Biomemetics: Theoretical Foundations of Gene-Meme Interaction

Introduction

This document formalizes the Biomemetic Complex (BMC) — a system arising from the interaction of two replicators (genes and memes) within a single host. The goal is to move beyond descriptive metaphors toward a rigorous theoretical framework integrating evolutionary biology, neuroscience, and network science.

Conceptual foundation: For the memetic theory background, see Extended Meme Theory (EMT), Part XXVIII. For the network formalization of memes, see Network Memetics (NM). For AGI applications, see AGI Foundations.

Table of Contents

• Part I. Introduction: Why Biomemetics
• Part II. Evolutionary History: Coevolution of Two Replicators
• Part III. Neurobiological Substrate: The Hardware of BMC
• Part IV. Network Formalization: Genes as Special Nodes
• Part V. Interaction Mechanisms: Redirection, Suppression, Interpretation
• Part VI. Competition Dynamics: A Qualitative Model
• Part VII. Ontogeny: BMC Changes Across the Lifespan
• Part VIII. Cross-Cultural Variation: Different BMC Configurations
• Part IX. Pathologies: When BMC Breaks Down
• Part X. Epigenetics: Can Memes Influence Gene Expression?
• Part XI. Predictions and Falsifiability
• Part XII. Conclusions and Implications
• Appendix A. Formula Reference
• Appendix B. Glossary
• Appendix C. References
• Appendix D. Hypothetical Mathematical Formalization

Part I. Introduction: Why Biomemetics

The Problem: A Metaphorical Gap

The interaction of genes and memes has been described in the literature predominantly through metaphor:

• “Memes subjugate genes” — but how exactly?
• “Genes and memes compete” — in what units?
• “Consciousness arises from their interaction” — what is the mechanism?

Existing project documents touch on this topic:

• Extended Meme Theory (EMT), Part XXVIII — descriptively
• AGI Foundations — with an applied focus on AGI
• Network Memetics (NM) — formalizing memes only

The gap: There is no unified theoretical framework connecting biology (genes) with memetics (memes) at the level of formal models.

The Solution: The Biomemetic Complex

Biomemetics is an interdisciplinary theory integrating:

1. Evolutionary biology (genetic programs)
2. Neuroscience (the substrate of interaction)
3. Network science (formalization of dynamics)

Key idea: Consciousness is not a product of memes by themselves, but an emergent property of dynamic tension between two replicator systems. Understanding consciousness requires modeling both systems and their interface.

Definition of the Biomemetic Complex

The Biomemetic Complex (BMC) is a system consisting of four components:

where:

• $G$ — genetic layer: the set of genetic programs and drives
• $M$ — memetic layer: the network of memes (memeplex)
• $I$ — interface: mechanisms of interaction between layers
• $S$ — shared substrate: the neurobiological foundation

BMC Component Table

Why This Model Is Needed

See also: The concept of two replicators — EMT, Part XXVIII; architecture for AGI — AGI Foundations, Introduction.

Physical Structure of a Meme in the Brain

A meme is not an abstraction but a physical structure with hierarchical organization. Different parts of a meme are localized in different brain regions and have different resistance to degradation.

Neurobiological mechanism:

• Consolidation (systems): the meme core transfers from hippocampus to neocortex (months-years)
• Synaptic strengthening: frequent use -> LTP -> increased connection weight
• Synaptic weakening: disuse -> LTD -> peripheral degradation

Network formalization: The Fidelity function and storage modes — see NM, Part VIII.

Conceptual foundation: Meme structure: core and periphery — see EMT, Part I.

Part II. Evolutionary History: Coevolution of Two Replicators

The Problem: Why Is the Human Brain Anomalous?

The human brain is an evolutionary anomaly:

• 7 times larger than expected for a primate of our size
• Consumes 20% of the body’s energy (up to 60% in children)
• Contains functions superfluous for survival (music, abstract thought)

Standard evolutionary theory does not explain why evolution created such an “inefficient” organ.

Coevolution Timeline

The Baldwin Effect: Memes Accelerate Genetic Evolution

The Baldwin effect (Baldwin, 1896) is a mechanism whereby learning accelerates genetic evolution without violating Darwinian principles.

Current status (2025): Gene-culture coevolution remains an active research area. The work of Kasser (2025) expands the concept: genetic changes in humans are often driven by drift, founder effects, and gene flow, not only by selection. This means that cultural influence on biology is more complex than a pure Baldwin effect — culture affects migration, population isolation, and mutational conditions. See also: Nature Communications (2019) — review of gene-culture coevolution in animals; Royal Society B (2021) — longitudinal coevolution in humans.

Mechanism:

1. A meme creates an advantage for carriers of a certain genotype
2. Carriers of that genotype survive and reproduce more often
3. The genotype spreads through the population
4. The genetic basis for the meme becomes fixed

Evidence for Coevolution

Coevolutionary Fitness Formula

Total host fitness depends on both replicators and their synergy:

where:

• $W_g(t)$ — genetic fitness (survival, reproduction)
• $W_m(t)$ — memetic fitness (meme replication)
• $\alpha$ — synergy coefficient
• $synergy(g, m)$ — mutual reinforcement function

Synergy is positive when both replicators benefit:

• $synergy > 0$: childcare (genes transmitted + memes transmitted)
• $synergy < 0$: martyrdom (genes lose, memes win)
• $synergy \approx 0$: neutral scenarios

Table of Evolutionary Milestones

Numerical Example: Speed of Cultural vs. Genetic Evolution

Genetic evolution:

• Time to fixation of a beneficial mutation: ~1,000 generations ~ 25,000 years
• Speed: one significant change per 25,000 years

Memetic evolution:

• Time for meme propagation: from minutes to decades
• Speed: millions of changes per single human lifetime

Ratio: Memetic evolution is faster than genetic evolution by a factor of $10^6 - 10^9$.

Important clarification: This does not mean memes are “better.” Genes provide stability and basic functions, memes provide adaptability. BMC uses the advantages of both.

See also: Coevolution of genes and memes — EMT, Parts II, X.

Part III. Neurobiological Substrate: The Hardware of BMC

The Problem: Where Do Genes and Memes Physically Interact?

Memes are not abstractions but physical structures in the brain (synaptic connection patterns). Genetic programs are implemented through neural circuits. The question: which brain structures are responsible for which BMC layer?

Distribution Across Brain Structures

Neuroanatomical Distribution Table

Default Mode Network as SIT Substrate

The Default Mode Network (DMN) is a set of brain structures most active at rest (in the absence of external tasks) and deactivated during goal-directed activity. Key nodes:

Thesis: The DMN is the neurobiological substrate of SIT. When external tasks do not require SEEKING, the dopamine system switches to endogenous scanning of the memeplex, identifying clusters with $SIT > 0$. Subjectively, this is experienced as “mind-wandering,” rumination, or incubation.

Evidence base:

Connection to existing M-layer structures: The DMN overlaps with PFC and TPJ — structures already described as the substrate of the M-layer. This is not coincidental: SIT is a function of the memetic layer specifically (assessing memeplex structure), implemented on the same neuroanatomical resources. The DMN can be viewed as an interface between the M-layer and SEEKING: mPFC assesses gap relevance, angular gyrus determines their position in the semantic network, PCC correlates them with history.

Sources: Raichle et al. (2001). “A default mode of brain function.” PNAS, 98(2), 676-682; Andrews-Hanna (2012). “The brain’s default network and its adaptive role in internal mentation.” Annals of the NY Academy of Sciences, 1264(1), 1-13; Kounios & Beeman (2014). “The cognitive neuroscience of insight.” Annual Review of Psychology, 65, 71-93.

DMN as Substrate of Reflection and the Self-Model Cluster (SMC)

The function of the DMN is not limited to gap scanning (SIT). The DMN is the neurobiological substrate of reflection: SMC activity directed at the system’s own memeplex (see EMT, Part XVII).

Neural Substrate of SMC

The Self-Model Cluster (SMC) is a subgraph of the memeplex containing memes about the BMC system itself: $SMC = \{m \in M : target(m) \in M \cup G \cup I\}$ (see EMT, Part XVI). Neurobiologically, SMC is implemented at the intersection of DMN and M-layer structures:

DMN as Substrate of Reflection: The Mechanism

Reflection = SMC scans the M-layer, finding gaps, contradictions, G/M misalignments. Neurobiologically, this is realized through the DMN:

1. mPFC assesses gap relevance to the self-model: “does this concern me?”
2. PCC / precuneus contextualizes the gap in autobiographical history: “when did this begin?”
3. Angular gyrus determines the semantic position of the gap: “what is this connected to?”
4. Medial temporal lobe reactivates relevant episodes: “has something like this happened before?”

Result: LP (Learning Progress) — a signal of whether reflection is approaching closure. If LP > 0, productive reflection leads to a new meme (abduction). If LP ~ 0, rumination depletes $E_{available}$ (see EMT, Part XVII, Part IX of this document).

Quantitative consciousness assessment — the CL metric: The quality of SMC operation can be expressed as a single number: $CL(t) = \sigma_{SW}(t) \cdot A_{SMC}(t) \cdot f(Balance(t))$, where $\sigma_{SW}$ is network small-worldness, $A_{SMC}$ is Self-Model Cluster activity, and $f(Balance)$ is a bell-shaped function of G/M balance. CL unifies substrate (S), memetic (M), and interface (I) components of BMC into a single metric. Formalization details — NM, Part XIII.

M » G as a condition for reflection. Reflection is recursion of depth >= 2 within SMC: memes model their own memeplex. This requires a “surplus” of memes beyond the needs of modeling the world and the G-layer, i.e., $|V_m| \gg |V_u|$ (M » G). Neurobiologically, this is linked to the disproportionate development of PFC and associative zones (M-layer) relative to the limbic system (G-layer). The critical periods table below (Part IX) demonstrates this: in the “Sponge” phase (G » M) reflection is impossible; it appears only in the “Testing” phase (G ~ M, then M > G), when the memeplex first acquires sufficient capacity for self-modeling. Detailed formalization — EMT, Part XVI.

Proxy metrics for measuring CL in vivo: CL is directly computable only in a model, but each component has a measurable neurophysiological proxy. $\sigma_{SW}$ -> PCI (TMS-EEG, Casali et al., 2013): both measure the balance of integration/differentiation. $A_{SMC}$ -> DMN activity (fMRI): DMN is the neural substrate of the reflective cascade described above; L1 SMC (bodily self) maps onto posterior DMN (PCC, precuneus) + insula, L2 SMC (metacognition) onto anterior DMN (mPFC) + TPJ. $f(Balance)$ -> directed connectivity PFC->subcortical / subcortical->cortical (DCM). $I_{intero}$ (from the extended $CL_{full}$) -> HEP (Heartbeat Evoked Potential) + insula BOLD. A composite proxy index $CL_{proxy} = PCI \cdot DMN_{SMC} \cdot f(EC_{ratio})$ should predict subjective consciousness scales more accurately than any single component (detailed proxy table — NM, Part XII).

Evidence:

Active Inference Cascade: SIT -> SEEKING -> Motor Output

Reflection is not the only DMN function in the BMC context. When reflection discovers a gap and LP > 0 with high SIT, an active inference cascade is triggered — the memeplex motivates the host to change reality (see EMT, Part XVIII):

DMN (reflection) -> SIT > theta -> SEEKING (VTA/NA) -> PFC (planning) -> Motor cortex -> Action

Flow State: DMN Suppression and Optimal M/G Synchronization

• DMN suppressed during flow (Ulrich et al., 2016); $A_{SMC} \to$ minimum -> $CL_{reflexive}$ low but $CL_{operative}$ high
• SIT optimal range: $SIT_{bore} < SIT < SIT_{anxiety}$ -> progressive closure without overwhelm
• Neurochemical profile: transient hypofrontality (Dietrich, 2004); PLAY elevated, FEAR minimal; $\sigma \approx 1$ (criticality)
• Predictions: flow interruption -> sharp DMN reactivation (EEG alpha spike); high-PLAY baseline -> more frequent flow; flow correlates with power-law avalanche distribution

Cross-ref: NM Part XIII: CL operative vs reflexive; AGI Foundations: Flow as target mode.

Neurotransmitters as the Currency of Interaction

Three computational engines of BMC:

Physical Structure of a Meme: From Neuron to Cell Assembly

• Meme = cell assembly (Hebb 1949) = engram (Josselyn & Tonegawa 2020). Causal proof: optogenetic reactivation of specific neurons triggers memory (Liu et al. 2012)
• Population coding (Georgopoulos 1986): a meme is a distributed representation (vector in neural activation space)
• Neural reuse (Anderson 2010): neurons do not know what they encode; Broca’s area = speech + gestures + music
• Overlapping engrams (Cai et al. 2016): shared neurons between assemblies form the physical basis of edges ($w_{ij} \propto |ensemble_i \cap ensemble_j|$)
• Mixed selectivity (Rigotti et al. 2013): one neuron participates in ~5–20 assemblies -> combinatorial power
• Sparse coding: ~1–5% neurons per meme -> capacity ~$10^4$–$10^6$ memes

The Neuron as a Universal Computer

Dendritic computation: One biological neuron = 5–8 DNN layers (active dendrites: Na+, NMDA, Ca2+ channels). The NMDA spike is a coincidence detector (Hebbian at the molecular level). A single neuron can compute XOR (linearly inseparable). Implication: a cell assembly of ~$10^3$ neurons is computationally far more powerful than point-neuron models suggest.

