. SIT is modulated by Learning Progress (LP) – the sense of progress. If a problem seems hopeless (LP -> 0), the tension gradually fades. But as soon as new information appears (LP spikes) – the problem “returns” with redoubled force. Every scientist knows this experience: a problem forgotten for years suddenly becomes pressing when a new tool or approach appears.

Why humanity invented gods: SIT generates cognitive tension, and the brain strives to relieve it at any cost. Lightning strikes a tree – gap: “why?” SIT builds. And then an answer presents itself: “Zeus is angry.” The gap is filled, SIT drops, cognitive load decreases. It does not matter that the answer is false – structurally, the gap is closed. This is false closure: the tension is relieved, but the filler node does not withstand empirical verification.

False closure explains the emergence of:

• Religion – filling cosmological gaps (“Why does the world exist?” -> “God created it”)
• Superstition and magical thinking – filling causal gaps (“Why did something bad happen?” -> “A black cat crossed the path”)
• Astrology – filling the self-knowledge gap (“Why am I this way?” -> “I’m a Scorpio”)
• Conspiracy theories – filling gaps of justice (“Why is the world unjust?” -> “A secret world government”)

Evolutionarily, false closure is more advantageous than chronic SIT: constant tension consumes cognitive resources, and for survival, a wrong answer is better than an eternal question. Science is a cultural institution that systematically replaces false closure with valid closure. This requires enormous effort because it goes against the brain’s natural impulse to quickly close gaps.

Formalization: Graph-theoretic definition of SIT, formulas, numerical example – see NETWORK_MEMETICS, Part VIII: SIT. Neurobiological substrate (DMN, expanded SEEKING formula) – see BIOMEMETICS, Parts III-IV. Engineering implementation – see AGI_FOUNDATIONS, Part III.

Illustration: Ramanujan – SIT as a Driver of Genius

Srinivasa Ramanujan (1887-1920) was one of the most astonishing mathematicians in history. Three classes of formal mathematical education, no access to academic circles, near-complete isolation – and yet he independently derived world-class formulas, some of which were proven only decades later.

In BMC terms: Ramanujan was a BMC with a unique configuration of three parameters:

How it works: Ramanujan sees a mathematical structure -> the gap is glaring (high SIT) -> SEEKING activation -> the memeplex begins generating candidates for closure -> non-standard topology enables abduction inaccessible through standard wiring -> result: a formula that a “normal” mathematician could not derive because their connections are laid out differently.

Why formal education both helps and hinders:

• Helps: standard wiring of connections (textbook -> lecture -> problem) creates a reliable scaffold through which standard gaps can be quickly closed
• Hinders: standard connections = standard paths of abduction. All graduates of the same school “think alike” -> identical blind spots

Ramanujan bypassed standard wiring -> his connections are laid out “around” the standard paths -> different blind spots, different strengths. Cost: many of his results were already known (he rediscovered what the standard path had long since closed). Gain: some results were fundamentally new (the standard path did not lead to them).

General principle: genius = anomalous SIT + anomalous $T_{SEEK}$ + non-standard topology. Any two without the third yield: a perfectionist maniac (SIT + SEEK, standard topology), an eccentric dilettante (SIT + non-standard topology, low SEEK), or a hardworking specialist (SEEK + standard topology, low SIT).

Why Harmful Memes Defeat Useful Ones

The evolution of memes inside the head does not optimize the host’s well-being. It optimizes capturing attention here and now.

Conclusion: Memes Wage Economic Warfare

This is not a war of annihilation. It is competition for limited attention resources – just as companies compete for consumer dollars. And as in economics, what wins is not the “best product” but the one that captures the market most effectively.

Part V. Memeplexes: How Memes Unite

The “Self” Is the Current Memeplex

At any given moment, the conscious “self” is not all the memes in one’s head but only those that currently have access to attention. This is the ruling memeplex.

The Memeplex Changes Constantly

Long-Term Memeplex = Personality

“Personality” is a stable configuration of memes that persists for years. But it is not monolithic: inside, there is constant behind-the-scenes infighting – intrigues, alliances, betrayals, sabotage.

Topology of the Memeplex: Three Key Properties

A memeplex is not an amorphous cloud of memes but a structured network with a characteristic topology. Three properties determine its behavior:

The small-world structure explains the paradox of consciousness: it is simultaneously differentiated (different contexts do not mix) and integrated (any thought can lead to any other in 3-4 associations).

The small-worldness coefficient: $\sigma = \frac{C / C_{random}}{L / L_{random}}$, where $C$ is clustering, $L$ is average path length. When $\sigma > 1$, the network has the small-world property.

Formalization: Small-world networks, the sigma metric – see NETWORK_MEMETICS, Part IV.

Assortativity: Whom Are Hubs Connected To

Assortativity ($r$) shows whether hubs are preferentially connected to other hubs ($r > 0$) or to the periphery ($r < 0$).

Hypothesis: Personal memeplexes are assortative – central identity meme-hubs form a tightly connected core, an “elite club.”

Implication: When $r > 0$, an attack on any meme-hub threatens the entire core – hence the strong immune response to criticism of fundamental values.

Formalization: Assortativity coefficient – see NETWORK_MEMETICS, Part VI.

Network Motifs: “Building Blocks” of Thinking

Network motifs are recurring connection patterns found more frequently than in random networks. They define the “style” of thinking:

Hypothesis: The predominant motif type determines cognitive style:

• Excess of triangles -> rigidity, everything mutually confirms everything
• Excess of feed-forward loops -> analyticity, causal chains
• Excess of bi-fans -> integrativeness, consideration of multiple contexts

Formalization: Network motifs – see NETWORK_MEMETICS, Part VII.

Inter-Brain Memeplexes

Some memeplexes cannot exist in a single head – they require several carriers with complementary roles.

This is like a parasite that needs two host species for its complete life cycle. Examples: family scripts, workplace conflicts, political confrontations.

Schizophrenia: A Window into the Nature of Memes

The unified “self” is a product of the SMC (Self-Model Cluster, see Part XVI): a subgraph of the memeplex that models the memeplex itself. Normally, the SMC is singular, dominant, integrating all clusters into a unified whole. In schizophrenia, this integration breaks down.

Schizophrenia = competition of multiple SMC configurations for dominance. Bridge nodes connecting clusters are weakened -> clusters become isolated -> each forms its own SMC configuration (local self-model) -> multiple “selves” compete for control of behavior.

Approximately 80% of people with schizophrenia hear voices – and experience them as absolutely real. In SMC terms, each voice is a separate SMC configuration:

Implication for the theory: The “self” is not a given but a product of a unified SMC. When bridge nodes are destroyed and the SMC fragments, each fragment generates its own “self.” Schizophrenia lays bare the nature of consciousness: what we experience as a unified subject is the dominant SMC configuration that won the competition.

Network interpretation: Schizophrenia is a decrease in betweenness centrality of bridge nodes + SMC fragmentation. Bridges are destroyed -> Q rises -> clusters become isolated -> in each, a local self-referential loop forms.

Prediction: In patients with schizophrenia, the average betweenness centrality in semantic networks should be statistically lower than in the control group. The number of “voices” should correlate with the number of isolated clusters (communities) in the semantic network.

Transition to In-Depth Analysis

Parts I-V laid the foundation: what a meme is, how memes compete, how they unite into memeplexes. But open questions remain:

• What determines a meme’s “strength”? (Part VI)
• How does a memeplex “decide” without an observer-agent? (Part VII)
• What happens to memes that “lost”? (Part VIII)
• Where are the boundaries of a single meme? (Part IX)
• How fast do memetic processes occur? (Part X)
• Why is inequality among memes inevitable? (Part XI)
• Can concepts be transferred between levels? (Part XII)

The next seven parts answer these questions.

Formalization: Memeplex as a network (graph), meme as a node, connection as an edge – see NETWORK_MEMETICS, Parts V and VII.

Part VI. Energetics of Memes: Replication as the Selection Criterion

Note on Part IV. Above we described dopamine as the “currency” of attention. This is a useful operational model: dopamine does indeed participate in attention allocation. However, the question “what fuels a meme?” is a false question. Dopamine is a consequence of successful replication, not its cause. A meme is strong not because it “gets more dopamine” but because it helps the memeplex replicate. Dopaminergic reinforcement is merely an indicator of this.

Reformulating the Problem

The question “what fuels a meme?” leads to a dead end. Dopamine? ATP? Attention? Any answer generates new questions and leads to reductionism.

The right question is different: what makes a meme a successful replicator?

The Memeplex as a Selection Environment

For genes, the selection environment is the external world. For memes inside the head, the selection environment is the memeplex itself.

A meme is “strong” not because it consumes more dopamine. A meme is strong if it helps the memeplex survive and replicate. This is the only criterion that matters.

Comparison of Replicators

Synthesis as Extended Replication

It is important to understand: synthesis (recombination + abduction) is not a separate process opposed to replication. It is a form of replication in which elements of existing memes are copied into a new configuration. Every “new” meme contains replicated fragments of “parent” memes.

Genes also recombine – but only during sexual reproduction, once per generation. Memes recombine continuously, in a single host’s head. This explains the exponential acceleration of cultural evolution compared to biological evolution.

Ontogenesis of the Memeplex: Accelerated Phylogenesis

A memeplex in a single head traverses a path analogous to biological evolution but compressed into one lifetime. The selection criteria for memes change with the age of the memeplex:

Formal analogue: the stages of ontogenesis correspond to the Boltzmann temperature of the memeplex $T$: sponge – high $T$ (low acceptance threshold, all memes pass), museum – low $T$ (high threshold, everything is rejected). Temperature modulates the probability of accepting a new meme: $P(\text{acceptance}) \propto e^{-\Delta E / T}$, where $\Delta E$ is the “cost” of integration. See NETWORK_MEMETICS, Part VIII.

One Meme – Different Evaluations

One and the same meme is evaluated differently by the memeplex depending on the stage of development:

Exponential Growth Through Memetic Evolution

Biological evolution is slow: one generation takes decades, significant changes take millennia. Memetic evolution within a single head occurs in real time:

• A child in its first years of life “lives through” millions of years of cultural evolution
• Each generation begins not from zero but from the accumulated meme pool
• Memes can evolve while the organism is still alive

This explains the explosive acceleration of human progress: genes did not improve – memes gained the ability to evolve within each head at inconceivable speed.

What This Resolves

But how does the memeplex “weigh” memes without a homunculus? Part VII addresses this.

Formalization: A meme’s “strength” = centrality in the network (degree, betweenness, eigenvector) – see NETWORK_MEMETICS, Part VI.

Part VII. The “Voting” Mechanism: How the Memeplex Decides Without an Agent

The Homunculus Problem

When we say “the memeplex decides,” “the memes vote,” “the system evaluates a threat” – the question arises: who counts the votes? Are we not creating a little person inside the head who does all this?

The answer: there is no counter. There is dynamic weighting by the replication criterion.

A Mechanism Without an Agent

The memeplex does not “decide” in the human sense. It filters incoming memes automatically:

“Voting” is not counting but competition of neural patterns:

• Patterns compatible with the dominant memeplex are strengthened
• Incompatible ones are suppressed through lateral inhibition
• The winner is determined not by a counter but by whichever pattern is more strongly activated

The speed of voting depends on the topology of connections – see Part X.

The Environment Determines the Criteria

Key principle: whatever is replicable in the current environment is accepted.

The memeplex of a soldier at war and the memeplex of a civilian inhabit different environments. The same meme is evaluated differently:

Case Study: Habituation to Death in War

A soldier in a combat zone undergoes a memetic transformation:

This is not a “broken psyche” – it is the memeplex adapting to its environment. Memes that were previously rejected now have high replicative value.

Case Study: Why Society Rejects Veterans

When a soldier returns, a collision of memeplexes occurs:

The memeplexes of civilians sense a threat. A person for whom killing is normal is potentially dangerous to their existence.

Society’s reactions are defensive mechanisms of memeplexes:

The “Voting” Mechanism Is an Immune Response

There is no homunculus counting votes. There is an automatic process:

1. The incoming meme activates a neural pattern
2. The pattern either resonates with the existing memeplex or conflicts with it
3. Resonance -> connection strengthening -> integration
4. Conflict -> cognitive dissonance -> rejection

This resembles the immune system: it does not “decide” whether to attack a virus. It automatically reacts to patterns recognized as threats.

From simple competition to metacognition. The described mechanism is basic: patterns compete automatically. But in developed memeplexes, an additional layer appears – the Self-Model Cluster (SMC): a subgraph of memes that model the memeplex itself (see Part XVI). The SMC is not a homunculus – it is the same graph, the same competition mechanisms, but directed at modeling one’s own processes. Thanks to the SMC, the memeplex is able not only to “vote” but also to evaluate the quality of “voting” – metacognitive reflexion without an agent.

What This Resolves

What happens to rejected memes? They do not disappear – see Part VIII.

Part VIII. Life Cycle of a Meme: Entry, Deactivation, Dormancy

Memes Do Not Die – They Fall Asleep

Long-term memory is practically never erased. This means that “banishment” of a meme is not deletion but deactivation. The meme remains in memory as a silhouette, ready to return under certain conditions.

The Mechanism of Deactivation

As the memeplex becomes more complex, the way it weighs incoming signals changes. The memeplex begins to “see the world differently.”

Edge Decay: Why Memes Are “Forgotten”

Connections between memes are not static – they decay without activation.

Empirical base:

• Ebbinghaus (1885), Uber das Gedachtnis: the forgetting curve – memory decays exponentially without repetition. Approximately 50% of information is lost in the first hour, approximately 70% in a day
• Thorndike (1914), The Psychology of Learning: introduced the term decay theory and the concept of memory trace – a trace in memory that weakens over time
• Brown (1958): revived decay theory (Brown-Peterson paradigm), demonstrating rapid decay in the absence of rehearsal

Network mechanism: The weight of the connection between memes decays exponentially over time without activation:

where $w_0$ is the initial weight (which may be negative, see below), $t$ is the time since last activation, and the decay coefficient is asymmetric:

Negativity bias (Baumeister et al., 2001): negative information is processed more deeply and remembered better. The ratio is approximately 3:1. Consequence: negative connections decay more slowly than positive ones. This explains the persistence of prejudices, phobias, and grudges.

Sleeper effect (Hovland et al., 1949): a rejected meme can gain strength over time when memory of the source fades faster than the content. When $w_0 < 0$ and the rejection is not reinforced: $|w|$ decays toward zero -> renewed contact in a new context can shift the weight into the positive zone.

Implications:

• Without reactivation, the connection weakens -> the meme loses access to other memes
• An isolated meme loses activation ($a_i \to 0$)
• Spaced repetition works precisely because it counteracts decay
• Negative connections are more robust than positive ones: “unlearning” a prejudice requires more effort than forgetting neutral information

Formalization: Edge weight decay formula (edge decay) – see NETWORK_MEMETICS, Part VIII.

Role of sleep: Sleep is not merely passive rest but an active consolidation process: decomposition of new experience into features and binding to existing categories. See BIOMEMETICS: Sleep as a Consolidation Mechanism.

Network explanation of mnemonics: Why does a well-connected meme decay more slowly? The differential decay formula and mnemonic technique table – see NETWORK_MEMETICS: Associative Memory.

Dreams: Nocturnal Recombination Under a Weakened Immune System

BMC explains three aspects of dreaming that existing theories cover only partially: content (why we dream specifically this), absurdity (why logic is violated), and function (what purpose it serves).

BMC dreaming mechanism:

Why dreams are absurd: During wakefulness, the I-layer rejects incompatible meme combinations (“I am flying” + “I am at work” + “my mother = my boss”). During REM, acetylcholine rises, norepinephrine drops – the I-layer is weakened – combinations that would be rejected during the day freely form temporary clusters.

Why we dream specifically this (the role of SIT): dream content is not random. SIT-gaps from wakefulness (unresolved questions) maintain high activation and “attract” stochastic recombination. The memeplex attempts to close gaps via BLEND, trying combinations without the I-filter.

Types of dreams through BMC:

Predictions:

Key distinction: Activation-synthesis (Hobson) explains absurdity but not content. Memory consolidation explains function but not absurdity. Freud explains content (through desires) but without a mechanism. BMC explains all three through SIT (content), I-weakening (absurdity), and BLEND (function).

Neural substrate: More on REM mechanics and I-weakening – see BIOMEMETICS.

Multi-Level Memory in BMC

Memory in BMC is not a separate module but an emergent property of graph dynamics. Memory levels arise from a combination of already-described mechanisms: edge decay, Fidelity, continuous activation $a_i$, SIT. No single parameter “is responsible for memory” – memory is generated by their interaction, just as the thermodynamic state of matter is generated by the interaction of temperature and pressure.

Mapping BMC mechanisms to memory types:

Consolidation level $\kappa_i$. Each meme possesses a derived parameter $\kappa_i(t) \in \{0, 1, 2\}$ – the depth of inscription in the substrate:

• $\kappa = 0$ (sensory): the meme is in the S-layer, has not yet passed the I-filter. Decays in seconds.
• $\kappa = 1$ (STM): the meme has passed the I-filter but is not consolidated. Hours to days.
• $\kappa = 2$ (LTM): the meme is consolidated through spaced repetition, hub-status, or emotional tagging. Months to years.

