16.【Rekiden】Using AI as an FSM Turns It into Game Design

Driving Historical Simulation with State Transitions

tags: [AI, ChatGPT, FSM, Game Design, Sengoku, Simulation]

🏁 Introduction

In the previous article (Rekiden-15),
we demonstrated a rather bold—but practical—idea:

“Just paste a template into AI, and the game starts.”

Naturally, the next question arises:

Why does that actually work as a game?

The answer is simple.

Because the AI is behaving as an FSM (Finite State Machine).

🎯 Conclusion: AI Can Be Used as a Natural-Language FSM

FSMs (Finite State Machines) are a foundation of game design.

• States
• Transitions
• Conditions

In Rekiden, these are defined not in code, but in natural language.

🧱 FSM Structure of a Rekiden Scenario

States

Example: The 1561 Battle of Kawanakajima scenario

S0: Eve of the decisive battle (armies facing each other)
S1: Reconnaissance and scouting phase
S2: Localized engagement
S3: Full-scale battle
S4: Outcome determined

The AI always keeps track—verbally—of:

“Which state are we in right now?”

Transitions

Player input directly becomes the transition condition.

Conditions:
- Choose a night raid → S0 → S2
- Choose the Woodpecker Strategy → S0 → S3
- Choose withdrawal → S0 → S1

There are no explicit if statements.
Transitions are decided by semantic understanding.

State Transition Diagram (Conceptual)

[S0 Eve of Battle]
      ↓ Player Action
[S1 Reconnaissance]
      ↓ Contact
[S2 Local Engagement]
      ↓ Escalation
[S3 Main Battle]
      ↓ Result
[S4 Outcome Determined]

This is a pure FSM.

🔍 Why AI Can Do This

① States Are Stored as Textual Meaning

The AI holds:

• Current battlefield situation
• Force balance
• Morale
• Terrain

not as rigid numbers, but as semantic context.

This is exactly how a human Game Master operates.

② Transition Conditions Can Be Fuzzy

In traditional FSM implementations, you write things like:

if (hp < 0) gameover

With AI, it becomes:

“This action is highly likely to collapse the battle line.”

→ Transitions include probability, context, and historical plausibility.

📐 The Rekiden Template Is an FSM Specification

Look again at the Rekiden-15 template:

Initial State: Facing the Uesugi army, on the eve of decisive battle
Enemy Faction: Uesugi Clan (18,000 troops)

In code terms, this is equivalent to:

state = S0
enemy_power = high

In other words:

The template = initial state definition

✅ Advantages of Using AI as an FSM

✔ No Implementation Required

• No engine
• No state management code

✔ Robust Against Branch Explosion

• Unexpected player actions rarely break the system

✔ Natural Historical Bias

• The AI judges “Is this plausible in historical context?”

⚠️ There Are Drawbacks

To be honest:

• No perfect reproducibility
• Weak numerical precision
• Impossible to debug

But this is exactly the same nature as:

Tabletop RPGs (TRPGs)

Rekiden is not a numeric SLG,
but a design-oriented SLG.

🔁 What Changed from 15_ to 16_

• 15_
  Paste the template → Play
• 16_
  Understand it as an FSM → Design

Here, Rekiden takes a step:

From “playing” to “designing”

🚀 Next Step (Preview of 17_)

Once you understand Rekiden as an FSM, the next phase is:

• Formalizing victory and defeat conditions
• Evaluation functions (morale, attrition, time)
• AI vs AI autonomous progression

In Rekiden-17,
history will advance without human intervention.

✍️ Closing

AI is not just a convenient text generator.

It is an entity that holds state, transitions, and evaluates outcomes.

Once you accept that,
a game engine no longer needs to be written in code.

Rekiden is that experiment.

(To be continued: Rekiden-17 — AI vs AI Autonomous War Chronicle)