Narrative Engineering · 52/147

Part 2: Information Flow in Neural Systems

Chapter: Chapter 4: Cognitive Frameworks in Narrative Construction

Having established how individual neurons process information, we must now understand how information flows through neural networks to create narrative processing capabilities. This flow follows precise mathematical principles that determine both the possibilities and limitations of narrative processing in the brain.

Signal Propagation Through Neural Networks

Information propagates through neural networks according to a wave equation modified by active processing:

∂2V∂t2=c2∇2V+f(V)∂V∂t+g(V)\frac{\partial^2 V}{\partial t^2} = c^2\nabla^2V + f(V)\frac{\partial V}{\partial t} + g(V)

where:

  • V represents the propagating signal

  • c is the propagation speed

  • f(V) represents active amplification

  • g(V) represents non-linear processing

This equation reveals several fundamental properties of neural information flow:

  1. Wave-like propagation (c²∇²V term)

  2. Active modification (f(V)∂V/∂t term)

  3. Non-linear processing (g(V) term)

The propagation speed c varies with axon properties:

c=d4ρCmc = \sqrt{\frac{d}{4\rho C_m}}

where:

  • d is the axon diameter

  • ρ is the axoplasm resistivity

  • C_m is membrane capacitance

Information Transformation Through Layers

As signals move through neural layers, they undergo systematic transformation:

y⃗l+1=σ(Wly⃗l+b⃗l)\vec{y}_{l+1} = \sigma(W_l\vec{y}_l + \vec{b}_l)

where:

  • y_l represents layer l activity

  • W_l is the weight matrix

  • b_l is the bias vector

  • σ is the activation function

The information content at each layer follows:

Il=∫p(yl)log⁡2p(yl)q(yl)dylI_l = \int p(y_l)\log_2\frac{p(y_l)}{q(y_l)}dy_l

where:

  • p(y_l) is the actual distribution

  • q(y_l) is the prior distribution

This shows how each layer:

  1. Transforms representations

  2. Extracts features

  3. Builds abstractions

  4. Manages information flow

Information Bottlenecks

Neural systems face fundamental bottlenecks that constrain information flow. The information bottleneck principle states that each layer must optimize:

L=I(Y;T)−βI(X;T)\mathcal{L} = I(Y;T) - \beta I(X;T)

where:

  • I(Y;T) is relevant information preserved

  • I(X;T) is compression of input X

  • β balances compression vs preservation

  • T is the compressed representation

This creates a natural hierarchy where:

  1. Lower layers maintain detail

  2. Middle layers compress information

  3. Higher layers extract meaning

  4. Top layers represent abstractions

Processing Hierarchies

The hierarchical organization of neural processing follows:

∂y⃗∗l∂t=−y⃗l+Fl(y⃗l−1)+Gl(y⃗∗l+1)\frac{\partial \vec{y}*l}{\partial t} = -\vec{y}l + F_l(\vec{y}{l-1}) + G_l(\vec{y}*{l+1})

where:

  • F_l represents bottom-up processing

  • G_l represents top-down influence

This bidirectional flow enables:

  1. Feature extraction upward

  2. Expectation guidance downward

  3. Integration across levels

  4. Dynamic reconfiguration

The effective connectivity between layers follows:

Ceff=C0exp⁡(−∣l1−l2∣λ)C_{eff} = C_0\exp\left(-\frac{|l_1-l_2|}{\lambda}\right)

where:

  • C₀ is base connectivity

  • λ is the characteristic length

  • l₁,l₂ are layer indices

Capacity Limits and Resource Management

The total processing capacity of the system follows:

Ctotal=∑lCl−∑l1,l2Il1,l2C_{total} = \sum_l C_l - \sum_{l_1,l_2} I_{l_1,l_2}

where:

  • C_l is layer capacity

  • I_l₁,l₂ is inter-layer interference

This capacity must be allocated across competing demands:

dRidt=ai(Rmax−Ri)−∑jbijRiRj\frac{dR_i}{dt} = a_i(R_{max} - R_i) - \sum_j b_{ij}R_iR_j

where:

  • R_i represents resources for process i

  • a_i is the recovery rate

  • b_ij is the competition coefficient

Implications for Narrative Processing

These principles reveal several key insights for narrative construction:

  1. Information Flow Management Narratives must:

  • Match propagation speeds

  • Work with natural hierarchies

  • Respect bottlenecks

  • Manage resources

  1. Layer-Appropriate Processing Different layers handle:

  • Sensory details (low)

  • Pattern extraction (middle)

  • Meaning construction (high)

  • Abstract concepts (top)

  1. Bidirectional Integration Effective narratives support:

  • Bottom-up feature building

  • Top-down interpretation

  • Cross-level integration

  • Dynamic reconfiguration

  1. Resource Optimization Design must consider:

  • Capacity limits

  • Competition effects

  • Recovery rates

  • Interference patterns

Practical Design Principles

These mathematical insights suggest specific strategies:

  1. Match Natural Hierarchies

  • Build from concrete to abstract

  • Support level-appropriate processing

  • Enable bidirectional flow

  • Maintain cross-level coherence

  1. Manage Information Flow

  • Control propagation rates

  • Optimize compression

  • Balance preservation

  • Support integration

  1. Resource Management

  • Monitor capacity usage

  • Minimize interference

  • Allow recovery time

  • Optimize allocation

Understanding these principles helps explain why certain narrative structures are more effective than others and provides guidance for optimal design. The mathematics reveals both constraints and opportunities in narrative construction through neural systems.

The key insight is that effective narratives must work with rather than against natural patterns of neural information flow. This requires understanding both the capabilities and limitations of neural processing hierarchies while designing for optimal information transmission and integration.