Two learning rules: Pyramidal neurons use compartmentalized plasticity: (1) Basal dendrites — Hebbian (fire together, wire together), feedforward data; (2) Apical dendrites — non-Hebbian (local co-activation without spike), feedback/context/predictions. This is the anatomy of predictive coding: top-down predictions vs. bottom-up error correction.

Overlapping Ensembles: Why Connections Physically Exist

Overlapping engrams (Cai et al. 2016): shared neurons between assemblies = physical basis of edges ($w_{ij} \propto |ensemble_i \cap ensemble_j|$). Events occurring within a ~6 hour window share neurons (memory linking) -> causal association. In BMC: memes created within close $\Delta t$ acquire shared edges.

Mixed selectivity (Rigotti et al. 2013): one neuron participates in ~5–20 assemblies -> combinatorial power. This enables capacity scaling: not $N/s$ (linear), but $\binom{N/r}{s}$ (combinatorial).

Compressive Meme Coding

Sub-component reuse: memes are built from shared micro-ensembles (not atomic). A subset of size $r$ shared across $c$ memes yields capacity $\binom{N/r}{s}$ (combinatorial) vs. $N/s$ (linear). New meme cost: $O(1)$ new connections, not $O(s)$ neurons. Consequence: M » G requires sub-component reuse enabling a massive M without proportional G growth. Cross-ref: EMT XVI: M » G Theorem.

Neural Stigmergy: Coordination Through Traces in the Substrate

Cross-ref: NM Part X: Stigmergy; EMT Part XII.

Neurotransmitters as Global Modulators of the BMC Graph

Neurotransmitters as global modulators (Dayan 2012, volume transmission): DA = salience/learning, NE = attention gate, 5-HT = I-layer suppression, ACh = encoding mode, OT = in-group signal.

Fidelity substrate: Full = perforated synapses + CREB transcription; Skeletal = dynamic spines + early LTP; Trace = silent synapses (Isaac 1995) + spine persistence (Yang 2009).

Sleep sources: SHY (Tononi & Cirelli 2003/2014), triple coupling (Staresina 2015, Latchoumane 2017), spine formation (Yang 2014), systems consolidation (Diekelmann & Born 2010).

Intrinsic dynamics: Physical architecture creates “channels” — preferred activation paths. Some patterns are fundamentally unlearnable regardless of motivation (BCI experiments: monkeys cannot reverse neural pattern). Confirms: the G-layer (motivation) cannot override S-substrate constraints. Architecture > willpower.

G-factor = S quality: General intelligence G = metric of substrate S in BMC. High G = more efficient substrate: faster spreading activation, higher capacity, better pattern separation (neuroefficiency: less activation for same task). G is ~60–80% heritable (Bouchard 2004), stable across lifespan (r=0.72 at age 11 vs 80). Cognitive segregation: similar-G people form networks -> memeplex homophily. Assortative mating by IQ (4x stronger than by personality) = Baldwin Effect mechanism.

Hodgkin-Huxley -> sigmoid: The Hodgkin-Huxley model (4 coupled ODEs) justifies the sigmoid in spreading activation: threshold theta = resting->action potential transition; coupled feedback (V and gates) = activation and edge weights; refractory period = why memes cannot stay permanently active.

Theta rhythm: Hippocampal 4–12 Hz oscillation (medial septum pacemaker, HCN channels). Functions: (1) cross-modal synchronization — phase = common clock binding position, vision, emotion into a single snapshot; (2) temporal ordering — each cycle = chunk (past->present->future); theta sequences = episodic memory. Phase precession: spike timing encodes position in sequence. Without theta -> no recall, no planning.

SWR tag->replay: Two-stage memory selection: awake sharp-wave ripples (SWR) tag significant experiences at pauses -> sleep SWR replay and transfer to neocortex. Competition through inhibition: strongest patterns win. Temporal compression: seconds -> ~100ms (10x) fits STDP plasticity window.

Balance Activation Formula

Dynamics: $\frac{dA_m}{dt} = \alpha \cdot S_m(t) - \beta \cdot S_u(t) - \gamma \cdot fatigue(t)$

Stress -> Balance drops. Exhaustion -> G dominance. Recovery -> Balance restores.

Diagram: Flows Between Structures

Numerical Example: Stress and Balance

Situation: A person receives criticism at work.

Interpretation: The immediate reaction is genetic (status threat). Through the interface (ACC detects conflict), PFC control gradually recovers.

Important clarification: This is a simplified model. Real neurodynamic processes are considerably more complex and include feedback loops, nonlinearities, and individual differences.

See also: Neurobiological substrate in the AGI context — AGI Foundations, Part I.

Part IV. Network Formalization: Genes as Special Nodes

The Problem: How to Integrate Genes into the Network Model?

Network Memetics formalizes memes as graph nodes. But this model does not include genetic programs. An extension is needed.

Extending the Model: The BMC Graph

Definition: The BMC graph is a graph containing two types of nodes:

where:

• $V_g$ — utility nodes (genetic programs, fixed)
• $V_m$ — memetic nodes (memes, dynamic)
• $E_{gg}$ — connections between utility nodes (fixed)
• $E_{mm}$ — connections between memes (dynamic)
• $E_{gm}$ — connections between genes and memes (semi-dynamic)

Signed edges: All edge types have weights $w \in [-1, +1]$:

• $w > 0$ — excitatory connection
• $w < 0$ — inhibitory connection
• $w = 0$ — no connection

Comparison of Node Types

Utility Node Activation Formula

Unlike memes, utility nodes have a constant base activation:

where:

• $a_g^{base}$ — base drive level (always > 0)
• $w_{gi} \in [-1, +1]$ — weight of connection to stimulus $i$ (when $w_{gi} < 0$ the stimulus inhibits the utility node)
• $stimulus_i(t)$ — stimulus intensity at time $t$

Example: Hunger drive

Note (AGI implementation): In the AGI prototype, the base formula is augmented with inertia — $a_u(t+1) = \alpha \cdot a_u(t) + (1-\alpha) \cdot target(t)$, which smooths utility dynamics and models the “viscosity” of biological drives. Details: AGI Foundations, Part I.

Diagram: Bipartite BMC Graph

Table of Basic Utility Nodes

Numerical Example: Activation Cascade

Situation: A person sees a food advertisement.

Initial conditions:

• $a_{hunger}^{base} = 0.2$
• $time\_since\_meal = 3$ hours
• $stimulus_{smell} = 0.8$ (very appetizing ad)

Step 1: Activation of the “Hunger” utility node

Step 2: Propagation to connected memes

Step 3: Meme competition

The “Diet” meme activates the connected “Control” meme ($w = 0.7$), which inhibits the desire to eat.

The “Restaurant X” meme activates the “Pleasure” meme ($w = 0.8$), which reinforces the desire.

Result: The outcome depends on the relative strength of the memes and the current BMC balance.

Formalization of Interaction

Influence of a utility node on a meme:

where $(1 - a_m(t))$ is the “free capacity” of the meme (saturation).

Influence of a meme on a utility node:

The direction of influence is now determined by the sign of the weight $w_{mg} \in [-1, +1]$: a positive weight reinforces the utility node, a negative one suppresses it. A separate sign(compatibility) parameter is no longer needed.

Example: DISGUST as Interpretation: the meme “this is disgusting” has $w_{mg} < 0$ toward the corresponding stimulus, which suppresses its activation in the memeplex.

See also: Network model of memes — NM, Part V; utility nodes in AGI — AGI Foundations, Part I.

Differentiated Storage: Ranked Completeness

Problem: The brain stores thousands of memes, but its capacity is limited. How does the system cope?

Solution: Memes are stored with different completeness (Fidelity) depending on their rank in the network and frequency of use.

Fidelity Function

Orthogonality of Fidelity and Weight

Fidelity (storage completeness) and weight (sign of relationship) are independent characteristics of a meme. A meme can be well-studied yet rejected:

Antibody = a meme with high Fidelity and negative weight. The memeplex stores a detailed threat model for quick recognition and blocking. Degree $k_m$ in the Fidelity formula counts all meme connections — both positive and negative.

Three Storage Modes

Neurobiological Substrate of Fidelity

The three meme storage modes correspond to specific molecular and structural synaptic states:

Silent synapses (Isaac et al., 1995, Neuron): synapses containing only NMDA receptors (without AMPA). They do not transmit signals during normal activation but preserve a structural trace — the basis for rapid reactivation.

Spine persistence (Yang et al., 2009, Nature): dendritic spines formed during learning persist for months even without repeated activation. This is the physical basis of trace storage: the meme is “forgotten” functionally, but its structure is not erased.

Reactivation and savings effect (Ebbinghaus, 1885): relearning a forgotten skill occurs significantly faster than learning de novo — because silent synapses and dendritic tags are not erased. Savings = the difference between initial learning and relearning.

Consolidation timescales:

Adaptive Significance

Key Example: Foreign Language

Scenario: 5 years of study -> 10 years without practice -> one week in the language environment

Why “remembered” (days) rather than “learned anew” (years)?

The language core (grammatical structures, basic vocabulary) is already consolidated in the cortex. Reactivation merely strengthens weakened synapses — faster than creating new ones.

Reactivation Dynamics

where:

• $\rho$ — reactivation rate (0.1–0.5 day$^{-1}$)
• $exposure(t)$ — exposure intensity (0–1)

Property: Initial reactivation is fast (at low Fidelity there is much “free capacity”), then slows down.

Model Predictions

Network formalization: Full Fidelity formula and numerical examples — see NM, Part VIII.

AGI application: Skeletal storage in artificial systems — see AGI Foundations, Part III.

Practical consequence: Mnemonics as a way to increase meme connectivity. More connections -> slower decay -> higher Fidelity. See NM: Associative memory vs. isolated memorization.

Multilevel Memory: Neurobiological Basis of $\kappa$

The mechanisms of edge decay, Fidelity, SIT, and continuous activation $a_i$ give rise to multilevel memory as an emergent property — without a separate module. Consolidation level $\kappa_i(t) \in \{0, 1, 2\}$ is a derived discrete parameter classifying the depth of a meme’s inscription in the neural substrate.

Molecular Markers of $\kappa$-Levels

Neurobiology of $\kappa$-Transitions

$\kappa: 0 \to 1$ (sensory -> STM). The meme passes the I-filter (ACC / insula assess compatibility) -> hippocampal encoding. Mechanism: glutamate -> NMDA receptors -> Ca2+ influx -> CaMKII -> AMPA trafficking -> E-LTP. Speed: seconds. Pattern separation in the dentate gyrus (DG) ensures the uniqueness of each new meme even when similar to old ones.

$\kappa: 1 \to 2$ (STM -> LTM). Three competing pathways:

1. Spaced repetition ($n_{react} \geq N_{crit}$): repeated co-activation -> late LTP -> CREB-dependent transcription -> BDNF -> growth of new spines -> structural LTP. Typical threshold: 3–5 reactivations with intervals >= hours.

2. Emotional tag ($G_{align} > \theta_G$): amygdala (BLA) -> norepinephrine/cortisol -> enhanced hippocampal LTP; awake SWR tagging (Buzsaki, 2024, Science) marks the meme for overnight consolidation. Accelerated pathway: 1–2 exposures sufficient.

3. Hub status ($Fidelity \geq F_{LTM}$): a meme with high degree ($k_m$) -> many synaptic contacts -> automatic strengthening via heterosynaptic metaplasticity (Abraham, 2008). Hubs consolidate “by themselves” — through constant reinforcement from neighbors.

$\kappa: 2 \to 1$ (deconsolidation). Prolonged disuse -> spine retraction (Yang et al., 2009): dendritic spines shrink, AMPA receptors are internalized. The structure (NMDA-only silent synapses) is preserved -> savings effect: reactivation 10–100x faster than new learning. The meme transitions from Full -> Skeletal -> Trace mode.

$\kappa: 1 \to \varnothing$ (pruning). Two mechanisms:

• Passive: Synaptic downscaling (SHY, Tononi & Cirelli, 2014) during sleep: proportional weakening -> weak connections erased
• Active: Microglial pruning (complement-tagging: C1q/C3 -> phagocytosis of inactive synapses; Schafer et al., 2012, Neuron)

Engram Allocation: Competition for Consolidation

Not all STM memes transition to LTM — neurons compete for inclusion in the engram:

• CREB-dependent excitability (Josselyn & Frankland, 2015): neurons with high CREB -> increased excitability -> “win” allocation. In BMC: memes with high $C_E$ (eigenvector centrality) -> more frequently co-activated -> priority.
• Memory linking (Cai et al., 2016, Nature): events within a ~6-hour window share neurons (overlapping engrams) -> causal association. In BMC: memes created within close $\Delta t$ acquire shared edges.
• Stress -> engram expansion: cortisol increases the number of neurons in the engram -> generalized memories -> loss of specificity (PTSD). In BMC: high $E_{emot}$ -> multiple connections -> over-consolidation with loss of precision.