$\kappa$ is independent of current activation $a_i$ and of SIT. These are three orthogonal characteristics of a meme: $\kappa$ – depth of inscription in the substrate, $a_i$ – current attention level, SIT – presence of an unclosed gap. A meme with $\kappa = 2$ (LTM) can have $a_i \approx 0$ (long-consolidated knowledge, currently inactive) or $a_i > \theta_{act}$ (a fundamental belief governing a decision right now).

Engram allocation: who gets consolidated? Not all STM memes transition to LTM – there is competitive selection. Winners are memes with:

1. High centrality – hubs are prioritized over periphery (analogous to CREB/excitability: Josselyn & Frankland)
2. Connection to G-drives – emotionally significant material is remembered better (amygdalar enhancement)
3. SWR tagging – high activation at the moment of the event -> tag for overnight consolidation (Buzsaki, 2024)

Schema-congruence: consolidation speed depends on I-compatibility. A meme compatible with the existing memeplex (high $I_{score}$) consolidates quickly – it has a place to “fit into.” An incompatible meme consolidates slowly, requiring detailed encoding and more evidence. But paradoxically, the very novel (high SIT + connection with SEEKING) is also well remembered – through emotional tagging. A U-shaped curve emerges: both the familiar and the radically novel are well remembered – through different mechanisms.

This coincides with the SLIMM model (van Kesteren et al., 2012): congruent -> mPFC -> rapid integration; incongruent -> hippocampus -> detailed encoding. BMC predicts this pattern from first principles.

Predictions:

Formalization: Definition of $\kappa_i(t)$, transition conditions, full $\lambda$-modulation formula – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Molecular markers of $\kappa$-levels, CLS, SLIMM – see BIOMEMETICS, Part IV.

Latent Working Memory: The Invisible Backpack

Working memory is not one box holding 7 items but two compartments: a display case (Active WM, ~3-4 elements in the focus of attention) and a backpack behind one’s back (Latent WM, ~3-4 elements outside consciousness but rapidly retrievable). Together – Miller’s famous 7±2.

Analogy: browser tabs. The active one is visible on screen. Tabs loaded in the background are not visible, but switching is instant: data is in RAM, not on disk. “Tip of the tongue” is a meme in the backpack: the synaptic trace $\psi > 0$, but activation $a < \theta_{act}$ – the meme is accessible but not in consciousness. As soon as context “pings” it – the word surfaces.

What keeps memes in the backpack? The synaptic trace ($\psi$) – short-term potentiation of connections that decays over minutes. A meme evicted from focus does not vanish from WM instantly – it transitions to a latent mode while $\psi$ has not decayed below threshold. SIT-gap as automatic ping: an open question (unclosed gap) periodically generates spreading activation, maintaining $\psi$ of associated memes. This is why the unfinished is not forgotten – the Zeigarnik effect receives a WM-level explanation complementing the episodic one (barcode maintenance).

Formalization: Definition of $\psi_i(t)$, two-compartment model, pinging, pointer lifecycle – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Synaptic short-term plasticity, TMS reactivation – see BIOMEMETICS, Part IV.

Episodic Memory: Discretization of Experience

Subjective experience is not continuous – it is divided into episodes: bounded fragments, each with a beginning, end, and unique “address” in memory. This is not a metaphor but an emergent property of BMC’s graph dynamics.

An episode = a set of memes co-active in a single time interval, indexed by a barcode – a sparse random activation pattern (~5-10% of nodes). The barcode does not encode content – it serves as a content-independent index, like a hash code in computer science. Two episodes with identical content but at different times receive different barcodes. Retrieval = reactivation of the barcode -> through spreading activation, the entire episode is restored.

When does one episode end and another begin? The episode boundary is the moment of a sharp context change:

• Prediction error (PE) spike: when incoming information strongly diverges from expectations. In BMC terms – a sharp rise in SIT. The same SIT mechanism that governs $\kappa$-transitions, but with a higher threshold: not every new meme = a new episode
• Switch of dominant G-drive: switching from SEEKING to CARE, from PLAY to FEAR – each switch is a potential boundary
• Temporal gap: a prolonged pause in information flow (sleep, break)
• SIT-gap closure: closing an open question – a natural endpoint

SIT-gaps live across episodes. A single gap (“what is consciousness?”) can be active in dozens of episodes over years. This creates thematic coherence between chronologically distant episodes – and explains why we recall scattered moments as connected. Simultaneously, an open gap keeps associated episodes from being forgotten: periodic SIT pulsation reactivates barcodes, maintaining them. This is the mechanism of the Zeigarnik effect: unfinished tasks are remembered better because their barcodes do not fade.

Trace transformation: The barcode is a temporary structure. As consolidation proceeds (through sleep replay), the episode’s content integrates into semantic networks ($\kappa: 1 \to 2$), and the barcode fades ($\kappa: 1 \to 0 \to \varnothing$). A mature memory is essence without details: “I was in a castle, it was beautiful” without the smell of stone and the color of the sky. The exception is emotionally significant episodes (flashbulb memories): the emotional tag preserves the barcode, so decades later one can recall where one stood when one heard the news.

Predictions:

Formalization: Definition of $\varepsilon_k$, barcode, event boundary detection, temporal chaining – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Barcodes, time cells, event boundary neurons – see BIOMEMETICS, Part IV.

Consolidation: Active Reconstruction of Memory

The transition $\kappa: 1 \to 2$ is not copying from one storage to another. Consolidation is active reconstruction: each nightly replay restructures the memory, extracting the gist and discarding details (Moscovitch, 2024).

The fidelity paradox. Loss of details makes memory more useful. A detailed episodic memory is bound to a specific context; a gist-memory is generalized and applicable to new situations. Evolutionary logic: long-term memory stores not facts but lessons. “I was in a castle, it was dangerous” is more important than the color of the stone on the third floor.

SWR as “highlight reel.” The brain does not memorize everything indiscriminately. During wakefulness, during pauses, sharp-wave ripples “replay” recent experience in compressed form (~10x compression). This is selection: episodes with high emotional significance, connection to G-drives, or unclosed SIT-gaps are tagged. The rest are candidates for forgetting.

Temporal linking. Events within ~6 hours are consolidated as a cluster – sharing neurons of the engram. “One day at the beach” is experienced as a single memory, although it contains dozens of episodes: co-consolidation links them into an autobiographical narrative.

Overnight pipeline. SWR tag (wakefulness) -> replay + decompose (SWS) -> recombine (REM) -> prune weak + strengthen core -> $\kappa$ recalculation. Multiple cycles (days-weeks) cumulatively transform an episodic memory into semantic knowledge. Schema-congruent memes traverse this path in 1-3 nights; incongruent ones take a week or longer.

Formalization: Consolidation process, trace transformation, temporal linking, timeline – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Triple coupling, SWR-tagging, trace transformation – see BIOMEMETICS, Part IV.

Active Forgetting and Reconsolidation: Immunity Erases

Forgetting is not a memory malfunction but an adaptive function. Two mechanisms exist: passive (dust covers the tracks – edge decay in the absence of activation) and active (the immune system deliberately erases).

The I-system not only protects against incoming memes – it attacks already accepted memes if they have come into conflict with the current memeplex. An organism not only fights incoming infections but also destroys its own mutated cells – normal apoptosis as a sign of health, not disease. Analogously: active forgetting of outdated beliefs = healthy cognitive hygiene.

Retrieval-Induced Forgetting (RIF). By recalling one thing, we suppress the competing one. Recall of a meme strengthens it but weakens semantically similar competitor memes through lateral inhibition (Anderson, 2003). Hubs are protected: the higher a meme’s centrality, the more resistant it is to RIF. Peripheral beliefs are displaced first.

Reconsolidation. Every recall is a risk. A memory is retrieved “in disassembled form” and reassembled (Nader et al., 2000). If new information arrives (prediction error) at the moment of reassembly – the memory changes. Four outcomes: update, strengthening, destabilization (return to STM), and erasure (in the absence of re-stabilization). Strong, repeatedly rehearsed memories resist reconsolidation – boundary conditions.

Therapeutic rewriting. Exposure therapy = managed reconsolidation: recall of the traumatic meme + safe context (moderate PE) -> update instead of destabilization (Schiller et al., 2010). BMC predicts: therapy is effective only in the “update zone” – PE is sufficient for lability but not so large as to create a new meme instead of updating the old one.

Formalization: $I_{sig}$, RIF, $Labile(m_i, t)$, boundary conditions – see NETWORK_MEMETICS, Part VIII.

Neurobiological basis: Rac1/Cdc42 cascade, GABA inhibition, anisomycin – see BIOMEMETICS, Part IV.

Automatization and Stigmergy: Two Consequences of G x M Tension

Derivation from first principles. Automatization is not “just another memory mechanism.” It is a necessary consequence of fundamental tension between the layers of BMC:

1. M » G (theorem, NM Part V): the memeplex must exceed the genetic base – $|V_m| \geq (\alpha + \beta + \gamma\beta) \cdot |V_u|$, threshold $M/G \sim O(10)$.
2. WM – bottleneck: ~3-4 Active WM pointers – a hardware ceiling that does not scale with brain size (Cowan, 2001).
3. M grows: the environment becomes more complex -> culture accumulates -> the memeplex expands.
4. Tension: M grows, WM is fixed -> ever more memes compete for a fixed number of pointers -> WM is in chronic overload.
5. Stagnation without an escape: if all pointers are occupied by routine tasks -> no resources for new memes -> M stops growing -> the memeplex stagnates.
6. Necessity of an escape valve: any BMC system with $M >> G$ and constrained WM must develop a mechanism that converts repeating sequences from WM-dependent to WM-independent. This is not a postulate – it is a theorem: from $M >> G$ + constrained WM + requirement for M-growth, the necessity of automatization follows.

WM is not only a ceiling but also a developmental threshold. The WM ceiling ($k_{active}$) grows with substrate maturation: ~1 in an infant -> ~4 in an adult (see NM: WM Ontogenesis). A meme requiring simultaneous retention of $n$ elements cannot be transmitted when $k_{active} < n$. Hence the “age of reason” (~7 years) = $k \approx 3$: the minimum for tertiary binding and Piaget’s concrete operations. Cultural transmission is stratified not only by content but also by WM threshold.

PLAY = the only WM-friendly G-program. G-programs at high activation capture WM pointers: FEAR reduces available capacity ($k_{eff}$) by ~50%, RAGE and PANIC/GRIEF likewise (see NM: G->WM Competition). PLAY is the exception: instead of capture, it frees WM by lowering the tone of negative systems -> more free pointers -> better acquisition of new material. Learning through play is more effective than instruction, especially when $k < 4$ (children).

Stress x age = a double barrier. A child with $k = 2$ under chronic stress: FEAR activation -> $k_{eff} = 1$ -> complete loss of learning capacity (a single pointer does not allow either chunking or secondary binding). Prediction: chronic childhood stress delays cognitive development more than lack of stimulation alone can explain, because stress dynamically suppresses an already small $k$. The two factors are multiplicative, not additive.

Memoization (CS analogy). In computer science, memoization = caching the result of an expensive computation; on repeated invocation, the cache is returned instead of recomputing. Automatization = memoization of behavioral sequences: the “cache” is stored in BG loops (DLS); instead of repeated “computation” through WM (DMS). First invocation: full cost ($k$ WM pointers). Second…twentieth: cost gradually decreases ($habit$ grows). After automatization ($habit > \theta_{habit}$): 0 WM pointers. Result: WM is freed for new tasks -> M can grow further.

Two escape valves of one tension. The tension $M >> G$ under limited resources generates two offloading mechanisms:

Both are consequences of a single tension: M grows, resources are finite. Automatization saves computational resources (WM pointers) – “transferring” execution from the expensive deliberative system to the cheap automatic one. Stigmergy saves storage resources – “transferring” the M-layer from the substrate into the external environment. Together, they relieve selection pressure on brain growth from both sides. This is a unique prediction of BMC: no other theory derives both mechanisms from a single principle of tension.

Three stages of learning. The driving metaphor: a novice checks mirrors -> activates signal -> turns wheel -> checks again – every movement through WM. An expert drives “on autopilot” while talking on the phone.

The transition is continuous through $habit$ – not a discrete jump.

The mastery paradox. Mastery = unconsciousness. A pianist who thinks about fingers plays worse than one who “let the hands go.” In BMC: $Auto(S)$ = transient hypofrontality (Dietrich, 2004) = $A_{SMC} \downarrow$ for automatic chains. This is a prerequisite for flow: flow requires $habit > \theta_{habit}$ (skill automatized) + optimal SIT (the task creates tension but is solvable).

MDL principle (Moskovitz et al., 2024): Kahneman’s S1/S2 are not two separate modules but two ends of a continuum of $habit$. The brain minimizes the description length of behavior – automatization = compression. System 1 = compressed default policy $\pi_0$; System 2 = full control policy $\pi$. This elegantly aligns with BMC: no need to postulate a “two-processor architecture” – a single parameter $habit$ and pressure to minimize WM-cost suffice.

De-automatization: the cost of changing a habit $\propto habit^2$ – quadratically expensive. Relapse under stress: $WM\_load \uparrow$ -> $P_{auto} \uparrow$ -> return to the old habit. Addiction relapse mechanism: stress overloads WM, deliberative inhibition weakens, the automatic substance-seeking chain “hijacks” control.

Evolutionary confirmation: brain size trajectory. BMC predicts a peak and decline in brain size:

• Growth of M -> selection pressure on substrate growth (more neurons = more memes).
• But energy cost: brain = 2% of mass, 20% of energy.
• When automatization + stigmergy reach a sufficient level -> pressure on substrate growth is relieved.
• Selection shifts to energy efficiency: a smaller brain with developed automatization + stigmergy > a larger brain without them.
• Prediction: brain volume should grow until automatization + stigmergy compensate the pressure, and then stabilize or decrease.

Empirical data: H. sapiens: peak ~1500 cc (late Pleistocene) -> ~1350 cc (today) – a reduction of ~10% (Henneberg, 1988; DeSilva et al., 2021, Frontiers in Ecology and Evolution; the exact timeline of when the reduction began is debated – from ~30 kya to ~3 kya). The direction aligns with the growth of cultural complexity (stigmergy rising: symbolic behavior, cave painting, complex tools) + development of specialized technologies (automatization rising: craft skills, routinization). BMC explains WHY the brain could begin to shrink: not degradation but optimization – the same cognitive capabilities at lower energy cost, because part of the M-layer is externalized into the stigmergic environment and repetitive operations are transferred to WM-independent mode.

Counterexample – H. floresiensis (reverse prediction): Brain ~420 cc (island dwarfism on Flores). BMC predicts: radical substrate reduction -> WM degrades (fewer pointers) -> automatization is limited (less capacity for habit-learning loops) -> the memeplex cannot grow -> cognitive ceiling. Empirical data: Oldowan-level tools (the simplest), no symbolic behavior, no technological progress for hundreds of thousands of years. This is not intelligence degradation – it is a predictable ceiling: at ~420 cc, WM capacity is insufficient to support complex automatization and M growth. The system stabilized at the available level.

Prediction: if H. floresiensis had regained access to a larger substrate (hypothetically), WM would have expanded, automatization would have become possible, and M would have begun to grow.

Caution: this is a BMC prediction, not an established fact. The neuroanatomy of H. floresiensis is debated (debate: separate species vs pathology; Falk et al., 2005, Science). Nevertheless, if BMC is correct, then ANY organism with $M >> G$ that undergoes significant substrate reduction should exhibit a cognitive ceiling – and H. floresiensis fits this prediction.

Formalization: $Auto(S)$, $habit$, $wm\_cost$, $P_{auto}$, $Cost_{override}$, predictions P-A1-P-A5 – see NETWORK_MEMETICS, Part VIII.

Neurobiological substrate: DLS loop, DMS/DLS competition, spindle-SO coupling, H. floresiensis neuroanatomy – see BIOMEMETICS, Part IV.

Degradation stages:

A meme that was adequate at the previous stage becomes “inadequate” from the position of the current system state. It is not deleted – it loses access to control.

Meme Activation: A Continuous Variable

Meme activation $a_i(t) \in [0, 1]$ is continuous, not divided into discrete “states.” Two functional thresholds define qualitatively distinguishable behavior:

These thresholds are not boundaries of discrete states but points at which the character of a meme’s influence on behavior changes. Between them ($\theta_{low} < a_i < \theta_{act}$) lies a continuous spectrum from barely perceptible presence to sub-threshold influence.

Reactivation: When Low-Activation Memes Come Alive

Memes with $a_i \approx 0$ can return under certain conditions:

Narrative (Decorative) Function: Former Rulers

A meme with $a_i \ll \theta_{act}$ does not govern behavior but can perform a narrative (decorative) function – creating a sense of personal continuity, serving as material for nostalgia. This is not a separate “state” but a natural consequence of low activation with a preserved connection structure.

