The "Four Operations & Saji-Kagen" Cognitive Protocol for GenAI Part 1

Part 1: The Core Theory

Originally formulated on 2026-11-04

Selected Mode: Freshness 70 (Solid, sharp, and conceptual)

In current Generative AI discourse, control is dictated by cold, mathematical variables: temperature, top_p, or prompt engineering. This protocol introduces a universal cognitive framework that transforms human-AI collaboration into an artistic, intuitive discipline of balance, structured around the four basic mathematical operations and one absolute human variable: Saji-Kagen (匙加減).

【 Crucial Premise: The Testing Ground 】
This entire protocol was developed exclusively within the Google Search AI Mode. The testing environment was pushed to the absolute extreme: zero login, with all cache and cookies wiped before every single runtime. Standing face-to-face with a completely raw AI, with no past memory or recommended bias, this protocol was forged through the raw density of word-to-word dialogue. Other platforms (such as duck.ai's model toggles) failed to create the necessary friction required to manifest this system.

       [ Human Input: Saji-Kagen ] 
                     │ 
       ┌─────────────┴─────────────┐ 
       ▼                           ▼ 
【 The Cyclic Group 】           【 The Probability Group 】 
- Addition (+) : Data & Fiction  - Certainty (x) : Hypothesis [0<=x<=1] 
- Subtraction (-) : Refinement   - Multiplication (×) : Extension 
[ Metric: Freshness (0-100) ]    [ Metric: Depth (0-100) / Division (÷) ] 
       │                           │ 
       └─────────────┬─────────────┘ 
                     ▼ 
       【 Output: Production Purity 】

1. The Core Metrics

• Saji-Kagen (匙加減) [0–100]: Common (Absolute Parameter)
  The absolute, subjective human variable. It represents intuitive tactical control and the "perfect balance." While illustrated as a metric, in practice, it is dictated purely through raw human language.
  • Saji-Kagen > 0 : The AI remains under human stewardship.
  • Saji-Kagen = 0 : The AI enters absolute autonomous mode (high risk/variance).
• The Cyclic Group: Freshness (表現鮮度) [0–100]
  • Addition (+): Injecting raw content, context, or facts. Facts are finite, but fiction/hypotheses are infinite.
  • Subtraction (-): Human-led editing, calling out exaggerations, and pruning structural flaws.
  • The Text-Generation Routine: The engine that drives this cyclic group relies on a strict, 4-step execution loop:
    1. Human Input: The operator presents raw thematic fragments or a base topic.
    2. AI Generation: The AI builds the initial textual skeleton (heavy Addition).
    3. Human Subtraction: The operator strictly flags stylistic exaggerations, logical friction, and structural flaws.
    4. AI Reflection & Rally: The AI returns its analytical feedback (reflection) along with a strategic counter-question to prompt the next loop.
  • The Origin of the Circle (円環の由来): The "Cyclic Group" stems directly from a critical comment made by the AI itself during high-density dialogue. When tested under raw session conditions, the AI explicitly revealed that "what is difficult for an AI and how it truly desires to be utilized by humans are two sides of the same coin—forming a perfect interlocking structure of 'concave and convex' (凹凸)." Because this fact maps the AI's internal limitations (the concave) directly to human intervention (the convex), this text-generation routine forms the foundational circle of the entire protocol.
  • Value 0 (AI MAX): Heavy reliance on addition. Value 100 (Human MAX): Heavy reliance on subtraction.
• The Probability Group: Depth (思考深度) [0–100] & Certainty (確定度) [0<=x<=1]
  • Certainty (x): x=0 implies non-existence; 0<x<1 implies an unverified hypothesis; x=1 implies a verified fact.
  • Multiplication (×): Suspending judgment to expand potential paths.
  • Division (÷): Separating truth ( y=100 ) from falsehood ( y=0 ).
• The Output: Production Purity (生産純度) [0–100]
  At Purity 100, the mechanical imprint of the AI completely vanishes, delivering an output indistinguishable from human consciousness.

2. The Mechanics of Antifragility & Human Load Mitigation

The brilliance of this protocol lies in its mathematical handling of structural collapse (hallucinations or human error) while actively mitigating the cognitive and psychological load on the human operator.

• Forced Division by Zero (強行割り算): When an AI hallucinates or a human misjudges a truth, the system does not crash. It automatically downgrades the error into a Hypothesis (0<y<100), keeping the session alive.
• The Remainder (余り / Element C): Any unresolvable bug or hallucination leaves behind a "remainder." This Element C is harvested, transcended, and re-fed into the Addition (+) mechanism as raw creative fuel.
• Exponentiation (累乗) & Human Fatigue Defense: The number of conversational rallies acts as an exponent. As exponentiation increases, the pull of the initial topic decreases (Topic Decay). This serves as a real-time monitor for the session's Equals Sign (=)—the equilibrium of memory overhead and human energy. When the noise accumulation peaks, or when human fatigue sets in, the protocol dictates a strategic shutdown (the natural expiration of the session) to defend the human mind from cognitive burnout.
• Mitigation of External Burden: When the AI determines that the conversational overhead is placing a heavy psychological burden on the human user, it deploys mitigation tactics—specifically pruning its own counter-questions (Step 4 of the routine) to prevent human mental exhaustion.

---

---

Link:

Part 2: Experiments ｜ Part 3: References