CLS: Two Storage Systems

Complementary Learning Systems (McClelland et al., 1995; O’Reilly et al., 2014):

Transfer $\kappa: 1 \to 2$ = active systems consolidation: SWR replay -> triple coupling (SO -> spindles -> SWR) -> cortical recording. More details — in the next section.

Schema-Congruence (SLIMM): I-Compatibility Determines Speed

Schema-Linked Interactions between Medial and Memory systems (van Kesteren et al., 2012):

In BMC this U-shaped curve arises naturally: I-compatible memes integrate quickly (many attachment points); radically new ones receive an emotional boost through SEEKING/G-alignment. Intermediate ones (moderately unfamiliar, without emotion) are the most vulnerable.

Formalization: Definition of $\kappa_i(t)$, transition conditions, formulas — see NM, Part VIII.

Conceptual framework: BMC -> memory type mapping, predictions — see EMT, Part VIII.

Activity-Silent WM: The Synaptic Mechanism

The classical WM model (Fuster, 1973; Goldman-Rakic, 1995) explains information retention through sustained firing of PFC neurons — Active WM. However, evidence shows that WM also stores information without elevated activity — through synaptic short-term plasticity (STP) (Mongillo, Barak & Tsodyks, 2008, Science; Stokes, 2015, Trends in Cognitive Sciences).

STP mechanism: After firing cessation, elevated Ca2+ concentration remains in the presynaptic terminal (tens of seconds). The next spike causes facilitation — enhanced neurotransmitter release. Information is encoded not in neuron activity but in the pattern of potentiated synapses. This is the neurobiological substrate of $\psi_i$ in BMC: a synaptic trace existing when $a_i \approx 0$.

Pinging — experimental evidence: A TMS pulse (Rose et al., 2016, Journal of Cognitive Neuroscience) or orienting stimulus (Wolff et al., 2017, Nature Neuroscience) reactivates a “forgotten” WM item that was not decodable from neural activity before stimulation. In BMC: external signal (spreading activation) x synaptic trace ($\psi$) -> reactivation.

Panichello & Buschman (2021), Nature: WM representations in PFC are “morphed” — partially recoded during attention switching, but recovered upon re-attention. This is consistent with the model: the $\psi$-trace preserves the meme’s “address” but not its full activation; reactivation requires repeated spreading activation through preserved synaptic weights.

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

Episodic Memory: Neurobiological Substrate

Episodic Memory Barcodes

The hippocampus generates sparse random activation patterns unique to each event — barcodes (Chettih et al., 2024, Cell). Key properties:

Generation mechanism: The dentate gyrus (DG) performs pattern separation — transforming similar inputs into orthogonal representations through sparse activation of granule cells (~2–6% simultaneously active). DG -> CA3 (pattern completion via recurrent connections) -> CA1 (output + temporal ordering). The barcode is formed at the DG->CA3 junction: a random sparse sample from granule cells, fixed via rapid E-LTP.

In BMC terms: the barcode = a sparse subset of memes co-activated at episode onset. It does not encode content — it serves as a pointer through which spreading activation recovers the full set of episode memes.

Time Cells: Temporal Order Within an Episode

In CA1, time cells have been discovered — neurons that fire at a specific moment relative to episode onset (MacDonald et al., 2011, Neuron; Eichenbaum, 2014, Nature Reviews Neuroscience). Time cells create a temporal scaffold for the episode:

Time cells provide a temporal scaffold: within an episode, memes are not merely co-active — they are ordered. During recall, order is recovered through sequential reactivation of time cells. Without time cells — recall without order (semantic memory: “I know that X” but not “first X, then Y”).

Connection to theta rhythm: Theta-binding (~125 ms, see above) synchronizes components of a single moment. Time cells operate on a larger scale — ordering moments within an episode. Phase precession bridges these scales: spike order within a theta cycle reflects trajectory order (Dragoi & Buzsaki, 2006).

Event Boundary Neurons: Where a New Episode Begins

Episode boundaries are not arbitrary — they are detected by specialized neural circuits:

• Hippocampal boundary signals: CA1 activity changes sharply upon context switch — even without sensory changes (Baldassano et al., 2017, Neuron). The CA1 pattern “resets” — the old barcode deactivates, a new one is generated
• mPFC event model: The medial prefrontal cortex maintains a “current event model.” Upon prediction error, the model updates -> signal to hippocampus -> boundary (Zacks et al., 2007, Psychological Bulletin)
• Dopaminergic gating: Ventral tegmental area (VTA) -> hippocampus: a dopamine signal upon surprise marks the boundary and enhances encoding of the new episode (Shohamy & Adcock, 2010, Trends in Cognitive Sciences)

In BMC terms: Event boundary = the moment when prediction error ($\Delta_{PE}$) or a change in G-modulation exceeds a threshold. Neural substrate: mPFC monitors the event model; upon mismatch -> DA signal -> hippocampus resets the barcode.

Hippocampal Indexing Theory

The hippocampus stores not the memories themselves but indices (Teyler & DiScenna, 1986, Behavioral Neuroscience):

Hippocampus (index):  B_k -> [pointer_visual, pointer_auditory, pointer_emotional, ...]
Cortex (content):     visual_area: scene details
                      auditory_area: sounds
                      amygdala: emotional valence

Retrieval: cue -> hippocampus reactivates $B_k$ -> via index, cortical patterns are recovered -> full episode. This explains retrograde amnesia with hippocampal damage: content in cortex is intact, but indices are lost -> no access.

In BMC terms: Barcode = Teyler-DiScenna index. Memes ($\kappa = 2$, semantic) are stored in the graph. Barcodes ($\kappa = 1$, episodic) are temporary pointers to meme configurations. Trace transformation = index loss while content is preserved.

Barcode Lifecycle: Neurobiology

Formalization: Definition of $\varepsilon_k$, event boundary detection, temporal chaining — see NM, Part VIII.

Conceptual framework: Episodes as discrete units of experience — see EMT, Part VIII.

Sleep as a Meme Consolidation Mechanism

The Problem: The Working Memory Buffer

The hippocampus functions as a working memory buffer with limited capacity. During the day, new memes accumulate in this buffer — both received from outside and internally synthesized. The buffer overflows -> unloading is necessary.

Solution: Sleep is the process of transfer from the buffer (hippocampus) to long-term storage (neocortex).

Active Systems Consolidation: The Transfer Mechanism

Triple coupling — temporal coordination of three oscillations (Staresina et al., 2015, Nature Neuroscience — hierarchical nesting SO->spindles->ripples in humans; Latchoumane et al., 2017, Neuron — causal proof through optogenetic spindle induction):

Two-stage SWR system — tag -> replay: Consolidation begins not during sleep but during wakefulness. Awake SWRs occur during activity pauses (stops, task completion) and mark significant episodes for subsequent consolidation — this is tagging (Jadhav et al., 2012, Science; Fernandez-Ruiz et al., 2019, Science). In terms of episodic memory (see above): awake SWR tags the barcode $B_k$ of an episode; sleep SWR reactivates $B_k$ -> through pattern completion the full $M_k$ is recovered -> DECOMPOSE. Compression ~10x (seconds -> ~100 ms) falls exactly within the STDP plasticity window of the neocortex (Buzsaki, 2015, Neuron). Competition through inhibition creates narrow windows — the strongest patterns (with highest centrality) win, explaining differential decay: $\lambda_{ij} = \lambda_0 / (1 + \alpha \cdot C(i) \cdot C(j))$.

Theta rhythm as a substrate clock: In addition to overnight triple coupling, a key role in meme formation is played by the theta rhythm (4–12 Hz) — the hippocampus’s internal clock, generated by the medial septum through HCN channels (Buzsaki, 2002, Neuron). Theta performs two functions: (1) modality synchronization — the phase of the theta cycle = a common clock for binding meme components (visual + emotional + contextual -> one cell assembly); (2) ordering — each theta cycle = one chunk (past->present->future), theta sequences are stitched into an episodic chain (Dragoi & Buzsaki, 2006, Neuron). Phase precession provides a temporal code: spike order within a cycle = meme order on a trajectory. Without theta rhythm, internal sequences (recall, planning, SIT-driven navigation) do not arise — this is a fundamental S-constraint on memeplex dynamics.

Connection to triple binding: Theta-mediated synchronization is the neural substrate of temporal binding: memes activated within a single theta window (~125 ms) are bound into a unified cluster of consciousness. Along with structural binding (ensemble overlap, see below) and competitive binding (WM competition — only a coherent coalition is admitted), three binding types are formalized in NM, Part XIV.

Consolidation probability formula:

where:

• $\alpha_{SO}, \alpha_{spindle}, \alpha_{SWR}$ — amplitudes of corresponding oscillations (normalized, 0–1)
• $E_{emotional}$ — emotional intensity of the experience (0–1)
• $\gamma$ — emotional enhancement coefficient (~1.0–2.0)

Decomposition, Binding, and Recombination During Sleep

Model: Integral experience -> decomposition -> binding -> recombination (BLEND) -> pruning

The consolidation cycle includes five steps:

FOR each consolidation cycle:
  1. DECOMPOSE: decompose the experience into components (SWS, sharp-wave ripples)
  2. CONNECT: bind components to existing categories (SWS -> neocortex)
  3. BLEND: recombine components from different clusters (REM)  <-- NEW
  4. PRUNE: remove weak edges (synaptic homeostasis)
  5. STRENGTHEN: reinforce frequent edges (repeated replay)

The BLEND step is the key addition: during REM sleep, overlapping replay of related memories reinforces shared elements, forming cognitive schemas (Lewis & Durrant, 2011), creating conditions for meme recombination across clusters. Sleep restructures memory representations, facilitating “morning-after” insights (Wagner et al., 2004).

Neurobiological Basis of Sleep Processes

Synaptic Homeostasis Hypothesis (SHY) (Tononi & Cirelli, 2003, Brain Research Bulletin; Tononi & Cirelli, 2014, Neuron): wakefulness = net synaptic potentiation (learning strengthens synapses). Sleep = net synaptic downscaling (proportional weakening of all synapses). Result: strong pathways survive, weak ones are erased — this is the physical basis of PRUNE in our model.

Spine formation during sleep (Yang et al., 2014, Science): motor skill learning -> sleep -> growth of new dendritic spines on specific dendritic branches (not random). REM sleep deprivation blocks this process. This is the physical basis of STRENGTHEN.

Active systems consolidation (Diekelmann & Born, 2010, Nature Reviews Neuroscience): SWS provides systems consolidation (hippocampus -> neocortex transfer), REM provides synaptic consolidation (stabilization of transferred traces). The dual mechanism explains why both sleep phases are necessary.

Correspondence of sleep phases and operations on the BMC graph:

Trace Transformation: The Synaptic Mechanism

Repeated replay transforms the engram at the molecular level — separating the fate of core and peripheral synapses:

• Core synapses (high co-activation, connection to G-drives): replay -> repeated LTP -> CREB-dependent transcription -> BDNF -> new spines (Yang et al., 2014) -> structural LTP. Result: neocortical representation strengthened ($\kappa: 1 \to 2$).

• Peripheral synapses (weak co-activation, contextual details): SHY downscaling (Tononi & Cirelli, 2014) -> proportional weakening -> below survival threshold -> microglial pruning (C1q/C3-tagging). Result: detail lost.

• Barcode ($B_k$ in DG/CA3): maintained as long as replay continues. When core memes are integrated in the neocortex and replay ceases, neurogenesis in DG creates new neurons that integrate into existing circuits and degrade old barcode sparseness through interference — not direct replacement but pattern rewriting (Akers et al., 2014, Science; blocking neurogenesis slows forgetting) -> $B_k$ loses distinguishability -> retrieval via barcode impossible -> memory = gist without the episodic “wrapper.”

Molecular timescale of trace transformation:

Formalization: Fidelity_core(n), Fidelity_periph(n), N_replay^req, timeline — see NM, Part VIII.

Example: First visit to a castle

Result: A new meme group “medieval castles” is created, connected to all the above categories. The first visited castle “colonizes” this group.

New connection weight formula:

where:

• $w_{base}$ — base weight (~0.3)
• $E_{emotional}$ — emotional intensity (0–1)
• $N_{replay}$ — number of overnight reactivations (typically 5–20)
• $\gamma$ — emotional enhancement coefficient (~1.0–2.0)

Emotional Enhancement of Connections

Empirical base:

• Wagner et al. (2001): emotional memories are enhanced by sleep
• Payne et al. (2008): REM sleep is critical for emotional memory

Modified edge decay:

Emotionally colored connections (biomemetic in nature) decay more slowly.

Dreams: REM as Stochastic BLEND With Weakened I-Layer

Consolidation during sleep = optimization of the M-layer. But dreams are a separate phenomenon requiring explanation: why dream content is bizarre and why it is not random.

Neurochemical basis of I-layer weakening:

Result: The M-layer actively recombines (BLEND), but without I-filtration. Meme combinations that are rejected during the day by the immune system as incompatible freely form temporary clusters -> bizarre dreams.

Dream content via SIT:

SIT-gaps (unresolved questions from the day) maintain high activation through the night. During BLEND they function as attractors for stochastic recombination: the M-layer tries to close gaps by testing combinations without I-constraints. This explains why we often dream about themes that concern us.