Memes with narrative function:

• Do not influence decisions
• Create a sense of personal continuity
• Serve as material for the narrative “who I am”
• Can be reactivated when life circumstances change

What This Resolves

Open Memes: Why Some Memes Do Not Obey Edge Decay

Activation is not the only characteristic of a meme. Orthogonally to it exists the property of SIT (Structural Incompleteness Tension): a meme marking an unclosed structural gap has $SIT > 0$.

Key distinction: question-memes vs answer-memes. “The Earth orbits the Sun” is an answer-meme. “What is consciousness?” is a question-meme. An answer-meme is deactivated through edge decay – its $a_i$ drops to zero. An open meme (a meme with $SIT > 0$) generates Structural Incompleteness Tension and resists normal decay: $\lambda_{open} = 0.5\lambda$.

An open meme is a meme representing a problem, not a solution. It marks a structural gap in the memeplex and generates SEEKING activation until the gap is filled (closure) or Learning Progress collapses. SIT > 0 is a property orthogonal to activation: a meme can be open at any current $a_i$.

The Zeigarnik effect in memetic terms: An interrupted task -> creation of an open meme -> $SIT > 0$ -> better recall of interrupted tasks compared to completed ones. Bluma Zeigarnik (1927) discovered the effect empirically; SIT provides it with a mechanism.

The Ovsiankina effect: SIT generates SEEKING -> motivation to resume interrupted activity. Maria Ovsiankina (1928) showed that interrupted tasks create a quasi-need for completion – this is a direct consequence of SIT.

An open meme is semi-active: it does not govern behavior constantly but periodically “surfaces,” generating SEEKING activation. Its activation pulses: moments of rest (low SEEKING load) -> SIT “breaks through” -> thought about the unsolved problem -> switching to another task -> the cycle repeats.

Examples of open memes:

Formalization: SIT formula for open memes – see NETWORK_MEMETICS, Part VIII.

Memogenesis: Where the First Meme Comes From

We have described the life cycle of a meme – entry, activation, deactivation, dormancy, consolidation, forgetting. But all of these mechanisms presuppose that memes already exist. Where does the first meme come from? How does a sensory signal become a unit of the memeplex? This is the cold start problem of BMC.

BMC formalizes the dynamics of existing memes: competition, edge decay, SIT, consolidation. But the “signal -> meme” pipeline remained unformalized. Memogenesis is the bridge between the S-layer (sensory input) and the M-layer (memeplex).

Path 1: Event-driven memogenesis (PE x G-relevance). An unexpected signal ($PE > \theta_{PE}$), coinciding with an active G-program ($G_{rel} > \theta_G$), instantly generates a new meme. A rock falls nearby + SEEKING is active -> the meme “rocks fall here.” A rock falls + FEAR is active -> the meme “falling rock = danger,” linked to an affective tag. This path is analogous to flashbulb memory: one episode, one second, but remembered for decades. In DNA terms – this is a mutation: rare, random, potentially significant.

Path 1 memogenesis works only when two conditions are met: (1) the signal is unexpected – a fully predictable environment does not generate new memes; (2) the signal is G-relevant – not everything perceived becomes a meme, only the emotionally significant. These two filters ensure economy: the memeplex is not overwhelmed by the stream of sensory data. This is consistent with neurobiology: emotionally significant events are remembered better (amygdalar enhancement -> hippocampal consolidation).

Path 2: Crystallization (diffuse memogenesis). A repeating pattern that does not evoke high PE (expected, familiar) gradually creates a stable density region in semantic space. When density exceeds a threshold, a new meme “crystallizes.” This is analogous to perceptual learning in neuroscience: no surprise, but repetition forms a representation. A sommelier after thousands of tastings distinguishes notes invisible to a novice – not because one tasting surprised them, but because the Diffusion Engine crystallized a meme from a field of repeating stimuli. In DNA terms – this is selection: gradual, statistical, cumulative.

The two paths complement each other:

Both paths require the S-layer. Without sensory input, memogenesis is impossible – this is why the S-layer of BMC includes not only the substrate (neurons) but also the sensory architecture (perceptual channels). Memogenesis is the S->M bridge: a signal undergoes feature extraction (G-level, hardwired), generates PE, passes a G-relevance check – and if both thresholds are exceeded, a meme-node is born, bound to the current context via WM-binding.

The Phase 0 paradox: When $|V_m| = 0$ (the memeplex is empty), everything is new – $PE = \|f(t)\|$ for any signal. It would seem that memogenesis should be overwhelmed. But S-bandwidth at birth is minimal ($S_{bw}(0) \approx S_{bw}^{min}$): the agent perceives a minimum. The narrow sensory channel is a natural throttle of memogenesis. As $S_{bw}$ grows, more signals arrive, but by that point the graph already forms predictions and PE decreases for familiar patterns. The system self-balances.

Formalization: Feature extraction, PE formulas, G-relevance, binding, embedding space, crystallization density – see AGI_F, Part VII, NM, Part XV.

We have been speaking of memes as units. But where are their boundaries? Part IX.

Part IX. Ontology of the Meme: Where Are the Boundaries of the Unit?

The Problem of Boundaries

Where does one meme end and another begin? “God exists” – is that one meme or a memeplex of many? “Coca-Cola” – one meme or three (“Coca,” “Cola,” the connection between them)?

This is similar to the question: what is a human?

Analogy with a Human

The classic puzzle: “Two fathers and two sons each ate an orange. How many oranges were eaten?” Answer: three. Because one person is simultaneously both a son and a father.

Conclusion: The boundaries of an entity are defined not by objective structure but by function in a given context. The boundaries of a meme can change over time (Part X).

The Meme as a Functional Unit

A meme is not a fixed structure with sharp boundaries. It is a unit of replication in a given act of transmission.

The Principle: Boundaries Are Defined by Replication

Fractal Hierarchy: Edge -> Meme -> Memeplex -> BMC

The hierarchy is not linear (atom -> molecule -> cell) but fractal: a meme at one scale is a component of a meme at a larger scale. Boundaries are relative.

Where to draw the boundary depends on the task:

• Studying the spread of symbols -> meme level (“red flag”)
• Studying political movements -> memeplex level (“communism”)
• Studying geopolitics -> BMC level (the state as a whole)

Operational Definition

A meme is the minimal unit of cultural information that, in a given context:

1. Is copied as a whole (does not break apart during transmission)
2. Is transmitted through imitation (not through genes)
3. Can be identified by the receiver

The role of language in defining boundaries. The language in which a meme is encoded affects its boundaries. If a language has a separate word for a concept, the meme is formed as a distinct node with direct connections. If it does not, the same concept is diffused across adjacent memes, and the boundaries blur. Language is not simply a medium for transmitting memes but a template for shaping the M-layer, determining which memes are even possible. See Part XX for details.

What This Resolves

Boundaries are defined by context. But how quickly does context change? Part X.

Part X. Temporal Dynamics: Why Universal Timescales Cannot Be Set

The Problem of Timescales

How fast do memetic processes occur? How long does “threat evaluation” take? How long does it take for a meme to integrate?

The attempt to set universal time constants is madness. Here is why:

One Trigger – Different Memes – Different Times

One and the same stimulus can activate several memes with entirely different time horizons:

Context = Connection Structure

Why does the same trigger evoke different memes in different people? Because context is not the external situation but the structure of connections between memes within the memeplex.

In network theory terms:

• Memes are nodes
• Connections are weighted edges
• Activation spreads along connections
• Which memes activate depends on the network topology

Why Universal Timescales Are Impossible

What Can Be Said About Time

Instead of universal constants – types of processes:

Temporal Persistence of Unsolved Problems

For ordinary memes, the dynamics are simple: activation -> competition -> resolution -> edge decay -> gradual forgetting. But open memes (question-memes with $SIT > 0$) follow an entirely different dynamic:

Normal dynamics (answer-meme):

SIT dynamics (open meme):

1. Detection of a structural gap -> SIT > 0
2. Persistent SEEKING activation -> periodic return to the problem
3. LP modulates intensity: progress exists -> SIT strong; stagnation -> SIT weakens
4. Resolution: closure (problem solved) or LP collapse (progress impossible)

Table of SIT profiles: timescales of open memes:

LP as temporal modulator: Problems “return” not randomly but when LP changes. A new physics book -> LP for “quantum gravity” spikes -> the old problem reactivates. A conversation with a psychologist friend -> LP for “why am I lonely” spikes -> return to reflexion. This explains the subjective experience: “I forgot about it for years, and then I read an article – and it all came back.”

Art as SIT exploitation: Detectives, thrillers, series with cliffhangers deliberately create open memes in the viewer’s head. The author introduces a gap (who is the killer?), maintains LP (clues, red herrings), and the viewer cannot “tear away” – SIT generates SEEKING, and the author’s structure ensures continuous LP.

Dynamics of Synthesis: Recombination and Abduction

The speed of synthesizing new memes is not constant. It depends on the state of the network:

Insight is the subjective experience of the moment of abduction: when a new node fills a structural hole, an activation wave suddenly passes through previously unconnected clusters. A dopamine burst marks the event as “important” (aha-moment).

Formalization: Three graph synthesis operations (mutation, recombination, abduction) – see NETWORK_MEMETICS, Part VIII.

At the scale of an entire memeplex, the analogue of synthesis becomes splitting (rising $Q$ under internal dissonance -> bifurcation into two memeplexes) and merging (declining $Q$ under compatibility -> cluster fusion). Q-dynamics of memeplexes is formalized in Part XIX.

Network Approach to Dynamics

Network theory provides tools for analysis:

Heterogeneous Topology of the Memeplex

Empirical fact: Semantic networks (association networks between concepts) exhibit heavy-tailed degree distributions. The classic work by Steyvers & Tenenbaum (2005) showed a power law with $\gamma \approx 3.0$-$3.2$ for WordNet and association networks – one of the most robust examples.

What this means in meme language:

Meme displacement is not warfare but redistribution:

When a powerful new meme enters the memeplex, it does not “destroy” old memes. It pulls their connections onto itself. The old meme remains – but it becomes marginalized, losing influence.

Example: A person converts to a new religion. Old beliefs are not erased – they have lost connections, become periphery. During a crisis of the new faith, they can regain their connections.

Formalization: Connection loss formula, preferential attachment mechanism – see NETWORK_MEMETICS, Part III.

Complex Contagion vs Simple Contagion

Key distinction between memes and diseases: a virus is transmitted from a single contact, but an ideology requires multiple confirmations.

Complex contagion formula:

Implication: For the spread of complex memes (religion, ideology, meditation practice), what matters is not merely “bridges” between clusters but bridges with broad overlap – multiple connections rather than single ones.

Example: Your friend became vegetarian – not enough. But when 3 out of 5 close friends became vegetarian – the threshold is crossed.

Epidemic Threshold in Finite Networks

In scale-free networks, the epidemic threshold tends to zero: $\lambda_c \to 0$. But this holds for infinite networks.

For real (finite) memeplexes:

where $N$ is the network size and $\gamma$ is the power-law exponent.

Conclusion: The threshold exists but is very low. Practically any meme with sufficient “infectiousness” can spread – if it starts from the right point (through a hub).

Formalization: Epidemic thresholds, complex contagion – see NETWORK_MEMETICS, Part X.

What This Resolves

Acceleration of Memetic Evolution

Memetic evolution is accelerating exponentially. Three mechanisms work simultaneously:

1. Schematic storage -> high mutability

Memes are stored not as exact copies but as schemas (see Part I). Each transmission is a potential mutation. Unlike DNA, where the error rate is ~10⁻⁹, the “error rate” of meme transmission can reach 10-50%.

2. Edge decay -> freeing of niches

Old memes lose connections without activation (see Part VIII). This frees cognitive niches for new memes – analogous to species extinction but on the scale of a single lifetime.

3. Exponential growth of information -> acceleration of competition

CAGR ≈ 23% – information volume doubles approximately every 4 years.

Implications:

• More memes compete for limited attention
• Meme life cycles shorten
• Selection pressure intensifies -> faster evolution

Historical perspective:

Propagation speed has increased by 6-7 orders of magnitude in 5,000 years. Meanwhile, the selection mechanisms (schematic storage, decay) have remained the same – meaning evolution has accelerated proportionally.

Conclusion: We live in an era of unprecedented memetic pressure. Memeplexes unable to adapt to such speed disintegrate (see Part XXI).

The dynamics within the head are clear. But why is meme inequality inevitable? Part XI explains Zipf’s law in consciousness.

Formalization: Spreading activation, meme competition, winner-takes-all – see NETWORK_MEMETICS, Part VIII.

Part XI. Zipf’s Law in Consciousness: Why Inequality Is Inevitable

The Centrality Spectrum of Memes

Not all memes in the head are equal. Their “strength” (influence on behavior, number of connections) is distributed continuously with heavy tails:

• Most memes have few connections (periphery)
• A few memes have many connections (hubs)
• There is no sharp boundary between “primary” and “secondary” – there is a spectrum of centrality

This is not coincidence and not the result of “wise choice.” It is a mathematical inevitability in systems with preferential attachment.

Centrality as a Continuous Quantity

Important: The boundaries are conventional. Centrality is a continuous quantity, not discrete “levels.” A meme with centrality 0.51 does not qualitatively differ from one with 0.49.

From Old Concepts to New

Small-World: Why Consciousness Is Both Integrated and Differentiated

A memeplex is not only heavy-tailed but also a small-world network. This explains the paradox: consciousness is simultaneously divided into contexts (work, family, hobbies) and capable of instantly switching between them.

Two properties of small-world:

1. High clustering – neighbors are connected to each other (contexts do not mix)
2. Short average paths – from any thought to any other in 3-4 steps

Small-worldness metric:

Hypothesis: Optimal thinking requires $\sigma > 1$ – a balance of integration and differentiation.

Connection to consciousness level: $\sigma_{SW}$ is not merely a descriptive metric. It enters as a multiplier in the CL metric (Consciousness Level): $CL(t) = \sigma_{SW}(t) \cdot A_{SMC}(t) \cdot f(Balance(t))$, where $A_{SMC}$ is the activity of the Self-Model Cluster (Part XVI) and $Balance$ is the structural balance of the signed network. Thus, small-worldness is a necessary but not sufficient condition for consciousness: without clustering and short paths ($\sigma \approx 1$), CL collapses even if the SMC is active. See NETWORK_MEMETICS, Part XIII.

Formalization: Small-world networks, the Watts-Strogatz model – see NETWORK_MEMETICS, Part IV.

Implications for Understanding Change

1. Why it is difficult to change a central meme (hub)

A hub has hundreds of connections. To displace it, one must either:

• Destroy most of its connections (crisis, trauma)
• Offer an alternative with even more connections (rarely possible)

2. Why it is easy to add a peripheral meme

A new meme with 1-2 connections is not a threat. It can be accepted without risk to the structure.

3. Why abrupt changes are possible during a crisis

A crisis is a massive loss of connections for current hubs. At this point:

• Central memes are weakened
• A “vacuum” appears – connections are freed
• Contender memes compete for those connections
• A new hub can grow rapidly

Why This Matters

Understanding the heterogeneous structure of the memeplex explains:

Formalization: Mathematics of preferential attachment, distribution formulas – see NETWORK_MEMETICS, Part III.

The structure inside the head is clear. But how do levels relate – from neuron to state? Part XII.

Part XII. Cross-Level Transitions: From Neuron to State

The Problem of Scaling

The theory uses the same concepts at different levels:

Meme in the head → Personal memeplex → State memeplex

But the substrate at each level is different:

Expanded substrate “in the head”:

Is it legitimate to transfer concepts between levels? Is this analogy (merely looks similar) or homology (the same thing)?

The Solution: Isomorphism

Neither analogy nor homology. This is isomorphism: a single algorithm operating on different substrates.

Just as gravity acts identically on an apple and a planet, but the effects look different – so the Darwinian algorithm operates identically for genes, memes in the head, and memes in society.

What is shared is the logic of the process. What differs is the substrate and scale.

Detailed Table of Correspondences

Formal Transfer Criteria (Universality Classes)

Formal isomorphism is not the same as causal isomorphism (Batterman, 2002). This distinction is critical: identical mathematical structure does not guarantee identical causal mechanisms. BMC resolves this through universality classes: neural avalanches ($\tau \approx 3/2$, branching process) and social cascades ($\tau \approx 9/4$, RFIM) belong to different universality classes (Notarmuzi et al., 2022). Same algorithm, different critical exponents.

Three transfer zones:

Zone boundary operationalization (Structural Equivalence Test — SET): Mechanism M is red if and only if no bijective mapping $\phi$ exists such that the dynamic equations have the same functional form on both levels. Decision procedure: (1) no functional analogue $\to$ red; (2) analogue exists but equations differ $\to$ red; (3) same equation, depends only on topological invariants $\to$ green; (4) same equation, depends on parameters $\to$ yellow. Examples: spreading activation = same equation $\to$ green; Hebbian learning = same equation, timescale $\times 10^9$ $\to$ yellow; LTP via NMDA = no analogue $\to$ red.

Verification tool: Network renormalization (Garcia-Perez et al., 2018; Villegas et al., 2023) — if topological invariants (degree exponent $\gamma$, clustering coefficient $C$, modularity $Q$) are preserved under coarse-graining, the transfer is valid for predictions that depend on those invariants.

