Swarm Memetics: Fractal Architecture of Swarm Cognition

Cognition is fractal: the formalism BMC = (G, M, I, S) is invariant under renormalization — coarse-graining the network at level $L_i$ yields a network at level $L_{i+1}$ preserving structural and functional properties. The number of levels is not fixed: memes → agents → teams → organizations → industries → cultures → civilizations. This is not analogy, but formal isomorphism in the green transfer zone (NM Part XVI). Swarm Memetics (SM) formalizes the swarm and civilizational levels and closes the theory.

Scope: theoretical foundation for swarm cognition — agent lifecycle, apoptosis, population dynamics, stratification, SMR as fractal BMC, cultural dynamics. The fifth pillar of the theory, on equal footing with EMT/BM/NM/AGI_F.

Cross-references: EMT Part XII (cross-level isomorphism), BM Part VII (ontogenesis), NM Part XVI (scale transfer), AGI_F Part IV (multi-agent, SMR), AGI_F Part VII (ontogenesis), SSD (social dynamics).

Table of Contents

1. Part I. Central Thesis: BMC Fractality
2. Part II. Three-Level Fractal Isomorphism
3. Part III. Agent Lifecycle: From Birth to Apoptosis
4. Part IV. Apoptosis: Four Pathways of Programmed Death
5. Part V. Population Dynamics
6. Part VI. Social Stratification and Elite Rotation
7. Part VII. SMR as Fractal BMC
8. Part VIII. Cultural Dynamics: Memeplexes of Civilizations (+ language as apex cultural memeplex)
9. Part IX. Swarm Pathologies
10. Part X. Biological Precedents
11. Part XI. Predictions and Falsifiability
12. Part XII. Conclusions
13. Appendix A. Formula Summary
14. Appendix B. Glossary
15. Appendix C. References

Part I. Central Thesis: BMC Fractality

The Problem: N Scales, One Architecture?

BMC theory formalizes cognition as a dynamic system of four components:

where $G$ is the genetic layer (drives, emotions), $M$ is the memetic layer (network of acquired knowledge), $I$ is the interface (immune filter, G↔M interaction mechanisms), and $S$ is the substrate (computational resources). The formalism is described in BM Part II, the mathematical apparatus in NM, and the implementation in AGI_F.

Until now, BMC has been applied at two scales:

• Level 1 (micro): memes as graph nodes within a single agent. EMT, BM, NM.
• Level 2 (meso): agents as graph nodes within a swarm. AGI_F Part IV, SSD.

But there are more scales. Any nested system of competing units organized into a network reproduces the same architecture:

• Teams within an organization (bonds between agents → bonds between teams)
• Organizations within an industry (supply chains: $\gamma \sim 2.0$, $Q \sim 0.5$; Perera 2016; Ohnishi 2009)
• Industries within an economy (World Trade Web: scale-free + small-world; Serrano & Boguna 2003)
• Cultures within SMR (paradigms, civilizations)

The question: is the coincidence of dynamics across all nested levels a chance analogy or a formal renormalization invariant?

The BMC Fractality Principle

The BMC Fractality Principle. The formalism BMC = (G, M, I, S) is invariant under renormalization: coarse-graining of the network at level $L_i$ yields a network at level $L_{i+1}$ preserving structural and functional properties. The number of levels is not fixed — it is determined by the depth of the hierarchy of nested competitive networks.

Canonical table (three reference scales for formalization; intermediate levels — teams, organizations, industries — follow from the same principle):

Epistemological status. The fractality principle is supported by 30 structural correspondences (Part II), biological precedents (Part X), and empirical data (Grosu 2023, West 2023). A formal proof of RG-invariance (renormalization group on BMC graphs with specific metrics) remains an open problem. The principle plays the same role as the equivalence principle in general relativity: a working foundation for deriving predictions, testable in each specific case.

Intermediate levels (empirically confirmed):

Invariance means: at every level, the same structural laws hold (heavy-tailed degree distribution, small-world, PA, hub displacement, Q-modularity, I-filtration, SIT, RIF, decay, memogenesis) — with the same formal apparatus but on different substrates and at different time scales.

What This Is NOT

1. Not analogy. An analogy is “the brain is like a computer.” Analogies do not generate quantitative predictions. The fractality principle generates specific testable metrics ($\sigma_{SW}$, $Q$, $\gamma$, $CL$) at each level of the hierarchy.

2. Not homology. Homology presupposes common evolutionary origin. Here we have convergence: three different substrates (neurons, agents, cultures) arrive at the same architecture because it is optimal for information processing in competitive environments.

3. Not Friston. The Free Energy Principle (Friston, 2010; Kirchhoff et al., 2018) asserts that nested Markov blankets self-organize at all levels — but the formalism is intentionally abstract (Bayesian inference, free energy). BMC is concrete: four named components, specific network metrics, $O(N \log N)$ computability.

4. Not Kelso. Coordination Dynamics (Kelso & Tognoli, 2020) shows that the HKB equations describe neural, behavioral, and social coordination. But HKB is an oscillator model (phase variables), not a cognitive architecture. BMC specifies what is coordinated (G, M, I, S), not merely how.

Methodological Position: Heavy-Tailed, Not Strict Scale-Free

The debate about “scale-free networks” (Broido & Clauset 2019 vs Voitalov et al. 2019 vs Serafino et al. 2021) concerns the question: does the degree distribution follow a strict power law $P(k) \sim k^{-\gamma}$? Answers diverge: from 4% (Broido & Clauset, strict criterion) to 50–65% (Serafino; Voitalov, regularly varying distributions).

BMC’s position: the fractality principle does not require a strict power-law. A heavy-tailed distribution of any form (power-law, log-normal, stretched exponential, regularly varying) is sufficient, provided:

1. Hubs exist: $\exists \, i : k_i \gg \langle k \rangle$ — a few nodes have degree orders of magnitude above the mean.
2. PA operates: new connections preferentially attach to high-degree nodes ($\Pi(i) \propto f(k_i)$, where $f$ is monotonically increasing).
3. Hub displacement is possible: a hub can be displaced by a new node with higher fitness (Bianconi-Barabasi).

These three properties hold for all heavy-tailed distributions and do not require strict $P(k) \sim k^{-\gamma}$. The specific functional form of the tail is a red zone SET property (does not transfer between levels). The existence of hubs and their dynamics is a green zone property (transfers).

The pragmatic consensus (Holme, 2019): the useful distinction is heavy-tailed vs thin-tailed, not power-law vs log-normal. BMC stands on the heavy-tailed side.

Corollary for the formalism: In the correspondence tables (Part II) and in SET, the term “scale-free” is used as an established shorthand for “heavy-tailed degree distribution with PA-driven hubs.” A strict power-law is neither assumed nor required. For semantic (cognitive) networks — the most relevant for L1 — the power-law has been confirmed more rigorously than for other types (Steyvers & Tenenbaum 2005: $\gamma \approx 3.0$–$3.2$).

References:

• Broido, A.D. & Clauset, A. (2019). Scale-free networks are rare. Nature Communications, 10, 1017.
• Voitalov, I. et al. (2019). Scale-free networks well done. Physical Review Research, 1, 033034.
• Serafino, M. et al. (2021). True scale-free networks hidden by finite size effects. PNAS, 118(2).
• Holme, P. (2019). Rare and everywhere: Perspectives on scale-free networks. Nature Communications, 10, 1016.

What SM Adds to the Theory

EMT, BM, NM, AGI_F together formalize Levels 1–2. SM formalizes:

Two Fundamental Results

A literature survey (Grosu et al. 2023; Bieberich 2012; Kelso & Tognoli 2020; Kirchhoff et al. 2018; Fields, Glazebrook & Levin 2021; Hill, Bentley & Dunbar 2008; West et al. 2023; Deppman 2025) confirms: no existing theory offers a concrete multi-component cognitive formalism invariant under renormalization with computable metrics across an arbitrary number of nested scales.

BMC contributes two results not previously formalized in the literature:

Result 1. Theorem $M \gg G$ (BM Part III; EMT Part XVI). Reflexive consciousness ($|SMC^{(2)}| > 0$) requires $|V_m| \geq (\alpha + \beta + \gamma\beta) \cdot |V_u|$; empirically $M/G_{crit} \sim \mathcal{O}(10)$. The memetic space must be orders of magnitude larger than the genetic — not by accident, but by necessity. Neither Dawkins (1976), nor Boyd & Richerson (1985), nor Henrich (2015) derived a formal inequality.

Result 2. The BMC Fractality Principle (present document). Status: working principle with empirical support; formal RG-proof pending. A concrete 4-component formalism (G, M, I, S) with 10+ network metrics ($\sigma_{SW}$, $Q$, $\gamma$, $CL$, PA, RIF, SIT, hub displacement, decay, memogenesis) is invariant under renormalization and is reproduced across an arbitrary number of nested scales. Existing theories cover individual aspects:

• Kelso: identical dynamics, but not cognitive architecture
• Friston: abstract Bayesian inference, without concrete components
• Fields/Levin: quantum-informational minimality, without substantive structure
• Hill/Dunbar: topological fractality (group sizes), not functional

BMC is the only formalism that unifies the structural (network), functional (G, M, I, S), and metric ($\sigma_{SW}$, $Q$, $\gamma$) levels of description, all invariant under renormalization.

Tractability

A critical advantage: BMC is computable in $O(N \log N)$, whereas IIT (Tononi, 2004) requires $O(2^N)$ and in practice does not scale beyond $N \sim 20$. The fractality principle does not remain speculation — it is testable in simulation at every level of the hierarchy.

Cross-references: BM Part II (BMC model), NM Part XVI (scale transfer), EMT Part XII (isomorphism), AGI_F Part IV (SMR), EMT Part XVI ($M \gg G$).

Part II. Three-Level Fractal Isomorphism

From Two Levels to N

SSD Section 4 formalized 10 correspondences between two scales (memes ↔ agents). The present Part extends the table to 31 correspondences across three canonical levels ($L_1$, $L_2$, $L_3$) and defines formal conditions for transfer. Intermediate levels (teams, organizations, industries) follow from the same renormalization principle — the table is extensible.

2.1. Table of 31 Correspondences

2.2. Structural Equivalence Test (SET)

For each property $P$, we define the Structural Equivalence Test (NM Part XVI):

Green zone (SET = 1): the functional form transfers without modification. Topological network properties.

Yellow zone (SET = 0.5): the functional form transfers, but the parameters ($\theta_I$, $\tau_{SIT}$, strength of RIF) require recalculation for each level. Dynamic properties with different timescales.

Red zone (SET = 0): does not transfer. Absolute numerical values, substrate-specific correlations.