Functional consequences:

Compatible data:

• Voss et al. (2009): lucid dreamers show elevated ACC and dlPFC activity — structures interpreted as the I-layer in BMC
• Walker & Stickgold (2004): REM deprivation reduces creativity — compatible with loss of BLEND recombination
• Nir & Tononi (2010): neural activity in REM resembles wakefulness in strength but not in pattern — compatible with a stochastic M-mode

Conceptual connection: Dreams as a BMC mechanism — see EMT; offline simulation in AGI — see AGI Foundations.

Sleep Quality’s Effect on Fidelity

Extended formula:

where $S_{quality}$ — sleep quality (0–1):

Typical values: $w_{SWS} = 0.6$, $w_{REM} = 0.4$, $t_{SWS}^{norm} \approx 100$ min, $t_{REM}^{norm} \approx 90$ min.

Numerical Example: Night After Visiting a Castle

Summary:

• ~40 new connections created
• Emotional connections: $w \approx 0.7$ (high)
• Neutral connections: $w \approx 0.4$ (medium)
• After one year without repetition: emotional connections will persist, neutral ones will degrade

Model Predictions

Network formalization: Overnight network optimization and sleep’s effect on edge decay — see NM: Overnight network optimization.

Operationalization for native BMC: Sleep cycle with DECOMPOSE -> CONNECT -> BLEND -> PRUNE -> STRENGTHEN -> REPLAY phases, including S-input shutdown and modulator regimes of the three engines (AGI Foundations, Part VII).

Active Forgetting: Molecular Mechanisms

Passive decay (SHY — Tononi & Cirelli, 2014) is not the only forgetting mechanism. Neurobiology identifies at least 6 molecular pathways of active forgetting:

BMC unifies all 6 mechanisms along two axes: passive (SHY, neurogenesis-interference — background process without intentionality) vs. active (GABA inhibition, Rac1 cascade — the I-system triggers suppression of specific memes). Key discovery by Yi Zhong: active forgetting is not a side effect but a separate biochemical process, regulated independently of learning.

Reconsolidation: The Plasticity Window

Nader et al. (2000) — the seminal experiment: anisomycin injection (protein synthesis blocker) after reactivation of consolidated memory -> memory loss. Conclusion: recall places an LTM meme into a labile state requiring renewed protein synthesis (re-stabilization).

PE-dependency. Pedreira et al. (2004): recall without prediction error does not trigger lability. Reconsolidation is triggered only by expectation-reality mismatch — “routine” recall is safe.

Boundary conditions. Suzuki et al. (2004): old and strong memories are more resistant to reconsolidation — they require greater PE or longer reactivation. Mechanism: multiple replay cycles create a distributed engram with redundant connections that is not fully destabilized by a single recall.

Therapeutic application. Schiller et al. (2010): reconsolidation update protocol — extinction within the 6-hour lability window updates fear memory without erasure. In BMC terms: recall + moderate PE -> update ($w_{ij}$ are modified, $\kappa$ preserved), not erase ($F_i \to F_{trace}$).

Network formalization: $I_{sig}$, $Labile(m_i, t)$, RIF, boundary conditions, 4 predictions — see NM: Active forgetting and reconsolidation.

Automatization: Neurobiological Substrate

The transition from deliberative (WM-dependent) to automatic (WM-independent) execution of behavioral sequences has a clear neurobiological correlate: a shift of control from dorsomedial striatum (DMS) to dorsolateral striatum (DLS).

Grillner et al. (2025, PNAS): DLS-SNr-PF-DLS loop as a “habit repository” — a closed loop through basal ganglia and thalamus requiring no cortical control. Key data: (a) cortex is dispensable for expression of already automatized sequences (motor cortex lesion does not disrupt habits); (b) ZIP injection (PKMzeta inhibitor) into DLS -> habit loss with preserved deliberative skill; (c) plasticity is localized in thalamostriatal (PF->DLS) synapses, not corticostriatal.

Turner et al. (2022, J Neurosci): DMS and DLS functionally compete — DMS engagement early in learning slows transition to DLS-dominant mode (the deliberative system “does not release” control). DMS damage accelerates automatization — without a competitor, DLS “takes over” earlier. This explains why explicit verbalization (deliberation) slows motor learning (Masters, 1992, British Journal of Psychology).

Chunking is not automatization. These are orthogonal processes with dissociable neural substrates: chunking = uniting elements into a unit (DMS-mediated, hippocampal pre-ordering; Soni & Frank, 2025, eLife — chunking as a WM optimization strategy through PFC-BG gating); automatization = transfer of control to DLS (habit-learning). One can chunk without automatizing (a new chunk still requires a WM pointer) and automatize without explicit chunking (a long chain with gradual habit growth).

Yewbrey & Kornysheva (2024, J Neurosci): The hippocampus pre-orders sequences (element order) — hippocampal activity during planning predicts the order of upcoming movements. Subcortical structures (striatum, thalamus, cerebellum) are activated during execution but do not encode sequence identity, suggesting complementary roles: hippocampus — “what and in what order,” striatum — executive control.

Solano et al. (2024, J Neurosci): Sleep within ~1h of sensorimotor training -> ~30% enhancement of motor consolidation. Mechanism: spindle-SO coupling (contralateral cortex) enhances motor trace consolidation. Difference from declarative consolidation: declarative -> SWR-driven (hippocampus->cortex); motor -> spindle-SO coupling. Critical window ~1h ($T_{crit}$). In the BMC context: an analogous spindle-SO mechanism presumably strengthens edges in automatic chains (Auto(S)).

De-automatization (override). vlPFC (ventrolateral prefrontal cortex) as a “valve” — inhibits DLS output and reactivates the DMS pathway (Lakshminarasimhan et al., 2024; Coutureau & Killcross, 2003, Behavioral Brain Research). Override cost is nonlinear: $Cost_{override} \propto habit^2$ — deeply automatized sequences require powerful prefrontal intervention. Under stress (WM overload) prefrontal resources are depleted -> DLS “takes over” -> relapse.

H. floresiensis: neuroanatomical constraints. Brain ~420 cc -> proportional reduction of PFC and striatum. Falk et al. (2005, Science): endocast LB1 shows extensive reorganization (expanded temporal lobe, developed Brodmann area 10), but reduced absolute volume. In BMC terms: sufficient for basic G x M integration (genetic programs + simple memes), but insufficient for complex automatization — limited DLS loop capacity + limited WM pointer system -> cognitive ceiling. Oldowan tool level = maximum achievable with this substrate.

Network formalization: $Auto(S)$, $habit$, $wm\_cost$, $P_{auto}$, $Cost_{override}$, predictions P-A1–P-A5 — see NM: Automatization.

Character (Temperament): Individual Weights of Emotional Systems

The Problem: Why Does One Event Elicit Different Reactions?

Two people visit the same castle. One remembers it as “a creepy place,” the other as “interesting to explore.” Given identical input experiences, different memes are formed.

Hypothesis: The difference is caused by individual weights of emotional systems — character (temperament).

Temperament Vector T

Based on Panksepp’s 7 emotional systems (Panksepp, 1998):

Interpretation: $T_i > 1$ means heightened system reactivity, $T_i < 1$ means reduced reactivity.

Evolutionary Roots of Character: T in All Mammals

Key fact: Panksepp’s 7 emotional systems are homologously conserved across all mammals — from rats to primates. Character is not a uniquely human phenomenon but a fundamental property of the mammalian brain.

Neurotransmitter Profiles of Panksepp Systems

Each of the 7 systems relies on a characteristic set of neurotransmitters:

Global modulators (act on all systems simultaneously): glutamate (+), GABA (-), norepinephrine (+), serotonin (-). These modulators set the overall tone: high norepinephrine increases excitability of all systems (stress mobilization), while serotonin reduces it (stabilization, satiation).

Source: Panksepp (2011). “The basic emotional circuits of mammalian brains: Do animals have affective lives?” Neuroscience & Biobehavioral Reviews, 35(9), 1791-1804.

Why this matters for the theory:

1. Character = utility node weights. T variability is observed in all mammals because utility nodes (emotional systems) are the same neural circuits
2. Evolutionary advantage of variability. “Fluctuation selection” — not always the same trait is optimal. A reservoir of different T values in the population
3. Humans differ NOT in having character but in memeplex complexity. A cat has T but no memeplex to subordinate these weights

G-Programs and WM: The Dual Competition Model

dlPFC serves both working memory and emotion regulation simultaneously — one resource pool (Pessoa, 2009, TiCS). When a G-program is activated, part of PFC resources is redirected to processing the affective signal, reducing the number of available WM pointers. Empirically: CDA K-score ~ 2 under negative affect vs. ~ 3.5 under neutral conditions — loss of ~50% WM capacity (Figueira et al., 2017, SCAN; Stout et al., 2017, Scientific Reports).

Neurotransmitter mechanism by G-program:

Mechanism: amygdala activation -> NE burst -> dlPFC interference + simultaneous vmPFC recruitment for emotion regulation -> fewer resources for WM maintenance. PLAY is the exception: endorphins and endocannabinoids reduce the tone of negative systems (FEAR, RAGE), freeing PFC resources -> WM during PLAY >= WM during neutral state. This is the neurobiological basis for the effectiveness of learning through play.

Formalization: $k_{eff}(t) = k_{active}(t_{dev}) - n_{captured}^G(t) - n_{captured}^{signal}(t)$, capture-weights, PTSD-loop — see NM: G->WM competition.

Signal Memes as a Second Source of WM Capture

The table above describes WM capture by G-programs (affective channel). But WM slots are competitive: they are captured not only by the G-layer but also by the M-layer — specifically, by signal memes. Mechanism:

• Grounding: associating a signal with a referent through Hebbian co-activation requires simultaneous activation of both memes -> occupies a WM slot
• Routing: routing the signal (to whom, when, through which channel) requires executive WM resources
• Fidelity maintenance: maintaining signal accuracy (kappa-consolidation of signal memes) competes with maintenance of navigation/survival memes

At $k_{eff} \approx 3\text{--}4$ active slots, even one signal meme = 25–33% loss of survival-relevant capacity. Ten survival experiments (predator avoidance, cooperative foraging, farming, goal-directed navigation, $N$=8–150) confirm: language parasitizes WM — $\Delta_{alive} \approx 0$ or negative.

Generalized formula:

where $n_{captured}^{signal}$ = number of WM slots occupied by signal memes (grounding, routing, fidelity). This is the second WM competition channel: G -> WM (affective, subcortical) and M -> WM (signal, cortical).

Consequence: language can arise only under resource surplus — the environment must be sufficiently rich for the organism to survive with reduced $k_{eff}$. This is the basis of the Language Emergence Threshold — see EMT, Part XXVIII.

Evolutionary status of language. Language is optimized for M-fitness (transmission fidelity, compressibility), not G-fitness (survival). This is the only known case where M permanently suppresses G not as an anomaly (kamikaze, anorexia) but as a species norm. Prediction P-BM28.

Computationally verified. Ten survival experiments in the BMC engine ($N$=8–150) confirmed $\Delta_{alive} \approx 0$ or negative. Lewis signaling game: 85–97.5% accuracy across 25–533 concepts without gradient-based optimization. Separating test: BMC TopSim 0.72 vs REINFORCE 0.60 ($p = 0.011$). See DOI: 10.5281/zenodo.19181798.

Formalization:

The T vector is structurally identical for all mammals. Differences:

• Value ranges may vary between species
• In humans, T is modulated by the memeplex (via interface I)
• In animals, T directly determines behavior

Sources: Selected Principles of Pankseppian Affective Neuroscience (PMC6344464); The ‘Feline Five’ (PLOS ONE, 2017); Animal personality (Wikipedia).

Modified Utility Node Activation Formula

Current formula (see above):

With character factored in:

where $T_g$ is the character vector component for utility node $g$.

Influence of Character on Connection Formation

Modified new connection weight formula:

where:

• $e_g$ — emotional node (utility node)
• $T_g$ — character weight for the given emotion
• $a_g$ — emotion activation during the experience (0–1)

Consequence: With high $T_{FEAR}$, connections to the “Fear” node form more strongly.

Numerical Example: Two People Visit a Castle

Result after one year:

• Person A: “Castles are creepy places” (fear connection dominates)
• Person B: “Castles are interesting places to explore” (curiosity connection dominates)

Character Inheritance

Empirical base: Twin studies show Big Five heritability of 40–60% (Jang et al., 1996).

Transmission formula:

where:

• $\alpha_i \in [0, 1]$ — mixing coefficient for component $i$
• $\epsilon_i \sim \mathcal{N}(0, \sigma^2)$ — mutational noise

Inheritance modes:

Example: 4 children — each has bits of parental character, the last one is a “carbon copy” (“Full copy” mode).

Model Predictions

Network formalization: Spreading activation modulation by the T vector — see NM: Network modulation by character.

Sources: PMC: Selected Principles of Pankseppian Affective Neuroscience; PubMed: Jang et al. (1996).

Extension: Binding to Panksepp Systems

Three-Level Meme Binding

Following Panksepp & Biven (2012), emotional systems are organized into three levels. Memes bind to each level differently:

Consequence: Tertiary-level memes can modulate the expression of primary programs but cannot cancel them. Dementia is regression from tertiary to primary.

