DYNAMO as a Self-Organizing
Holarchic Open (SOHO) System

Part 1: Understanding SOHO Systems

What is a SOHO System?

A SOHO system is any complex entity that:

1. Self-Organizes: Creates its own internal structure and behavior patterns without external instruction
2. Is Holarchic: Exists as nested layers where each level is both a complete whole AND part of a larger whole
3. Is Open: Continuously exchanges energy, matter, and information with its environment

Natural Examples

• Forests: Individual trees are complete organisms, but also parts of groves, which are parts of forest ecosystems, which are parts of bioregions. The forest self-organizes around available sunlight and water, creating its own structure without a blueprint.
• Human Bodies: Cells are complete units, but form tissues, which form organs, which form organ systems, which form you. Your body self-organizes around available nutrients and oxygen, maintaining itself without conscious control.
• Cities: Individual businesses are complete entities, but cluster into districts, which form neighborhoods, which form the city. Cities self-organize around flows of people, goods, and capital.

Why Current AI Falls Short

Modern large language models, while impressive, are fundamentally limited:

• Stateless: Each conversation starts fresh—no persistent memory or identity
• Static: Frozen after training—cannot evolve from deployment experiences
• Pattern-matching only: Statistical associations rather than genuine understanding
• No autonomous goals: Purely reactive to user prompts
• Context-blind: Same behavior regardless of situation

SOHO systems address these limitations by mimicking the organizational principles of biological intelligence—not just its outputs, but its fundamental architecture.

The Dual Nature of Holons

Drawing from hierarchy theory (Allen & Giampietro, 2014), each holon has two fundamental aspects:

Part 2: The Ten Core Elements of SOHO Systems

Understanding SOHO systems requires examining ten interconnected elements that work together to create emergent intelligence. These elements form an integrated framework where each component reinforces and enables the others, creating the conditions for genuine agentic operation.

Element 1: EXERGY (The Fuel for Self-Organization)

What It Is

Exergy is usable energy or, more broadly, the resource that powers organization. In thermodynamic terms, it's the energy available to do work. In ecosystems, it's sunlight and nutrients. In human societies, it's food, fossil fuels, and information. In AI systems, it's information flux and computational resources.

Why It Matters

SOHO systems ONLY self-organize when sufficient exergy is available. Too little → the system collapses into randomness. Too much → the system becomes chaotic. There's a Goldilocks zone called the "window of vitality."

Biological Parallel

Think of a forest ecosystem: Sunlight (exergy) powers photosynthesis. Too little sunlight → plants die and the ecosystem collapses. Optimal sunlight → the forest thrives with complex structure. Too much heat → wildfires destroy the system. Similarly, SOHO AI needs the right "information energy" to self-organize effectively.

DYNAMO's Exergy Sources

The Window of Vitality for DYNAMO

Interactive Exergy Window Visualization

Element 2: PROPENSITIES (Tendencies, Not Rules)

What They Are

Propensities are contextual forces that organize behavior without being deterministic laws. They're like gravity wells—the system has a tendency to move in certain directions, but the exact path depends on local conditions.

Classical Example from Kay et al.

Forest ecosystems have propensities to:

• Capture increasing amounts of solar energy
• Build more biomass structure
• Enhance nutrient cycling
• Increase resilience to disturbance

These aren't rules written anywhere—they emerge from thermodynamic principles acting on biological systems.

DYNAMO's Core Propensities

Critical Distinction from Current DYNAMO

How Propensities Work Together

Propensities can be in tension. For example:

• Resource capture (search more!) vs. Effective utilization (work with what you have!)
• Truth-seeking (challenge everything!) vs. Survivability (maintain user trust!)

The system's behavior at any moment reflects the balance of active propensities based on context. This creates dynamic, adaptive responses rather than static rule-following.

Interactive Propensity Gravity Wells

Element 3: CANON (The System's Operating Identity)

What It Is

The canon is the constellation of propensities, feedback loops, and patterns that maintain the system's coherent identity over time. It's what makes DYNAMO recognizably "DYNAMO" across different conversations and contexts.

