Home

AI is not about prediction.
It is about decision.

Where AI comes from — and where it is going

When I was building AI solutions at Xerox,
we were pursuing a certain vision.

An AI like Astro Boy—

an entity that could think,
make decisions,
and act like a human.

But in reality, AI took a different path.

It evolved into systems that predict,
classify, and optimize—

machines that produce answers from data.

And along the way,
something began to disappear.

It was the question:

“Who is making the decision?”

What should be done
Where should the system stop
Who takes responsibility

Decisions that were once made by humans
became buried inside systems.

AI can generate answers.
But that alone is not enough.

What we need is:

to bring decision back.

To involve humans,
to take responsibility,
and to treat decision as a structure.

That is why we need:

Decision Trace Model

This is not about making AI smarter.

It is about transforming AI
from a system that produces outputs
into one that executes decisions.

And ultimately,

it is a design
to bring human decision-making
back to the center of the system.

Source: Fujifilm Business Innovation (Facebook, February 8, 2019)

This video represents a time when AI was imagined as a “dream.”

But in reality, AI has evolved in a different direction.

That is why now,
what we need is not more intelligence,

but a system.

Decision Trace Model
treats decision-making as a structured process:

Event → Signal → Decision → Boundary → Human → Log

The output of AI is only a “Signal.”

What matters is:

what decision is made,
how it is executed,
and where control is returned to humans.

Decision Trace Model
explicitly defines this process,

making decisions visible,
enabling human involvement,
and restoring accountability.

Introduction

AI today has made significant advances in prediction, classification, and generation.

In recent years, with the evolution of development environments such as Claude Code,
these capabilities can now be implemented by anyone, in a short period of time.

In other words,

AI outputs (Signals) are rapidly becoming commoditized.

However, what is truly required in real-world practice is not the output itself.

It is:

“What should be decided, and how should it be executed?”

This is where most AI systems stop.

They can predict.
But they do not define what to do.

In other words, the structure of decision-making is missing.

This site presents a way of thinking that reconstructs AI
—not as a mechanism for generating outputs—
but as infrastructure for decision-making.

Decision Structure

Decision-making inherently follows this flow:

Event → Signal → Decision → Boundary → Human → Log

• Event: What happened
• Signal: AI prediction
• Decision: What to do

This is how decisions actually work.

This is not just a concept.
Decision-making is already executed in this structure.

Decision Trace : the core concept of treating decision-making as a structured process

AI is not for prediction.
It is for decision-making.

This demo shows how decisions are structured, executed, and recorded.

This system transforms real-world changes into a clear and traceable decision-making flow.

Raw Change → Signal Extraction → Decision → Boundary → Human → Log

Unlike traditional AI, where decision logic is hidden inside the model,
this approach externalizes decision-making logic and treats it explicitly.

Each step is:

• Traceable — You can see how the decision was made
• Explainable — The reasoning is visible as a structure
• Executable — Decisions directly lead to actions
• Governable — Constraints and human intervention can be integrated

This is not just AI.
It is a decision-making system.

Multi-Agent : a decision-making system based on role-based decomposition

This demo shows how multiple AI agents, each with a specific role, collaboratively generate a single decision.

• Signal Agent: Extracts context and intent
• Decision Agent: Selects the optimal action
• Risk Agent: Evaluates risks and constraints
• Execution Agent: Initiates the execution process

The key point is that AI is not making decisions as a single entity.
Instead, decision-making is decomposed and handled by specialized roles.

This structure significantly improves:

• Transparency of decisions
• Reproducibility
• Ability to continuously improve

Architecture : the system structure for implementing decision-making

This section organizes how AI performs decision-making as a system architecture.

Traditional AI focuses on prediction and generation by individual models.
However, in real-world operations, what matters is the process of deciding what to do.

In this architecture, decision-making is enabled by combining the following components:

• Ontology — Defines meaning
• DSL (Domain-Specific Language) — Describes decision conditions
• Behavior Tree — Controls execution flow
• GNN (Graph Neural Network) — Learns relationships

By integrating these elements, the system enables decision-making to be structured, described, and executed.

Use Cases : real-world applications in business operations

With Decision Trace Model × Multi-Agent,
AI evolves from a tool for analysis and prediction
into a system that supports and executes decision-making.