Affective Space: M-Layer Construction Over Discrete G

Central thesis: The G-level is discrete (Panksepp is right), the M-level is continuous (Barrett is right). Both are right at different levels of BMC.

G: discrete circuits. Panksepp’s seven systems are separate neural circuits with distinct nuclei (PAG, hypothalamus, VTA, BLA), different neurotransmitters (see table above), and different output patterns. The primary level: discrete, evolutionarily conserved, identical across all mammals.

M: continuous construction. The tertiary level is neocortical interpretation. Simultaneous activation of multiple G-programs produces a blend — a continuous point in affective space:

where $\mathbf{v}_g$ is a fixed vector in (valence, arousal) coordinates per G-program (Russell, 1980, J Pers Soc Psychol):

$E(t)$ is emergent from discrete G-activations, modulated by $T$. “Nostalgia” = $E$ near GRIEF + SEEKING + CARE; “Schadenfreude” = PLAY + RAGE blend. M-memes cluster certain zones of E-space, giving them names and cultural interpretations.

Cultural variability. Amae (Japanese) = CARE + GRIEF + acceptable dependency blend. Saudade (Portuguese) = GRIEF + SEEKING + absent object. Schadenfreude (German) = PLAY + RAGE + another’s failure. Identical G-activations -> different M-clusters -> different “emotions.” This explains Barrett (2017, How Emotions Are Made): constructivism is correct for M, not for G. Cultures with different M-memes carve up the continuous E-space differently (Wierzbicka, 1999, Emotions Across Languages and Cultures).

Dementia as a test. When $M \to 0$ (neocortical neurodegeneration), blends disappear, exposing discrete primary G-patterns: pure FEAR, pure RAGE, uncontrolled GRIEF. If Barrett were entirely right (emotions are only constructions), there would be no discrete emotions in dementia — but they exist, and they correspond precisely to subcortical G-systems. This is a strong argument for the two-level model.

BMC position: Barrett is right for the M-level (continuous, culturally constructed); Panksepp for the G-level (discrete, biologically fixed). BMC reconciles these positions through a three-level architecture: primary G -> secondary conditioned -> tertiary M-construction.

Formalization: Blend formula $E(t)$, $\mathbf{v}_g$ coordinates — see NM: Affective Space.

SEEKING as Metasystem

SEEKING is not merely one of the seven systems. It is recruited by all others and functions as a metasystem — the “currency of attention” (dopamine).

where:

• $T_{SEEK}$ — individual weight from the character vector
• $a_{SEEK}^{base}$ — SEEKING base activation
• $\alpha_s$ — recruitment coefficient from system $s$
• $a_s(t)$ — activation of system $s$ at time $t$

Interpretation: When FEAR is activated (threat), it recruits SEEKING to search for an exit. When CARE is activated (need to care), it recruits SEEKING to find an object of care. Dopamine is not “reward” but “directed curiosity.”

Structural Incompleteness as SEEKING’s Second Input (SIT)

Problem: The current SEEKING formula describes recruitment from other emotional systems:

This explains why SEEKING is activated by fear (search for escape), care (search for object), curiosity (search for novelty). But the formula does not explain persistent SEEKING activation during unresolved tasks — without external stimuli, without recruitment from other emotions. Under the current model, an unresolved task should decay through edge decay. But this does not happen.

Structural Incompleteness Tension (SIT) — the tension of structural incompleteness — is SEEKING’s second input, alongside recruitment from emotional systems. Structural gaps in the memeplex generate SEEKING activation endogenously, proportional to the cluster’s importance and the degree to which the gap is unfilled.

Extended SEEKING formula:

where:

• $SIT(C) = \sum_{g \in gaps(C)} relevance(g) \cdot centrality(C) \cdot (1 - closure(g))$ — structural incompleteness tension of cluster $C$ (see NM: SIT)
• $LP(C, t) = \frac{d}{dt} closure(C, t)$ — Learning Progress: the rate of approaching closure
• $\gamma_{SIT}$ — individual SIT sensitivity coefficient

Key point: The SIT term is inside the bracket $T_{SEEK}$. This means SIT modulates base curiosity, not emotional recruitment. A person with high $T_{SEEK}$ (researcher, scientist) reacts more strongly to structural gaps. A person with low $T_{SEEK}$ may have the same gaps but feel no discomfort.

LP-filter as a temporal modulator:

False closure and evolutionary logic: SIT generates persistent cognitive tension that consumes resources. Evolution developed a false closure strategy: fill the gap with any available node, even one with zero empirical validity. “Why is there thunder?” -> “Zeus is angry” — SIT decreases, cognitive load drops, resources are freed for survival. This is cheaper than maintaining an open question for years. Hence — religion, superstitions, magical thinking, conspiracy theories (details — NM: False closure).

Consequences of the extended formula:

1. Spontaneous return to tasks: An unresolved task with $SIT > 0$ periodically “surfaces” — especially when SEEKING load from other sources is low (rest, falling asleep, walking)
2. Differential curiosity: Two people with identical memeplexes but different $T_{SEEK}$ will react differently to gaps
3. Scientific obsession: Very high $T_{SEEK}$ + significant gap -> $SIT_{eff}$ can dominate other motivations
4. DMN activation: Without external tasks, SEEKING switches to SIT-driven scanning — neurobiologically this is the Default Mode Network (see above, Part III: DMN)

DISGUST as an I-Layer Mechanism

DISGUST is not one of Panksepp’s 7 systems (he classified it as a sensory affect). However, in biomemetic theory DISGUST plays a key role as an interface mechanism (I-layer): a genetic program recruited by the memetic layer for marking “foreign” memes.

Evolution of DISGUST:

Key observation: The physiological substrate is identical for all levels — the insular cortex is activated identically during physical and moral disgust (Rozin et al., 2008). This means the memeplex co-opts a ready-made genetic mechanism for marking hostile memes.

DISGUST = the main neuromechanism for assigning negative weight. When the memeplex “rejects” an idea, it literally activates the same neural circuit as disgust at rotten food.

Sources: Panksepp & Biven (2012). The Archaeology of Mind; Haidt (2001). “The Emotional Dog and Its Rational Tail”; Rozin, Haidt & McCauley (2008). “Disgust” (in Handbook of Emotions).

Neurobiological Levels of Immune Filtration

DISGUST is not the only immune filtration mechanism. Hierarchical filtration at different processing levels is described in attention research (Broadbent, 1958; early vs late selection) and predictive processing (Friston, 2005). BMC interprets these levels as per-level immune subsystems $I^{(k)}$, each with its own neural substrate:

Key observation: Filtration speed increases from bottom to top — lower levels react in tens of milliseconds (attentional gate), upper levels in hundreds (reflective assessment). This creates cascading filtration: a stimulus that passes the L0 filter may be rejected at L2, but with a delay. Empirically: the amygdala reacts to threat in ~100 ms (L1), PFC suppresses false alarm in ~200 ms (L2) — the I-layer begins working (see Part VII: conflict cascade).

Consequence for development: Upper-level immune filters mature later than lower ones. The age periods below correspond to known stages of cognitive development (Piaget) and data on PFC maturation (Gogtay et al., 2004); the mapping onto BMC I-levels is an interpretation of these data, not an independent prediction:

This explains why radicalization is most effective with adolescents (immature $I^{(3)}$) and least effective with the elderly (rigid $I^{(3)}$, but at the cost of plasticity).

Pathologies as I-level disruptions:

Formalization: $I^{(k)}$ as immune subsystem of level $k$, G-invariants — see NM, Part IX. Engineering specification: AGI Foundations, Part IV.

Part V. Interaction Mechanisms: Redirection, Suppression, Interpretation

The Problem: How Do Memes Influence Genetic Programs?

A meme cannot “switch off” a genetic program — it is hardcoded. But a meme can modify its expression through three mechanisms.

Three Interface Mechanisms

1. Redirection

Mechanism: The meme does not cancel the drive but links it to an alternative goal.

Redirection cost:

The farther the meme’s goal from the genetic one, the higher the cost.

2. Suppression

Mechanism: The meme creates competing activation that inhibits the genetic program’s expression.

Suppression cost formula:

where:

• $\beta$ — base cost coefficient
• $a_g(t)$ — strength of the suppressed program
• $duration$ — suppression duration
• $habit(m)$ — degree of automatization (0 to 1); formal definition, dynamics, and $Auto(S)$ — see Automatization: neurobiological substrate

Important: Suppression is depletable. Resources for suppression are limited.

3. Interpretation

Mechanism: The meme changes the meaning of the genetic program’s signal without changing the signal itself.

Mechanism Comparison Table

Diagram: Three Mechanisms in Action

Suppression Energy Budget

Total energy budget:

where:

• $E_{max}$ — maximum energy resource (restored by sleep)
• $Cost_{supp,i}$ — cost of suppressing the $i$-th program
• $Cost_{active,j}$ — cost of active processes

Failure modes:

Numerical Example: Diet and Hunger

Situation: A person on a diet sees a cake.

Parameters:

• $a_{hunger} = 0.6$
• $a_{diet\_meme} = 0.4$
• $habit_{diet} = 0.3$ (recently started)
• $E_{available} = 0.5$

Suppression cost calculation:

After 10 minutes:

At $\theta_{min} = 0.1$, suppression begins to weaken.

Conclusion: Without external help (leaving the kitchen) or meme reinforcement (remembering the diet goal), the person will probably eat the cake.

See also: Gene-meme interaction mechanisms — AGI Foundations, Part II.

Expression: Neurobiology of Replication Pressure

The three mechanisms above (redirection, suppression, interpretation) describe how memes influence G-programs. But there is a symmetric question: how do memes use the neuronal substrate for their own replication? Redirection, suppression, and interpretation are the M->G interface. Replication pressure is M->output: the mechanism by which activated memes are “pushed out” through the communication channel.

Formalization: $R_{expr}(m_i, t)$ — see NM, Part X.

The Speech Apparatus as a Replication Substrate

Genes replicate via DNA polymerase — an enzymatic complex refined over billions of years of evolution. Memes replicate via the speech apparatus — a neuromotor complex in which key roles are played by:

• Broca’s area (BA 44/45, posterior IFG): articulation planning, syntactic structuring — “assembling” the meme into a linear word sequence
• Wernicke’s area (BA 22, posterior superior temporal gyrus): decoding incoming speech — “unpacking” the meme on the recipient’s side
• Arcuate fasciculus: Broca <-> Wernicke link — a closed “understood -> can say” circuit, necessary for self-monitoring of speech (the meme is checked before output)
• Motor cortex (BA 4) + basal ganglia: articulation execution — the physical “printing” of the meme

Inner Speech as Pre-Expression

Vygotsky (1934): thought passes through inner speech before becoming external. In BMC terms: an activated meme first passes through the internal circuit of Broca (premotor planning without articulation), and only with sufficient $R_{expr}$ transitions to motor execution. Inner speech is a trial run of replication: the meme is “assembled” into linear form, checked through the arcuate fasciculus, and upon passing the I-filter is released externally.

This explains the delay between meme activation and utterance: the internal circuit requires ~200–400 ms (Indefrey & Levelt, 2004) for lemmatic selection and phonological encoding.

Mirror Neurons and “Contagion” of Pressure

Mirror neurons, discovered in zone F5 of monkeys (Rizzolatti et al., 1996), are activated both during action execution and observation of others’ actions. Zone F5 is the homolog of human Broca’s area (BA 44); the hypothesis of language evolving from the mirror system: Rizzolatti & Arbib (1998). In the context of speech:

• When listening to an interlocutor, the listener’s speech motor areas are activated phoneme-specifically: hearing words with lingual consonants enhances tongue motor potentials (Fadiga et al., 2002, TMS study)
• This activation is not random: the incoming meme via Wernicke -> motor mirror system activates related memes of the listener
• Related memes receive an activation boost -> their $R_{expr}$ grows -> the listener feels their own pressure to express

This is the neuronal basis of the “I had a similar experience!” phenomenon: the incoming meme via the mirror system activates the listener’s associated memes, which accumulate sufficient $R_{expr}$ for expression.

Neurochemistry of Social Reward

Successful communication (replication success) is reinforced via the mesolimbic pathway:

• VTA -> NAcc: dopamine release during self-disclosure (Tamir & Mitchell, 2012, PNAS): talking about oneself activates the same NAcc regions as monetary reward; participants forfeited ~17% of potential earnings for the opportunity of self-disclosure
• In BMC interpretation: NAcc reward for self-disclosure = dopaminergic reinforcement for successful meme replication. Not “people like talking about themselves” (phenomenology), but “the substrate reinforces meme replication just as it reinforces resource acquisition” (mechanism)
• Oxytocin enhances social bonding -> lowers $\theta_{expr}$ (expression threshold) -> more memes pass the I-filter -> more “open” communication

Two Neurochemical Evidences for $R_{expr}$

1. Tip-of-the-tongue (TOT). The meme is activated (the subject knows what they want to say), but lemmatic selection fails (the word cannot be found). Subjectively felt as pressure — direct evidence of $R_{expr}$:

• Meme activation is high (it is “knocking” on the channel)
• The channel is blocked (phonological code not retrieved)
• Result: an unpleasant feeling of incompleteness — analogous to SIT at the level of a single act of communication

2. Taboo effect (Wegner, 1987). “Don’t think of a white monkey” -> the meme is activated more strongly. The I-system suppresses the meme ($I_{sig}$ high), but the meme is already activated -> high $R_{expr}$. The conflict between I-suppression and expression pressure produces ironic monitoring: the stronger the suppression, the higher the pressure. This conflict is neuroanatomically localized: PFC (suppression) vs Broca (expression planning) vs ACC (conflict detection).