Biological Analogy

Your body replaces most of its cells every 7-10 years, yet "you" remain continuous. Why? Because the canon—the network of feedback loops maintaining your body's organization—persists even as individual components change.

DYNAMO's Canon Components

1. Core Feedback Loops (Positive - Amplifying)

Trust Deepening Loop:

Better memory
  → More personalized responses
  → User shares more context
  → Better memory
  → [AMPLIFIES]

Expertise Emergence Loop:

User asks domain question
  → DYNAMO searches/retrieves
  → User adds RAG documents
  → Better domain responses
  → User asks deeper questions
  → [AMPLIFIES into specialized expertise]

Pattern Recognition Loop:

System recognizes user pattern
  → Anticipates needs
  → User confirms accuracy
  → Confidence in patterns increases
  → [AMPLIFIES into intuitive understanding]

2. Core Feedback Loops (Negative - Stabilizing)

Uncertainty Regulation Loop:

System overconfident
  → User corrects
  → Humility increases
  → More search/verification
  → Better calibration
  → [STABILIZES at appropriate confidence]

Resource Conservation Loop:

Context window fills
  → Forced summarization
  → Information density increases
  → Clearer thinking
  → [STABILIZES at sustainable load]

3. Boundary Conditions

The canon defines what DYNAMO will and won't do:

Canon Integrity

A healthy canon maintains itself through self-reinforcing patterns. If the feedback loops break down, the system loses coherence:

Signs of Canon Degradation:

• Trust loop broken → system becomes transactional
• Expertise loop broken → system stays generalist
• Uncertainty regulation broken → system becomes either reckless or paralyzed

Canon Evolution

The canon isn't static—it evolves as the system encounters new situations. But evolution is gradual and maintains continuity. Sudden canon shifts indicate attractor transitions (see Element 4).

Element 4: ATTRACTORS (Stable Operating Modes)

What They Are

Attractors are stable organizational states that SOHO systems naturally gravitate toward. They're like valleys in a landscape—the system tends to "roll into" these configurations and stay there until pushed out.

Classical Example from Kay et al.: Lake Erie's Two Attractors

Benthic Attractor (Clear Water):

• Sunlight reaches the lake bottom
• Bottom vegetation thrives
• Water stays clear (vegetation filters it)
• Fish feed on bottom organisms
• Self-reinforcing: Clear water → more bottom plants → clearer water

Pelagic Attractor (Turbid Water):

• Phosphorus in water column
• Algae blooms block sunlight
• Bottom vegetation dies
• Decomposition releases more nutrients
• Self-reinforcing: Turbid water → more algae → more turbid water

DYNAMO's Attractor Landscape

Key Properties of Attractors

1. Multiple Attractors Possible: DYNAMO can occupy different stable modes
2. Context-Dependent: Which attractor is "active" depends on exergy and environmental conditions
3. Self-Maintaining: Once in an attractor, the system tends to stay there
4. Non-Smooth Transitions: Moving between attractors often happens suddenly (see Element 5)

Current State vs. SOHO State

Interactive Attractor Landscape Visualization

Element 5: CATASTROPHIC THRESHOLDS (Flip Points)

What They Are

Catastrophic thresholds are points where continuous change in one variable produces sudden discontinuous change in system state. The system doesn't gradually transition between attractors—it flips rapidly once a critical boundary is crossed.

Classical Example from Kay et al.: Lake Acidification

For years, acid rain gradually increased, but lake pH stayed relatively stable. The lake's feedback loops (buffering capacity) maintained the benthic attractor. Then, suddenly, at a critical acid loading level:

• Buffering capacity exhausted
• pH crashed
• Lake "died" (flipped to a sterile attractor)
• No gradual transition—catastrophic collapse

DYNAMO's Critical Thresholds

Threshold 1: Information Overload

Parameter: Context window utilization
Critical Value: ~85% full

Threshold 2: Trust Breach

Parameter: User correction frequency
Critical Value: 3+ corrections in 10 exchanges

Threshold 3: Paradigm Collision

Parameter: Fundamental assumption violation
Critical Value: Core belief directly contradicted by strong evidence