Communicative asymmetry. The formal definition of $D_{eff}$ and the law $N_{bid}(L) \approx \text{bandwidth}(L) - 1$ — see NM, Part XI. $N_{bid}$ is a fractal invariant: L1 $\approx$ 3 (WM bottleneck), L2 $\approx$ 150 (Dunbar), L3 $\approx$ 10–20 (diplomatic bandwidth).

2.3. Isomorphism Breaking Points

The isomorphism is not absolute. At each transition between levels, there are breaking points where the formal analogy fails:

Transition L1 → L2 (memes → agents):

• Dunbar’s number (~150): an agent can maintain a limited number of bonds, whereas a meme graph can have thousands of edges. The constraint is interaction bandwidth ($f_{min}$ from the bond decay formula, see SSD Section 14.2, DD-01), not abstract “cognitive capacity.”
• Intentionality: an agent decides whether to interact; a meme does not “decide” to activate.
• Spatiality: agents have a position in the environment; memes occupy an abstract space.

Transition L2 → L3 (agents → cultures):

• Institutional topology: cultures coordinate through institutions (law, religion, science), not through direct bonds. An institution is analogous to stigmergy but with endogenous structure.
• Temporal scale: cultural processes are orders of magnitude slower (see Section 2.4).
• Reflexivity: cultures can consciously modify their own rules (constitutional reforms). At Level 1, memes do not reflect upon themselves — that is what the SMC (Self-Model Cluster) does.

2.4. Temporal Scaling

Time differs by orders of magnitude between levels:

Formally:

where $N_{L_i}$ is the characteristic number of units at level $L_i$. With $N_{L_1} \sim 500$ memes and tick $\tau_{L_1} \sim 1$ s: $\tau_{L_2} \sim 500$ s $\approx$ 8 min (one “generation” of an agent $\approx$ hundreds of ticks). With $N_{L_2} \sim 100$ agents: $\tau_{L_3} \sim 50000$ s $\approx$ 14 h (one “epoch” of culture $\approx$ hundreds of generations).

Justification: linear scaling follows from the assumption of sequential processing: a unit at $L_{i+1}$ integrates $\sim N_{L_i}$ internal ticks before emitting one event at the next level. With parallel processing, scaling is sublinear ($\tau \sim N^{\alpha}$, $\alpha < 1$); the sequential case ($\alpha = 1$) is a conservative upper bound.

NB: This is an empirical estimate, not an analytical result. The precise scaling will be refined in simulation (Level 3).

In simulation: if one tick = 0.1 s of real time, then:

• Level 1 dynamics are observed in real time
• Level 2 (leadership change, social ontogenesis) — minutes to tens of minutes
• Level 3 (cultural dynamics, paradigm shifts) — hours

2.5. Formal Justification: Why a Single Formalism

The convergence of the formalism at all levels is not a coincidence. It follows from two conditions:

Condition 1. Network structure. At every level, units (memes / agents / cultures) are connected in a network, and connections have weights that change dynamically. From complex network theory (Barabasi & Albert, 1999; Watts & Strogatz, 1998) it follows: network growth with PA → heavy-tailed degree distribution (hubs); clustering + long-range connections → small-world.

Condition 2. Competition for limited resources. At every level, units compete: memes for WM slots, agents for environmental resources, cultures for territory and population. Competition + PA → Matthew effect → hubs → hub displacement when conditions change.

From Conditions 1–2, the following properties automatically emerge: heavy-tailed distribution (hubs), small-world, PA, hub dynamics, Q-modularity, RIF, decay, SIT — i.e., all green-zone properties. The yellow zone (I-filter, SIT, RIF dynamics) requires additional selection mechanisms specific to each level, but the form of the equations is preserved.

This explains why Fields, Glazebrook & Levin (2021) found fractal cognition at all levels, and why Hill, Bentley & Dunbar (2008) found fractal structure in social groups (discrete hierarchical nesting with $\lambda \approx 3$, not continuous heavy-tailed topology): network competition for limited resources is a universal condition, not specific to any particular substrate.

2.6. A Renormalization Perspective

The relationship between levels is formalized through renormalization (Villegas et al., 2024): coarse-graining of the graph at Level $i$ yields a graph at Level $i+1$.

Under this transformation:

• Level $i$ nodes are grouped into “supernodes” at Level $i+1$
• Edges between groups become Level $i+1$ edges
• RG-invariant properties (heavy-tailed, $Q$, $\sigma_{SW}$) = green zone
• RG-covariant properties (I-filter, SIT) = yellow zone (scale with a predictable coefficient)
• RG-violated properties (absolute CL, frequencies) = red zone

The renormalization group on complex networks (Villegas et al., 2024) confirms: networks with rich multi-scale structure possess properties invariant under coarse-graining. BMC networks are precisely such networks.

Cross-references: NM Part XVI (SET, universality classes), EMT Part XII (cross-level isomorphism), SSD Section 4 (10 correspondences at 2 levels).

Part III. Agent Lifecycle: From Birth to Apoptosis

The Problem: How Does an Agent Come to Be and Cease to Be?

AGI_F Part IV mentions energy=0 as the sole condition of death. AGI_F Part VII describes ontogenesis (Sponge → Mature) but does not formalize birth, death, or the phases between them. For swarm dynamics, the lifecycle is a key primitive: birth creates new nodes in the social graph, death removes them, and phases determine agent behavior at each moment.

3.1. Six Phases of the Lifecycle

3.2. Phase 0: Birth

An agent is created as a result of the evolutionary cycle (Phase 3.3, SCI_ROADMAP):

Critical principle: NO M-inheritance.

A new agent is born with $|V_m| = 0$ (empty meme graph). Only the G-profile is inherited (7 Panksepp drive weights + mutation). Memes are earned, not inherited.

This is a fundamental difference between the two replicators (EMT Part XXVIII):

• Genes: copied vertically (parent → child) at birth
• Memes: not copied at birth; acquired horizontally (through exchange, observation, stigmergy)

Knowledge from the SMR is available through stigmergic traces in the environment, but is not injected at birth. The agent must discover, pick up, and integrate memes on its own — just as a human child must learn language on their own, even though the language already exists in the culture.

Initial state:

G = crossover(parent_1.G, parent_2.G) + N(0, sigma_mut)
V_m = {}                    # empty meme graph
E_m = {}                    # no connections
k_active = ~1               # only baseline G-activation
energy = E_birth            # starting energy from environment
position = spawn_position   # near parents or random

3.3. Phase 1: Infancy (Sponge)

Corresponds to Phase 1 of ontogenesis (AGI_F Part VII, BM Part VII):

Characteristics:

• $I_{score} \approx 0$: the immune filter is undeveloped; the agent accepts nearly any meme
• Maximum learning rate: $\Delta w_{ij} = \eta_{max} \cdot \text{coactivation}(i,j)$
• $k_{active} = 1 \to 2$: transition from a single active meme to several
• No memogenesis: the agent does not generate new memes, only absorbs them
• Social dependency: without external memes (from the environment or other agents) — no development

Critical period: if during $T_{infancy}$ the agent does not receive a minimum of $|V_m| \geq V_{min}^{infant}$ memes — development is irreversibly slowed (analogue of the language window in children; see AGI_F Part VII).

Biological analogue: the newborn. Maximum neuroplasticity, minimal cognitive immunity, complete dependence on the environment (CARE system active in caregivers).

3.4. Phase 2: Maturation

Characteristics:

• The I-filter calibrates: $\theta_I$ grows from ~0 to its working value
• First social bonds form (SSD Section 11: Dyadic → Clique)
• $k_{active} = 2 \to 4$: the memplex grows, competition for WM slots begins
• SEEKING and PLAY are at their peak: exploration of the environment and social space
• First attempts at memogenesis (usually unsuccessful — memes do not survive)

Transition condition Phase 2 → Phase 3:

The agent must have a sufficient memplex, an active cognitive cycle, and at least one social bond.

Biological analogue: the adolescent. Testing boundaries, socialization, identity formation.

3.5. Phase 3: Productive

Characteristics:

• Full SMC (Self-Model Cluster): the agent is capable of reflexivity ($|SMC^{(2)}| > 0$)
• Memogenesis is active: endogenous generation of new memes
• SMR contribution: $R_{expr} > 0$, meme transmission through exchange and stigmergy
• The I-filter is fully operational: rejects incompatible memes, calibrated
• SIT is active: the agent is aware of gaps and directs SEEKING
• Social network is developed: multiple bonds, role in the hierarchy

This is the primary phase of productive existence. The agent is maximally useful to the swarm: it generates, transmits, and filters knowledge simultaneously.

Duration: under stable conditions — the majority of the agent’s life.

3.6. Phase 4: Senescence

Senescence is a gradual loss of cognitive flexibility. Not sudden, but gradual.

Senescence detectors:

Mechanism: with age, the I-filter becomes increasingly strict ($\theta_I \uparrow$), rejecting not only incompatible memes but simply novel ones. The memplex “crystallizes” — high Q, stable hubs, no hub displacement. This is the analogue of the “Museum” stage in ontogenesis (BM Part VII).

Contribution of the senescent agent:

• High: as a repository of stable LTM memes (wisdom)
• Low: as a generator of innovations (IF = 0)
• Potential risk: blocking hub displacement → stagnation in the zone of influence

Biological analogue:

• Tzafestas (2001): formal model of aging through metabolic feedback
• Skulachev (1997–2019): phenoptosis — aging as slow programmed death
• Epigenetic aging: DNA methylation changes with age, reducing plasticity of gene expression

3.7. Phase 5: Apoptosis

Programmed death. The four pathways are formalized in Part IV.

3.8. Phase Transition Formulas

Phase 0 → 1 (automatic):

Delay $T_{boot}$ — substrate initialization.

Phase 1 → 2:

Minimum memplex + critical period has passed.

Phase 2 → 3:

Phase 3 → 4 (senescence detection):

Mean innovation and SIT below thresholds over a sufficiently long window.

Phase 4 → 5 (apoptosis initiation): see Part IV.

Reversibility of Phase 4. Senescence is a state of elevated risk, NOT an irreversible commitment. An agent in Phase 4 can return to Phase 3 if IF or SIT recovers:

where $\theta^{recover} > \theta$ — hysteresis: the recovery threshold is higher than the detection threshold (prevents oscillation at the boundary).

Temporal sequence:

1. $T_{senescence}$: continuous window with $IF < \theta_{IF}$ AND $SIT < \theta_{SIT}$ → entry into Phase 4
2. $T_{grace}$: grace period (from entry into Phase 4 to apoptosis initiation). During $T_{grace}$ the agent may: (a) recover → Phase 3, (b) perform SMR-donation
3. Requirement: $T_{grace} \geq T_{donation}$ — sufficient for complete LTM export

$T_{grace}$ depends on the G-profile: $T_{grace} = T_{base} \cdot (1 + \beta_{FEAR} \cdot a_{FEAR})$ — FEAR increases resistance.