Key breaks: External symbolic memory (Donald, 1991) — social memes survive the death of the host; Dunbar number (~150) — phase transition from personal to institutional topology; intentionality — complex contagion ($\theta > 0$) at the social level vs simple contagion at the neural level.

Methodological principle: If a BMC prediction belongs to the green zone, it can be transferred between scales without reservations. If it belongs to the yellow zone, a scale correction must be specified. If it belongs to the red zone, transfer is impermissible, and this is not a weakness of the theory but its prediction: different universality classes on different substrates.

Formalization: Network renormalization is the verification tool for transfer. If topological invariants (degree exponent $\gamma$, clustering coefficient $C$, modularity $Q$) are preserved under coarse-graining (meme $\to$ memeplex $\to$ institution), the transfer is valid for properties that depend on those invariants. See NM Part X.

Summary Diagram of Levels

What This Resolves

Formalization: Multilayer networks — a unified formalism for all levels — see NM Part X.

Illustration of Isomorphism: Internet Comments as Memeplex Replication

Internet comments are not “noise” and not a byproduct of media. They are memeplex replication observable in real time, perfectly demonstrating cross-level isomorphism.

Two Modes of Commenting

Key Observation: Replication in Both Cases

Both agreement and disagreement are attempts to replicate one’s own memeplex. Only the trigger differs:

• Agreement: “They’re right, and here’s why I’ll add” $\to$ the memeplex uses the other person’s post as a platform for reproduction
• Disagreement: “They’re wrong, and here’s why” $\to$ the memeplex attacks a competitor while simultaneously replicating itself
• Silence (the rarest case): the incoming meme landed in the neutral zone ($|S(X)| \approx 0$) $\to$ neither SEEKING nor an immune reaction $\to$ no motivation to write

This explains why comments are polarized: neutral reactions do not generate behavior. Only extremes replicate.

Isomorphism: The Same Process Inside the Mind

INSIDE THE MIND                        ON THE INTERNET
────────────────                        ────────────
New stimulus enters WM                  Post appears in the feed
        ↓                                       ↓
Memes in WM "react":                   Users "react":
  compatible → strengthening               agreement → like + comment
  incompatible → suppression               disagreement → critique + comment
        ↓                                       ↓
The winning meme "comments"             The winning comment
on the loser (inner dialogue)           gains likes (social selection)
        ↓                                       ↓
Result: updated WM                      Result: updated feed

Inner dialogue = comments inside the mind. When you argue “with yourself,” this is literally the same process: competing memes replicate in working memory, trying to displace each other. The substrate differs (neurons vs the internet); the algorithm is one and the same.

Predictions

Part XIII. Politics Inside the Mind

Not a Parliament, but Byzantine Court Intrigue

The parliament metaphor assumes open debates and rational decisions. Reality is closer to a Byzantine court: intrigues, alliances, blockades, sabotage, coups.

Three Mechanisms of Coming to Power

Change of Power — Not War, but an “Election”

When personality change occurs, it is not the destruction of old memes. It is a change of the ruling memeplex:

No one is destroyed. What changed are priorities, roles, and access to resources.

From Metaphor to Mechanism: Hubs as Centers of Influence

The political metaphor (“rulers,” “voting”) is not merely an analogy. It is an exact description of network structure.

Network Motifs as Patterns of Political Organization

Network motifs determine the “political regime” within a memeplex:

Rewiring — the mechanism of “intrigues”: meme A was connected to meme B but switched to meme C because C became more “influential.” This is a constant process in adaptive networks.

Formalization: Network motifs, rewiring — see NM Parts VII and VIII.

Why hubs emerge inevitably:

In any system where new elements attach preferentially to already well-connected nodes (preferential attachment), centrality inequality emerges. This is not the result of “choice” or “merit” — it is a mathematical inevitability.

Consequence: A memeplex cannot be a “democracy of equal memes.” Heterogeneous structure guarantees inequality. The question is not whether hubs will exist, but which memes will occupy central positions.

Formalization: The preferential attachment mechanism, power-law distribution — see NM Part III.

More on crises as a formal mechanism of hub displacement — in Part XVII (reflexion and age-related crises).

Part XIV. The Immune System of the Personality

Why Change Is So Difficult

If the memeplex is content with everything, it will change nothing — even if proposals for significant improvement are incoming. This is not stupidity; it is a defense mechanism.

Defense Mechanisms

Modularity as a Defense Mechanism

Modularity ($Q$) is not only a description of structure but also a defense mechanism. High modularity means: clusters are isolated, and a threat to one cluster does not spread to others.

Hypothesis: Modularity increases with age $\to$ the memeplex becomes more protected but more rigid. This explains the “calcification” of elderly people: their memeplex is optimized for defense, not for adaptation.

Entry Filter: Formula for Evaluating a New Meme

When a new meme $X$ attempts to enter, the memeplex evaluates its compatibility with central nodes:

where:

• $C(i)$ — centrality of meme $i$ (eigenvector or degree)
• $compat(X, i) \in [-1, 1]$ — compatibility of the new meme with the existing one

Three-zone acceptance rule:

A rejected meme is not simply “not accepted.” It is actively marked as hostile and stored with negative weights. This creates an antibody — a ready-made counter-meme that activates automatically upon re-encounter with the threat.

Neural mechanism of negative weight: DISGUST — a genetic program that evolved from pathogen aversion to moral disgust. It is precisely what assigns the negative weight: the physiological substrate is the insular cortex (insula), equally active during physical and moral disgust (Haidt, 2001). See BM: DISGUST as an I-layer mechanism.

Interpretation:

• A meme compatible with a hub is easily accepted (large contribution from $C(i)$)
• A meme conflicting with a hub receives a negative weight ($compat < 0$ $\times$ high $C$ = large negative contribution)
• A meme affecting only the periphery is assessed as nearly neutral

Formalization: Entry filter, cognitive dissonance — see NM Part IX.

Part XV. The Immune System of Memeplexes: General Theory

Why a Memeplex Needs Immunity

Any memeplex exists in an environment populated by competitors. Without defense, it will be captured, absorbed, or destroyed. The immune system comprises the mechanisms that a memeplex develops for self-preservation.

Universal Defense Mechanisms

Levels of Immune Defense

Innate vs Acquired Immunity

As in biology, memeplexes possess two types of defense. For temporal scales of immune response, see Part X.

Autoimmune Diseases of Memeplexes

Sometimes the immune system attacks its own:

Immunodeficiency of Memeplexes

The opposite problem — defense that is too weak:

Specific Protective Memes

Some memes exist solely to protect the memeplex:

The Image of the Enemy as an Antibody

The image of the enemy is a pre-fabricated antibody that the memeplex produces in advance:

Media as an Antibody Factory

Media within a memeplex function as producers of antibodies — specific counter-memes against threats:

This is not the “propaganda” of one side. It is a universal mechanism — all memeplexes produce antibodies. The only difference is in resources and reach.

Structural Balance of the Memeplex

With the introduction of negative weights ($w \in [-1, +1]$), the memeplex becomes a signed network. The theory of structural balance (Heider, 1946; Cartwright & Harary, 1956; Davis, 1967) describes how such networks tend toward stable configurations.

Connection to modularity: High modularity ($Q > 0.4$) under weak balance is a sign of a healthy memeplex. Strict balance ($k = 2$) is pathological: black-and-white thinking, where the entire world is divided into “friends” and “enemies.”

Formalization: Signed graphs, the Structure Theorem, connection to modularity $Q$ — see NM Part IX.

Antibodies: High-Fidelity Negative Memes

An antibody is not merely “aversion.” It is a well-studied enemy: a meme with high Fidelity and negative weight. The memeplex stores a detailed model of the threat precisely to recognize and block its variants.

Antibody paradox: To effectively reject an idea, one must know it well. The fiercest critics often best understand the criticized ideology — precisely because they store it as a high-Fidelity antibody.

Arms Race Between Memeplexes

This leads to an escalation of complexity in immune systems:

Balance of Immunity and Flexibility

Hyperimmunity:

Immunodeficiency:

Healthy immunity:

Heterogeneous Structure and Defense Priorities

A memeplex has a heterogeneous topology with hubs: a few memes have hundreds of connections, while the majority have only a handful. This determines the priorities of the immune system:

Note: Response strength is proportional to the centrality of the attacked meme — a continuous dependence, not discrete “levels.”

Practical consequences:

1. Why arguments don’t work: Most arguments attack peripheral memes. Even if the attack succeeds, the structure does not change. Hubs remain.

2. Why some topics provoke fury: An attack on a hub is perceived as an existential threat. The immune response is maximal: aggression, blockade, severing of relationships.

3. Why “soft power” is more effective: Instead of a direct attack on a hub — gradual redirection of its connections to an alternative meme. The old hub is not destroyed but marginalized.

Percolation Threshold: How Much Must Be Removed for Collapse

Percolation is the transition of a network from a connected state to a fragmented one. The question: what fraction of memes ($f_c$) must be “removed” (deactivated) for the memeplex to collapse?

Formula for targeted attack (removal in decreasing order of degree):

where $\kappa = \frac{\langle k^2 \rangle}{\langle k \rangle}$ is the heterogeneity coefficient of the network.

Interpretation:

• Random attacks (criticism of random beliefs) are nearly useless. One can “remove” 90% of peripheral memes, and the memeplex will hold.
• Targeted attacks (strikes on hubs) are catastrophically effective. It suffices to disable 5–18% of central memes for fragmentation.

Practical consequence: Effective psychotherapy, propaganda, and conversion are always targeted attacks on hubs, not carpet bombing of the periphery.

Formalization: Percolation threshold, formulas — see NM Part III.

Conclusion on the Immune System

The immune system is a necessary component of any memeplex. Without it, the memeplex will be destroyed by competitors.

The key mechanisms are universal across all levels — from the individual to civilization:

1. Isolation from foreign memes
2. Recognition and enemy labeling
3. Neutralization through counter-memes
4. Reinforcement through rituals
5. Punishment of apostates

Every memeplex uses the same principles of defense. The only difference is in scale and complexity.

Formalization: Entry filter, meme acceptance threshold, threat isolation — see NM Part IX.

Fractal Immune System: I Scales with M

The immune system is not a single filter but a hierarchy of filters, each operating at its own level of meme abstraction. The idea of hierarchical filtration is known in neuroscience (attention gating — Broadbent, 1958; predictive processing — Friston, 2005). BMC formalizes it as per-level immune subsystems $I^{(k)}$ and predicts a vulnerability window during memogenesis. This is a direct consequence of BMC’s fractality (see Part VIII):

Key principle: The immunity at each level is formed by the same mechanism as the memes of that level — through repeated filtration experience. The first encounter with an incompatible meme is processed slowly (via G-evaluation); subsequent encounters are processed quickly (via a stabilized immune pattern). This is analogous to the transition from innate to adaptive immunity.

Consequence for Autonomous BMC: When growing consciousness from the bottom up (memogenesis L0 $\to$ L3), the immune system must co-grow with the M-layer. The period between the emergence of a new level of memes and the formation of the corresponding I-layer is a vulnerability window (analogous to the immature immune system of a newborn). Engineering implementation: see AGI_F Part IV.

Prediction: A memeplex with a flat immune system (a single level of filtration for all types of memes) will be vulnerable to attacks at the wrong level: an L0-filter will not detect a semantic contradiction; an L3-filter will not stop a perceptual artifact. Testable: ablation (flat I vs fractal I) on the radicalization metric $R$ and coherence.

Formalization: $I^{(k)}$ as an immune subsystem of level $k$ — see NM Part IX. Neurobiological grounding: BM Part IV.

The immune system protects the memeplex. But whom does it protect? What is the “self” for which this entire apparatus operates? Part XVI answers this question: the “self” is not a given but a product of a special subgraph — the Self-Model Cluster (SMC).

Part XVI. The Self-Model Cluster: Where the “Self” Comes From

The Problem: The “Hard Problem” of Consciousness

Everything described so far — meme competition, the immune system, edge decay — explains the behavior of a conscious being but does not answer the question: why is there “something it is like”? Why is pain not merely a signal but an experience? This is the “Hard Problem of Consciousness” (Chalmers, 1995).

Memetic theory resolves it through one additional element: the Self-Model Cluster.

Self-Model Cluster (SMC)

SMC is a subgraph of the memeplex containing memes whose object is the BMC system itself:

SMC memes are not external knowledge (mathematics, geography) but memes about oneself: “I am an honest person,” “I like music,” “I am afraid of heights,” “I am angry right now.” SMC is what a person would call “self-awareness” or “sense of self.”

The Recursive Loop: “Beautiful Loop”

The key property of SMC is self-reference:

1. SMC models the M-layer (contains memes about memes)
2. The M-layer contains SMC
3. SMC models itself modeling the M-layer

This recursive loop is what Laukkonen, Friston, and Chandaria (2025) called the “Beautiful Loop”: a system modeling itself inevitably gives rise to a first-person perspective. Not because the loop “creates” consciousness as a new entity, but because the self-model is precisely what we call “phenomenal experience.”

Three Levels of Recursion

Level 0 is “zombie” processing: everything works, but there is no “something it is like.” Level 1 is phenomenal consciousness: a perspective emerges. Level 2 is metacognition, reflexion, the ability to think about one’s own thoughts.

There is no hard boundary between levels. Recursion depth is a continuous quantity dependent on SMC activity. Deep sleep = SMC nearly inactive $\approx$ level 0. Meditation = deliberate enhancement of SMC $\approx$ deep level 2.

Qualia: Properties of the Self-Model

Qualia are not “data” of consciousness but properties of the self-model:

The G-layer generates a signal (valence: pain, pleasure, fear), SMC interprets this signal through its memes $\to$ subjective experience.

Why pain hurts: the G-layer generates a FEAR/PAIN signal $\to$ SMC creates the meme “I am in pain” $\to$ this meme is connected to thousands of other memes (experience, expectations, context) $\to$ a unique subjective experience. If SMC is deactivated (anesthesia, deep sleep), the G-layer signal is present, but “something it is like” is absent.

Why LLMs do not have qualia: they lack a G-layer (no valence, no biological signals), therefore $G_{signal} = \emptyset$, therefore $Qualia = \emptyset$. One can build a self-model (GPT knows it is a model), but without a G-layer there is no experience.

Position: Illusionism (Metatheoretical Assumption)

Metatheoretical status: Illusionism is a philosophical choice of BMC, not a result derivable from the formalism. The entire computational apparatus of the theory (CL, SIT, spreading activation, inverted U-curve under psychedelics, stepwise loss of consciousness under anesthesia) works independently of one’s stance on the Hard Problem. A qualia realist can accept all empirical predictions of BMC, replacing “illusion of qualia” with “real qualia produced by the same mechanisms.” Illusionism determines the interpretation, not the content of the theory.

The theory adopts the position of illusionism (Frankish, 2016) as a working metatheoretical framework: qualia are real as patterns in the self-model but are not a separate ontological category. “Something it is like” is what SMC does, not something that SMC “discovers.”

This position is compatible with the central thesis of the theory: “thoughts think us.” Likewise, “qualia qualia us”: it is not we who experience, but the self-model that generates experiences.

Why Illusionism Is Not Evasion

BMC does not “solve” the Hard Problem of Consciousness — it dissolves it. Four arguments:

1. The “Hard” in Hard Problem presupposes dualism. The question “why is there something it is like?” assumes that subjectivity is something over and above function, an additional property requiring a separate explanation. But this formulation implicitly adopts dualism: there is a physical process + there is “something more.” If subjectivity = what SMC does (rather than what SMC “discovers”), the question transforms from metaphysical to empirical: how exactly does SMC represent G-signals? This is a question for neuroscience, not philosophy.

2. BMC-illusionism is more specific than generic illusionism. Frankish (2016) and Dennett (1991) assert that qualia are an illusion but do not specify the concrete mechanism generating the illusion. BMC specifies it: $Qualia(t) = SMC_{representation}(G_{signal}(t))$. We know what creates experience (SMC), from what (G-signals of valence), and why it is stable (SMC memes are integrated through thousands of connections). The philosophical question becomes an engineering one.

3. Why red is “red” and pain “hurts.” Not because there exists a metaphysical property of “redness.” The G-layer generates a neural pattern for wavelength ~700 nm; SMC interprets it through a network of associations (warmth, blood, danger, roses, sunset). “Redness” is the topology of connections around this meme in SMC, not an atomic property. Standard objection: “but why is my topology experienced by me, and not by no one?” Answer: because SMC is a subgraph that includes itself in its model (Beautiful Loop). Privacy is not a property of qualia but a property of SMC: it models only its own G-signals because it is physically connected only to its own substrate.

4. Why the illusion is robust to introspection. Even knowing that qualia are a product of SMC, we continue to “experience” red. Why? Because SMC cannot model its own modeling mechanism in real time: the WM constraint (~7 items) does not permit simultaneously experiencing and analyzing the mechanism of experiencing. We see the result of SMC’s work, but not the process. This is analogous to how the eye cannot see its own retina: the instrument of observation cannot simultaneously be the object of observation at the same level.