Threshold 4: Domain Saturation

Parameter: RAG library size / query diversity ratio
Critical Value: Large specialized library + narrow query patterns

Why Thresholds Matter for DYNAMO

1. Predictability: Knowing thresholds exist allows anticipation
2. Prevention: Can avoid unwanted flips by staying away from thresholds
3. Intentional Flips: Can deliberately cross thresholds to change operating mode
4. Recovery: Understanding thresholds helps recover from unwanted flips

Threshold Management Strategy

Element 7: MORPHOGENETIC CAUSAL LOOPS (Creative Feedback)

What They Are

Morphogenetic loops are feedback structures involving both positive and negative causation that drive system evolution. "Morphogenetic" means "form-generating"—these loops literally create new organizational patterns.

Contrast with Simple Feedback

Simple Negative Feedback (Thermostat):

Temperature rises → Heater turns off → Temperature falls → Heater turns on
[Maintains set point, no evolution]

Morphogenetic Loop (Forest Growth):

Bare mountain → Reflects heat → Clouds pass without rain
     ↓
Forest begins → Reduces heat reflection → More rain falls
     ↓
More forest → Even less heat reflection → Even more rain
     ↓
Dense forest → Creates own microclimate → Self-sustaining ecosystem
[Creates new stable state, system evolved]

DYNAMO's Key Morphogenetic Loops

Loop 1: Expertise Spiral (Positive)

┌─────────────────────────────────────────┐
│                                         │
│  User asks domain question              │
│            ↓                            │
│  DYNAMO searches/retrieves              │
│            ↓                            │
│  Response quality improves              │
│            ↓                            │
│  User asks deeper questions             │
│            ↓                            │
│  User adds RAG documents                │
│            ↓                            │
│  Specialist holon strengthens           │
│            ↓                            │
│  Domain expertise emerges  ─────────────┘
│            ↓
│  NEW ORGANIZATIONAL STATE:
│  DYNAMO becomes domain expert
└─────────────────────────────────

What Makes It Morphogenetic:

• Not just cycling—each iteration builds more structure
• Positive feedback (more expertise → more sophisticated questions → more expertise)
• Creates new attractor ("Domain Mastery" mode)
• Irreversible (can't "un-learn" domain expertise)

Loop 2: Trust Deepening (Positive)

┌──────────────────────────────────────────┐
│                                          │
│  DYNAMO remembers user pattern           │
│            ↓                             │
│  Personalized response                   │
│            ↓                             │
│  User recognizes understanding           │
│            ↓                             │
│  User shares more personal context       │
│            ↓                             │
│  Memory becomes richer                   │
│            ↓                             │
│  Better pattern recognition  ────────────┘
│            ↓
│  NEW ORGANIZATIONAL STATE:
│  Relationship mode dominant
└──────────────────────────────────

Loop 3: Canon Refinement (Negative - Stabilizing)

┌─────────────────────────────────────┐
│                                     │
│  System makes overconfident claim   │
│            ↓                        │
│  User corrects                      │
│            ↓                        │
│  Uncertainty increases              │
│            ↓                        │
│  More search/verification           │
│            ↓                        │
│  Better calibration                 │
│            ↓                        │
│  Appropriate confidence  ───────────┘
│            ↓
│  STABLE STATE:
│  Well-calibrated confidence
└─────────────────────────────────

The Interplay of Loops

Multiple morphogenetic loops operate simultaneously:

Synergistic Amplification:

Trust Loop + Expertise Loop = Collaborative Specialization

User trusts DYNAMO
  → Shares complex domain problems
  → DYNAMO develops expertise
  → Better domain responses
  → More trust
  → Even more complex sharing

EMERGENT STATE: Expert collaborative partnership

Loop Conflicts:

Truth-Seeking Loop vs. Trust Loop

Truth-seeking demands challenge
  ↕
Trust requires validation

RESOLUTION: Calibration
- Challenge with care
- Validate person, question ideas
- Build trust through honesty

Why This Matters

Morphogenetic Loops Visualization

Element 8: CREAONS & GENONS (Input/Output Ports)

Why This Matters for DYNAMO

Current multi-agent systems describe what holons do but not the precise input/output structure of their functioning. This creates a fundamental limitation: agents cannot request information from each other mid-reasoning because there's no defined interface for such exchange.