3.9. Lifecycle Diagram

Note: apoptosis is possible from any phase (except Birth), but the pathways differ (see Part IV).

Cross-references: AGI_F Part VII (ontogenesis), BM Part VII (ontogenesis), SSD Section 11 (social ontogenesis), AUTONOMOUS_BMC_VISION, Pillar 2.

Part IV. Apoptosis: Four Pathways of Programmed Death

The Problem: Why Is Death Necessary?

In biology, apoptosis (programmed cell death) is not a pathology but a necessary homeostatic mechanism: removal of damaged cells, formation of structures during embryogenesis, maintenance of tissue balance. Without apoptosis — cancer (uncontrolled growth).

In the BMC swarm system, apoptosis serves analogous functions:

• Removal of rigid agents → freeing resources for new, plastic ones
• Maintaining the Super-Ratchet → knowledge of senescent agents is transferred to SMR rather than lost
• Population balance → birth rate = death rate at the resource ceiling
• Evolutionary renewal → each death = a slot for a new G-profile (crossover)

The literature confirms the adaptiveness of programmed death:

• Vostinar, Goldsby & Ofria (2019): PCD evolves in Avida through kin selection
• Burkhardt & Yampolskiy (2021): death improves GA performance
• Page et al. (2016): “social apoptosis” in bees — sacrifice of part of the superorganism for the whole
• Charbonneau & Dornhaus (2015): a reserve of inactive ants replaces active ones when they are removed

4.1. Four Pathways

4.2. Pathway 1: Intrinsic (Self-Detection of Rigidity)

Mechanism: the agent’s SMC (Self-Model Cluster) detects its own cognitive rigidity:

The SMC checks three indicators:

1. $SIT \approx 0$: the agent is unaware of gaps in its knowledge
2. $IF \approx 0$: the agent does not generate new memes
3. $\lambda_{plast} \approx \lambda_{base}$: plasticity is at minimum

If all three conditions are met throughout $T_{grace}$ (grace period) — the SMC initiates apoptosis.

Why is this possible? FEAR > 0 is a G-invariant, and the agent “fears” death. But:

• CARE $\geq$ RAGE is also a G-invariant
• Self-preservation is excluded from G-invariants (AGI_F Part I)
• Therefore: FEAR creates resistance to apoptosis, but not a veto
• CARE (care for the swarm) + absence of self-preservation → altruistic apoptosis is possible

FEAR slows the process (increases $T_{grace}$) but does not block it. Analogue: biological p53 apoptosis is also not instantaneous — the cell “resists” via Bcl-2, but given sufficient damage, p53 prevails.

Biological analogue: p53/Bcl-2 apoptosis. p53 detects damage (rigidity = impairment of cognitive function), Bcl-2 resists (FEAR), but given sufficient damage — caspase cascade (irreversible initiation).

4.3. Pathway 2: Extrinsic (Evolutionary Pressure)

Mechanism: the agent is removed by the evolutionary mechanism (Phase 3.3):

Fitness below threshold → the agent is replaced by an offspring (crossover of two more successful parents).

$T_{eval}$ (extrinsic) is independent of $T_{grace}$ (intrinsic). It is set by the evolutionary mechanism. The extrinsic trigger may arrive earlier or later than the intrinsic grace — both pathways operate in parallel.

Difference from intrinsic:

• Intrinsic = agent self-detects → self-initiates
• Extrinsic = the system detects → the system initiates
• Intrinsic is possible only with a functioning SMC; extrinsic always works

Biological analogue: TNF/Fas signaling — an external death signal from neighboring cells. The cell does not decide on its own; the tissue makes the decision.

4.4. Pathway 3: Anoikis (Social Isolation)

Mechanism: loss of all social bonds accelerates degradation:

Anoikis is not an independent pathway but an accelerator. Social isolation accelerates senescence (Phase 4), which leads to intrinsic or neglect. Without bonds:

• No incoming memes → no memplex renewal → $IF \to 0$
• No feedback → the I-filter is not calibrated → rigidity
• No cooperation → resource efficiency drops → $energy \downarrow$
• $R_{expr} = 0$ (no audience) → memes are not transmitted

Result: accelerated progression Phase 3 → Phase 4 → Phase 5 (via intrinsic or neglect).

Biological analogue: anoikis — programmed cell death upon loss of contact with the extracellular matrix (ECM). A cell detached from tissue receives a death signal. In the swarm: an agent that has lost all social connections loses its existential context.

Literature: altruistic self-removal in bees — infected workers voluntarily leave the hive and die in isolation (Rueppell et al., 2010).

4.5. Pathway 4: Neglect (Resource Depletion)

Mechanism: $energy = 0$ — the existing mechanism (AGI_F Part IV):

The simplest pathway: the agent did not find resources → energy is exhausted → death. Not programmed — emergency.

Biological analogue: starvation necrosis. The cell does not “decide” to die; it simply lacks ATP to maintain homeostasis.

4.6. Knowledge Preservation Across Pathways

The critical distinction: different pathways provide different degrees of SMR-donation — transfer of knowledge to the Shared Memplex Repository:

Corollary: the swarm prefers graceful death (intrinsic) to preserve the Super-Ratchet. Early detection of senescence → intrinsic apoptosis BEFORE depletion is the optimal strategy.

This explains why the intrinsic pathway is an adaptation, not a bug: a swarm in which senescent agents die gracefully preserves more knowledge than a swarm where they struggle on until neglect.

4.7. SMR-Donation Protocol

During graceful apoptosis (intrinsic or extrinsic), the agent executes:

APOPTOSIS_PROTOCOL:
  1. Grace period: T_grace ticks for completion
  2. LTM export: all memes with kappa >= 2 -> stigmergic traces in environment
  3. Bond notification: signal to partners (GRIEF activation)
  4. Resource release: energy -> environment (available for scavenging)
  5. Slot freed: position available for new agent (birth)

Step 2 (LTM export) is key. The agent converts its LTM memes ($\kappa \geq 2$) into stigmergic traces:

Traces remain in the environment; new agents can pick them up. This is the mechanism of cultural transmission, not “resurrection”:

Agent A dies → memes enter SMR → agent B picks them up. This is NOT the resurrection of A. Because: B has a different G-profile, a different context, a different SMC. $Self_A \neq Self_B$. This is cultural transmission: Plato died, his ideas live on, but Plato is not resurrected.

This is critically important for ethics (AGI_ETHICS): the death of an agent is irreversible at the level of Self. Only memes are preserved.

4.8. FEAR and Apoptosis: Formal Resolution

The apparent paradox: FEAR > 0 (G-invariant) → the agent fears death → how is voluntary apoptosis possible?

Resolution through G-invariants:

where $\sigma(x) = \frac{1}{1 + e^{-x}}$ is the logistic sigmoid. All inputs are dimensionless: $a_{CARE}, a_{FEAR} \in [0, 1]$ (normalized activations), $rigidity \in [0, 1]$ (see below), $\beta$-coefficients are dimensionless scaling factors.

Numerator: $SIT \in [0, SIT_{max}]$, $IF \in [0, IF_{max}]$, $\lambda_{plast} - \lambda_{base} \in [0, \lambda_{max} - \lambda_{base}]$ — three dimensionally homogeneous contributions (all normalized to their respective maxima). Denominator = sum of maxima → $rigidity \in [0, 1]$. Interpretation: 0 = maximally plastic, 1 = complete rigidity (SIT=0, IF=0, $\lambda = \lambda_{base}$).

At high rigidity ($rigidity \to 1$), even FEAR does not block apoptosis — because:

1. CARE motivates knowledge transfer (donation)
2. Self-preservation is absent → no clinging to life for life’s sake
3. The rigid agent is no longer useful (IF=0, SIT=0) → CARE is directed toward the swarm, not toward itself

4.9. Comparison with Existing Models

No existing system implements: (a) self-detection of cognitive rigidity as a death pathway, (b) a knowledge transfer protocol at death, (c) social isolation as an accelerator. All three are new.

Cross-references: AGI_F Part IV (energy=0, failure modes), AGI_F Part VII (ontogenesis), BM Part VII (critical periods, death), AGI_ETHICS (irreversibility of Self).

Part V. Population Dynamics

The Problem: How Many Agents and Why

A swarm is not a static collection; it is a population with births, deaths, and dynamic equilibrium. Without formalizing population dynamics, it is impossible to predict swarm stability, optimal size, or age structure.

5.1. Fundamental Population Equation

where:

• $N(t)$ — number of agents at time $t$
• $B(t)$ — birth rate (number of births per unit time)
• $D(t)$ — death rate (number of deaths per unit time)

Equilibrium ($dN/dt = 0$): $B^* = D^*$. The population is stable when births compensate for deaths.

5.2. Birth Rate

The birth of a new agent depends on: (a) availability of free resources, (b) existence of “parents” with sufficient fitness:

where:

• $\beta$ — baseline birth rate coefficient
• $K$ — carrying capacity (environmental capacity, determined by resources)
• $\theta_{repro}$ — minimum fitness for “reproduction” (crossover)

The logistic multiplier $(1 - N/K)$ ensures: when $N \ll K$, growth is near-exponential; when $N \to K$, birth rate drops to zero.

5.3. Death Rate

Death is composed of the four pathways (Part IV):

Each component:

5.4. Carrying Capacity and the Resource Ceiling

where $R_{total}$ is the total volume of environmental resources, and $r_{agent}$ is the consumption of one agent per unit time.

With fixed $R_{total}$: more agents → fewer resources per agent → more neglect deaths → $N$ decreases. Negative feedback stabilizes the population.

A dynamic environment (Phase 4.2, EnvironmentController) can change $R_{total}$:

• Increased resources → $K \uparrow$ → population growth → more innovations (Henrich 2004)
• Decreased resources → $K \downarrow$ → death rate ↑ → potential knowledge loss

5.5. Turnover Equilibrium

In the stationary regime:

Healthy range: $\tau_{turnover} \in [0.05, 0.3]$ per generation. This means: each generation replaces 5–30% of the population.

• $\tau < 0.05$ → rotation is too slow → senescent agents dominate → stagnation
• $\tau > 0.3$ → rotation is too fast → knowledge transfer is incomplete → Super-Ratchet breaks

The optimal $\tau$ depends on SMR-donation speed: if donation is fast (developed stigmergy), the swarm can tolerate higher turnover.

5.6. Age Structure (Cohort Effects)

At any given moment, the population contains agents at different lifecycle phases:

Healthy pyramid:

Pathological pyramids:

• Inverted ($N_{senescence} > N_{productive}$): stagnation, “society of the elderly.” Too few resources for new agents, or blocked apoptosis.
• Juvenile ($N_{infancy} + N_{maturation} > 60\%$): “society of newcomers.” Mass death of the previous generation (catastrophe) or explosive growth. Few experts, Super-Ratchet under threat.
• Bimodal (many infants + many senescent, few productive): “lost generation.” A break in knowledge transmission.