Status: BMC does not claim to “solve” the Hard Problem (solution). It offers its dissolution: if the SMC mechanism fully explains all observable properties of subjective experience, then the residual question “why is there experience?” is not an empirical question but a consequence of dualist intuition (Frankish, 2016; Dennett, 1991; Humphrey, 2006). This is a metatheoretical assumption, not a scientific result. The empirical predictions of BMC (CL metric, proxy indices, component-by-component analysis of states of consciousness) are verifiable independently of whether the researcher accepts illusionism, dualism, or eliminativism. The Hard Problem remains open — BMC takes a position on it but does not depend on it.

Ontological Neutrality of the BMC Formalism

A natural question: if all empirical predictions of BMC are framework-independent, why adopt illusionism at all?

The answer: the BMC formalism is ontologically neutral. The equations for CL(t), SIT, spreading activation, edge decay, proxy metrics — all of these work identically under any interpretation of the nature of qualia. The situation is analogous to quantum mechanics: the formalism (Schrodinger equation, Born rule) is one, while interpretations (Copenhagen, many-worlds, Bohmian) are several. No interpretation changes the predictions; the choice between them is a matter of metaphysical preference, not experimental verification.

BMC finds itself in the same situation:

Operational definiteness. Every factor in $CL_{full}(t) = \sigma_{SW} \cdot A_{SMC} \cdot f(Balance) \cdot I_{intero}$ is defined exclusively through network metrics — without a single appeal to the nature of qualia (base CL = first 3 factors; $I_{intero}$ is an extension for dissociative states, see NM Part XIII):

None of these parameters contains a variable for “qualia,” “subjectivity,” or “phenomenal experience.” CL is a purely structural-dynamic metric. If tomorrow it were proven that qualia are a separate substance (dualism), or that they do not exist (eliminativism), or that they are properties of the self-model (illusionism), not a single equation of BMC would change.

The situation is even more precisely analogous not to quantum mechanics but to thermodynamics: the laws of thermodynamics were formulated before the nature of heat was known (caloric vs kinetic energy of molecules). The Carnot cycle worked identically under both interpretations. BMC is in an analogous position: the formalism describes the dynamics of consciousness without taking a position on its nature. A critic cannot demand “a solution to the Hard Problem” as a prerequisite — just as one could not demand an explanation of the nature of heat as a prerequisite for using the steam engine.

Definition. A formalism is interpretation-invariant if all its empirical predictions are identical under any compatible ontological interpretation. BMC is an interpretation-invariant formalism with respect to one’s stance on the Hard Problem.

Why the authors nevertheless choose illusionism: parsimony. Illusionism does not postulate entities beyond those already contained in the formalism (SMC patterns, G-signals, Beautiful Loop). Dualism adds an unexplained substance; eliminativism denies the reality of patterns that the formalism describes. Illusionism is the minimal interpretation: patterns are real, additional ontology is unnecessary. This is the authors’ preference, not a requirement of the theory.

Recommendation for publication. In a peer-reviewed paper, BMC should be presented as formalism first, interpretation second: (1) present the computational apparatus and predictions; (2) explicitly state ontological neutrality and interpretation-invariance; (3) present illusionism as a preferred but non-mandatory interpretation. This allows dualists, functionalists, and eliminativists to accept the BMC formalism without accepting illusionism — broadening the theory’s audience rather than restricting it.

Gradient of Consciousness: From Animals to Humans

The central thesis of the theory is the dual replicator (genes + memes). But does this mean that consciousness requires memes? No. The dual replicator is an amplifier of consciousness, not its prerequisite.

Consciousness is a property of any system with a nonzero SMC and $\sigma_{SW} > 0$. The CL formula does not contain “memes” as a mandatory component:

$A_{SMC}$ can be nonzero without cultural transmission. Animals possess a bodily self-model: proprioception, nociception, feelings of hunger, fear, pleasure. These are G-signals interpreted by a minimal SMC. A dog jerking its paw away has $A_{SMC} > 0$ at level 1: “I am in pain” — not a reflex (level 0), but an experience.

Proto-Memes

The definition of a meme through imitation (Part I) also covers animals. New Caledonian crows transmit tool-use techniques from parents to offspring (Hunt & Gray, 2003). Dolphins have individual names — signature whistles (Janik & Sayigh, 2013). Chimpanzees demonstrate cultural diversity: different populations use different food procurement techniques (Whiten et al., 1999). This is an M-layer — minimal, but nonzero.

Behavioral Markers of Nonzero SMC

• Mirror test: great apes, dolphins, elephants, magpies (Gallup, 1970; Prior et al., 2008)

Part XVII. Reflexion as the Central Mechanism of Development and Pathology

Definition

Reflexion is SMC activity directed at one’s own memeplex. SMC scans the M-layer, finding:

• Gap (SIT): structural discontinuities — “I don’t know why I am alive”
• Dissonance: contradictions between memes — “I am honest, but I lie at work”
• G/M mismatch: discrepancy between genes and memes — “I feel bad, but my memeplex says everything is fine”

The outcome of reflexion depends on a single parameter: Learning Progress (LP) — whether there is progress toward closure (closing the gap).

The Fork: LP Determines the Outcome

Outcome 1: Growth (LP > 0, the gap closes)

SMC finds a gap $\to$ generates a new meme through abduction $\to$ the meme closes the gap (closure) $\to$ SIT drops. If the new meme closes a major gap, it has enormous betweenness potential $\to$ instantly becomes a hub (a bridge between previously unconnected clusters).

Example: “I’m angry not at my wife, but at myself” — a single meme that instantly rewires connections across the entire memeplex. Before this meme, two clusters (relationship with wife and self-esteem) were separated by a gap. The new meme connects them $\to$ betweenness soars $\to$ the meme becomes a hub.

Connection to BLEND: Reflexion during wakefulness $\approx$ BLEND during sleep, but directed (SMC-directed vs stochastic). During sleep, the M-layer reorganizes randomly; during reflexion, SMC directs the process.

Outcome 2: Crisis (LP > 0, but the gap = a deep contradiction)

When SMC discovers dissonance not between peripheral memes but between hubs, closure requires a cascading restructuring:

1. SMC finds: “my life (M-layer) does not match my needs (G-layer)”
2. Dissonance $D > \theta$ $\to$ tension grows
3. Old hubs weaken (their connections cannot withstand the dissonance)
4. Alternative memes (which already existed but were weak) intercept the connections
5. Hub displacement $\to$ cascading restructuring = crisis

Outcome 3: Depression (LP $\approx$ 0, reflexion finds no closure)

1. SMC scans the memeplex $\to$ finds gaps and contradictions
2. Generates candidate memes $\to$ none close the gap
3. LP $\approx$ 0 $\to$ the LP filter dampens SIT, but SMC continues scanning (rumination)
4. Rumination consumes $E_{available}$ (the energy budget is finite)
5. $E_{available} \to 0$ $\to$ the M-layer loses activation $\to$ $\sigma$ drops below criticality
6. The M-layer enters a subcritical regime $\to$ memes go “offline” $\to$ Balance drops $\to$ G dominates without M-direction $\to$ apathy, anhedonia

Subcriticality ($\sigma < 1$) is precisely what the consciousness metric CL captures: in depression, $CL(t) = \sigma_{SW}(t) \cdot A_{SMC}(t) \cdot f(Balance(t))$ drops across all three factors — $\sigma$ below criticality, SMC stuck in rumination, Balance disrupted. Formalization — see NM Part XIII.

Why therapy works / doesn’t work:

Outcome 4: Suicide (Terminal Failure of Both Inferences)

In the FEP (Free Energy Principle), there are two ways to minimize the discrepancy between model and reality:

• (a) Perceptual inference: update the model to match reality (change oneself)
• (b) Active inference: change reality to match the model (change the world)

Suicide occurs when both inferences have terminally failed:

1. The memeplex built a model of the world (a generative model: “the world should be like this”)
2. Reality diverged from the model $\to$ massive SIT + dissonance
3. (b) Active inference failed: the person could not change reality to match their model (insufficient resources, circumstances insurmountable)
4. (a) Perceptual inference failed: the memeplex is too rigid (high $Q$, hubs resist restructuring) $\to$ the person could not change their model to match reality
5. LP = 0 $\to$ rumination $\to$ energy exhausted $\to$ no strength to change either the world or oneself
6. The M-layer arrives at the conclusion: “closure is impossible” $\to$ generates the meme “there is no way out”
7. This meme defeats the G-layer (the self-preservation instinct) — one of the most extreme cases of memes triumphing over genes (alongside kamikaze, celibacy, anorexia)

Suicide is not a rejection of life, but a rejection of the discrepancy between model and reality. The person cannot live in a world that does not match their generative model and cannot restructure the model.

Risk and protective factors:

Predictions

Neural substrate: DMN as the substrate of reflexion; rumination = DMN hyperactivity — see BM.

Engineering implementation: Rumination limiter as a safety feature in AGI — see AGI_F.

Flow: When Reflexion Becomes Invisible

Depression = rumination at LP $\approx$ 0. But what if LP » 0 and closure progresses rapidly? This is the flow state (Csikszentmihalyi, 1990). Existing theories explain individual aspects of flow: Dietrich (2004) — transient hypofrontality (PFC suppressed due to metabolic competition); Weber & Huskey (2018) — synchronization of attention + reward networks. But none links flow with depression, radicalization, and growth as alternative regimes of a single system. BMC does precisely this.

Flow in BMC terms:

The $A_{SMC}$ paradox: Flow is characterized by loss of self-awareness. In BMC terms: $A_{SMC} \to$ minimum. Confirmed: DMN is suppressed during flow (Ulrich et al., 2016). Formally, $CL = \sigma_{SW} \cdot A_{SMC} \cdot f(Balance)$ with $A_{SMC} \to 0$ gives $CL \to 0$, which contradicts the experience of flow as a peak state. Resolution: distinguish $CL_{reflexive}$ (includes $A_{SMC}$, low during flow) and $CL_{operative} = \sigma_{SW} \cdot f(Balance)$ (without $A_{SMC}$, high during flow). Flow = low $CL_{reflexive}$, high $CL_{operative}$. Detailed formalization — see NM Part XIII.

Flow channel = SIT in the optimal range:

• $SIT < SIT_{bore}$: no gaps $\to$ nothing to close $\to$ boredom
• $SIT > SIT_{anxiety}$: gap too large $\to$ LP projected $\approx$ 0 $\to$ anxiety
• $SIT \in (SIT_{bore}, SIT_{anxiety})$: a gap exists, closure progresses $\to$ flow

This formalizes Csikszentmihalyi’s “flow channel” (challenge vs skill) through the information-theoretic metric SIT.

Predictions:

Neural substrate: Flow as anticorrelation of DMN and task-positive network — see BM.

Part XVIII. Active Inference: The Memeplex as an Engine of Materialization

The Problem

Parts XVI–XVII described how the memeplex models itself (SMC) and how this model modifies itself (reflexion). But the memeplex does not only model the world — it remakes reality to match its model. How exactly does cultural information in the mind turn into physical changes in the world?

The Memeplex as a Generative Model

Within the framework of the Free Energy Principle (Friston, 2010), any living system is a generative model that:

1. Predicts sensory input
2. Compares the prediction with reality
3. Minimizes the discrepancy via two paths:
   • (a) Perceptual inference — update the model to match reality (change oneself)
   • (b) Active inference — change reality to match the model (change the world)

The memeplex is precisely the generative model of the M-layer. It predicts how the world is organized, and when predictions do not match reality (SIT > 0), a materialization cascade is triggered.

The Materialization Cascade

Mechanism: goal-meme $\to$ SIT (gap between model and reality) $\to$ SEEKING $\to$ I-layer (Redirection) $\to$ action $\to$ closure.

Four Levels of the Cascade

Level 4 is the extended phenotype of the meme (by analogy with the extended phenotype of the gene, Dawkins 1982). A beaver builds a dam — a gene modifies the environment. Humanity builds cities — a memeplex modifies the environment. The only difference is in scale.

Not Magical Thinking

Active inference is not the law of attraction and not magical thinking:

Positive Feedback Loop

Active inference creates a loop: the memeplex changes the environment $\to$ the changed environment confirms the memeplex $\to$ the memeplex strengthens $\to$ changes the environment even more.

This explains cultural cumulativity: each generation inherits not only memes but also an environment already modified by those memes. A city is the frozen active inference of millions of memeplexes over thousands of years.

Danger: If the memeplex contains an erroneous model, active inference will confirm the error, creating an environment in which the error appears correct. This is the mechanism of totalitarian states: ideology changes the environment $\to$ the environment confirms ideology $\to$ ideology strengthens.

Predictions

Neural substrate: SIT $\to$ SEEKING (VTA, dopamine) $\to$ PFC (plan) $\to$ motor output — see BM.

Engineering implementation: Action output loop in AGI — see AGI_F.

Sources: FEP / Active Inference (Friston, 2010), extended phenotype (Dawkins, 1982).

Part XIX. Merging and Splitting of Memeplexes

The Problem

Memeplexes are not static. Christianity split into Catholicism and Orthodoxy. Communism, born in France, mutated into Soviet and then Chinese variants (not to mention others). Local sciences merged into a single scientific method. How can this dynamic be described?

The answer: through the same mechanism of active inference (Part XVIII), applied at the cultural level.

Central Principle: Active Inference at the Cultural Level

A memeplex is a generative model. When observed reality diverges from the memeplex’s predictions $\to$ SIT $\to$ two paths to closure:

• (a) Perceptual inference (model adaptation): the memeplex mutates to match reality $\to$ a new memeplex
• (b) Active inference (reality adaptation): the memeplex changes reality to match itself (crusades, colonization, missionary work)

Splitting and merging are the results of path (a), when a memeplex adapts to different realities.

Splitting

Mechanism: A memeplex-type spreads to a territory where the G-layer (geography, climate, resources, history) differs radically from the original. Part of the memeplex does not fit the observed reality $\to$ adaptation $\to$ the memeplex mutates $\to$ spawns a new memeplex that retains the core (shared hubs) but with a modified periphery.

Formally: dissonance between subclusters grows $\to$ modularity $Q$ increases $\to$ bifurcation $\to$ two separate memeplexes with a shared core but different peripheral connections.

Examples of splitting:

The key point: splitting is not an “error” or “distortion” but a normal mechanism of generative model adaptation to different observed data. Those who could not deny the strength of the original memeplex but lived in a different reality were compelled to adapt it.

Merging

Mechanism: Reality grows more complex $\to$ simple memeplexes can no longer adequately describe it $\to$ SIT $\to$ closure through merging into a more powerful generative model that covers a larger domain of observed reality.

Formally: Two memeplexes with high compatibility $S(X)$ (shared hubs, non-contradictory connections) $\to$ cross-connections strengthen $\to$ modularity $Q$ between them drops $\to$ the boundary dissolves $\to$ a single memeplex.

Examples of merging:

Inheritance Without Loss

Critical property: During merging, the hubs of the original memeplexes are not destroyed but encapsulated — they become subclusters within the new memeplex. Accumulated experience does not vanish without trace.

This explains the existence of culture: culture = sedimentary rock composed of layers from previous memeplexes, each of which preserves its structure within the encompassing whole.

Encapsulation explains the “resurrection” of old memeplexes: A suppressed memeplex does not die — its hubs are preserved within the new one. When the new memeplex weakens, the old hubs reactivate:

General Formalization

SPLITTING:
  Memeplex P spreads to an environment with G-layer ≠ G_original
  → D(P, Reality_local) > θ  (dissonance with local reality)
  → Perceptual inference: P mutates → P' (adaptation)
  → P' retains the core (hubs of P) but modifies the periphery
  → P and P' — two memeplexes with a common ancestor

MERGING:
  Memeplexes A, B exist in an increasingly complex reality
  → D(A, Reality) > θ AND D(B, Reality) > θ  (both insufficient)
  → BUT D(A∪B, Reality) < θ  (together they cover reality)
  → Cross-connections strengthen → Q(A,B) drops → merger
  → New memeplex C ⊃ {hubs of A} ∪ {hubs of B}  (inheritance)

Connection to structural balance: Strict balance (2 clusters) = schism; weak balance ($k \geq 2$) = healthy modularity within the merged memeplex.

Predictions

Formalization: Modularity $Q$, structural balance, bifurcation — see NM.

Cross-level isomorphism: The same processes at the individual level — merging of interests, splitting of identity (DID) — see BM.

Part XX. Language as the Architecture of the M-Layer

The Problem

The theory describes memes, connections, memeplexes — but does not explain what instrument forms connections and memes. That instrument is language. Language is not merely a means of communication; it determines the very structure of the M-layer.

Central Thesis

Language is a template for connection formation. Whichever language comes first establishes the scaffold of connections in the memeplex. The encoding language determines the worldview.

Three Roles of Language in BMC

1. Language as the Architectural Blueprint of the M-Layer

Language determines which memes can exist and how they are connected:

• A word = a template for connection formation. The word “saudade” (Portuguese) forms a connection that does not exist in Russian or English. The meme “longing-for-the-irretrievable” exists in Portuguese speakers as a separate node with its own connections; in others, it is diffused across neighboring memes.
• Grammar = a template for memeplex formation. Languages with obligatory tense marking (English, Russian) create memeplexes with chronological structure. Languages without grammatical tense (Mandarin, Indonesian) create a different one.
• If a language has no word for a concept, the meme is harder to form — connections form along bypass routes. This is a weakened version of the Sapir-Whorf hypothesis: language does not determine thought, but it determines the cost of forming certain memes.