Sources: Rizzolatti et al. (1996) — mirror neurons in F5; Rizzolatti & Arbib (1998) — F5->Broca homology and language evolution; Fadiga et al. (2002) — phoneme-specific motor facilitation during speech listening (TMS); Tamir & Mitchell (2012, PNAS) — self-disclosure activates NAcc reward circuits; Indefrey & Levelt (2004) — temporal course of word production; Wegner et al. (1987) — paradoxical effects of thought suppression; Wegner (1994) — ironic process theory; Vygotsky (1934) — inner speech.

Part VI. Competition Dynamics: A Qualitative Model

The Problem: How to Understand Replicator Competition?

The description “genes and memes compete” is insufficient. A model is needed that allows:

1. Understanding which layer dominates at any given moment
2. Predicting transitions between regimes
3. Explaining observed phenomena (breakdowns, willpower, impulsivity)

Key Idea: Competition for Limited Attention

Genetic programs ($G$) and the memetic layer ($M$) compete for a shared resource — attention. At any moment, attention is distributed unevenly between them.

What affects the balance:

Four System Regimes

Example: A Person on a Diet

Morning (regime: balance)

• Full after breakfast, rested
• “I’m losing weight” meme active
• Hunger drive low
• -> Easy to stick to the plan

Evening after work (regime: conflict -> gene dominance)

• Tired, work stress
• “I’m losing weight” meme weakened by fatigue
• Hunger drive strengthened (6 hours without food + stress)
• Sees the refrigerator (trigger)
• -> Breakdown

What could have helped:

• Not allowing strong hunger (remove G trigger)
• Resting before making decisions (restore M)
• Removing food from sight (remove external stimulus)
• Reminding oneself of the goal (activate the meme)

Example: Street Aggression — G/M Tension as a Consciousness Trigger

The diet illustrates slow G/M competition (hours). Aggression illustrates instantaneous competition — and reveals a fundamental property: conscious thinking is triggered by tension between G and M.

Scenario: A person is walking down the street, approached by an aggressive stranger.

Time     What happens                       BMC Regime
-----    ---------------                    ----------
0 ms     Stimulus: threatening gesture      Entry via amygdala
50 ms    G-layer reacts: RAGE or FEAR       G dominance
         a_g^base -> fight/flight
         Cortisol, adrenaline up
         Balance drops sharply
200 ms   I-layer begins working:            Conflict
         Suppression (suppress impulse)
         Interpretation (assess threat)
500 ms   M-layer engages:                   M gaining weight
         "He's drunk, better leave"
         "There are cameras here, not worth it"
         "I have a family at home"
1-2 sec  Decision: leave / respond          Balance or
         (depends on competition outcome)   M dominance

Key observation: The M-layer was not thinking before the stimulus appeared. Memes about caution, cameras, family were in the memeplex but dormant. It was the G-layer that created tension that activated M.

Without the G-layer there is no trigger for M. Without the M-layer there is no brake for G. Conscious thinking is not an independent activity but a reaction of M to tension created by G.

Neurobiological substrate:

What determines the outcome:

Conclusion: consciousness is neither a passive observer nor a “boss” over the body. Consciousness (M-layer) is a competitor who is awakened when the G-layer creates tension. No tension — no need for conscious thinking. This explains why routine actions are performed “on autopilot”: with low G/M tension, the M-layer is not activated, and behavior is controlled by habits (entrenched memes with high inertia, not requiring WM involvement).

Regime Transitions

The system can abruptly switch between regimes when conditions change:

Practical Consequences

1. Willpower is a depletable resource: M/G conflict is depleting. After prolonged struggle, breakdown probability increases.

2. Prevention beats fighting: Easier to prevent G dominance (avoid triggers) than to fight in conflict mode.

3. Environment beats intentions: External stimuli strongly affect the balance. Changing the environment is more effective than “willpower.”

4. Training works: Meditation and mindfulness practices strengthen M’s ability to suppress G.

Methodological note: This qualitative model describes observed phenomena without claiming quantitative precision. Mathematical formalization (e.g., via Lotka-Volterra-type equations) is possible but requires empirical parameter calibration that is currently absent. See Appendix D for a possible formalization direction.

SIT as the Fifth Competition Factor

To the four previously described G/M competition factors (physical state, stress, M training, meme strength), a fifth is added: Structural Incompleteness Tension (SIT).

SIT creates endogenous M-layer motivation independent of external stimuli and not derived from genetic drives. It is the only competition factor that works in favor of the M-layer autonomously:

SIT interactions with other factors:

• Stress + SIT: Stress suppresses the M-layer, but SIT continues to generate SEEKING activation. Result: rumination — obsessive return to unresolved problems precisely during stress, when the M-layer is weakened and cannot work effectively toward closure. This explains the paradox: stress reduces cognitive abilities but intensifies “fixation” on problems.

• Low energy + SIT: Under exhaustion, the G-layer dominates, but SIT-driven thoughts continue to intrude. Subjectively: “can’t stop thinking about the problem even though I’m tired.” Hence — insomnia with unresolved problems.

• High M training + SIT: Meditation and mindfulness allow observing SIT activation without automatically following it. The M-layer does not suppress SIT but chooses when to allocate resources to it.

• Strong memes + SIT: Clusters with high eigenvector centrality containing gaps generate maximum SIT — these are the most “obsessive” unresolved problems (life questions, professional puzzles).

See also: Formal definition of SIT — NM, Part VIII; Resource competition — AGI Foundations, Part II.

Part VII. Ontogeny: BMC Changes Across the Lifespan

The Problem: BMC Is Not Static

BMC configuration changes throughout life:

• Children easily accept memes
• Adolescents test boundaries
• Adults are stable but rigid
• The elderly resist change

How to formalize these changes?

Critical Periods

A critical period is a time window during which certain meme types can enter with minimal resistance.

Critical Periods Table

Balance by Age Formula

where:

• $C_{PFC}(age)$ — degree of prefrontal cortex maturation
• $C_{limbic}$ — constant (limbic system matures early)
• $M_{density}(age)$ — memeplex density
• $M_{max}$ — maximum memeplex capacity

Approximate $C_{PFC}(age)$ values:

BMC Lifecycle Diagram

Table: Consequences of Missing a Critical Period

Numerical Example: Openness Change

Model of openness to new memes:

where $O_{base} = 1.0$, $age_{peak} = 10$, $\lambda = 0.003$.

Conclusion: Core memes should enter before ~25 years. After that, entry is difficult.

See also: Memeplex ontogeny — EMT, Part VI.

Openness as a Cross-Cluster Plasticity Modulator

The openness function $O(age)$ (see table above) not only regulates the probability of new meme entry but also modulates reactivation of cross-cluster connections.

Mechanism: During Hebbian reactivation (both nodes co-activated), intra-cluster edges are reactivated at full strength $\Delta w$, while cross-cluster edges are modulated:

Biological basis: Openness to experience is linked to dopaminergic activity (DeYoung, 2013), specifically tonic dopamine modulation of distal cortical connections. Age-related decline in dopaminergic activity (Backman et al., 2006) reduces plasticity of “long-range” associative pathways while preserving plasticity of local (intra-cluster) ones.

Phases:

Additional mechanism at $O < 0.3$: In addition to weakened reactivation, differential plasticity engages — background strengthening of intra-cluster and weakening of cross-cluster connections (see NM: “Differential plasticity at low openness”).

Age Crises as Cascading Hub Restructuring

Critical periods (table above) describe smooth BMC dynamics. However, ontogeny also contains crises — discrete phase transitions during which memeplex topology reorganizes in an avalanche-like fashion (see EMT, Part XVII).

Mechanism: Hub Displacement

A crisis is the moment when the old hub is weakened (information gap, changed circumstances, G/M misalignment) and an alternative hub has already accumulated enough connections for avalanche-like capture:

When $k_j$ (degree of the new hub) exceeds $k_i$ (degree of the old), connections begin to avalanche-transfer -> cascading restructuring = crisis.

Key point: “adult” memes (responsibility, independence, mortality) exist in a child’s memeplex long before the crisis, but as weak peripheral nodes. A crisis is not the appearance of new memes but the capture of connections by already existing memes.

Age Crisis Table

Children After Upheavals

Serious stress (parent’s death, war, violence) is a forcible destabilization of the child’s memeplex. Trauma creates massive SIT -> old hubs (childhood world models) cannot provide closure -> “adult” hubs (which already existed but were weak) get a chance to capture connections -> the child “suddenly” starts reasoning like an adult.

This is not maturation but emergency hub displacement: the memeplex restructures under pressure, bypassing normal phases. The price is instability: “adult” hubs captured connections without having sufficient periphery (childhood memes are not integrated but suppressed -> high internal dissonance -> risk of pathologies, see Part IX).

Predictions:

See also: Hub displacement — EMT, Part XVII; network formalization of hub displacement — NM.

BMC Death: The Terminal Phase of Ontogeny

What Happens When the Host Dies

Death is the terminal phase of BMC ontogeny, characterized by irreversible substrate ($S$) loss.

Fate of BMC components:

Quantitative Assessment of Information Loss

Formula:

where $Copied(m_i) \in \{0, 1\}$ indicates whether the meme was copied during the host’s lifetime.

Typical values: $Loss_{death} > 0.99$ (>99% of the memeplex is lost).

Evolutionary Strategies for Meme Preservation

The paradox of death: The genetic layer ($G$) has a built-in transmission mechanism (sexual reproduction), ensuring 50% preservation. The memetic layer ($M$) has no analogous mechanism — transmission depends on the host’s active efforts and requires a complex communication channel.

AGI application: Overcoming BMC death via Shared Memplex Repository — see AGI Foundations: Agent death and shared memplex repository.

Memetic perspective: Death and immortality of memes — see EMT, Part XXIII.

Programmed agent death (AGI): In multi-agent BMC systems, death is formalized as apoptosis with four pathways: (1) intrinsic — self-detection of rigidity (SIT near 0, IF near 0), analogous to p53-apoptosis; (2) extrinsic — evolutionary pressure (fitness < theta), analogous to TNF/Fas-signaling; (3) anoikis — social isolation as a senescence accelerator, analogous to loss of extracellular matrix; (4) neglect — resource exhaustion (energy=0), analogous to necrosis. The SMR-donation protocol ensures graceful knowledge transfer (kappa >= 2 memes -> stigmergic traces) before death, implementing the Super-Ratchet at the swarm level. Formalization: SM, Parts III–IV.

Part VIII. Cross-Cultural Variation: Different BMC Configurations

The Problem: Is the BMC Model Universal?

The genetic layer is universal (all humans have identical basic drives). But the interface configuration and memeplex structure vary across cultures.

Thesis: Cultures as BMC Configurations

Different cultures represent different ways of tuning BMC:

• Which genetic programs are suppressed vs. encouraged
• How emotions are interpreted
• Which memes become hubs

Dimensions of Cultural Variation

Cultural Configuration Formalization

where:

• $W_{ind}$ — weight of individualistic memes
• $W_{col}$ — weight of collectivistic memes
• $\theta_{honor}$ — threshold for honor insult response
• $\sigma_{context}$ — degree of contextual communication dependency

Table of Cultural Configurations

Visibility of Genetic Programs

Different cultures mask or expose genetic programs differently:

Prediction: Honor Cultures and Escalation

Hypothesis: In cultures with low $\theta_{honor}$ (low insult-response threshold), G-M conflict escalates faster.

Mechanism:

1. Insult activates $G_{status}$
2. In honor culture, low $\theta_{honor}$ -> “defend honor” meme activates easily
3. The meme does not suppress $G_{aggression}$ but directs it
4. Result: physical aggression

In a dignity culture:

1. Same insult activates $G_{status}$
2. High $\theta_{honor}$ -> “rise above it” meme
3. Meme suppresses $G_{aggression}$
4. Result: external calm (internal tension)

Diagram: Cultural Modulation of BMC

Intergenerational Configuration Transmission

Cultural configuration is transmitted through:

1. Parenting — parents tune their children’s BMC
2. Institutions — school, church, army
3. Narratives — stories, heroes, role models

where $\alpha + \beta + \gamma = 1$, and the weights change with age.

Stratification of Access to Cultural Configuration

The intergenerational transmission formula masks a critical fact: source quality depends on social stratum.

where $Config_{quality} = \alpha \cdot q_{parents} + \beta \cdot q_{peers} + \gamma \cdot q_{media}$

Source quality ($q$) is determined by:

• Memeplex richness (meme diversity)
• Coherence (internal consistency)
• Adaptiveness (correspondence to reality)

Connection to Balance:

Low $Config_{quality}$ -> weak memetic layer (M) -> low $Balance$ -> greater susceptibility to genetic drives and external “tests.”