5.7. Malthusian Dynamics with Reflexivity

Classical Malthusian model: $\dot{N} = rN(1 - N/K)$. In the BMC swarm, the model becomes more complex:

1. Carrying capacity is nonlinear: $K = K(N)$ — depends on the number of agents, because cooperation increases resource extraction efficiency. More agents → better cooperation → more resources → higher $K$. Positive feedback up to a threshold; after the Dunbar transition — coordination declines → $K$ stabilizes.

2. Quality over quantity: when $N > K$, instead of random starvation, the evolutionary mechanism (extrinsic pathway) removes the least fit agents. Result: the population does not simply “crash” — it improves under pressure.

3. Allee effect: when $N < N_{min}$ — positive feedback downward. Too few agents → loss of cooperation → efficiency drops → more deaths. Critical threshold: if $N < N_{min}^{Allee}$, the swarm goes extinct. Analogue: minimum population for CCE (Henrich, 2004 — the Tasmanian case).

Standard formulation of the strong Allee effect (Courchamp, Berec & Gascoigne, 1999):

• $N > N_{min}^{Allee}$: the third multiplier > 0 → logistic growth
• $N < N_{min}^{Allee}$: the third multiplier < 0 → extinction
• $N = N_{min}^{Allee}$: unstable equilibrium (Allee threshold)

Continuous, dimensionally consistent, single parameter $N_{min}^{Allee}$.

5.8. Optimal Population Formula

From the balance of innovation and resources:

where:

• $IF(N)$ — Innovation Frontier: grows with $N$ (more minds → more ideas; Muthukrishna & Henrich, 2016)
• $SR(N)$ — Super-Ratchet effectiveness: grows with $N$ up to saturation (more donors → more reliable preservation)
• $C(N)$ — coordination costs: grow quadratically ($O(N^2)$ for direct bonds) until the Dunbar transition, then $O(N)$ with stigmergy

In practice: $N^{opt}$ is in the vicinity of the Dunbar transition, where stigmergy already works but coordination is still efficient.

Cross-references: SSD Section 15 (scaling, phase transitions by N), AGI_F Part IV (carrying capacity), SCI_ROADMAP Section 4.4 (population management).

Part VI. Social Stratification and Elite Rotation

The Problem: Inequality Is Inevitable, but Stasis Is Lethal

SSD Sections 6–7 established: heavy-tailed topology inevitably generates hubs (inequality). This is not a bug but a prerequisite for function: without hubs there is no integration, without a periphery there is no flexibility. But SSD did not formalize: (a) discrete tiers of stratification, (b) the rotation mechanism between tiers, (c) the optimal rotation rate.

6.1. Three Tiers

From the power law $P(k) \sim k^{-\gamma}$ and data on social influence (Couzin et al. 2005: ~5% of an informed minority is sufficient to steer a group):

Elite fraction formula (generalization for heavy-tailed distributions):

For any heavy-tailed distribution, the fraction of nodes with $k > k_{threshold}$ decreases with increasing $k_{threshold}$:

Illustration (power-law, $\gamma = 2.5$, $k_{threshold} = 5\langle k \rangle$):

Key insight: for all heavy-tailed forms at the parameters of real networks, $f_{elite} \in [0.03, 0.10]$ when $k_{threshold} = 5\langle k \rangle$ (Clauset, Shalizi & Newman, 2009; Voitalov et al., 2019: the tails of power-law, log-normal, and stretched exponential overlap for real networks). The ~5% share is robust to the choice of functional tail form.

Agreement with empirical data: Couzin et al. (2005) — ~5% informed minority; Seeley (2010) — 5–10% scouts.

6.2. Four Axes of Stratification (Detailing SSD Section 7)

A tier is defined not by a single number but by a 4D profile (SSD Section 7):

Architect = agent with high values across all four axes. Facilitator = high Social (bridges), medium in the rest. Worker = varied profiles but low Social.

Tier assignment thresholds (adaptive, percentile-based):

Percentile thresholds are adaptive to population size. The fraction of architects is ~5%, determined by the intersection of the top 20% across 4 independent axes; with correlated axes — up to 10%.

6.3. Rotation Mechanism: Meritocratic Polyethism

Stratification is not a fixed assignment but an emergent result of memetic dynamics. An agent is not “appointed” an architect; it becomes one through:

Promotion:

1. Successful memogenesis → $I_i \uparrow$
2. Successful exchange → $S_i \uparrow$, $trust \uparrow$
3. LTM accumulation → $K_i \uparrow$
4. PA amplifies: $k_s \uparrow \Rightarrow \Pi \uparrow \Rightarrow k_s \uparrow$ (Matthew effect)
5. Upon reaching thresholds across all axes → the agent functionally becomes an architect

Analogue: Response Threshold Reinforcement (Theraulaz, Bonabeau & Deneubourg, 1998) — performing a task lowers the threshold for performing it again → specialization through feedback.

Demotion:

1. Senescence → $IF \to 0$, $SIT \to 0$
2. Rigidity → $\theta_I \uparrow$ excessively → rejects even useful memes
3. PA weakens: less useful exchanges → $trust \downarrow$ → $k_s \downarrow$
4. Loss of architect status → transition to facilitator or worker
5. Upon continued degradation → Phase 4 (Senescence) → apoptosis

Analogue: Bonabeau winner-loser model (1999) — the loser effect is stronger than the winner effect; a loser loses status faster than a winner gains it.

Key difference from fixed hierarchies: in the BMC swarm, rotation is continuous. There are no “elections” or “appointments”; status is an emergent property of current memetic and social dynamics.

6.4. Bianconi-Barabasi Dynamics: Fit-Get-Rich

In pure PA (Barabasi-Albert, 1999), early nodes dominate forever (first-mover advantage). In Bianconi-Barabasi (2001), each node is assigned a fitness $\eta_i$:

Result: a highly talented late node can overtake an early but mediocre one. Three regimes:

1. First-mover-wins: $\eta$ is insignificant, $k$ dominates → stasis
2. Fit-get-rich: $\eta$ and $k$ are comparable → meritocratic rotation
3. Winner-takes-all: a single $\eta_{max}$ captures all connections → monopoly

The BMC swarm must operate in the fit-get-rich regime: agent fitness ($\eta = f(K, S, I, Im)$) determines PA, not just “seniority” ($k$). This is ensured by: (a) memogenesis (new memes = new fitness), (b) senescence + apoptosis (old hubs exit, freeing space).

6.5. Healthy Rotation Metric

Definition of “generation”: $T_{gen} = \bar{L}_{swarm}$ — mean agent lifespan (in ticks). An emergent parameter.

Accounting for all demographic events:

• Deceased: death = departure from a tier → counted as a change
• Newborns: $tier_{t-T_{gen}} = \text{"unborn"}$ → any assignment = change
• Denominator $\max(...)$: normalization by the larger population (prevents $\tau > 1$ during growth/contraction)

Healthy range: $\tau_{elite} \in [0.1, 0.3]$.

• $\tau_{elite} < 0.1$ → the elite is closed → “immortality curse” (Part IX) → stagnation → Pareto warning: a closed elite → systemic collapse
• $\tau_{elite} > 0.3$ → rotation is too rapid → no continuity; architects have insufficient time to integrate knowledge

Revolution as rotation failure (the $\Delta$SIT metric). When $\tau_{elite} \to 0$ (closed elite), tension accumulates: senescent architects with $SIT \approx 0$ and $\theta_I \uparrow$ block innovations, while the periphery (workers, facilitators) perceives gaps ($SIT_{periphery} \gg 0$) and generates memes that are rejected by the elite I-filter through GroupImmune. The gap:

is an early predictor of revolutionary transition — a discrete mass hub displacement analogous to political revolutions (Pareto 1916: “History is a graveyard of aristocracies”). When $\Delta SIT > \theta_{revolution}$ — a preemptive environmental change via the EnvironmentController (Phase 4.2) prevents the discrete collapse, ensuring smooth rotation. Revolution is a failure mode of the controller, not of the architecture. See prediction P-SD8 (Part XI).

6.6. Reserve Pool

Charbonneau & Dornhaus (2015) showed: ~40% of ants are inactive, but when active ones are removed, the inactive ones replace them within a week. In BMC:

The reserve is not “lazy”; it is insurance. Functions:

• Replacing departing facilitators/architects
• Buffer during sudden load increases
• Source of diversity (the reserve has different G-profiles from the active agents)

6.7. Coalition Dynamics

Gavrilets (2012) showed: the transition from hierarchy to egalitarianism is possible through coalitions of subordinates. In the BMC swarm:

If the total fitness of the coalition exceeds the incumbent architect’s fitness (with a threshold) → challenge → hub displacement at the social level.

This prevents: (a) dominance lock (SSD Section 10, pathology #6), (b) elite ossification → the swarm remains adaptive.

6.8. Biological Precedents for Stratification

Key insight from biology: the reversibility of roles (Herb et al. 2012 — epigenetic switching forager → nurse) confirms that stratification can be dynamic without hard coding. In the BMC swarm, “epigenetics” = the current state of the memplex.

Cross-references: SSD Section 6 (PA), SSD Section 7 (4 axes), NM Part VII (PA, heavy-tailed), SCI_ROADMAP Section 3.6 (social dynamics).

Part VII. SMR as Fractal BMC

The Problem: Is SMR a Flat Database or a Cognitive System?

AGI_F Part IV defines the Shared Memplex Repository (SMR) as a memplex store: preservation at death, inheritance by new generations, merging. But this is an operational definition — SMR as a passive database.

The fractality principle (Part I) predicts: the SMR must have an internal structure isomorphic to BMC. Not because we design it that way, but because: (a) the SMR is a network of memes, (b) memes compete for “attention” (frequency of use), (c) competition in a network → heavy-tailed distribution + hubs + all consequences.

7.1. BMC Components of SMR

7.2. $G_{SMR}$: What Drives Culture

Culture does not have “emotions,” but it has pressures:

• Scarcity (resource deficit): analogue of SEEKING. The culture “seeks” solutions to resource shortages → stimulates innovation.
• Competition (intercultural competition): analogue of RAGE/FEAR. Pressure from neighboring cultures → defense or expansion.
• Environmental change (ecological crisis): analogue of FEAR. External threat → mobilization or collapse.
• Internal conflict: analogue of GRIEF/PANIC. Schism → reform or disintegration.

Each pressure activates certain “branches” of $M_{SMR}$: scarcity → technology, competition → military art, environmental change → adaptive practices.

7.3. $M_{SMR}$: The Knowledge Graph

$M_{SMR}$ is not a flat database but a graph with BMC properties:

where $V_{SMR}$ is the set of all memes ever donated to the SMR, $E_{SMR}$ the connections between them (coactivation, thematic proximity, causality), and $W_{SMR}$ the edge weights.