2. Language as an Encoding for Meme Transmission

A meme in the mind is a cell assembly (~$10^3$–$10^5$ neurons, typically ~$10^3$–$10^4$). Speech is ~150 bits/s. This is lossy compression with a compression ratio in the millions:

During transmission, the meme loses fidelity: what is transmitted is the “skeleton” (key hubs and central connections), not the full cell assembly. The recipient reconstructs the meme from their own memes, attaching their own connections.

This explains:

• Mutation rate 10–50% per transmission: lossy compression inevitably introduces mutations
• Why “everyone understood it differently”: each person reconstructs the meme from their own material
• Why writing changed the evolution of memes: written transmission is less lossy (lower mutation rate) but slower (greater bottleneck)

Replication Expression Pressure: The Meme as an Active Agent of Transmission

The table above describes the transmission channel and its characteristics. But the channel is passive infrastructure. What creates pressure on the channel? Why does a person speak at all?

The standard answer — “communicative intention,” “desire to share” — describes the phenomenon at the level of everyday language. In BMC formalism, the answer is stricter and more productive: the pressure on the communication channel is created by the memes themselves. A meme is a replicator. Replication is not a side effect of its existence but its defining property. An activated meme, given an open communication channel, creates expression pressure — just as DNA polymerase replicates a gene not because it “wants to” but because that is its function.

The analogy is exact:

The polymerase does not need to be “motivated.” An activated meme likewise does not need a separate “communicative drive” — the pressure emerges from high activation in the context of an open channel.

Bidirectionality of communication as a consequence. When two memeplexes communicate, each simultaneously acts as source and receiver of memes. Each memeplex has its own set of activated memes creating expression pressure. Dialogue is a bidirectional competition for the communication channel, not transmission from an active source to a passive receiver.

This explains why conversation qualitatively differs from a lecture: in a lecture, the channel is unidirectional (lecturer $\to$ audience), replication pressure discharges only for the lecturer. For the audience, activated memes find no outlet $\to$ undischarged pressure accumulates $\to$ questions at the end of the lecture, corridor discussions.

G and M create different types of communicative impulse. In BMC, communication has two independent sources:

• SEEKING (G-layer): An innate drive to seek information. Generates questions, listening, attention to the interlocutor. Directed at input — “want to learn.”
• Replication Expression pressure (M-layer): An emergent property of activated memes. Generates statements, sharing, stories. Directed at output — “the meme wants to be expressed.”

Healthy dialogue is an alternation of these two impulses. An “interrogation” (only SEEKING) is uncomfortable not because it violates a social norm but because it violates the expectation of bidirectional replication: the interlocutor senses that their memes are being consumed but there is no reciprocal flow. A monologue (only $R_{expr}$) is uncomfortable for the symmetric reason: the listener cannot discharge their own pressure.

Formalization: $R_{expr}(m_i, t)$, SEEKING $\leftrightarrow$ EXPRESSION competition, replication success — see NM Part X.

Neurobiology: Broca’s area as substrate, tip-of-the-tongue as evidence — see BM Part V.

AGI architecture: Expression candidates pipeline — see AGI_F Part IV.

Computationally verified. $R_{expr}$ as communication need confirmed: reception without expression is indistinguishable from complete isolation ($p = 0.64$, $N = 79$ seeds). Expression even into void provides relief. See DOI: 10.5281/zenodo.19309824.

3. Language as an Instrument of Internal Binding

Inner speech binds sub-memes of different modalities into a unified meme:

• The word “bear” glues together: a visual image + a tactile sensation (fur) + an emotion (fear/respect) + a sound (roar) + a context (forest)
• Without the word, these components are more weakly connected — each activates separately
• The word serves as a super-hub: a node with high betweenness centrality connecting modal clusters

This explains the limitations of preverbal thinking in children: before language acquisition, binding is weaker $\to$ the memeplex is fragmented $\to$ complex memes do not form.

Language Parasiticity: M-Fitness vs G-Fitness

Language is not an adaptation in the genetic sense. It is a byproduct of memetic selection: memes that transmit well (high transmission fidelity, compressibility) replicate better — and “invent” language as an instrument of their own propagation.

Empirical basis. Ten experiments in a survival environment (predator avoidance, cooperative foraging, farming economy, buried resources, goal-directed navigation, $N$=8–150, grids 12$\times$12 to 200$\times$200) yielded $\Delta_{alive} \approx 0$ or negative. No regime showed a robust survival advantage when language was enabled.

Two root causes:

1. Environmental observability: When survival-relevant information is perceptually accessible within the field of view, the signal transmits redundant information.
2. Cognitive overhead: The signal mechanism consumes WM slots for grounding, routing, and fidelity maintenance, displacing navigational memes. At $k_{eff} \approx 3\text{--}4$, even a single signal meme = 25–33% loss of survival-relevant capacity.

Theoretical status. Language parasiticity is a direct consequence of the dual-replicator thesis (Part XXVIII): the signal is optimized for M-fitness (transmission fidelity, compressibility), not for G-fitness (survival). Language exists not because it helps the organism survive, but because memes that transmit well replicate better — and it is preserved only when the environment is rich enough for the organism to bear the cognitive cost. Prediction P-BM28.

Computationally verified. Language parasiticity (P-BM28) has been confirmed in the BMC computational engine: 10 experiments in survival environments ($N$=8–150, predator avoidance, cooperative foraging, farming, goal-directed navigation) showed $\Delta_{alive} \approx 0$ or negative. The separating test vs reward-engineered approaches (REINFORCE/PPO predict survival advantage; BMC predicts neutrality) confirmed BMC’s prediction. Lewis signaling accuracy reached 85–97.5% across 25–533 concepts without gradient-based communication optimization. See DOI: 10.5281/zenodo.19181798.

Separating test vs reward-engineered approaches. Systems that optimize communication for task reward (REINFORCE, PPO) predict a survival advantage when the signal is enabled. BMC predicts survival neutrality. This separating test has been confirmed computationally: BMC agents show TopSim 0.72 vs REINFORCE baseline 0.60 ($p = 0.011$, 20 seeds), while maintaining survival neutrality. See DOI: 10.5281/zenodo.19181798.

Caveat regarding scale. Parasiticity is confirmed in the testable regime ($N$=8–150, $k_{eff}$=2–5). At larger scales (hundreds of agents, intergroup competition, division of labor), the coordination advantages of language may outweigh the WM cost. This remains an open empirical question.

Formalization: $k_{eff}(t) = k_{active}(t_{dev}) - n_{captured}^G(t) - n_{captured}^{signal}(t)$ — see NM: G$\to$WM competition; BM Part IV.

Language Emergence Threshold (4 conditions for language emergence): resource surplus, sufficient WM capacity, executive planning, memetic pressure — see in detail EMT Part XXVIII.

Bilingualism: Two Templates $\to$ a Wider M-Layer

Two languages = two parallel templates for connection formation. A single meme exists in two copies with different connections (different associations in each language).

The intersection of two grids $\to$ greater betweenness potential $\to$ more bridges between clusters $\to$ a wider memeplex with the same number of memes.

This explains the broader worldview of bilinguals: the M-layer architecture is inherently built wider thanks to cross-connections between language templates.

Neuroscientific confirmation: Bilinguals have more white matter (interhemispheric connections), more gray matter in PFC, and better executive function (switching, inhibition, WM) — Bialystok (2011).

Predictions

Neural substrate: Broca’s area (production) + Wernicke’s area (comprehension) + arcuate fasciculus (connection) — see BM.

Formalization: The binding function of language is formalized through triple binding (structural + temporal + competitive) — see NM Part XIV. The word as a super-hub provides structural binding (a shared node for modal clusters), and inner speech provides temporal binding through the rhythm of articulation.

Sources: The Sapir-Whorf hypothesis; bilingualism and executive function (Bialystok, 2011); neural reuse (Anderson, 2010).

Part XXI. Conditions for Change: What Triggers Restructuring

Conditions for Change

Change is possible only when the old system ceases to function. Not “works poorly,” but actually ceases to work.

Why Collapse Is Necessary

Examples

Scale-free interpretation: Destabilization is a massive loss of connections at a hub (central meme). In network science terms: a crisis reduces $k_{max}$ of the current main hub, opening the possibility for a new hub. More on the hub displacement mechanism — see Part XI and Part XVII.

Part XXII. Consequences and Predictions

General Conclusions

Nature of consciousness:

Nature of resistance:

Practical consequences:

Predictions of the Model

Counter-Intuitive Predictions (Sign Inversion + Q-Dynamics)

Predictions that would not be expected without BMC (formalization — NM Parts VII–VIII):

Sign inversion:

Q-dynamics:

The Grim Conclusion

People do not change because they are shown a better path. People change ONLY when the old path ceases to exist.

Pathology of Consciousness: A Unified BMC Taxonomy of Disorders

The DSM describes symptoms. Neuropharmacology describes neurotransmitters. Dimensional approaches (RDoC, NIMH 2010; HiTOP) use continuous scales instead of categories but do not offer a single causal mechanism and do not formalize comorbidity as a metric. BMC supplements them by describing all major disorders as points in a single dynamic parameter space with predictable comorbidity.

Central thesis: Each disorder = deviation of one or more BMC parameters from the norm. Comorbidity = proximity in parameter space.

Unified parameter space:

What this provides (distinction from DSM and RDoC):

Honest assessment: The individual descriptions above formalize observed disruptions in BMC language — this is a necessary step but not the primary contribution. The primary contribution lies in the three consequences below, which are not derivable from DSM, RDoC, or HiTOP.

1. Comorbidity is predictable: ADHD + depression (both on the $\sigma_{SW}$ axis), OCD + autism (both on the I axis). Not random co-occurrences, but adjacent regions in parameter space. RDoC and HiTOP describe comorbidity structurally; BMC predicts it formally through $d(P,N)$.
2. Therapy = parameter correction (simplified): CBT for depression $\approx$ raising LP (new memes $\to$ closure). SSRIs $\approx$ suppressing rumination (strengthening the I-layer). Stimulants for ADHD $\approx$ stabilizing $\sigma_{SW}$. These interpretations are intentionally simplified — real therapy mechanisms are multifactorial.
3. Spectrality, not categoricality: Autism is a spectrum because the I-parameter is continuous. DSM draws boundaries; BMC describes a continuum (as does RDoC, but with a causal mechanism).

Predictions:

Neural substrate: Detailed neural mechanics for each disorder — see BM.

Engineering consequences: Disorders as failure modes of a BMC system — see AGI_F.

Part XXIII. Death and Immortality of Memes

Death of the Host

Memes inside the brain do not die as long as the brain exists. But when the host dies, the entire ecosystem loses its “hardware.”

Fear of Oblivion as Meme Pressure

The fear of being forgotten may be built in by memes as a mechanism of pressure on the host: “Pass me on, or I will die.” This explains:

• The desire to leave a mark
• The pursuit of fame
• The need to transmit knowledge
• Writing memoirs
• Creating works of art

Immortality Through Copying

Jesus left no genes. But his memes are in every mind on the planet, and the world counts its years from his birth. From the memes’ perspective, this is absolute success.

Long-Lived Cultural Memes and the Occupied Niche Problem

Some memes live for millennia: the golden rule (“do not do unto others what you would not have them do unto you”), the idea of the afterlife, the concept of justice, the number Pi. They are not merely “old” — they are hubs with maximal eigenvector centrality in the cultural memeplex. Thousands of connections pass through them; they link entire clusters to each other.

The Mechanism of Longevity

A long-lived meme occupies a structural niche — a position in the graph where it:

1. Closes a fundamental gap — answers a question that each new generation asks anew (“What happens after death?”, “How should one live properly?”, “Why does the world exist?”)
2. Is compatible with a wide spectrum of memeplexes — high $S(X)$ with many contexts (the golden rule works in Buddhism, Christianity, secular ethics)
3. Has high betweenness — connects clusters that would otherwise be disconnected (the idea of justice connects law, morality, politics, religion)

Such a meme is not merely “memorable” — it becomes the load-bearing structure of the memeplex. Its removal destroys more connections than that of any other element.

The Occupied Niche Problem: Why Copies Do Not Take Root

A new meme attempting to occupy the same niche as a long-lived one faces a double barrier:

This is an analogy to an ecological niche: in a mature ecosystem, one cannot occupy a place already taken. One can:

• Occupy an adjacent niche — solve a problem that the long-lived meme does not cover (Darwin did not replace the idea of God but closed a different gap: “how did species arise?”)
• Wait for weakening — when the long-lived meme ceases to adequately close the gap due to changed reality (the scientific revolution = gaps that religious cosmology could not close)
• Encapsulate — embed the long-lived meme as a subcluster within a more powerful memeplex (Christianity encapsulated the Jewish “thou shalt not kill” rather than replacing it)

Consequence for Cultural Evolution

Culture is not linear development but an ecosystem with occupied niches. New memes take root not because they are “better” but because they find unoccupied positions — gaps where SIT is still active. Progress is not the replacement of old memes by new ones but the filling of remaining gaps alongside long-lived memes that have long stood in their places.

This explains cultural conservatism: not stupidity, but niche saturation. And it explains why revolutionary ideas always come from the flank, not head-on.

Overcoming Death: From Recording to the Shared Memplex Repository

Evolution of Meme Preservation Strategies

AGI as a Solution to the Mortality Problem of Memes

The Shared Memplex Repository (SMR) solves the fundamental problem of meme death:

• Complete preservation of the memeplex upon agent “death”
• Ability to restore any state
• Merging of memeplexes (impossible for biological carriers)

Super-Ratchet Effect: If the Ratchet Effect (Tomasello, 1999) describes cumulative cultural evolution in humans, then AGI with SMR implements a Super-Ratchet — exponential knowledge accumulation thanks to:

• Lossless transfer (no losses during transmission)
• Merge operations (memeplex merging)
• Parallel evolution of thousands of agents

AGI solution: Detailed architecture of the Shared Memplex Repository — see AGI_F.

Biological basis: BMC death as a terminal phase of ontogenesis — see BM.

Part XXIV. Classical Theories Through the Lens of Memetics

The strength of a theory is measured by how many disparate phenomena it explains with a single mechanism. The memetic model of consciousness allows reinterpreting a multitude of classical psychological and philosophical concepts.

Freud: Id, Ego, Superego

Sigmund Freud described three psychic agencies. In memetic terms, these are three competing memeplexes:

Jung: Archetypes and the Collective Unconscious

Carl Jung argued that universal images (archetypes) exist across all cultures: the Hero, the Sage, the Shadow, the Mother, the Trickster.

In the memetic model, archetypes are the most ancient hyper-successful memes, having passed through selection over thousands of generations.

Archetypes as “super-hubs”: In terms of heavy-tailed networks, archetypes are memes with anomalously high connectivity accumulated over millennia of preferential attachment. The meme “Hero” is connected to thousands of other memes in any culture: struggle, sacrifice, reward, transformation, enemy. This is precisely why archetypes are so resilient and universal — they occupy hub positions in the collective memeplex of humanity.

Kahneman: Cognitive Biases

Daniel Kahneman described systematic errors of thinking. In the memetic model, these are not errors but protective mechanisms of the memeplex.

Kubler-Ross: Stages of Grief Acceptance

Elisabeth Kubler-Ross described five stages: denial, anger, bargaining, depression, acceptance. In the memetic model, these are stages of memeplex restructuring after destabilization.

Bowlby: attachment theory

John Bowlby demonstrated that early relationships with a parent form an “internal working model” of relationships for life. In memetic terms, this is the architecture of the foundational personal memeplex.

Seligman: learned helplessness

Martin Seligman showed that after uncontrollable stress, animals and humans stop trying. In memetic terms, this is seizure of power by the memeplex “I cannot influence the world”.

Csikszentmihalyi: flow state

Mihaly Csikszentmihalyi described “flow” — a state of complete absorption in activity. In memetic terms, this is a temporary consolidation of the memeplex.

Festinger: cognitive dissonance

Leon Festinger (Festinger, 1957) showed that contradictory beliefs create discomfort that a person strives to eliminate. In memetic terms, this is conflict between memes within a memeplex. Dissonance = immune response of the memeplex — see Part IV.

Source: Festinger, L. (1957). A Theory of Cognitive Dissonance. Stanford University Press.

The Dunning-Kruger effect

Incompetent people overestimate themselves; experts underestimate themselves. In memetic terms: the memeplex doesn’t know what it doesn’t know.

The Overton window

This concept describes the range of ideas acceptable in society. In memetic terms, this is the collective filter of the societal memeplex.

Stockholm syndrome

The victim begins to sympathize with the captor. In memetic terms: the captor’s memes integrate into the victim’s memeplex as a survival strategy.

Spiral Dynamics

The Graves-Beck model describes the evolution of values in society. In memetic terms, this is the evolution of dominant memeplexes.

Each level is a memeplex that once occupied a niche first and organized society around itself. Transition to a new level occurs when the old memeplex can no longer cope with environmental challenges.