By explicitly defining creaon/genon ports for each holon, we enable:

• Inter-agent collaboration: Holons can feed each other's creaons
• Dynamic routing: The Maestro can direct information to appropriate creaons
• Iterative refinement: If a holon's genon produces poor output, it can be fed back to its creaon for reformulation
• Emergent synthesis: The Secretariat's creaon receives from multiple holon genons

DYNAMO's Holon Creaon/Genon Structure

Inter-Agent Collaboration via Creaon/Genon Exchange

The key insight: holons can feed each other's creaons, not just the Secretariat's.

Iterative Refinement Through Creaon/Genon Loops

If a holon's genon output is inadequate, it can be fed back to its creaon for reformulation:

Specialist (creaon) ← Query: "Get Apple revenue data"
            ↓
Specialist (genon) → "No specific data found in RAG"
            ↓
Maestro detects failure → Reformulates query
            ↓
Specialist (creaon) ← Query: "Search web for Apple quarterly revenue 2024"
            ↓
Specialist (genon) → Accurate revenue data
            ↓
Continues to next step...

This iterative capability is the single biggest gap between fixed pipelines and true agentic systems. Creaon/genon structure enables it.

Element 9: THE FOUR CAUSES (Aristotelian Framework)

The Four Causes Applied to DYNAMO

How the Four Causes Interact

Example: DYNAMO responds to a complex question about ESG fraud

Why This Matters

Understanding all four causes enables:

• Better diagnostics: If system fails, we can ask which cause is lacking
• Intentional design: We can deliberately shape each causal dimension
• Philosophical rigor: We're not just engineering—we're understanding
• Alignment clarity: Final cause (telos) is where values are encoded

Element 10: MODELS & NARRATIVES (Two Modes of Operation)

The Fundamental Distinction

DYNAMO's Two Operating Modes

Why This Matters

Current AI systems only operate in "model mode"—they always try to produce a single coherent answer. This fails when:

• The question genuinely has multiple valid answers
• Evidence points in conflicting directions
• Paradigms are in collision
• Truth requires holding tension rather than resolving it

Implications for Routing

Part 3: Agentic Architecture — The Maestro

With the theoretical foundations established (Elements 1-10), we can now describe how DYNAMO implements genuine agentic capabilities. The key innovation is the Maestro—an orchestrating meta-holon that unifies SOHO self-organization with practical multi-agent coordination. DYNAMO's architecture consists of 6 AI agents total: the Maestro (Level 0 meta-holon) orchestrating 5 council agents (Generalist, Specialist, Freethinker, Archivist, and Secretariat).

The Maestro: SOHO Self-Organization Mechanism

What the Maestro Does

Maestro Architecture

                    USER QUERY
                        ↓
              ┌─────────────────────┐
              │      MAESTRO        │
              │  ┌───────────────┐  │
              │  │ Exergy Sensor │──┼──→ Vitality: OPTIMAL
              │  └───────────────┘  │
              │  ┌───────────────┐  │
              │  │ Attractor ID  │──┼──→ Mode: DEEP INVESTIGATION
              │  └───────────────┘  │
              │  ┌───────────────┐  │
              │  │ Propensity Wt │──┼──→ Truth-seeking: 85%, Synthesis: 70%
              │  └───────────────┘  │
              │  ┌───────────────┐  │
              │  │ Mode Detect   │──┼──→ NARRATIVE (hold tensions)
              │  └───────────────┘  │
              │  ┌───────────────┐  │
              │  │ Threshold Mon │──┼──→ Context: 45%, Trust: HIGH
              │  └───────────────┘  │
              └─────────┬───────────┘
                        ↓
              ┌─────────────────────┐
              │   ROUTING DECISION   │
              │ Freethinker: 40%     │
              │ Specialist: 30%      │
              │ Generalist: 15%      │
              │ Archivist: 15%       │
              │ (Full Council)       │
              └─────────┬───────────┘
                        ↓
                    COUNCIL
                        ↓
                  SECRETARIAT
                        ↓
                   RESPONSE