Across various domains, the following transformations occur:

Manufacturing

• Anomaly detection → Automated response decisions
• Quality inspection → Shipment approval support
• Equipment maintenance → Decision-making for repair timing

On-site decisions become structured and reproducible

Retail

• Demand forecasting → Ordering decisions
• Customer analysis → Action selection
• Campaigns → Optimization of distribution and execution

Moving from revenue-only to ROI and LTV-based decision-making

Call Centers / Customer Support

• Inquiry handling → Response generation
• Knowledge search → Decision support
• Escalation → Branching decisions

Context-aware responses with clear accountability

Education

• Learning logs → Next learning decisions
• Understanding analysis → Timing of intervention
• Content recommendation → Design of learning paths

Dynamically optimizing what to teach

Legal / Compliance

• Contract review → Risk assessment
• Regulatory compliance → Identification of impact scope
• Approval processes → Visualization of decision flows

Toward explainable and auditable decision-making systems

What Changes

Traditional AI has been designed around a model-centric structure
that generates outputs from inputs.

Data is provided as input
The model processes it
An output (prediction or generation) is returned

This approach is powerful.

And today, such systems themselves have become easy to implement.

However, there is a critical problem.

AI can produce outputs,
but it does not define decisions.

As a result:

We do not know why a decision was made
Accountability becomes unclear
The same situation can lead to different outcomes

In practice, what is required is not just prediction,

but decision-making —

determining what to do.

In the Decision Trace Model,
decision-making is treated not as something embedded within a model,
but as an explicit structure.

Event: What happened
Signal: Interpretation of the situation (e.g., LLM outputs)
Decision: What action to take
Boundary: Constraints and policy checks
Execution: Action is carried out
Log: The decision process is recorded

By decomposing decision-making in this way,

• Transparency of decisions
• Reproducibility
• Continuous improvement

become possible.

AI shifts from something that predicts,
to a system that executes decisions.

■ Beyond Knowledge: The Accumulation of Decisions

Traditionally, AI adoption in enterprises has focused on:

• Data accumulation
• Knowledge sharing
• Improving searchability

In other words, managing knowledge.

However, what truly matters in real-world operations is:

What decisions were made,
based on that knowledge.

In the Decision Trace Model,

• Under what circumstances
• Based on which Signals
• What decisions were made
• Where human intervention occurred

are all captured as structured data.

This is not just knowledge.

It is the history of decisions.

And this accumulation of decisions
becomes a true organizational asset.

Decisions that can be reproduced
Decisions that can be improved
Decisions that can be transferred

As these accumulate,

the organization’s decision-making capability itself evolves.

Latest Insights

• What is Cognitive Orchestration? — A Cognitive Architecture of Stability × Creativity × Variation Extending Reinforcement Learning
  Introduction AI has traditionally been discussed within two primary frameworks: Prediction Reinforcement Learn […]
• Why the Same AI Leads to Different Decisions — The Invisible Design of Decision Priorities
  Introduction In the past, I implemented AI using Light DTM (Decision Trace Model) in two very different domain […]
• Lightweight DTM for Building “Decision-Capable AI” — Can Customer Support Truly Be Automated?
  ■ Introduction When I talk about the Decision Trace Model (DTM), I often hear: “That kind of rich design is no […]
• What Multi-Agent Orchestrator Core Can Do — The Execution Layer of a Decision System
  Introduction To realize a Decision System,it is not enough to simply separateSignal (predictions, scores) and […]
• Why AI Can’t Make Decisions — The Reason We Built Decision Trace Ledger Core
  Introduction Most modern AI systems are designed around: Prediction Scoring Classification In other words, the […]
• Is Kaizen Only Possible by Humans? Decision Trace Model × Multi-Agent Systems Unlock the Next Evolution of Improvement
  In manufacturing, improvement (Kaizen) holds a special meaning. Insights from the shop floor Ingenuity of work […]

Technical Reference

The Decision Trace Model is not just a concept.
All of its components can be implemented as code.

• Ontology (definition of meaning)
• DSL (decision conditions)
• Behavior Tree (execution control)
• Multi-Agent (role-based decomposition)
• LLM integration (signal generation)

All of these are explained with actual code examples.

AI is not only about predicting the future,
but about supporting decision-making in the present moment.