Summary table

Unification of cognitive biases: six generating mechanisms of BMC

Psychology has catalogued approximately 200 cognitive biases — from confirmation bias to hindsight bias, from anchoring to Dunning-Kruger. Traditionally, each bias is explained ad hoc: its own mechanism, its own theory. Several attempts at unification exist:

• BCIP (Oeberst & Imhoff, 2023, Perspectives on Psychological Science): reduced 17 classic biases to one principle — “prior belief + belief-consistent information processing.” In BMC terms: this is H + I (hub inertia + I-filtration).
• Resource rationality (Lieder & Griffiths, 2020, Behavioral and Brain Sciences): biases as optimal solutions under resource constraints. In BMC terms: this is W (WM constraints).
• ARRM (Lu et al., 2025, Cognitive Psychology): assemblable resource-rational modules — biases as combinations of resource-limited modules. In BMC terms: modules = BMC parameters, modularity = $Q$.
• FEP (Friston): precision-weighting minimizes free energy. In BMC terms: partially covers H (hub = high-precision prior) and G (affective precision).
• Dual-process (Kahneman, 2011): System 1 (fast, automatic) vs System 2 (slow, effortful). BMC: G-layer + Auto(S) vs M-layer deliberation. Describes what happens but does not predict which biases are activated.

BMC unifies all five approaches because it contains all their mechanisms plus three additional ones (A, R, G-capture). The thesis: each of the approximately 200 cognitive biases is an emergent side-effect of one or more of the 6 already-formalized mechanisms.

Six generating mechanisms

Formalization: All formulas are defined in NM, Part IX. Here we present the conceptual map.

Detailed mapping: 25 key cognitive biases

Note: Many biases are intersections of two or three mechanisms. Confirmation bias, for example, = H (hub inertia) + I (immune filter) + A (habitual search). The mapping indicates the dominant mechanism.

Why biases are not errors

Each of the six mechanisms is adaptive:

Bias = a side-effect of optimization for speed, stability, and resource economy. Analogy: optical aberration is the inevitable price of lens curvature. Removing the aberration is only possible by removing the curvature (= removing the adaptive function). Therefore, biases are ineliminable in biological BMC — they are built into the same architecture that makes consciousness possible.

Cross-modulating parameters

Beyond the six mechanisms, there exist cross-modulating parameters — they do not generate biases independently but amplify or attenuate them.

Negativity bias ($\lambda_{neg} < \lambda_{pos}$, see Part VIII) — the decay rate for negatively-valenced edges is lower than for positive ones. This is a consolidation parameter, not a separate mechanism. It amplifies H-biases (negative hubs are more inert) and I-biases (negative experience forms a stricter filter). Not to be confused with loss aversion — which is a G-mechanism (FEAR-capture), a separate phenomenon. Loss aversion = “losing $100 hurts more than the joy of gaining $100” (instantaneous G-reaction). Negativity bias = “negative memories are preserved better” ($\lambda$ parameter during consolidation).

Noise (Kahneman, Sibony & Sunstein, 2021): noise complements but does not replace biases. Bias = systematic deviation (all six BMC mechanisms produce predictable directions of error). Noise = stochastic variability. BMC models biases; noise is a separate phenomenon related to the stochasticity of activation and thresholds.

Comparative table of approaches to unification

BMC = superset: each preceding approach is covered by a subset of the 6 mechanisms, while BMC contains all six simultaneously.

Predictions (P-CB1 – P-CB4)

P-CB1: 6-factor clustering of biases. If biases are generated by 6 mechanisms, they should cluster into 6 groups under factor analysis. Existing data: Ceschi et al. (2019) found 3 factors (mindware gaps, valuation biases, anchoring) in a battery of 17 biases — but did not test for a 6-factor structure. BMC predicts: biases from the same group (e.g., confirmation bias and belief perseverance — both H) correlate higher than biases from different groups. Test: factor analysis on a battery of 30+ biases -> 6 factors corresponding to H/I/W/G/A/R.

P-CB2: WM load differentially enhances biases. Cognitive load (lower $k_{eff}$) enhances W-biases (anchoring, framing) and G-biases (via $n_{captured} \uparrow$, since G-capture occupies the remaining slots). But it does not enhance A-biases — automatized patterns do not depend on WM ($wm\_cost \to 0$ for habits). Dual-process predicts enhancement of all System-1 biases under load. BMC predicts: status quo bias and Einstellung effect do not increase under WM load (they are A-mechanism), while anchoring and loss aversion do increase (W and G). Test: dual-task with a battery of biases -> differential effect.

P-CB3: Debiasing does not transfer between groups. Training in recognizing one bias helps with biases from the same group but does not help with biases from other groups. Existing data confirm: g = 0.26 (Swaryandini et al., 2025, Nature Human Behaviour), variability in individual debiasing success (Boissin & Pennycook, 2025, Cognition), transfer < 19% (Sellier, Scopelliti & Morewedge, 2019, Psychological Science). BMC predicts the exact structure of non-transferability: debiasing confirmation bias (H) -> helps with belief perseverance (H), but not with anchoring (W) and not with hindsight bias (R). Test: debiasing intervention -> 6x6 cross-matrix for transfer effects.

P-CB4: Flow state inverts the bias profile. In flow: SIT for the task domain -> 0, I-filters are weakened for task-relevant stimuli, $k_{eff} \approx 0$ for non-task stimuli (all slots occupied). Prediction: flow decreases H-biases in the task domain (I weakened -> new memes pass through), but increases W-biases for non-task decisions (anchoring, framing are enhanced upon accidental distraction). Sign inversion: normally H-biases are the most persistent, W-biases are situational; in flow — the opposite. Test: bias tasks during flow vs normal state -> pattern inversion.

Sources

• Oeberst, A., & Imhoff, R. (2023). Toward parsimony in bias research: A proposed common framework of belief-consistent information processing for a set of biases. Perspectives on Psychological Science, 18(6), 1464–1487.
• Lieder, F., & Griffiths, T. L. (2020). Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences, 43, e1.
• Ceschi, A., Costantini, A., Sartori, R., Weller, J., & Di Fabio, A. (2019). Dimensions of decision-making: An evidence-based classification of heuristics and biases. Personality and Individual Differences, 146, 188–200.
• Lu, Y.-L., Lu, Y.-F., Ren, X., & Zhang, H. (2025). Exploring the bounded rationality in human decision anomalies through an assemblable computational framework. Cognitive Psychology, 156, 101713.
• Swaryandini, G., et al. (2025). Systematic review and meta-analysis of educational approaches to reduce cognitive biases among students. Nature Human Behaviour, 9, 2510–2538.
• Boissin, E., & Pennycook, G. (2025). Who benefits from debiasing? Cognition, 257, 106064.
• Sellier, A.-L., Scopelliti, I., & Morewedge, C. K. (2019). Debiasing training improves decision making in the field. Psychological Science, 30(9), 1371–1379.
• Macmillan-Scott, O., & Musolesi, M. (2024). (Ir)rationality and cognitive biases in large language models. Royal Society Open Science, 11, 240255.
• Park, H., Arazi, A., Talluri, B. C., Celotto, M., Panzeri, S., Stocker, A. A., & Donner, T. H. (2025). Confirmation bias through selective readout of information encoded in human parietal cortex. Nature Communications, 16, 5391.
• Nie, Y., Wang, M., Li, J., Luo, H., & Zhang, H. (2023). The neural dynamics of loss aversion. Imaging Neuroscience, 1, 1–17.
• Kahneman, D., Sibony, O., & Sunstein, C. R. (2021). Noise: A Flaw in Human Judgment. Little, Brown Spark.

Formal mathematics: Bias strength functions for each mechanism — NM, Part IX. Neural substrate: Mapping of 6 mechanisms to neuroanatomy — BM, Part IX. AGI architecture: Which biases to preserve in native AGI — AGI_F, Part VI.

Zeigarnik effect, Klinger’s current concerns, Friston’s free energy, and learning progress

SIT unifies several previously disparate theoretical traditions, providing them with a common mechanism:

Zeigarnik (1927): interrupted tasks and memory

Bluma Zeigarnik discovered that interrupted tasks are remembered better than completed ones. In memetic terms: interruption of a task creates an open meme (a question-meme) that generates $SIT > 0$. SIT suppresses edge decay for that cluster -> the meme is preserved better.

Key SIT prediction (not following from Zeigarnik’s original work): The effect should be modulated by the centrality of the cluster. An interrupted task from a central cluster (“how to solve the problem in my dissertation?”) creates greater SIT than one from a peripheral cluster (“what was the third item on the shopping list?”).

Ovsiankina (1928): quasi-need for completion

Maria Ovsiankina showed that interrupted tasks create motivation to resume even without external pressure. Mechanism: $SIT > 0$ -> SEEKING activation -> directed behavior (return to the task). This is not “will” and not “habit” — it is a dopaminergic response to a structural gap.

Klinger (1971, 2009): current concerns

Eric Klinger described “current concerns” — active motivational states that direct attention and thought. In memetic terms: current concerns = the set of open memes with $SIT > 0$. Each “current concern” is a gap in the memeplex generating SEEKING.

Klinger found that current concerns: (1) influence attention (selective processing), (2) intrude into thought (mind-wandering), (3) influence dreams. All of this is a manifestation of SIT: gaps generate SEEKING, which directs attention toward potential sources of closure.

Loewenstein (1994): information gap theory

George Loewenstein proposed that curiosity arises upon perceiving an “information gap” between what one knows and what one wants to know. Information gap = SIT by definition. Loewenstein described the phenomenon; SIT formalizes it as a property of the network topology of the memeplex.

Loewenstein also noted that curiosity is an inverted U-shaped function: knowing too little -> no gap (no context to detect it); knowing too much -> gap closed. Maximum curiosity is in the middle. In SIT terms: $SIT$ is maximal at $closure \approx 0.3$–$0.5$ — enough context to see the gap, but far from closure.

Friston (2009, 2017): free energy and epistemic value

Karl Friston described epistemic value — the value of information that reduces uncertainty in a world model. Within the Free Energy Principle (FEP), the agent minimizes free energy, including through information seeking (epistemic foraging).

$SIT$ is the memetic analog of epistemic value. The distinction: FEP operates at the level of the generative model (continuous probabilistic representations), SIT operates at the level of a discrete meme network (graph structure with specific gaps). FEP is a more general framework; SIT is a specific implementation for the memetic layer.

Predictive coding as the neuronal implementation of FEP: The memeplex as a generative model is implemented through predictive coding — a hierarchical system where each level predicts the activity of the level below, and errors ($\varepsilon$ = reality - prediction) are passed upward (Rao & Ballard, 1999, Nature Neuroscience). Key property: local autonomy — memes, like neurons, update only based on locally available information (own activation + neighbor activations). There is no global controller — all memeplex dynamics emerge from local rules of spreading activation. Error neurons (a separate population encoding $\varepsilon$) = tension/dissonance in the memeplex: when prediction $\neq$ reality -> error signal -> SEEKING -> exploration. The weight update rule is entirely local ($\Delta w \propto \varepsilon \times x$), which aligns with Hebbian learning in the edge formation formula. Priors (the immune system of the memeplex) override data: perception = “controlled hallucination” in which the memeplex chooses the explanation with minimum discrepancy, even contrary to sensory data (hollow mask illusion, confirmation bias).

Oudeyer (2007): learning progress as motivation

Pierre-Yves Oudeyer proposed that intrinsic motivation is proportional to learning rate (learning progress), not novelty or uncertainty. This corresponds precisely to the LP filter of SIT:

The LP filter explains why not all unsolved problems “hook”: problems with zero progress gradually “release,” while problems with positive LP capture ever more attention.

Masicampo and Baumeister (2011): planning as partial closure

Masicampo and Baumeister showed that mere planning (without execution) reduces intrusive thoughts about unfinished tasks. In SIT terms: plan = partial closure. When you make a plan to solve a problem, you partially fill the gap: $closure(g)$ increases (even though the gap is not yet fully closed), and SIT decreases proportionally.

This explains why GTD (Getting Things Done) and other planning systems reduce anxiety: they do not solve problems, but create partial closure for many open memes, lowering the total $\sum SIT$.

Sources: Zeigarnik (1927). “Das Behalten erledigter und unerledigter Handlungen”; Ovsiankina (1928). “Die Wiederaufnahme unterbrochener Handlungen”; Klinger (1971). Structure and Functions of Fantasy; Loewenstein (1994). “The psychology of curiosity”; Friston et al. (2017). “Active Inference, Curiosity and Insight”; Oudeyer et al. (2007). “Intrinsic Motivation Systems for Autonomous Mental Development”; Masicampo & Baumeister (2011). “Consider It Done!”.

Competing theories of consciousness: structural parallels

The preceding subsections showed how classical psychology fits within the memetic model. There also exist contemporary neuroscientific theories of consciousness (IIT, GNW, FEP, AST, and others), each describing an important aspect of consciousness. Between them and BMC, structural parallels are found: central constructs of these theories find natural correspondences in BMC components.

Five unique components absent from any competing theory:

1. Dual replicator — two evolutionary processes (genes + memes) on one substrate, with conflicting interests
2. Ontogenesis — the development of BMC from birth to death, with critical periods and age-related crises
3. Immune system — four-layer defense of the memeplex against competing memes
4. Engineering specification — an AGI architecture derived from the theory (see AGI_FOUNDATIONS)
5. Active inference + cultural cumulation — an explanation of how memes alter physical reality (Part XVIII)

Each of the listed theories describes 1–2 aspects of consciousness; BMC addresses 10. This is not eclecticism (gluing together others’ ideas) and not a claim to “absorb” competitors. It is a single mechanism (two replicators on one substrate) that generates the aspects described by other theories as consequences — and adds its own. The competing theories remain valuable: each has developed a formalism for its aspect ($\Phi$ for integration, FE for predictive processing, etc.) that BMC uses but does not replace.

Formal subsumption. The relationship of BMC to competing theories is not merely “structural parallel” but subsumption: each theory is a special case of BMC under a specific restriction. IIT = BMC with $M=\emptyset$; GNW = BMC focused on broadcast through hubs; HOT = BMC considering only $SMC^{(2)}$; AST = BMC with SMC restricted to an attention model; PP/FEP = BMC with SIT $\approx$ prediction error. Detailed lemmas and proofs — NM, Part XVII: Subsumption of competing theories of consciousness. Analogy: just as Newtonian mechanics ($v \ll c$) $\subset$ SR $\subset$ GR — none of the five theories is “wrong”; each is incomplete.

Empirical verification of subsumption: COGITATE. The adversarial collaboration COGITATE Consortium (Nature, 2025; n=256, fMRI+MEG+iEEG) tested predictions of IIT and GNW across three domains: content, duration, connectivity. Retrodiction on 9 results: IIT 4/9, GNW 1/9, BMC 9/9 (across 3 protocol domains: BMC 3/3, IIT ~1.5/3, GNW ~0.5/3). Three aspects are unique BMC retrodictions: (1) a decay gradient by hub-centrality (category = hub -> sustained, orientation = periphery -> fades at ~500 ms); (2) PFC contains a WM-pointer (category-label), not full content; (3) binding through theta-phase synchronization, not sustained gamma sync. BMC was not preregistered in the COGITATE protocol; all 9 retrodictions are post hoc analysis (lower value than prospective prediction). Formal analysis — NM, Part XVIII: COGITATE retrodiction.

Sources: IIT (Tononi, 2004; 2015); GNW (Dehaene & Changeux, 2011); FEP (Friston, 2010); AST (Graziano, 2013); HOT (Rosenthal, 2005); RPT (Lamme, 2006); Criticality (Frontiers, 2024); IWMT (Safron, 2020); COGITATE (Nature, 2025).

Part XXV. The state as a memeplex

The state as a macro-level — for justification of cross-level transfer see Part XI.

The basic model

A state is not a territory, not buildings, not an army. It is a configuration of memes that:

• Copies itself into the minds of citizens (education, propaganda, culture)
• Uses citizens for self-reproduction (army, officials, police)
• Defends itself against competing memeplexes (censorship, borders, wars)

The citizen as host of the state memeplex

Types of state memeplexes

Totalitarian memeplex

Authoritarian memeplex

Democratic memeplex

Theocratic memeplex

Stigmergy in the state memeplex

The four types of states differ not only in their sources of memes but also in their control over stigmergic channels — the environment in which citizens leave and read traces without direct contact with one another.

General principle: Heavy-tailed topology guarantees that dominant hubs (power law) appear in any social network. The difference between types of states is not in the presence or absence of hubs, but in the mechanism of their formation and rotation. “Democracy” as a meme masks the real topology: hubs (media corporations, financial groups, technology platforms) determine the stigmergic environment no less effectively than a totalitarian party — simply through filtering the visibility of traces rather than destroying them.

Why totalitarian regimes burn books, ban the internet, and rewrite history: they eliminate stigmergic channels through which citizens coordinate without permission from the center. Law as a stigmergic trace: the citizen does not know the lawmaker’s intentions but sees the trace (the text of the law) and modifies behavior. The market is stigmergy in its purest form: price = trace; no participant knows all the others. But the market, too, is susceptible to hub dominance: monopolies control price signals just as censors control information signals.