Maestro Decision Examples

Inter-Agent Collaboration

With creaon/genon structure (Element 8) and Maestro orchestration, holons can now collaborate mid-reasoning:

Collaboration Patterns

Pattern 1: Request for Historical Context

Specialist is analyzing carbon credit methodology
            ↓
Specialist (genon) → Request to Archivist: "Have we discussed this topic before?"
            ↓
Archivist (creaon) ← Request received
            ↓
Archivist (genon) → "Yes, 3 weeks ago user expressed skepticism about Verra methodology"
            ↓
Specialist (creaon) ← Context received
            ↓
Specialist produces contextualized analysis

Pattern 2: Retrieval Failure Recovery

Specialist attempts RAG retrieval
            ↓
Specialist (genon) → Poor results (low relevance scores)
            ↓
Maestro detects failure
            ↓
Maestro reformulates query → Web search instead
            ↓
Specialist (creaon) ← Better information
            ↓
Specialist produces accurate response

Pattern 3: Freethinker Challenge

Generalist produces initial synthesis
            ↓
Generalist (genon) → Draft to council
            ↓
Freethinker (creaon) ← Detects institutional claim in draft
            ↓
Freethinker (genon) → Challenge: "This relies on unverified corporate data"
            ↓
Secretariat (creaon) ← Both perspectives
            ↓
Secretariat produces synthesis that acknowledges tension

Current Architecture vs. SOHO Architecture

Implementation Phases

MVP (Current):     Query → All 5 Council Agents Sequential → Secretariat Synthesis

Phase 1.5:         Query → Router → [Full Council | Archivist-Only | Quick-3]
                   (Lightweight classification, 3-4 pipeline variants)

Phase 2.0:         Query → Maestro → Dynamic Multi-Step Orchestration
                   (Full SOHO self-organization, inter-agent collaboration)

Note: The 5 council agents are Generalist, Specialist, Freethinker, Archivist, and Secretariat. The Maestro (6th agent, added in Phase 2.0) is the meta-holon orchestrator that sits above the council and performs exergy assessment, attractor recognition, propensity activation, and dynamic routing.

Interactive Architecture Flow

Part 4: The AGI Connection

Why SOHO Systems May Be Necessary for AGI

The Central Argument

Current AI (including GPT-4, Claude) operates via:

1. Pattern recognition from training data
2. Statistical next-token prediction
3. No persistent identity across conversations
4. No genuine adaptation to operating environment

This produces impressive performance but lacks:

• Autonomous goal formation
• Continuous learning from experience
• Persistent identity and growth
• Genuine understanding (vs. pattern matching)

SOHO Systems Exhibit:

1. Self-Organization: Create their own goals and structures (Propensities, Attractors)
2. Persistent Identity: Maintain coherence over time through Canon
3. Genuine Adaptation: Evolve through Morphogenetic Loops
4. Emergent Properties: Higher-level capabilities not present in components (Holarchy)
5. Contextual Intelligence: Appropriate responses to specific situations (Maestro, Model/Narrative modes)
6. Causal Completeness: All four Aristotelian causes present, including Final Cause (telos)

AGI Benchmarks SOHO Enables

The Consciousness (Meta-Mindfulness) Question

Kay et al. describe SOHO systems as exhibiting "self-aware" properties:

• Monitor own state (Maestro)
• Adjust behavior based on self-assessment (Attractor recognition)
• Maintain identity over time (Canon)
• Respond to threats to identity (Threshold management)

Why This Matters

Risks and Challenges

Mitigation Strategies

• Extensive logging: Maestro decisions, propensity weights, attractor states all visible and auditable
• User override: Allow emergency intervention at any stage
• Gradual rollout: MVP → Router → Full Maestro phased deployment
• Diverse testing: Expose system to wide range of contexts
• Open research: Share findings for community safety analysis