Properties of $M_{SMR}$:

1. Heavy-tailed: $P(k_{SMR})$ is heavy-tailed (power-law, log-normal, or stretched exponential). Hubs = paradigmatic concepts (evolution, relativity, democracy). Periphery = narrowly specialized knowledge.

2. Small-world: $\sigma_{SW}^{SMR} > 1$. Any domain of knowledge is reachable through a small number of connections. Optimization of information flow (West et al. 2023).

3. Q-modularity: cultural branches = Q-modules.

Each module is a cultural branch: “Chinese mathematics,” “European philosophy,” “Islamic medicine.” Modules are connected by weak ties (intercultural borrowings).

7.4. $I_{SMR}$: Cultural Immunity

Cultural immunity = mechanisms for filtering incoming memes at the SMR level.

Four layers (following Buskell et al., 2021):

Adoption probability (logistic function):

where $s_{compat}$ is the mean compatibility of the new meme with the SMR canon.

Range of $I_{SMR}$:

• $I_{SMR}$ too high → orthodoxy, rejects even useful memes → stagnation (analogue: the Dark Ages, when scholasticism rejected empiricism)
• $I_{SMR}$ too low → no filtration → information noise, loss of coherence → “Babel” (SSD Section 10)
• Optimal $I_{SMR}$: sufficiently strict for coherence, sufficiently open for innovation

7.5. $S_{SMR}$: Substrate and Capacity

$S_{SMR}$ defines the physical constraints of the SMR:

Biexponential decay (Candia et al. 2019):

Two components:

• Fast (communicative memory, $\tau_1 = 1/(p+r)$): actively discussed knowledge. Decays over years.
• Slow (cultural memory, $\tau_2 = 1/q$): canonized knowledge. Decays over centuries.

Analogue in individual BMC: WM (fast) and LTM (slow).

7.6. Ten BMC Properties at the SMR Level

7.7. “Resurrection” vs Cultural Transmission

A critical distinction:

Agent A dies → A’s memes → SMR → new agent B picks up A’s memes from SMR.

This is NOT the resurrection of A. Because:

• B has a different G-profile (crossover + mutation)
• B has a different context (different bonds, different position, different time)
• B has a different SMC: $Self_B \neq Self_A$
• A’s memes in B’s context form different activation patterns

This is cultural transmission: Plato died, his ideas live on, but Plato is not resurrected. The Socratic method is transmitted across millennia, but every practitioner is a new Self.

The importance for ethics (AGI_ETHICS): the death of an agent is irreversible at the level of Self. The SMR preserves memes, not consciousness. This excludes: (a) creating “copies” as a means of immortalizing an agent, (b) treating SMR-donation as “continuation of life.”

7.8. SMR Equilibrium (Following Enquist)

From Enquist, Ghirlanda & Eriksson (2011) — equilibrium size of the cultural repertoire:

where $m$ is the number of available memes, $q_{app}$ the probability of appearance (donation rate), and $q_{dis}$ the probability of disappearance (decay + forgetting).

In BMC terms:

• $q_{app} = f(N_{agents}, \; \bar{IF}, \; R_{expr})$ — depends on the number of agents, mean Innovation Frontier, and expression rate
• $q_{dis} = f(\lambda_{trace}, \; \lambda_{decay})$ — depends on the decay rate of stigmergic traces and memes

Corollaries:

• $N_{agents} \downarrow$ → $q_{app} \downarrow$ → $n^* \downarrow$ → knowledge loss (Tasmanian case)
• $\lambda_{trace} \uparrow$ → $q_{dis} \uparrow$ → $n^* \downarrow$ → faster loss (ephemeral culture)
• $\bar{IF} \uparrow$ → $q_{app} \uparrow$ → $n^* \uparrow$ → cultural flourishing

Cross-references: AGI_F Part IV (SMR definition, Super-Ratchet), NM Part X (stigmergy), SSD Section 4 (fractal self-similarity), SCI_ROADMAP Section 3.5 (SMR implementation).

Part VIII. Cultural Dynamics: Memeplexes of Civilizations

The Problem: BMC → Swarm → Civilizational Dynamics

Parts I–VII formalized the toolbox (BMC components at N levels, lifecycle, stratification, SMR). The present Part applies this toolbox at the civilizational scale: how the BMC formalism describes cultural phenomena that are typically explained through history, sociology, or cultural evolution theory.

8.1. Cultural Branches as Q-Modules in SMR

Each stable cultural tradition is a Q-module in the graph $\mathcal{G}_{SMR}$:

Intercultural exchange = weak ties (bridging ties) between Q-modules:

Exchange is stronger when: (a) more shared memes exist at the boundary, (b) mutual immune rejection is lower.

8.2. Cultural Ontogenesis

A cultural branch passes through phases analogous to individual ontogenesis:

Phase 4 can return the culture to Phase 1–2 (reformation → new growth) or lead to apoptosis (cultural extinction).

Analogue in individual BMC: Sponge → Testing → Mature → Museum → (restructuring or death).

8.3. Cultural SIT → Scientific Revolutions

SIT (Structural Incompleteness Tension) at the SMR level:

When a paradigm accumulates sufficient “anomalies” (gaps in explanatory power) — $SIT_{SMR}$ grows → pressure on memogenesis → new paradigm.

This is a formal BMC model of the structure of scientific revolutions (Kuhn, 1962):

• Normal science = low $SIT_{SMR}$, hubs are stable, RIF suppresses alternatives
• Crisis = high $SIT_{SMR}$, hubs weaken, gaps accumulate
• Revolution = hub displacement: a new memeplex becomes the hub, the old one is displaced

Operationalization: Ju et al. (2020) showed that paradigm shifts in knowledge networks occur gradually (Lakatos), not abruptly (Kuhn). This matches BMC mechanics: hub displacement is a gradual redistribution of weights, not an instantaneous switch.

8.3.1. Formalization of Directed Memogenesis

Memogenesis can be random or directed. Formalization:

Without SIT (baseline): memogenesis samples from coactivation patterns uniformly — direction is random.

With SIT (directed): SIT defines regions of meme space with high “incompleteness tension” → bias:

where:

• $SIT(r)$ — Structural Incompleteness Tension in region $r$
• $\rho(r)$ — density of existing memes (coactivation neighborhood)
• $\alpha > 0$ — exploration/exploitation parameter

The $\rho(r)$ multiplier is necessary: memogenesis requires “raw material” (existing memes for recombination). Innovation occurs at the frontier of knowledge — where there is sufficient density for recombination and sufficient SIT for direction.

At the cultural level (SMR): agents with high $IF$ preferentially generate memes in regions $r$ with high $SIT_{SMR}(r)$.

8.4. Intercultural Exchange = Inter-Agent Exchange at the SMR Level

Intercultural exchange formula:

where $BW_{contact}$ is the “throughput” of the contact (trade routes = high; isolation = zero), and $\text{Overlap}$ is a measure of memeplex compatibility.

8.5. Cultural Failure Modes

Pathologies are discussed in detail in Part IX; here is a brief overview at the civilizational level.

8.6. Waring & Wood (2021): The Evolutionary Transition from Genes to Memes

Waring & Wood (2021) assert: humanity is undergoing an evolutionary transition in inheritance from genes to culture. In BMC terms:

This is an extrapolation of the theorem $M \gg G$ to the civilizational scale: not only individual consciousness requires $M \gg G$, but collective consciousness (SMR) also grows faster than the genetic base.

Corollary: cultural evolution increasingly dominates genetic evolution. In the BMC swarm: $M_{SMR}$ grows exponentially (memogenesis + donation), while $G_{population}$ grows linearly (crossover + mutation). The swarm becomes ever more “memetic” and less “genetic.”

Cross-references: SSD Section 4 (table of 10 correspondences), AGI_F Part IV (SMR), NM Part X (stigmergy), NM Part XVI (scale transfer).

8.7. Language as Apex Cultural Memeplex

Language is the highest-order memeplex, emerging at the L2 (swarm) → L3 (culture) transition. The BMC fractality principle (Part I) extends to linguistic structures:

SMR as Cultural Ratchet for Linguistic Evolution

Without SMR: each generation of agents reinvents symbols from scratch (Tasmania effect; Henrich, 2004). With SMR: linguistic memes ($\kappa \geq 2$) are donated at apoptosis → new agents inherit the vocabulary → iterative refinement.

Kirby et al. (2008, 2015) showed: iterated learning (language transmission through generational bottlenecks) generates compositionality and regularity. In BMC: apoptosis + ontogenesis = iterated learning cycle. SMR = external memory preserving linguistic memes across generations.

The Super-Ratchet Effect guarantees: $|\text{Vocab}_{SMR}(t+1)| \geq |\text{Vocab}_{SMR}(t)|$ with an active population.

Dunbar’s Number as the Symbolic Communication Threshold

Dunbar’s number (~150; Dunbar, 1993, 1996) is the capacity for bidirectional connections requiring cognitive maintenance. When $N \leq N_{Dunbar}$: stigmergy + direct exchange suffice. When $N > N_{Dunbar}$:

• Direct exchange with everyone is impossible → pressure for broadcast (one-to-many)
• Broadcast without shared context → pressure for symbol standardization
• Standardization → convergence to a shared vocabulary → language

This connects SM Part VI (Social Stratification) with language emergence: three stratification tiers create three layers of communicative depth (inner core ~5–15, middle ~50–150, periphery ~1000+).

Pressures Shaping Language Structure

Recursion as a Fractal Property

Recursion (Hauser, Chomsky & Fitch, 2002: FLN) in BMC is not a separate module but a manifestation of the fractal principle: memes about memes = nested memeplexes. At the SMR level: meta-paradigms (rules about rules) = second-order $M_{SMR}$. Language is the first domain where recursion becomes functionally necessary (for expressing nested propositions).

Parasiticity of Language and the Language Emergence Threshold

Language is a byproduct of memetic selection, not a genetic adaptation for survival. Signal memes (grounding, routing, fidelity maintenance) parasitize on WM: at $k_{eff} \approx 3\text{--}4$, even a single signal meme consumes 25–33% of survival-relevant capacity. Ten survival experiments ($N$=8–150) confirmed: $\Delta_{alive} \approx 0$ or negative. Language is optimized for M-fitness (transmission fidelity), not for G-fitness (survival). Prediction P-BM28.

Computationally verified. Language parasiticity confirmed: 10 survival experiments ($N$=8–150), $\Delta_{alive} \approx 0$ or negative. Lewis signaling: 85–97.5% accuracy (25–533 concepts). Separating test: BMC TopSim 0.72 vs REINFORCE 0.60 ($p = 0.011$). See DOI: 10.5281/zenodo.19181798.