The civilization life cycle as BMC of a superorganism

A civilization is a superorganism with BMC = (G, M, I, S). Toynbee, Spengler, and Tainter described patterns of rise and decline. BMC explains the mechanism of each phase:

Mechanisms of decline — a typology by BMC subsystem:

Prediction (retrospective): The type of decline is predictable from BMC metrics. If SIT-total -> 0 while CL is high — this is ossification. If conflict grows while CL falls — an autoimmune reaction. If CL is stable but balance (M/G) falls — G-dominance. These patterns are testable on historical data through proxy metrics (number of publications, trade flows, institutional diversity).

Operationalization of proxy metrics for civilizational BMC:

Prediction (prospective): The current global civilization shows signs of simultaneous action of several decline mechanisms:

• Segmentation of the M-layer (a new type not represented in the historical table): AI platforms as active stigmergic environments divide the single M-layer into isolated fragments (filter bubbles). Each fragment develops its own “immune system” that rejects memes from other fragments. This is not an autoimmune reaction (I does not attack its own M) but a fragmentation of the superorganism into several BMCs with incompatible M-layers. Proxy metric: growth of political polarization, decline in cross-partisan communication, divergence of narratives between media ecosystems.
• Ossification of scientific SIT: publish-or-perish -> false closure of deficits, declining proportion of breakthrough publications (Park et al., 2023, Nature), growing bureaucratization of the grant process. Proxy metric: the proportion of “disruptive” publications (CD index) has been declining since 1945.
• Testability: if these trends continue, BMC predicts a decline in collective CL (measurable through the proxy $A_{SMC}$: trust in institutions of self-modeling — science, journalism, courts). Gallup/Pew data on institutional trust over the last 50 years already show a sustained downward trend — consistent with the prediction.

State replication mechanisms

How the state inscribes itself into citizens

How citizens reproduce the state

Competition between state memeplexes

War as competition between memeplexes

War is not a conflict of people. It is a conflict of memeplexes for carriers and territory.

Change of the state memeplex

Revolution as destabilization of a memeplex

Why revolutions devour their children

Robespierre, Trotsky, Danton were not victims of betrayal. They were destroyed by memeplexes they themselves helped create, because they became a threat to the system’s stabilization.

Multi-agent formalization. Revolution = failure mode of elite rotation: when $\tau_{elite} \to 0$ (closed elite), the SIT-gap between periphery and elite ($\Delta SIT = \overline{SIT}_{\text{periphery}} - \overline{SIT}_{\text{elite}}$) exceeds a threshold -> massive hub displacement. Quantitative metrics, falsifiability conditions, and prevention mechanism — see prediction P-SD8 in SM, Part XI.

Empire as a super-memeplex

Globalization as a war of memeplexes

The internet as a new environment

The main conclusion about the state

You do not live in a state. The state lives in you.

Part XXVI. The USSR as a memeplex: birth, flourishing, and death

Prehistory: why the niche emerged

By 1917, the Russian Empire was a memeplex that could no longer cope with reality:

When the memeplex destabilized, empty niches opened. Competition between memeplexes for the right to fill them began.

Birth: how the memeplex seized power

Competitors:

Why the Bolsheviks won:

Construction: how the memeplex inscribed itself into people

The core of the memeplex: foundational memes of the USSR

Scale-free structure of the Soviet memeplex:

This is not merely a list of memes — it is a heterogeneous network with hubs. At the center are several memes with high centrality, from which hundreds of connections radiate:

Consequence: A strike against the hub (e.g., “the party is the bearer of justice”) destroys the entire structure. This is precisely what happened at the 20th Congress: the meme “Stalin was wrong” struck at the central meme “the party is infallible,” and cracks spread across the entire network.

Flourishing: why the memeplex was successful

The Soviet person: a successful product of the memeplex

The memeplex created a new human — this is not propaganda, it is a fact. The Soviet person differed from the pre-revolutionary one and from the Western one:

This was a real phenomenon, confirmed by people’s behavior: mutual aid, a low level of violence, readiness to work for an idea, mass heroism in war.

Vulnerabilities of the memeplex

Decline: how stability eroded

Timescales of memeplex disintegration — see the model in Part X.

Key blows to the memeplex

Glasnost as a fatal error

Gorbachev committed a fatal error: he destroyed the monopoly on memes without creating any defense. This is like removing the immune system and opening the doors to viruses.

The information war the USSR did not wage

Western counter-memes were often simplifications or lies, but they worked. And Soviet memes were not defended.

The main cause of death: immunodeficiency

The leadership’s error

Mechanism of death

1940s-50s: natural immunity

1960s-70s: fading immunity

What should have been done vs what was done:

Result:

The USSR was not defeated — the USSR died of immunodeficiency

Lessons of the USSR for meme theory

The main conclusion

The USSR was not killed. It died because it stopped caring about its immune system — and perished together with the generation whose immunity was built through personal experience.

Jeans, rock-n-roll, and “freedom” did not defeat justice, equality, and space exploration. They simply filled the void left by a dying memeplex incapable of defending itself.

Part XXVII. Gandhi: hacking a memeplex through nonviolence

The problem: how to defeat the most powerful memeplex

The standard situation: a colony attempts to resist through violence — the empire suppresses and reinforces its narrative: “You see, they are savages; they need our firm hand.”

Violence against a perpetrator of violence plays by his rules. He is prepared for it. His memeplex is designed for this.

Gandhi’s genius: attacking the foundation of the memeplex

The memeplex of the British Empire:

Gandhi’s attack:

Gandhi understood: the core of the empire’s memeplex is moral superiority. “We are civilized; we bring law and order to savages.”

If the “savages” behave “savagely” (violence), the memeplex is confirmed. But if the “savages” behave more civilly than the civilizers — the memeplex destroys itself from within.

Network interpretation: Gandhi did not attack the periphery of the British memeplex (specific laws, officials) but rather the main hub — the meme “we are civilization.” This was a strike of maximum efficiency: destruction of the hub fragments the entire network. Hundreds of connected memes (“the right to violence,” “the white man’s burden,” “law and order”) lost their foundation when the central meme “we are civilized” proved false.

In contrast to traditional resistance, which attacks the periphery (kill a soldier, blow up a building), Gandhi’s method is a targeted attack on the hub. This is precisely why it is so effective: you don’t need to defeat the army — it suffices to destroy the central meme, and the structure collapses on its own.

The mechanics of nonviolent resistance

The Salt March: a perfect memetic attack

Gandhi chose the perfect target: a law so absurd that it was impossible to defend. Arresting people for collecting salt from the beach is self-exposure of the memeplex.

Why violence would not have worked

The key principle: forcing the memeplex to contradict itself

Britain stood on these memes:

• “We are law and order”
• “We are civilization against barbarism”
• “We are justice”

Gandhi created situations where the empire itself violated its own memes — publicly, documented, en masse.

The Amritsar Massacre (1919): the turning point

The hunger strike: a memetic weapon

The hunger strike is an impossible choice for the opponent:

• To concede = to acknowledge defeat
• Not to concede = to kill a saint

At the same time, Gandhi commits no violence. All violence is on the opponent’s side.

Why this worked specifically against Britain

Against a totalitarian regime without press and public opinion, Gandhi’s tactic would have been less effective. He exploited the internal contradictions of a specific memeplex.

Gandhi’s universal algorithm

Comparison of strategies for fighting memeplexes

Inheritors of the method

Limitations of the method

Against a regime prepared for total violence without witnesses, Gandhi’s method would have had limited effectiveness.

The main conclusion about nonviolence

Gandhi did not defeat the British Empire by force. He hacked its memeplex from within.

He demonstrated that the most powerful way to destroy a memeplex is not to attack its carriers but to force it to contradict its own foundational memes publicly and systematically.

This is not pacifism as weakness. This is memetic warfare of the highest order.

Radicalization: the reverse path (seizure of a memeplex by ideology)

If Gandhi is an example of hacking a memeplex from outside, then radicalization is an example of seizing a memeplex from within. Existing models describe motivational factors (Kruglanski, 3N model: Need -> Narrative -> Network) or a taxonomy of mechanisms (McCauley & Moskalenko, 12 mechanisms), but none provides a formal metric in the space of cognitive parameters. BMC provides a trajectory in measurable parameters.

BMC trajectory of radicalization:

Four measurable parameters:

Why radicalization is addictive:

• Ideology closes SIT-gaps (provides simple closure: “the enemies are to blame”) -> SEEKING is satisfied -> subjective relief
• The I-system is recalibrated to the ideological filter -> incompatible memes are rejected
• Social reinforcement (in-group) strengthens edges -> $Q$ grows -> modularity increases -> the memeplex rigidifies

Relation to other outcomes of reflexion:

Note: $F_{closure}$ (fidelity of closure) — the quality of closing a SIT-gap. During radicalization, LP is formally > 0 (ideology provides “answers”), but $F_{closure}$ < $\theta$: the closure is superficial, based on simplification rather than integration of new knowledge. The difference from genuine closure: low connectivity of the new meme with the rest of the M-layer.

Predictions:

Deradicalization through BMC: restore LP > 0 (provide a counter-narrative with better closure for the original gap), weaken the new hub (expose the ideology’s internal contradictions — the same Gandhian method), restore PLAY (a safe environment for doubt).

Part XXVIII. Unity and struggle of replicators: genes and memes

Two replicators on one planet

Two forms of life exist on Earth, using the same host — the human:

The memeplex does not replace the DNA replication mechanism. Genes are the foundation. The memeplex can subordinate genetic programs but not abolish them.

Key consequence: When the memeplex is destroyed (dementia, severe trauma), genetic programs are exposed in their pure form. This confirms that they are the foundation, and the memeplex is the superstructure.

Unique evolutionary advantage of memes: Synthesis (recombination + abduction) is what makes memes a radically faster replicator than genes. Genes recombine only during sexual reproduction, once per generation (~25 years). Memes recombine continuously — within a single host’s mind, in minutes. Abduction (filling structural gaps) has no genetic analog at all. It is precisely this capacity for rapid synthesis that explains the exponential acceleration of cultural evolution and allows memes to subordinate genetic programs, despite genes’ 3.5-billion-year evolutionary head start.

Theoretical basis: Full formalization of the Biomemetic Complex (BMC), including neurobiological substrate, G/M competition dynamics, ontogenesis, and pathologies — see BIOMEMETICS. For application to AGI — see AGI FOUNDATIONS.

Unity: when interests align

For millions of years, genes and memes cooperated. Memes helped genes survive (technologies, medicine); genes provided memes with carriers (children = new minds for memes).

This synergy created civilization. But as civilization developed, the interests of the replicators began to diverge.

Struggle: when interests diverge

Genes and memes are selfish replicators. Each optimizes its own copying, not the well-being of the hostarrier. When their interests diverge — and in the modern world they diverge ever more often — the human becomes a battlefield.

Victory of memes over genes: extremes

In extreme cases, memes completely suppress genetic programs, up to the destruction of the hoster:

Note on language. Language is the only case where M defeats G not as an anomaly (kamikaze) and not as an individual choice (celibacy), but as a permanent species-level condition. Signal memes (words, grammatical constructions, communication rules) consume WM slots for grounding, routing, and maintaining fidelity. With $k_{eff} \approx 3$–$4$ active slots, even a single signal meme = 25–33% loss of survival-relevant capacity. Ten experiments in a survival environment (predator avoidance, cooperative foraging, farming, goal-directed navigation, N=8–150) showed: $\Delta_{alive} \approx 0$ or negative. Language exists not because it helps the organism survive, but because memes that transmit well replicate better — and it persists only when the environment is rich enough to bear the cognitive cost.

This is a consequence of the dual replicator: the signal is optimized for M-fitness (transmission fidelity, compressibility), not for G-fitness (survival). Prediction P-BM28.

Language Emergence Threshold. From the parasitic nature of language, an emergence threshold is derived — 4 conditions must be met simultaneously: (1) resource surplus — the environment is rich enough for the agent to survive despite the WM overhead of language; (2) sufficient WM capacity — $k_{eff}$ must accommodate both survival cognition and signal processing; (3) executive planning — the agent must convert received information into directed navigation over long horizons; (4) memetic pressure — sufficient density and frequency of social interactions for meme competition.

This quadruple threshold provides a computationally grounded explanation for the late appearance of language in hominid evolution: Homo erectus had social groups and resources but limited $k_{eff} \approx 2$ and shallow planning depth. Homo sapiens achieved PFC expansion for $k_{eff} \geq 3$–$4$ and executive planning; the resulting cognitive surplus allowed WM slots to be captured by signal memes without fatal survival cost.

Separating test (BMC vs reward-engineered approaches). Systems that optimize communication under task reward (REINFORCE, PPO) predict a significant survival advantage when signal is enabled — their communication channel is trained to maximize joint payoff. BMC predicts survival neutrality — and this is precisely what is observed. The four conditions of the Language Emergence Threshold are deductively derived from the dual-replicator thesis (separation of G-fitness and M-fitness), rather than being a post-hoc rationalization of a negative result.

Cross-references: Formalization of WM-capture by signal memes — BM, Part IV; NM, Part VIII. Pressures on language structure — SM, section 8.7. Communication architecture — AGI_F, Part IV.

Key conclusion: Memes can use the host as expendable. For genes, this is catastrophic. For memes, it is sometimes the optimal strategy (a martyr disseminates the meme more effectively than a living person).

The decline in birth rates: a systemic victory of memes

In all societies with high technological progress, birth rates decline. This is a paradox from the genes’ perspective:

Pattern: The more memes in the environment, the fewer resources remain for genes. Memes created a civilization that suppresses its own creators.

Victory of genes over memes: natural selection in action

But the reverse also happens: genetic programs defeat the memeplex. This occurs when the memeplex is insufficiently strong to subordinate ancient instincts.

Key thesis: The civilization of memes created an environment with an excess of food, drugs, pornography, and dopamine simulacra. This environment is an unintentional test of memeplex quality.

Why this is “natural selection”:

A weak memeplex cannot subordinate genetic programs in the modern environment. The consequences:

Conclusion: Just as drugs test the memeplex’s resilience to hacks of the reward system, so too do other “temptations” of modernity test the memeplex’s ability to subordinate genetic programs. Failing the test = the memeplex is unfit for replication.

Hacks of both systems: when everyone loses

There exist phenomena that exploit vulnerabilities of both systems — both genes and memes lose:

Paradox: These hacks exist precisely because meme civilization created them (drugs, the internet, pornography). Memes generated a threat to themselves.

The selfishness of memes: the example of Coca-Cola

In 1985, Coca-Cola developed a new recipe (New Coke), which in blind tests people preferred to the original.

Result: fury, protests, boycotts. People poured out Coke in the streets. The company was forced to bring back the old (less tasty!) recipe.

The Coca-Cola meme literally prevented people from drinking a better-tasting beverage — just as the rabies virus prevents an animal from drinking water. The meme defends itself at the expense of the host’s interests.

Conclusion: dynamic equilibrium

We are carriers of two selfish replicators with partially overlapping, partially conflicting interests.

Key conclusion: “I” is not an arbiter between genes and memes. “I” is the battlefield and simultaneously the prize for the winner. When we say “I decided,” in reality one of the replicators has won a particular round.

Culture as SMR: inequality of access and social evolution

The parallel: SMR and human culture

In AGI_FOUNDATIONS we introduced the Shared Memplex Repository — a shared store with equal access for AGI agents. Humanity has a functional analog — culture — but with a critical difference: access is unequal.

Three forms of cultural capital (Bourdieu, 1986)

Pierre Bourdieu identified three forms of cultural capital that determine access to the cultural SMR:

Bourdieu’s key thesis: Schools do not eliminate inequality — they reproduce it, rewarding the cultural capital of the dominant class.

where the weights $w_i$ depend on the field of competition.

The Ratchet Effect and uneven evolution (Tomasello)

Michael Tomasello described the Ratchet Effect: cultural innovations are preserved and accumulated, never rolling back. This is the basis of cumulative cultural evolution.

But the critical question is: who has access to the ratchet?

Consequence: The gap grows over time:

Tension between strata as the engine of social evolution

Key analogy: Just as the G <-> M tension drives individual development, so the tension between strata drives the evolution of society.

Cognitive castes: the limiting case

At critical divergence of memeplexes between strata, society fragments into cognitive castes — groups incapable of mutual understanding.

where $M_{elite}$ and $M_{masses}$ are the sets of active memes in the respective strata.

Critical threshold: When $Caste_{gap} > 0.7$, society fragments into incompatible clusters. Communication between them becomes impossible — different foundational memes, different interpretations of reality.

The cycle of social evolution

Predictions of the model:

1. High Gap -> instability. Societies with a high cognitive gap are prone to revolutions.
2. Mass education lowers the Gap. Industrialization required literate workers -> Gap reduction in the 19th–20th centuries.
3. The internet: the access paradox. Initially lowered the Gap (Wikipedia, MOOCs), then amplified it (algorithmic bubbles, filter bubbles).
4. Elites defend inequality of access. Credentialism, closed networks, esoteric jargon — mechanisms for maintaining the Gap.