Introduction
Manufacturing is now at a major turning point.
- Retirement of veteran engineers
- Severe labor shortages
- Increasing complexity of products and processes
What has supported the field until now is:
👉 Human judgment embedded in individuals
However, this foundation is now being lost.
The Limits of Traditional Knowledge Transfer
Many companies have long worked on knowledge transfer:
- Manuals
- Procedural documents
- Training videos and educational content
Yet, the following challenges remain:
① Unwritten Decisions
- Why was this action taken?
- Under what conditions would the decision change?
👉 Tacit knowledge is lost
② Knowledge That Is Not Used
- Too busy to check
- Hard to find
👉 Knowledge becomes unused assets
③ Non-Reproducible Decisions
- Same procedure, different results
👉 Knowledge fails to function
④ Expert “Intuition” Cannot Be Reproduced
- Subtle anomalies
- Experience-based judgment
👉 Cannot be formalized
The Core Problem
The root cause of these issues is:
👉 Knowledge is stored as information,
not as a process of judgment
Most knowledge captures:
- Results (what was done)
- Procedures (how it was done)
But in reality, what matters is:
- How the situation was interpreted
- What options were considered
- Why a specific action was chosen
👉 The flow of thinking and judgment
Ultimately, this leads to:
👉 Decision-making
However, current knowledge systems store:
- Results → retained
- Decision process → lost
As a result:
- Decisions cannot be reproduced
- Knowledge is not utilized
Decomposing “Intuition” into Structure
Expert judgment is often called “intuition.”
But in reality, it is not vague.
👉 It is the result of multiple evaluation processes integrated at high speed
For example, in equipment failure handling:
What happens in an expert’s mind:
- “This vibration is unusual”
- “I’ve seen something similar before”
- “This may lead to failure”
- “Stopping now will impact production”
- “Safety limits are near”
👉 All evaluated simultaneously
Thus, intuition is:
👉 Not a single decision,
but a multi-layered integration of evaluations
Key Insight
These evaluations are:
👉 Integrated and compressed in the human mind
- Individual judgments exist
- But cannot be externally decomposed
👉 This “compressed judgment” is what we call intuition
Why It Cannot Be Reproduced
Traditional knowledge systems:
- Record only final actions
- Do not capture intermediate evaluations
👉 Only compressed results are stored
👉 Internal structure is lost
Thus:
👉 The reasoning process becomes a black box
Resulting in:
- Inconsistent decisions
- Inability for newcomers to reproduce
- No continuous improvement
Decomposition via Multi-Agent
Decision Trace Model × Multi-Agent:
👉 Decompresses “intuition” into structured processes
Expert judgment is:
👉 A compressed integration of multi-layer evaluations
DTM × Multi-Agent:
👉 Expands it into role-based processes
Decomposed Structure
- Signal Agent
State recognition (anomaly detection, feature extraction) - Diagnosis Agent
Cause estimation - Decision Agent
Action selection - Policy Agent
Constraints (safety, quality rules) - Risk Agent
Risk evaluation - Execution Agent
Action execution
👉 These were originally integrated unconsciously in experts
What Changes
Through decomposition:
👉 Compressed judgment becomes visible structure
Traditional:
👉 Intuition = too integrated to analyze
DTM × Multi-Agent:
👉 Intuition = a set of decomposable processes
👉 Invisible structure becomes manageable
The Most Important Point: Logging
The key transformation is:
👉 The decomposed process is recorded as-is
Traditional:
- Only final actions remain
- Thought process disappears
DTM:
- What signals were observed
- What hypotheses were considered
- What constraints were applied
- What risks were evaluated
- Why a decision was chosen
👉 Everything is recorded as a process
Fundamental Differences from Traditional Knowledge
DTM × Multi-Agent transforms knowledge itself.
① Structuring Tacit Knowledge
Traditional:
- Experience remains personal
- Reasoning is not recorded
New:
- Signal
- Evaluation
- Decision
- Policy / Risk
👉 Fully recorded
② Reproducible Decisions
Traditional:
- Decisions vary by person
New:
- Same input → same decision process
👉 Reproducibility ensured
③ Real-Time Knowledge Utilization
Traditional:
- Search manuals
New:
- AI retrieves relevant decision traces
👉 Knowledge used in real-time
④ Continuous Learning
Traditional:
- Static knowledge
New:
- Logs accumulate
- Patterns evolve
- GNN optimizes structure
👉 Decisions evolve
⑤ Robust Operations
Traditional:
- Black-box decisions
- High risk
New:
- Policy + Risk + Human boundary
👉 Safe and accountable decisions
Practical Use Cases
Equipment Failure
Event → Signal → Diagnosis → Risk → Policy → Decision
👉 Reproducible responses
Quality Issues
Signal → Diagnosis → Decision → Log
👉 Structured knowledge accumulation
Process Improvement
Signal → Diagnosis → Evaluation → Decision
👉 Reusable improvement logic
Training
- Decision traces shown
- Reasoning visualized
👉 Learn how to decide
Business Impact
- Faster knowledge transfer
- Stable quality
- Increased productivity
- Reduced risk
- Competitive advantage
Conclusion
Decision Trace Model × Multi-Agent:
- Visualizes decisions
- Enables reuse
- Drives continuous evolution
It transforms knowledge from:
👉 Something to read
into:
👉 Something to act upon
The essence is:
👉 Redefining knowledge itself
Traditional:
👉 Knowledge = documents
New:
👉 Knowledge = decision systems
What matters is not:
👉 What you know
but:
👉 How you decide
Ultimately:
👉 Intuition becomes a reproducible organizational asset
Final Message
The future of manufacturing is not about preserving knowledge.
👉 It is about reproducing decisions.
Decision Trace Model × Multi-Agent transforms manufacturing into:
👉 A system where decisions are accumulated, reused, and evolved.
This is the structural transformation it brings.
AIシステム設計・意思決定構造の設計を専門としています。
Ontology・DSL・Behavior Treeによる判断の外部化、マルチエージェント構築に取り組んでいます。
Specialized in AI system design and decision-making architecture.
Focused on externalizing decision logic using Ontology, DSL, and Behavior Trees, and building multi-agent systems.