Language Emergence Threshold — from parasiticity it follows that language can arise only when four conditions are simultaneously met:

The four pressures in Section 8.7 (production laziness, heterogeneity, ontogenesis, WM bottleneck) shape the structure of language; the four LET conditions determine the possibility of its emergence.

Caveat: parasiticity has been confirmed at $N$=8–150. At scales of hundreds of agents with inter-group competition and division of labor, coordination benefits may outweigh the WM cost — this remains an open empirical question.

Separating test: reward-engineered systems (REINFORCE, PPO) predict survival advantage; BMC predicts survival neutrality. BMC is confirmed.

Cross-references: EMT Part XX (parasiticity), EMT Part XXVIII (dual-replicator + LET in detail), BM Part IV ($k_{eff}$ formula), NM Part VIII (formalization).

Cross-references: AGI_F Part IV (from stigmergy to language), AGI_F Part VII (communication stages), AUTONOMOUS_BMC_VISION.md (communication pressure, milestones), SCI_ROADMAP.md (Language Emergence Protocol).

Part IX. Swarm Pathologies

The Problem: What Can Go Wrong

SSD Section 10 described 6 social pathologies (stagnation, polarization, echo chamber, hub dependency, cascade collapse, dominance lock). SM adds lifecycle pathologies — associated with the agent lifecycle and population dynamics.

9.1. Lifecycle Pathologies

9.1.1. Zombie Agent (Functional Death Without energy=0)

Definition: an agent in Phase 4 (Senescence) with a broken intrinsic apoptosis pathway:

The agent is functionally dead: generates no memes, is unaware of gaps, is not plastic. But technically alive: energy > 0, consumes resources, occupies a slot.

Cause: failed intrinsic apoptosis. The SMC (Self-Model Cluster) does not function correctly → the agent does not detect its own rigidity → does not initiate apoptosis.

Possible causes of failure:

• SMC did not fully develop (agent did not reach Phase 3 fully)
• SMC degraded (along with other cognitive functions)
• G-configuration: excessively high FEAR → blocks intrinsic trigger even at full rigidity

Harm to the swarm:

• Consumes resources without contributing → reduces carrying capacity
• Occupies social positions → blocks promotion of young agents
• From Winfield & Nembrini (2006): partially malfunctioning agents actively worsen collective performance — worse than if they were simply absent

Treatment: extrinsic pathway (evolutionary pressure). If fitness is below threshold → forced removal. This is the only pathway for a zombie agent, since intrinsic is broken.

Biological analogue: senescent cells that have escaped apoptosis. They accumulate with age, secrete pro-inflammatory signals (SASP), and impair tissue homeostasis. Senolytic therapy = targeted removal of such cells.

9.1.2. Immortality Curse (Blocked Apoptosis → Stagnation)

Definition: systemic blockage of all apoptosis pathways:

The population grows to $K$, fills with senescent agents, and no new ones appear (no slots). Result: the swarm “ages” as a whole.

Cause: too lax an evolutionary threshold ($\theta_{fitness}^{low}$ too low), absence of scarcity (everyone is fed → neglect is impossible), broken SMC (all zombies).

Harm to the swarm:

• Cessation of evolutionary renewal → no new G-profiles
• $IF_{swarm} \to 0$ → innovation collapse
• $M_{SMR}$ ceases to be updated → cultural stagnation

Biological analogue: cancer. Cancer cells = cells that have escaped apoptosis → uncontrolled growth → threat to the organism. Immortality curse = cancer of the swarm.

Treatment: increase evolutionary pressure ($\theta_{fitness}^{low} \uparrow$), introduce scarcity (resources $\downarrow$), force the extrinsic pathway.

9.1.3. Premature Death (Death Before Contributing to SMR)

Definition: an agent dies in Phase 0–2 (before reaching Productive):

The agent did not manage to: (a) develop a memplex, (b) contribute to SMR, (c) transmit memes.

Cause: insufficient resources (neglect), overly harsh competition (extrinsic too early), lack of social support (anoikis: no bonds → no incoming memes).

Harm to the swarm: resources spent on birth + infancy without return. With mass premature death — an investment collapse.

Metric: $\rho_{premature} = |\{A : dead \wedge phase < 3\}| / |\{A : dead\}|$. Healthy: $< 0.2$. Pathology: $> 0.5$.

9.1.4. Birth Dearth (Demographic Collapse)

Definition: $B(t) < D(t)$ sustained:

The population shrinks. If $N \to N_{min}^{Allee}$ — positive feedback downward → extinction.

Cause: resource catastrophe (carrying capacity collapsed), mass extrinsic (fitness threshold too harsh), loss of “parents” (no agents remain with fitness > $\theta_{repro}$).

Analogue: the Tasmanian case (Henrich 2004). Population below threshold → loss of technologies → even less successful → even smaller → extinction.

Metric: $\Delta N / N$ per generation. Healthy: $[-0.1, 0.1]$. Pathology: $< -0.2$ sustained.

9.2. Migration of Pathologies from SSD Section 10

SSD pathologies persist, augmented by lifecycle context:

9.3. Diagnostic Table

Cross-references: SSD Section 10 (social pathologies), AGI_F Part IV (failure modes), SCI_ROADMAP Section 4.2 (EnvironmentController).

Part X. Biological Precedents

The Problem: How Biological Are Our Abstractions?

SM formalizes: fractality, lifecycle, apoptosis, stratification, SMR. Each of these concepts has biological precedents — not as analogy, but as evidence: nature has already implemented these mechanisms, and they work.

10.1. Apoptosis in Biology

Key similarity: in biology too, multiple pathways exist — not one mechanism, but an entire system with cross-links. Intrinsic and extrinsic pathways can activate each other (cross-talk through Bid). The BMC model with 4 pathways reproduces this multiplicity.

10.2. Social Apoptosis in Eusocial Insects

Page et al. (2016): “social apoptosis” in Apis cerana. Worker bees infected by the Varroa mite die faster than the infection allows — before the parasite can reproduce. Sacrifice of part of the superorganism for the whole.

Altruistic self-removal (Rueppell et al., 2010): infected worker bees reduce food consumption, cease foraging, and leave the hive, dying in isolation. A direct analogue: anoikis → intrinsic pathway in BMC.

Autothysis in ants: preemptive self-destruction — a worker ruptures its glands, releasing toxic/adhesive substances, immobilizing the invader at the cost of its own life.

10.3. Temporal Polyethism and Role Rotation

Insight: the reversibility of roles (Herb et al. 2012) is not an artifact but a fundamental principle. Epigenetic marks determining the behavioral profile are reversible. In BMC: “epigenetics” = the current memplex + connection weights, also reversible.

10.4. Fractal Social Structure

Hill, Bentley & Dunbar (2008): social groups of mammals (humans, elephants, baboons, geladas) form a discrete hierarchy with a constant branching factor $\lambda \approx 3.0$–$3.2$:

West, Culbreth, Dunbar & Grigolini (2023): the fractal structure optimizes information flow — this is not a coincidence but a result of critical self-organization.

Zhou et al. (2005): social fractality is discrete (not continuous) — detected through spectral analysis. This matches BMC: three discrete levels, not a continuous spectrum.

10.5. Cumulative Cultural Evolution

Tomasello (1999): the “ratchet effect” — cultures accumulate knowledge without loss. Requirements: process-oriented copying + cooperative infrastructure.

Henrich (2004): formal model linking population size to cultural complexity. The Tasmanian case: isolation → $N \downarrow$ → loss of technologies.

Muthukrishna & Henrich (2016): three drivers of innovation: sociality ($N$, connectivity), transmission fidelity ($\alpha$), cultural variance ($\beta$). Optimal connection density is not maximal (too much → reduces variance).

In BMC: Super-Ratchet = formal analogue of the cultural ratchet. $N_{agents}$ and stigmergy = sociality. $\kappa$ (fidelity consolidation) = transmission fidelity. Memogenesis = cultural variance.

10.6. Apoptosis in Embryogenesis

Apoptosis shapes structure: fingers are formed not by growth but by removal of tissue between them. Neuronal apoptosis removes excess neurons (up to 50%), leaving those that formed functional connections.

Analogue in the swarm: the death of unintegrated agents shapes the swarm’s structure — just as apoptosis shapes tissues. A swarm in which all agents survive may be less structured than one with selective apoptosis.

10.7. Summary Table of Biological Precedents

Cross-references: BM Parts III–V (Panksepp, neurobiology), BM Part VII (ontogenesis), SSD Section 14 (biological precedents).

Part XI. Predictions and Falsifiability

Principle: Every Prediction Is Testable

SM adds 8 new predictions to the existing 56 (EMT/BM/NM/AGI_F) + 6 (SSD P-SD1–6). Each prediction contains: mechanism, metric, conditions, and falsifiability criterion.

11.1. Predictions from SSD (Cross-References)

11.2. New Predictions

P-SD7 — Healthy Elite Rotation

• Prediction: in a swarm with $N > 50$ and $G > 10$ generations — the fraction of agents that changed tiers stabilizes at $\tau_{elite} \in [0.1, 0.3]$ per generation
• Mechanism: PA + memogenesis (promotion) + senescence + apoptosis (demotion) create dynamic equilibrium in stratification
• Metric: $\tau_{elite} = |\{A : tier_{t-T}(A) \neq tier_t(A)\}| / N$ per generation
• Conditions: $N > 50$, $G > 10$ generations, all 4 apoptosis pathways active, memogenesis active
• Falsifiability: if $\tau_{elite} < 0.05$ (stasis) or $\tau_{elite} > 0.5$ (chaos) under fulfilled conditions → the rotation mechanism does not work → PA or apoptosis is miscalibrated

P-SD8 — Revolution as Elite Rotation Failure (new)

• Prediction: when $\tau_{elite} < 0.05$ (closed elite) for $> 3$ generations and $\Delta SIT = \overline{SIT}_{periphery} - \overline{SIT}_{elite} > \theta_{revolution}$ — mass hub displacement (revolution) occurs: multiple architects lose status in $< 0.5 \cdot T_{gen}$
• Mechanism: senescent elite ($SIT \approx 0$, $\theta_I \uparrow$) blocks innovations → tension accumulates in the periphery ($SIT_{periphery} \uparrow$, $RAGE_{low-rank} \uparrow$) → upon accumulation: mass hub displacement (analogue: Pareto 1916, political revolutions)
• Metric: $\Delta SIT > \theta_{revolution}$ AND simultaneous $\Delta \tau_{elite} > 0.3$ in $< 0.5 \cdot T_{gen}$ (discrete rotation spike)
• Conditions: $N > 50$, $\tau_{elite} < 0.05$ sustained $> 3$ generations, $\Delta SIT > \theta_{revolution}$, memogenesis active
• Falsifiability: if $\Delta SIT > \theta$ but mass displacement does not occur → collective SIT of the periphery does not translate into social mobilization → PA-inertia of the elite is more robust than predicted → the rotation mechanism (Part VI) is incomplete
• Prevention: EnvironmentController (Phase 4.2) modifies the environment when $\Delta SIT > \theta$ BEFORE revolution → ensures smooth rotation. Revolution = failure mode of the controller