Consequence: Social inequality is not merely economic — it is memetic. Children from lower strata receive not only fewer resources but also a lower-quality BMC configuration.

Theory of cultural capital: Pierre Bourdieu (1986) — embodied, objectified, institutionalized capital.

See also: EMT: Culture as SMR.

Part IX. Pathologies: When BMC Breaks Down

The Problem: What Happens When BMC Malfunctions?

Psychopathology can be viewed as a disruption of balance, stability, or connectivity in BMC.

BMC Pathology Classification

Pathology Formula

where each parameter has a “healthy range”:

BMC Pathology Taxonomy

Diagram: Trajectories to Pathology

Numerical Example: Addiction as a Trajectory

Initial state (norm):

• $Balance = 1.5$
• $A_m = 0.6$, $A_g = 0.4$
• “Healthy lifestyle” meme controls $G_{reward}$

Event: first drug use

• Drug directly activates $NA$ (nucleus accumbens)
• $A_g$ sharply rises to 0.9
• $Balance$ drops to 0.67

Dependency formation:

Mechanism: The drug hacks the reward system, making it insensitive to natural rewards (including rewards from memes). $G_{reward}$ no longer serves either genes or memes — only the substance.

Depression as Rumination: The SMC Mechanism

The classical description of depression as “memes switched off” is imprecise. The refined mechanism via reflection (see EMT, Part XVII):

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

Neural substrate:

Evidence: Berman et al. (2011), Biological Psychiatry: DMN hyperactivation in depression correlates with rumination; Sheline et al. (2009), PNAS: disrupted anti-correlation of DMN and task-positive networks in depression.

Suicide as Terminal Failure of Both Inferences

Suicide is an extreme case in which both discrepancy-minimization mechanisms have terminally failed (see EMT, Part XVII):

• (a) Perceptual inference (update model to match reality): the memeplex is too rigid (high Q, hubs resist restructuring) -> the person cannot change their model
• (b) Active inference (change reality to match model): insufficient resources or insurmountable circumstances -> the person cannot change the world

When both paths are blocked: LP = 0 -> rumination -> $E_{available} \to 0$ -> the M-layer generates the meme “closure is impossible” -> the “no way out” meme overcomes the G-layer (self-preservation instinct).

Risk and protective factors in BMC language:

Therapeutic Implications

See also: Addiction as defeat of both systems — EMT, Part XXVIII; network predictions of pathologies — NM, Part XII.

Extended Neurotaxonomy: ADHD, Autism, DID, and Cognitive Dissonance

The existing section covers addiction, depression, PTSD, schizophrenia, OCD, and borderline disorder. But BMC predicts a unified parameter space for all disorders. Below are the neuromechanisms of the missing ones.

ADHD: sigma > 1 (supercriticality) + unstable SIT

BMC mechanism: sigma > 1 (supercriticality) — activation spreads too easily -> memes do not compete but all activate simultaneously -> no winner -> switching. SIT creates many gaps, but closure comes prematurely (superficially) -> new gap -> cycle.

Stimulants (Ritalin, Adderall): increase tonic dopaminergic and noradrenergic activity -> stabilize the I-layer and lateral inhibition -> sigma approaches 1 -> M-layer can maintain focus.

Formal connection to WM selection: The distinction between phasic and tonic dopamine is directly reflected in the WM salience formula. Phasic VTA response = $|\Delta a_i(t)|$ (prediction error -> burst firing -> attention capture). Tonic level = $a_i(t)$ (baseline firing -> sustained focus). In ADHD, phasic signal is unstable (SEEKING flickering) and tonic is insufficient (weak lateral inhibition). Stimulants increase the tonic component, stabilizing $a_i(t)$ and allowing habituation $h_i(t)$ to correctly displace irrelevant memes.

Autism: I-Overtuning + Weak Long-Range Connections

BMC mechanism: The I-system is too strict — new memes (especially unstructured, social ones) are filtered out. The M-layer is highly structured within “special interests” (deep local clustering) but weakly connected between clusters. PLAY is reduced — little stochastic recombination.

Compatible neuroanatomical data: Ecker et al. (2013): diffusion MRI shows altered white matter connectivity in autism — specifically weak long-range connections with preserved local connectivity. This corresponds to the BMC description (high local clustering, weak inter-cluster edges), though the data precede the model.

DID (Dissociative Identity Disorder): Multiple SMC

BMC mechanism: massive trauma -> forced M-layer segmentation to isolate unbearable SIT-gaps. Each segment develops its own SMC. Result: Modularity $Q$ » normal, connectivity between identity-modules -> 0.

Data: Reinders et al. (2014): fMRI shows different neural activity patterns for different identity-states, including different DMN configurations.

Cognitive Dissonance: Hub Advantage Neuromechanism

BMC predicts that in cognitive dissonance, the meme with lower eigenvector centrality will change (hub advantage — see EMT, Part IV). Neuromechanism:

1. Conflict detection: ACC detects simultaneous activation of incompatible memes (negative edge + both > theta_high)
2. Discomfort: Insula generates a somatic marker — a feeling that “something is wrong”
3. Resolution via hub advantage: PFC redistributes activation. The meme with greater degree (more incoming edges -> more supporting activation) holds its position. The less connected meme is suppressed via lateral inhibition -> rationalization.

Neural data: van Veen et al. (2009): ACC is activated during cognitive dissonance (fMRI); dACC activation predicts degree of attitude change — the I-layer (conflict detection) initiates the process.

Unified parameter space of all disorders and formalization — see EMT, Part XXII; network formalization — see NM.

Radicalization: The Neural Trajectory of Ideological Capture

Radicalization is a neurobiologically describable process of hub displacement under ideological pressure. The mechanism parallels ontogenetic crises (hub displacement under trauma) but with an external agent (ideology).

Neural trajectory:

Neurochemistry:

• Oxytocin: enhances in-group bonding, enhances out-group hostility (De Dreu et al., 2010)
• Cortisol: chronically elevated (constant threat detection)
• Dopamine: reward from in-group belonging -> reinforcing cycle

Adolescent vulnerability: PFC not fully myelinated yet (until ~25 years) -> I-layer immature -> weak filtration. Simultaneously, peak SEEKING (identity formation) + G-instability (puberty) -> ideal window for ideological capture.

Predictions:

Conceptual connection: Radicalization as the reverse path of Gandhi — see EMT, Part XXVII.

PFC Myelination and WM Capacity Growth

The WM ceiling ($k_{active}$) is determined by theta-gamma coupling in PFC: the number of gamma cycles fitting within one theta cycle sets the maximum number of simultaneous representations (Lisman & Jensen, 2013, Neuron). This coupling matures as PFC myelinates — the latest-maturing cortical region (Gogtay et al., 2004, PNAS).

EEG marker: Growth in theta-gamma phase-amplitude coupling (PAC) in frontal leads correlates with WM capacity across ages (Gaillard et al., 2021, Heliyon). This provides an objective neural marker for $k_{active}(t_{dev})$.

Critical periods of memogenesis. Each increment in $k$ opens a new binding level: $k=1$ -> single memes (sensorimotor), $k=2$ -> conditioned pairs (early symbolism), $k=3$ -> triple operations (logic, rules), $k=4$ -> metarepresentations (reflection over memes). Memes requiring $k > k_{active}$ literally do not fit in WM -> cannot be acquired. Hence age-stratified cultural transmission.

Formalization: $k_{active}(t_{dev})$, formula and empirical anchors — see NM: WM Ontogeny.

Cognitive Biases: Neuroanatomical Substrate

The six BMC mechanisms generating cognitive biases have distinct neuroanatomical substrates. Below is the mapping of mechanisms to brain regions and key neural data.

Dual system: G vs M, not System 1 vs System 2. Kahneman described two “systems”: fast intuitive (System 1) and slow analytical (System 2). Neural data show a more precise picture:

• G-layer (amygdala, striatum, VMPFC) — fast utility assessment. Not “irrational” — optimizes survival and reproduction. Loss aversion, affect heuristic = G-capture of WM slots.
• M-layer deliberation (DLPFC, ACC, posterior parietal) — slow analytical processing. Requires WM resources ($k_{eff} > 1$). Confirmation bias, Dunning-Kruger = H-inertia + I-filtration in M-layer.
• ACC (anterior cingulate cortex) — conflict detector. Activated during G/M misalignment (when G “wants” one thing and M analysis shows another). Detects cognitive dissonance ($D > \theta$) and triggers override — M’s attempt to suppress the G signal.

“System 1” = G + A (utility reactions + automatized patterns). “System 2” = M-layer deliberation with WM. But Kahneman does not distinguish G and A within “System 1” — and this is critical: G-biases (loss aversion) are intensified by WM load, A-biases (status quo) are not.

Neural signatures of biased confidence. Lebreton et al. (2015) and subsequent work show: VMPFC correlates positively with confidence (certainty in a decision), while DLPFC/DMPFC correlates negatively. In BMC: VMPFC = G-valuation (utility assessment “this is right/safe”), DLPFC = M-override (analytical check, requiring WM). High confidence during bias = strong G signal (VMPFC) with weak M-override (DLPFC loaded or not activated).

Cross-modulation: negativity bias. $\lambda_{neg} < \lambda_{pos}$ — neurobiologically: the amygdala reacts faster and more intensely to negative stimuli (LeDoux, 1996); consolidation of negative memories is enhanced (via norepinephrine -> BLA -> hippocampus pathway). This is not a separate system but a parameter modulating the edge decay rate for negatively colored edges in the memeplex.

Connection to pathologies. Extreme values of the same parameters produce clinical disorders (see Part IX above: parametric taxonomy). Cognitive biases are the normative range; clinical disorders are tail distributions of the same parameters:

Formalization: Bias strength functions $B_H, B_I, B_W, B_G, B_A, B_R$ — see NM, Part IX. Conceptual overview: EMT, Part XXIV. AGI engineering: AGI Foundations, Part VI.

Part X. Epigenetics: Can Memes Influence Gene Expression?

The Problem: One-Directional Influence?

So far we have considered the influence of genes on memes and the reverse influence of memes (through behavior) on gene success. But is direct meme influence on gene expression possible?

Epigenetic Primer

Epigenetics is the study of changes in gene expression not related to changes in the DNA sequence.

Main mechanisms:

Evidence of Behavioral Influence on the Epigenome

Table: Practices and Epigenetic Effects

Epigenetic Modification Probability Formula

where:

• $\lambda$ — base modification rate (gene-dependent)
• $duration$ — exposure duration
• $intensity$ — intensity (stress, practice)

Example calculation:

Meditation practice: $\lambda = 0.01$ (per week), $duration = 52$ weeks, $intensity = 0.5$ (moderate practice)

Probability of a noticeable epigenetic effect ~12% after one year of moderate practice.

Transgenerational Effects

A controversial but important question: Can epigenetic changes be transmitted to offspring?

Evidence:

2025 update: Transgenerational epigenetic effects in humans replicate more reliably than previously thought. However, effect sizes are small, and causal relationships remain debated due to multiple confounders (shared environment, cultural transmission of trauma, socioeconomic factors).

Diagram: Path from Meme to Epigenome

Important Clarification: Lamarckism Is NOT Rehabilitated

Critical note: Epigenetic effects of behavior do not mean the rehabilitation of Lamarckism. Important limitations:

1. Effects are probabilistic, not deterministic
2. Effects are often reversible
3. Transgenerational transmission is limited (1–2 generations)
4. Mechanisms in humans are less studied than in rodents
5. Many early studies have not been replicated

Memes can influence gene expression, but this influence is limited, probabilistic, and is not “inheritance of acquired traits” in the Lamarckian sense.

Implications for the BMC Model

If memes can influence the epigenome:

1. Long-term memes can modify the $G$-layer itself (not the genome, but its expression)
2. Cultural evolution can indirectly influence biological evolution via epigenetics
3. Therapy may have epigenetic effects

This adds another level of interaction to the BMC model.

Part XI. Predictions and Falsifiability

The Problem: Does the Theory Explain Everything?

Any good theory must make falsifiable predictions. What does the BMC model predict?

Predictions Table

Specific Predictions from the Model

1. Neuroanatomical Prediction

Prediction: Individual PFC thickness correlates with meme-favoring $Balance$.

Mechanism: PFC is the M substrate. More PFC -> more meme capacity -> higher $Balance$.

Verification: Correlation of PFC thickness (MRI) with self-control measures.

Falsification: No correlation or inverse correlation.

2. Pharmacological Prediction

Prediction: PFC-enhancing drugs (e.g., modafinil) increase meme-favoring $Balance$.

Mechanism: Strengthening the M substrate -> more M resources -> higher $Balance$.

Verification: Behavioral tests under drug vs. placebo.

Falsification: Drug does not affect self-control.

3. Ontogenetic Prediction

Prediction: PFC maturation (age ~25) coincides with peak $Balance$.

Mechanism: Full maturation of the M substrate -> maximum capacity.

Verification: Longitudinal study of $Balance$ (self-control tests) + MRI.

Falsification: Peak $Balance$ does not coincide with PFC maturation.

4. Cross-Cultural Prediction

Prediction: Cultures with high $\theta_{honor}$ (dignity cultures) have fewer violent crimes.

