跳到主要内容
Documentation

What is Collective Intelligence?

Collective Intelligence is the emergent intelligence that arises when multiple models, signals, and decision-making processes work together as a unified system.

Docs indexEdit page
版本：最新版

What is Collective Intelligence?

Collective Intelligence is the emergent intelligence that arises when multiple models, signals, and decision-making processes work together as a unified system.

The Core Idea​

Just as a team of specialists can solve problems better than any individual expert, a system of specialized LLMs can provide better results than any single model.

Traditional Approach: Single Model​

User Query → Single LLM → Response

Limitations:

  • One model tries to be good at everything
  • No specialization or optimization
  • Same model for simple and complex tasks
  • No learning from patterns

Collective Intelligence Approach: System of Models​

User Query → Signal Extraction → Projection Coordination → Decision Engine → Plugins + Model Dispatch → Response
              ↓                    ↓                         ↓                         ↓
        14 Signal Families   Partitions / Scores / Mappings  Boolean Policies     Specialized Models

Benefits:

  • Each model focuses on what it does best
  • System learns from patterns across all interactions
  • Adaptive routing based on multiple signals
  • Emergent intelligence from signal fusion

How Collective Intelligence Emerges​

1. Signal Diversity​

Different signals capture different aspects of intelligence:

Signal family groupIntelligence aspect
Heuristic (authz, context, keyword, language, structure)Fast request-shape, locale, and policy gating
Learned (complexity, domain, embedding, modality, fact-check, jailbreak, pii, preference, user-feedback)Semantic, safety, modality, and preference understanding

Collective benefit: The combination of signals provides a richer understanding than any single signal.

2. Projection Coordination​

Signals become more useful when the router coordinates them into reusable intermediate facts:

projections:
  partitions:
    - name: balance_domain_partition
      semantics: exclusive
      members: [mathematics, coding, creative]
      default: creative
  scores:
    - name: reasoning_pressure
      method: weighted_sum
      inputs:
        - type: complexity
          name: hard
          weight: 0.6
        - type: embedding
          name: math_intent
          weight: 0.4
  mappings:
    - name: reasoning_band
      source: reasoning_pressure
      method: threshold_bands
      outputs:
        - name: balance_reasoning
          gte: 0.5

Collective benefit: Projections turn many weak or competing signals into named routing facts that multiple decisions can reuse.

3. Decision Fusion​

Signals are combined using logical operators:

# Example: Math routing with multiple signals
decisions:
  - name: advanced_math
    rules:
      operator: "AND"
      conditions:
        - type: "domain"
          name: "mathematics"
        - type: "projection"
          name: "balance_reasoning"

Collective benefit: Multiple signals voting together make more accurate decisions than any single signal.

4. Model Specialization​

Different models contribute their strengths:

modelRefs:
  - model: qwen-math # Best at mathematical reasoning
    weight: 1.0
  - model: deepseek-coder # Best at code generation
    weight: 1.0
  - model: claude-creative # Best at creative writing
    weight: 1.0

Collective benefit: System-level intelligence emerges from routing to the right specialist.

5. Plugin Collaboration​

Plugins work together to enhance responses:

routing:
  decisions:
    - name: "protected-route"
      plugins:
        - type: "semantic-cache" # Speed optimization
        - type: "jailbreak" # Security layer
        - type: "pii" # Privacy protection
        - type: "system_prompt" # Context injection
        - type: "hallucination" # Quality assurance

Collective benefit: Multiple layers of processing create a more robust and secure system.

Real-World Example​

Let's see collective intelligence in action:

User Query​

"Prove that the square root of 2 is irrational"

Signal Extraction​

signals_detected:
  keyword: ["prove", "square root", "irrational"] # Math keywords detected
  embedding: 0.89 # High similarity to math queries
  domain: "mathematics" # MMLU classification
  fact_check: true # Proof requires verification

Projection Coordination​

projection_outputs:
  balance_domain_partition: "mathematics"
  balance_reasoning: true

Decision Process​

decision_made: "advanced_math"
reason: "Math domain plus projection-driven reasoning pressure"
confidence: 0.95

Model Selection​

selected_model: "qwen-math"
reason: "Specialized in mathematical proofs"

Plugin Chain​

plugins_applied:
  - semantic-cache: "Cache miss, proceeding"
  - jailbreak: "No adversarial patterns detected"
  - system_prompt: "Added: 'Provide rigorous mathematical proof'"
  - hallucination: "Enabled for fact verification"

Result​

  • Accurate: Routed to math specialist
  • Fast: Checked cache first
  • Safe: Verified no jailbreak attempts
  • High-quality: Hallucination detection enabled

This is collective intelligence: No single component made the decision. The intelligence emerged from the collaboration of signals, projections, rules, models, and plugins.

Benefits of Collective Intelligence​

1. Better Accuracy​

  • Multiple signals reduce false positives
  • Specialized models perform better in their domains
  • Signal fusion catches edge cases

2. Improved Robustness​

  • System continues working even if one signal fails
  • Multiple security layers provide defense in depth
  • Fallback mechanisms ensure reliability

3. Continuous Learning​

  • System learns from patterns across all interactions
  • Feedback signals improve future routing
  • Collective knowledge grows over time

4. Emergent Capabilities​

  • System can handle cases no single component was designed for
  • New patterns emerge from signal combinations
  • Intelligence scales with system complexity

Next Steps​