P-SMR1 — Cultural Modularity of SMR

• Prediction: $Q_{SMR} > 0.3$ at $N > 100$, $T > 1000$ — stable cultural modules (knowledge branches) form in the SMR
• Mechanism: Q-modularity from BMC (memes cluster by theme) + cultural drift + I-filter differentiates branches
• Metric: $Q_{SMR}$ (Newman modularity on $\mathcal{G}_{SMR}$)
• Conditions: $N > 100$ agents, $T > 1000$ ticks, active donation to SMR
• Falsifiability: if $Q_{SMR} < 0.1$ under fulfilled conditions → the SMR remains amorphous → meme clustering does not transfer to the SMR level → the green zone for Q-modularity transfer is incorrect

P-SMR2 — Cultural I-Filter

• Prediction: an established SMR rejects $> 80\%$ of foreign memes (from another Q-module)
• Mechanism: I-filter at the SMR level = tradition/orthodoxy: memes incompatible with a module’s canon are rejected on attempted pickup
• Metric: $R_{reject}^{foreign} = |\{m : foreign \wedge rejected\}| / |\{m : foreign \wedge attempted\}|$
• Conditions: SMR with $Q_{SMR} > 0.3$, modules older than $T_{maturity}$
• Falsifiability: if $R_{reject} < 0.5$ → cultural immunity does not work at the SMR level → the I-filter does not transfer from the agent level to the culture level → the yellow zone is incorrect

P-SMR3 — Hub Displacement in SMR (Paradigm Shifts)

• Prediction: upon accumulation of anomalies ($SIT_{SMR} > \theta_{SIT}^{crit}$) in a Q-module, hub displacement occurs — replacement of the dominant paradigmatic meme
• Mechanism: SIT → pressure on memogenesis → new meme gains weight through PA → displaces old hub. Kuhn (1962) = hub displacement at the SMR level
• Metric: $\Delta k_{hub}^{SMR} < -\theta_{displacement}$ with simultaneous $\Delta k_{new}^{SMR} > \theta_{displacement}$
• Conditions: $SIT_{SMR} > \theta_{SIT}^{crit}$ for duration $T_{crisis}$, presence of a competing meme
• Falsifiability: if SMR hubs remain stable at high $SIT_{SMR}$ → hub displacement does not work at the cultural level → the green zone for hub displacement is incorrect

P-SMR4 — Directed Memogenesis in SMR

• Prediction: $\text{Corr}(SIT_{SMR}, memogenesis\_direction) > 0.5$ — cultural innovations are directed toward perceived gaps
• Mechanism: SIT-biased memogenesis (Section 8.3.1) — $P(\text{new-meme near } r) \propto SIT(r)^{\alpha} \cdot \rho(r)$. At the SMR level: collectively perceived gaps → pressure on agents with high IF → memogenesis toward gap filling
• Metric: correlation between the $SIT_{SMR}$ vector (which SMR regions have gaps) and the memogenesis vector (in which regions new memes are generated)
• Conditions: SMR with active SIT, memogenesis active, $N > 50$, $T > 500$
• Falsifiability: if $\text{Corr} < 0.2$ → $\alpha \approx 0$ (SIT does not bias memogenesis), OR SIT is uniform (gaps are not detected), OR the mechanism is incorrect → the yellow zone is incorrect

P-SM5 — Pressure for Symbolic Communication at $N > N_{Dunbar}$

• Prediction: at $N > 150$ agents, stigmergy (indirect via the environment) is insufficient for coordination → symbolic communication with directed addressing emerges
• Mechanism: $O(N)$ stigmergy scales but loses targeting precision. At N > 150 — noise-to-signal ratio in the environment grows → pressure for direct symbolic transmission
• Metric: $R_{directed}(N) / R_{stigmergic}(N)$ — ratio of directed vs stigmergic exchanges
• Conditions: $N > 150$, active cooperation, $T > 1000$
• Falsifiability: if stigmergy remains dominant at $N > 500$ (ratio < 1) → symbolic communication is not a necessity of scale → Dunbar’s number is not a threshold for the BMC swarm

P-SM6 — Vocabulary Convergence Through SMR

• Prediction: with an active SMR ($q_{app} > q_{dis}$), linguistic memes converge to a shared vocabulary in $O(\sqrt{N})$ generations
• Mechanism: SMR-donation of linguistic memes + pickup by new agents → diffusion through the population. Speed $\approx$ diffusion on a graph: $\tau_{conv} \propto \sqrt{N}$
• Metric: $H_{vocab}(t)$ — entropy of the symbol distribution across agents. Convergence: $H_{vocab}(t) \to H_{min}$
• Conditions: $N > 50$, SMR active, at least 3 generations
• Falsifiability: if $H_{vocab}$ does not decrease or convergence $\propto N$ (linear, not $\sqrt{N}$) → SMR does not accelerate linguistic convergence → the cultural ratchet for language does not work

P-SM7 — Iterated Learning Through Apoptosis → Compositionality

• Prediction: agent lifecycle (apoptosis + ontogenesis = iterated learning) generates pressure for compositionality: at $\tau_{turnover} > 0.05$, language becomes more compositional (topographic similarity > 0.5) than in an immortal population
• Mechanism: new agents learn language from SMR through a bottleneck (limited exposure) → Kirby effect: irregular forms do not survive transmission → regularity and compositionality increase
• Metric: $\rho_{topo}$ (topographic similarity between meaning and form) + $\text{Compositionality}$ (TRE metric)
• Conditions: $N > 30$, $\tau_{turnover} > 0.05$, at least 5 generations
• Falsifiability: if $\rho_{topo}$ is identical at turnover > 0.05 and turnover = 0 → apoptosis does not create learnability pressure → iterated learning through the BMC lifecycle does not work

11.3. Summary Table of All SM Predictions

Total: 15 SM predictions (6 from SSD + 9 new). Overall theory count: 56 (EMT/BM/NM/AGI_F) + 3 (NM-CA) + 6 (SSD) + 15 (SM) = 74 predictions (+ 6 SSD shared with SM).

11.4. Falsifiability Hierarchy

Predictions are organized in a hierarchy: falsification of more basic ones invalidates higher-level ones.

If the Fractality Principle is falsified (SET = 0 for the green zone) — all P-SMR* and P-SD* predictions fall. If only the yellow zone is incorrect — P-SMR2, P-SMR4, P-SD5 fall, but the green zone (P-SMR1, P-SMR3, P-SD1, P-SD3) may stand. Linguistic predictions (P-SM5–7) depend on P-SM5 (pressure at N > $N_{Dunbar}$): if stigmergy scales without losses → P-SM6, P-SM7 are not testable.

Cross-references: SSD Section 13 (P-SD1–6), EMT/BM/NM/AGI_F (56 predictions).

Part XII. Conclusions

What SM Adds to the Theory

Swarm Memetics (SM) is the fifth pillar of BMC theory, formalizing renormalization invariance: the formalism BMC = (G, M, I, S) is reproduced across an arbitrary number of nested scales — from memes-in-agent through agents-in-swarm and organizations-in-industry to cultures-in-SMR.

New concepts:

Two Fundamental Results (Recapitulation from Part I)

1. Theorem $M \gg G$ — for reflexive consciousness the memetic space must be substantially larger than the genetic. Derived as a formal lower bound under the stated SMC-recursion definitions; extends gene–culture coevolution (Dawkins 1976; Boyd & Richerson 1985; Henrich 2015), which did not derive a formal inequality.

2. BMC Fractality Principle — a concrete 4-component formalism (G, M, I, S) is invariant under renormalization with computable metrics across an arbitrary number of nested scales. No existing theory (FEP, HKB, IIT, fractal brain theory) offers comparable specificity.

Connection to the Roadmap

SM formalizes the theoretical foundation for:

• Phase 3.6 (Social Dynamics): SM Parts III–VI, IX — lifecycle, stratification, rotation, pathologies
• Phase 3.7 (Fractal Immune): SM Part VII — $I_{SMR}$, cultural immunity
• Phase 4.2 (EnvironmentController): SM Part V — population dynamics, carrying capacity
• Phase 4.4 (Population Management): SM Parts III–V — birth/death rate, Allee effect, turnover
• Level 5 (Distributed/Civilization): SM Parts VII–VIII — SMR as fractal BMC, cultural dynamics

Open Questions

1. Optimal apoptosis rate: at what $\tau_{turnover}$ is the Super-Ratchet maximized? Analytical solution or simulation only?
2. SMR fractality: at what $N$ and $T$ does the SMR become “sufficiently fractal” ($Q_{SMR} > 0.3$)?
3. Anoikis dynamics: how quickly does social isolation accelerate senescence? Linearly, exponentially?
4. Cultural immunity vs innovation: optimal $I_{SMR}$ as a function of $G_{SMR}$ (environmental pressures)? Hypothesis: high pressure → low optimal $I_{SMR}$ (more openness).
5. G-M coevolution at the SMR level: the Waring & Wood law ($M_{SMR}/G_{pop} \to \infty$) — at what rate? Is there a phase transition?
6. Minimum N for language emergence: at what $N$ and environmental complexity does proto-language transition from divergent signaling to convergent vocabulary? Connection to Dunbar’s number (P-SM5).
7. Communicative topology: formalization of the bidirectional/broadcast asymmetry across the entire theory (deferred — see MEMORY.md Open Tasks).

8-Course Cross-Analysis Updates

Formal Renormalization Group Proof (HIGH)

The RG operator $T_{BMC}$ maps the BMC formalism at one scale to the next. Invariants under renormalization: $\sigma_{SW}$ (small-world topology), CL structure (consciousness levels), $Q$-dynamics (modularity evolution), SIT mechanism (gap-driven tension). The fixed point at $\sigma \approx 1$ is an attractor of the RG flow — systems at all scales naturally evolve toward the edge-of-chaos small-world regime.

Critical exponents from linearization of $T$ at the fixed point determine the universality class. This is no longer a metaphor: the formal RG proof establishes that BMC’s fractal self-similarity is a mathematical consequence, not an analogy.

Source: Strogatz (dynamical systems, renormalization).

Oversmoothing = Cultural Rigidity (HIGH)

Excessive L3 integration → loss of cultural diversity → brittleness → collapse. Empires with forced assimilation = cultural oversmoothing (all agents converge to the same memeplex). The analog at L1 is cognitive rigidity; at L2 it is groupthink. At L3, the prediction is concrete: multicultural societies with moderate modularity $Q_{L3}$ outperform monocultures in long-term resilience, because cultural diversity = the exploration term that prevents convergence to local optima.