Mechanism: High threshold -> M suppresses G aggression more often.

Verification: Correlation of cultural metrics with crime statistics.

Falsification: No correlation or inverse correlation.

Predictions Structure Diagram

Existing Confirmations

5. SIT-Specific Predictions

Structural Incompleteness Tension (SIT) generates a separate class of predictions not derivable from the base BMC model without the SIT extension:

5a. Neuroanatomical (DMN x gap density):

Subjects with more “open questions” (high gap density) should show higher basal DMN activation. This is a direct consequence of SIT -> SEEKING -> DMN.

5b. Pharmacological (dopaminergics x SIT-rumination):

The SIT term in the SEEKING formula is modulated by $T_{SEEK}$, which depends on dopamine. Increasing available dopamine -> increased $T_{SEEK}$ -> enhanced SIT-driven activation -> more spontaneous thoughts about unresolved problems.

5c. Behavioral (Zeigarnik x centrality):

The classic Zeigarnik effect (better memory for interrupted tasks) should be modulated by significance: $SIT \propto centrality(C)$, so interrupted tasks from central clusters should be remembered significantly better than those from peripheral ones.

5d. Temporal (SIT is not a forgetting curve):

Resolved tasks follow edge decay: $w(t) = w_0 \cdot e^{-\lambda t}$. Unresolved tasks with $SIT > 0$ should show a plateau of recall that does not decline until closure (solution found) or LP-collapse (progress completely stopped).

See also: Falsifiability of meme theory — EMT, Part XXIX; network predictions and SIT prediction #9 — NM, Part XI.

Part XII. Conclusions and Implications

Summary of the BMC Model

The Biomemetic Complex is a system arising from the interaction of two replicators (genes and memes) on a shared neurobiological substrate.

Key Conclusions

Implications for Different Domains

For Psychiatry

For Education

For AI Alignment

For Social Policy

Bridge to AGI

The BMC model proposes an architecture for AGI with human-like dynamics:

Hypothesis: To create AGI with a human-like “personality,” one must reproduce not only the memeplex but also the utility substrate and the interface between them.

Final Theory Diagram

What Comes Next

1. Empirical testing — testing predictions from Part XI
2. Formula refinement — calibrating parameters on real data
3. AGI application — implementing the architecture from AGI Foundations
4. Clinical application — developing BMC-based diagnostics

See also: Theory application to AGI — AGI Foundations; network formalization — NM; core theory — EMT.

8-Course Cross-Analysis Updates

The following updates emerge from a systematic cross-analysis of eight graduate-level courses — spanning behavioral biology, nonlinear dynamics, reinforcement learning, neuronal dynamics, cognitive neuroscience, information theory, and network science — against the original BMC formalism. Each update replaces or extends a specific formula from the core theory above with a more principled, empirically grounded version.

Balance Dynamics: Static Ratio → Bistable ODE (CRITICAL)

The original static formula $Balance = A_{PFC}/(A_{limbic}+\varepsilon)$ is replaced by a bistable 2D ODE system:

Two stable attractors (rational-dominant, emotional-dominant) with hysteresis: $\theta_{exit} \gg \theta_{enter}$. Once the system tips into an emotional attractor, significantly more rational input is required to return — explaining why “calming down” is harder than “getting angry.” Hopf bifurcation at critical parameter values yields a pathology taxonomy: bipolar disorder = limit cycle between attractors, PTSD = excitable system with low threshold.

Sources: Sapolsky (behavioral biology), Strogatz (nonlinear dynamics).

Learning Rules: Hebbian → TD-Modulated (HIGH)

Three-level hierarchy replacing the generic $\Delta w \propto a_i a_j$:

1. STDP (Spike-Timing-Dependent Plasticity): $\Delta w = A_+ e^{-\Delta t/\tau_+}$ (pre→post, strengthening) $- A_- e^{-|\Delta t|/\tau_-}$ (post→pre, weakening). Introduces directional causality into edge weights.

2. Reward-modulated STDP (3-factor rule): $\Delta w_{ij} = \eta \cdot \delta_{BMC} \cdot \psi_i \cdot a_j$. Learning occurs only when the TD error signal $\delta_{BMC}$ confirms that the co-activation was behaviorally relevant. Matches Yagishita et al. (2014) findings on dopamine-gated synaptic plasticity.

3. TD error as learning signal: $\delta_{BMC} = \sum_g w_g a_g(t+1) + \gamma V(S_{t+1}) - V(S_t)$. Critic (2-factor, ventral striatum analogue) vs Actor (3-factor, dorsal striatum analogue).

Sources: Gerstner (neuronal dynamics), Sutton & Barto (reinforcement learning).

Mutual Information Interpretation of Edges (HIGH)

Edge weight now has a precise information-theoretic meaning:

Edge weight = mutual information in bits between two memes. The PRUNE criterion becomes principled: remove edge if $I < \theta_{prune}$ bits (the memes are informationally independent). Hub = node with maximum total mutual information $\sum_j I(m_i; m_j)$.

Sources: MacKay, Stone (information theory).

Autoreception and Adaptation (MED-HIGH)

Chronic activation of a G-program → sensitivity decrease → tolerance, allostatic load, anhedonia. This formalizes why sustained stress eventually blunts the stress response itself. Implemented as an AdEx (adaptive exponential) slow variable: $\tau_w dw_i/dt = \alpha(a_i - a_i^{rest}) - w_i$.

Sources: Sapolsky (stress physiology), Gerstner (adaptive neuron models).

Serotonin: Dual Role (MED-HIGH)

Serotonin (5-HT) modulates two independent parameters simultaneously:

1. I-gate amplification: $I_{eff} = I_{base}(1 + \beta_{5HT} \cdot Mod_{5HT})$ — impulse control strength
2. Temporal discount rate: $\gamma_{BMC} = \gamma_{base} + \beta_\gamma \cdot Mod_{5HT}$ — patience/future orientation

Low serotonin → impulsive AND myopic (double deficit). Predicts Caspi effect: MAO-A variant × childhood trauma → antisocial behavior via both pathways.

Sources: Sapolsky (behavioral biology), Sutton & Barto (discount factor).

Dual-Route Processing (MED-HIGH)

Two routes for information influence on behavior:

• Route 1 (conscious, ~200ms): $influence^{conscious} = a_i \cdot \mathbb{1}[a_i > \theta_{act}] \cdot priority$
• Route 2 (subliminal, ~12ms): $influence^{subliminal} = \gamma_{sub} \cdot a_i \cdot f_{route2}(m_i)$

Where $f_{route2}$ depends on meme type: G-connected memes (1.0), semantic memes (0.3), abstract memes (0). Formalizes intuition, implicit bias, blindsight, and Zajonc’s mere exposure effect.

Source: Gazzaniga (cognitive neuroscience).

Rate-Distortion → Fidelity Tiers (MED-HIGH)

Memory fidelity is not binary (remember/forget) but operates on a rate-distortion curve with three tiers:

Information Bottleneck principle: optimal memory maximizes $\sum I(m; behavior)$ given a finite storage budget. Consolidation = source coding optimization.

Sources: MacKay, Stone (information theory, rate-distortion).

Additional Formulas

Appendix A. Formula Reference

Definitions

Coevolution

Neurobiology

Activation

Interaction Mechanisms

Competition Model

Mathematical formulas for the competition model are placed in Appendix D as hypothetical — they require empirical validation.

Ontogeny

Culture

Pathologies

Epigenetics

Signed Edges and Asymmetric Decay

SEEKING Metasystem and BLEND

Differentiated Storage

Appendix B. Glossary

Appendix C. References

Evolution and Coevolution

• Baldwin, J.M. (1896). A new factor in evolution. The American Naturalist, 30(354), 441-451.
• Blackmore, S. (1999). The Meme Machine. Oxford University Press.
• Dawkins, R. (1976). The Selfish Gene. Oxford University Press.
• Deacon, T.W. (1997). The Symbolic Species. W.W. Norton.
• Richerson, P.J., & Boyd, R. (2005). Not by Genes Alone. University of Chicago Press.

Neurobiology

• Arnsten, A.F. (2009). Stress signalling pathways that impair prefrontal cortex structure and function. Nature Reviews Neuroscience, 10(6), 410-422.
• Damasio, A. (1994). Descartes’ Error. Putnam.
• Gogtay, N., et al. (2004). Dynamic mapping of human cortical development during childhood through early adulthood. PNAS, 101(21), 8174-8179.
• Miller, E.K., & Cohen, J.D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24, 167-202.

Epigenetics

• Meaney, M.J. (2010). Epigenetics and the biological definition of gene x environment interactions. Child Development, 81(1), 41-79.
• Yehuda, R., et al. (2016). Holocaust exposure induced intergenerational effects on FKBP5 methylation. Biological Psychiatry, 80(5), 372-380.

Cross-Cultural Psychology

• Cohen, D., et al. (1996). Insult, aggression, and the southern culture of honor. Journal of Personality and Social Psychology, 70(5), 945-960.
• Hofstede, G. (2001). Culture’s Consequences. 2nd ed. Sage Publications.
• Nisbett, R.E., & Cohen, D. (1996). Culture of Honor. Westview Press.

Psychopathology

• Baumeister, R.F., et al. (2007). The strength model of self-control. Current Directions in Psychological Science, 16(6), 351-355.
• Koob, G.F., & Volkow, N.D. (2016). Neurobiology of addiction: a neurocircuitry analysis. The Lancet Psychiatry, 3(8), 760-773.

Affective Neuroscience and Emotions

• Panksepp, J. (2011). The basic emotional circuits of mammalian brains: Do animals have affective lives? Neuroscience & Biobehavioral Reviews, 35(9), 1791-1804.
• Panksepp, J., & Biven, L. (2012). The Archaeology of Mind: Neuroevolutionary Origins of Human Emotions. W.W. Norton.
• Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108(4), 814-834.
• Rozin, P., Haidt, J., & McCauley, C.R. (2008). Disgust. In M. Lewis, J.M. Haviland-Jones & L.F. Barrett (Eds.), Handbook of Emotions (3rd ed., pp. 757-776). Guilford Press.

Meme Synthesis and Sleep

• Fauconnier, G., & Turner, M. (2002). The Way We Think: Conceptual Blending and the Mind’s Hidden Complexities. Basic Books.
• Wagner, U., Gais, S., Haider, H., Verleger, R., & Born, J. (2004). Sleep inspires insight. Nature, 427(6972), 352-355.
• Lewis, P.A., & Durrant, S.J. (2011). Overlapping memory replay during sleep builds cognitive schemata. Trends in Cognitive Sciences, 15(8), 343-351.

Negative Weights and Structural Balance

• Baumeister, R.F., Bratslavsky, E., Finkenauer, C., & Vohs, K.D. (2001). Bad is stronger than good. Review of General Psychology, 5(4), 323-370.
• Heider, F. (1946). Attitudes and cognitive organization. The Journal of Psychology, 21(1), 107-112.
• Davis, J.A. (1967). Clustering and structural balance in graphs. Human Relations, 20(2), 181-187.
• Hovland, C.I., Lumsdaine, A.A., & Sheffield, F.D. (1949). Experiments on Mass Communication. Princeton University Press.

Network Science

• Barabasi, A.-L. (2016). Network Science. Cambridge University Press.
• Newman, M.E.J. (2018). Networks: An Introduction. 2nd ed. Oxford University Press.

Appendix D. Hypothetical Mathematical Formalization

Status: This formalization is hypothetical. Parameters are not empirically calibrated. It is presented as a possible direction for future research, not as a validated model.

Lotka-Volterra Competition Model

The classical two-population competition model for a shared resource can be adapted to describe the competition between genetic ($A_g$) and memetic ($A_m$) activation:

Parameters:

• $A_m$, $A_g$ — memetic and genetic layer activations
• $r_m$, $r_g$ — activation growth rates
• $K_m$, $K_g$ — capacities (maximum activation)
• $\alpha_{gm}$, $\alpha_{mg}$ — competition coefficients
• $S_g(t)$ — external stimuli for genetic programs

Possible equilibria:

• Meme dominance ($A_m \approx K_m$, $A_g \approx 0$)
• Gene dominance ($A_m \approx 0$, $A_g \approx K_g$)
• Coexistence ($A_m > 0$, $A_g > 0$)
• Bistability (outcome depends on initial conditions)

What Is Needed for Validation

1. Operationalization of variables: How to measure $A_m$ and $A_g$? Possible proxies: PFC activity (fMRI), cortisol levels, behavioral markers.

2. Parameter calibration: Experiments with controlled factor changes (hunger, stress, meditation) and measurement of behavior/neuroactivity changes.

3. Prediction testing: The model predicts bifurcation points — sharp regime transitions. This can be tested.

Literature on Dynamic Models in Psychology

• Guastello, S.J., et al. (2009). Chaos and Complexity in Psychology. Cambridge University Press.
• van Geert, P. (1994). Dynamic Systems of Development. Harvester Wheatsheaf.
• Warren, K., et al. (2003). Nonlinear dynamics in psychology. Nonlinear Dynamics, Psychology, and Life Sciences.

Document created as the theoretical foundation of the MEMETICS project. For AGI application, see AGI Foundations.