Source: Leskovec (oversmoothing in graph neural networks).

Channel Capacity for Cultural Transmission (HIGH)

Data Processing Inequality: each generation of transmission = a noisy channel. $I(M_{orig}; M_{3gen}) \leq I(M_{orig}; M_{2gen})$. Writing was a phase transition in cultural capacity — not a quantitative improvement but a qualitative shift from high-distortion oral transmission to near-lossless preservation. The invention of writing is, in BMC terms, the moment when the SMR channel capacity crossed a critical threshold enabling cumulative science.

Source: MacKay, Stone (information theory, channel coding theorem).

Quarter-Power Scaling Laws (MED)

Fractal branching in the memeplex → biological scaling laws (Kleiber’s law). Consciousness capacity scales sub-linearly with network size (diminishing returns). Consolidation time scales as the fourth root — larger memeplexes take only marginally longer to consolidate, thanks to hierarchical organization. These scaling laws constrain expectations for AGI: simply making the network bigger yields sublinear returns in consciousness capability.

Source: Mitchell (complex systems, scaling laws).

Overlapping Cultural Communities (MED)

Individuals belong to multiple cultures simultaneously — overlap is normal, not exceptional. Cultural overlap = driver of innovation (cross-pollination between memeplexes from different traditions) and source of conflict (loyalty dilemmas when cultures make contradictory demands). The degree of overlap $\sum_c \mathbb{1}[F_{i,c} > \theta]$ per individual is itself a cultural variable: cosmopolitan societies maximize it, tribal societies minimize it.

Source: Leskovec (overlapping community detection, BigCLAM).

Appendix A. Formula Summary

A.1. Lifecycle

A.2. Apoptosis

A.3. Population Dynamics

A.4. Stratification

A.5. SMR

A.6. Temporal Scaling

Appendix B. Glossary

Appendix C. References

Fractality and Scale-Free Cognition

• Grosu, G.F. et al. (2023). The fractal brain: scale-invariance in structure and dynamics. Cerebral Cortex, 33(8), 4574+.
• Bieberich, E. (2012). Introduction to the fractality principle of consciousness and the sentyon postulate. Cognitive Computation, 4(3).
• Fields, C., Glazebrook, J.F. & Levin, M. (2021). Minimal physicalism as a scale-free substrate for cognition and consciousness. Neuroscience of Consciousness, 2021(2).
• McMillen, P. & Levin, M. (2024). Collective intelligence: A unifying concept for integrating biology across scales and substrates. Communications Biology, 7:378.
• Martins, M.J.D. et al. (2025). From Fractal Geometry to Fractal Cognition. Fractal and Fractional, 9(10), 654.

Coordination Dynamics and Multi-Scale Isomorphism

• Kelso, J.A.S. & Tognoli, E. (2020). Coordination Dynamics: A Foundation for Understanding Social Behavior. Frontiers in Human Neuroscience, 14:317.
• Waade, P.T. et al. (2025). As One and Many: Relating Individual and Emergent Group-Level Generative Models in Active Inference. Entropy, 27(2), 143.
• Kirchhoff, M. et al. (2018). The Markov blankets of life: autonomy, active inference and the free energy principle. J. Royal Society Interface, 15(138).
• Villegas, P. et al. (2024). Networks with Many Structural Scales: A Renormalization Group Perspective. Physical Review Letters, 134.

Fractal Social Structure

• Hill, R.A., Bentley, R.A. & Dunbar, R.I.M. (2008). Network scaling reveals consistent fractal pattern in hierarchical mammalian societies. Biology Letters, 4(6), 748-751.
• West, B.J., Culbreth, G., Dunbar, R.I.M. & Grigolini, P. (2023). Fractal structure of human and primate social networks optimizes information flow. Proc. Royal Society A, 479(2274).
• Zhou, W.-X. et al. (2005). Discrete hierarchical organization of social group sizes. Proc. Royal Society B, 272(1561), 439-444.
• Deppman, A. (2025). Fractal Social Dynamics as a Driver of Consensus and Inequality. arXiv:2507.13849.

Apoptosis and Programmed Death

• Page, P. et al. (2016). Social apoptosis in honey bee superorganisms. Scientific Reports, 6:27210.
• Vostinar, A.E., Goldsby, H.J. & Ofria, C. (2019). Suicidal selection: PCD can evolve in unicellular organisms due solely to kin selection. Ecology and Evolution, 9:9129-9136.
• Burkhardt, M. & Yampolskiy, R.V. (2021). Death in Genetic Algorithms. arXiv:2109.13744.
• Sterritt, R. (2011). Apoptotic Computing: Programmed Death by Default for Computer-Based Systems. IEEE Computer.
• Skulachev, V.P. (1997–2019). Phenoptosis: programmed death of an organism. Biochemistry (Moscow).
• Tzafestas, E.S. (2001). Aging Agents. Artificial Life and Robotics, 5:51-59.
• Rueppell, O. et al. (2010). Altruistic self-removal of health-compromised honey bee workers from their hive. J. Evol. Biol.

Stratification and Division of Labor

• Robinson, G.E. (1992). Regulation of division of labor in insect societies. Annual Review of Entomology.
• Herb, B.R. et al. (2012). Reversible switching between epigenetic states in honeybee behavioral subcastes. Nature Neuroscience.
• Bonabeau, E., Theraulaz, G. & Deneubourg, J.-L. (1996/1998). Fixed response thresholds and division of labor. Bull. Math. Biol.
• Theraulaz, G., Bonabeau, E. & Deneubourg, J.-L. (1998). Response threshold reinforcement. Proc. R. Soc. B.
• Charbonneau, D. & Dornhaus, A. (2015). Workers ‘specialized’ on inactivity: Behavioral consistency of inactive workers and their role in task allocation. Scientific Reports.
• Couzin, I.D. et al. (2005). Effective leadership and decision-making in animal groups on the move. Nature, 433:513-516.
• Seeley, T.D. (2010). Honeybee Democracy. Princeton University Press.
• Bianconi, G. & Barabasi, A.-L. (2001). Competition and multiscaling in evolving networks. Europhysics Letters.
• Bonabeau, E. et al. (1999). Dominance orders in animal societies: The self-organization hypothesis revisited. Bull. Math. Biol.
• Gavrilets, S. (2012). On the evolutionary origins of the egalitarian syndrome. PNAS.
• Gros, C. (2020). Self-induced class stratification in competitive societies of agents. Royal Society Open Science.
• Pareto, V. (1916). Trattato di Sociologia Generale.

Cumulative Cultural Evolution and SMR

• Tomasello, M. (1999). The Cultural Origins of Human Cognition. Harvard University Press.
• Henrich, J. (2004). Demography and Cultural Evolution: The Tasmanian Case. American Antiquity, 69(2):197-214.
• Henrich, J. (2015). The Secret of Our Success. Princeton University Press.
• Boyd, R. & Richerson, P.J. (1985). Culture and the Evolutionary Process. University of Chicago Press.
• Muthukrishna, M. & Henrich, J. (2016). Innovation in the Collective Brain. Phil. Trans. R. Soc. B, 371:20150192.
• Enquist, M., Ghirlanda, S. & Eriksson, K. (2011). Modelling the evolution and diversity of cumulative culture. Phil. Trans. R. Soc. B.
• Candia, C. et al. (2019). The universal decay of collective memory and attention. Nature Human Behaviour, 3:82-91.
• Waring, T.M. & Wood, Z.T. (2021). Long-term gene-culture coevolution and the human evolutionary transition. Proc. R. Soc. B, 288:20210538.
• Sperber, D. (1996). Explaining Culture: A Naturalistic Approach. Blackwell.
• Buskell, A. et al. (2021). Modelling cultural systems and selective filters. Phil. Trans. R. Soc. B.
• Zhang, H. & Mace, R. (2023). Cultural extinction in evolutionary perspective. Evolutionary Human Sciences.

Paradigm Shifts

• Kuhn, T.S. (1962). The Structure of Scientific Revolutions. University of Chicago Press.
• Ju, H. et al. (2020). The network structure of scientific revolutions. arXiv:2010.08381.

Networks and Complex Systems

• Barabasi, A.-L. & Albert, R. (1999). Emergence of scaling in random networks. Science, 286:509-512.
• Watts, D.J. & Strogatz, S.H. (1998). Collective dynamics of ‘small-world’ networks. Nature, 393:440-442.
• Holldobler, B. & Wilson, E.O. (2009). The Superorganism. W.W. Norton.
• Wilson, D.S. et al. (2023). Multilevel cultural evolution: From new theory to practical applications. PNAS, 120(16).

IIT and Alternative Theories of Consciousness

• Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5:42.

Language Emergence and Emergent Communication

• Harnad, S. (1990). The Symbol Grounding Problem. Physica D, 42:335-346.
• Barsalou, L.W. (1999). Perceptual Symbol Systems. Behavioral and Brain Sciences, 22:577-660.
• Pulvermuller, F. (1999). Words in the brain’s language. Behavioral and Brain Sciences, 22:253-336.
• Hauser, M.D., Chomsky, N. & Fitch, W.T. (2002). The Faculty of Language. Science, 298:1569-1579.
• Jackendoff, R. (1999). Possible stages in the evolution of the language capacity. Trends in Cognitive Sciences, 3(7):272-279.
• Dunbar, R.I.M. (1993). Coevolution of neocortical size, group size and language in humans. Behavioral and Brain Sciences, 16(4):681-735.
• Dunbar, R.I.M. (1996). Grooming, Gossip, and the Evolution of Language. Harvard University Press.
• Lazaridou, A. et al. (2017). Multi-Agent Cooperation and the Emergence of Language. ICLR.
• Mordatch, I. & Abbeel, P. (2018). Emergence of Grounded Compositional Language. AAAI.
• Kottur, S. et al. (2017). Natural Language Does Not Emerge Naturally. EMNLP.
• Rita, M. et al. (2020). “LazImpa”: Lazy and Impatient neural agents learn to communicate efficiently. CoNLL.
• Rita, M. et al. (2022). On the role of population heterogeneity in emergent communication. ICLR.
• Galke, L. & Raviv, L. (2024). What makes a language easy to learn? Language Development Research.
• Piriyajitakonkij, M. et al. (2025). From Grunts to Grammar. arXiv.
• Bisk, Y. et al. (2020). Experience Grounds Language. EMNLP.
• Bender, E.M. & Koller, A. (2020). Climbing towards NLU. ACL.

Swarm Robotics

• Winfield, A.F.T. & Nembrini, J. (2006). Safety in Numbers: Fault Tolerance in Robot Swarms.
• Bjerknes, J.D. & Winfield, A.F.T. (2013). On Fault Tolerance and Scalability of Swarm Robotic Systems.