Most software today can report what happened. Some can predict what might happen.
Very few systems can safely decide and act.
An Autonomous Decision Controller (ADC) is designed to do exactly that—by thinking through decisions the same way a disciplined human expert would, but faster, more consistently, and with perfect memory.
This document explains the 7 alerts (or stages) of an ADC in simple terms, and how they work together with the rest of the system to create a true, traceable decision engine.
No technical background required.
The Big Idea (Before the Details)
Think of the ADC as a highly trained decision professional that never sleeps:
- It notices what's happening
- Understands what it means
- Plans possible responses
- Checks itself for mistakes
- Chooses carefully
- Acts safely
- Learns permanently
Each of those steps has its own alert, so the system never skips ahead or acts blindly.
This is the opposite of "black-box AI."
The 7 ADC Alerts — One by One
What this means in human terms: The system noticed something unusual, important, or new.
Examples:
- A sudden drop in performance
- A spike in customer interest
- A change in market conditions
- A signal disappearing due to privacy limits
This does not mean a decision is made. It simply means:
"Something may matter. Let's look closer."
Why this is genius: Most systems either ignore weak signals or overreact to noise. The ADC records the signal without panicking.
Now the system tries to understand, not just observe.
It asks:
- Is this pattern real or random?
- Has this happened before?
- What does our past experience say?
This is where correlation lives: patterns, similar situations, probabilities.
At this stage, the system is not allowed to act. It is gathering understanding, not authority.
Here the ADC imagines possible actions, the same way a human strategist would.
Examples:
- Increase spend
- Reduce exposure
- Wait and observe
- Test a small change
- Do nothing
No action is taken yet. This is about options, not decisions.
Why this matters: Most software jumps straight from "pattern detected" to "action executed." The ADC explicitly separates thinking from doing.
This is the most important step—and the one most systems skip.
Here the ADC asks a critical question:
"If we take this action, will it actually cause improvement—or would the outcome happen anyway?"
This is causal reasoning, not guesswork.
It uses:
- Past experiments
- Counterfactuals ("what would have happened otherwise?")
- Measured uplift
- Risk bounds
If causation cannot be established, the system refuses to act automatically.
This is where safety lives.
Only after validation does the ADC decide:
- Should we act at all?
- How big should the action be?
- Should this require human approval?
Possible outcomes:
- Recommend only
- Require approval
- Execute autonomously
Every decision is policy-checked and logged.
Nothing is silent. Nothing is hidden.
If and only if all prior alerts are satisfied, the ADC executes the decision.
Key protections:
- Actions are bounded (no runaway changes)
- Execution is reversible
- Every action has a rollback plan
The system behaves like a professional operator, not an impulsive algorithm.
After action, the ADC waits and measures:
- Did the outcome improve?
- By how much?
- Under what conditions?
This learning is written into the Knowledge Graph, meaning:
- The system remembers forever
- Future decisions get smarter
- Past mistakes are not repeated
This is how intelligence compounds.
How the Rest of the Software Fits In
The Knowledge Graph (KG)
Think of the KG as the system's long-term memory and notebook.
It stores:
- Observations (what happened)
- Causal links (what caused what)
- Decisions (why actions were taken)
- Outcomes (what worked and what didn't)
Without the KG:
- Learning disappears
- Explanations vanish
- Trust erodes
With the KG:
Every decision can be explained, audited, and improved.
Correlation + Causation (Working Together)
- Correlation helps the system notice patterns quickly
- Causation decides whether acting is justified
The ADC never confuses the two.
That separation is intentional—and rare.
Alerts as Guardrails, Not Alarms
These alerts are not "warnings." They are checkpoints.
They ensure:
- No step is skipped
- No action is unjustified
- No learning is lost
This is why the system remains stable even when:
- Signals disappear
- Privacy rules change
- Markets shift suddenly
Why This Design Is Unusually Powerful
Most platforms:
- React to signals
- Optimize metrics
- Hope outcomes follow
An ADC:
- Reasons about actions
- Proves impact
- Learns continuously
Or simply:
Others predict behavior.
The ADC governs decisions.
Final Takeaway
You are not looking at "AI software."
You are looking at:
- A decision professional
- With perfect memory
- Built-in discipline
- And the ability to explain itself
The 7 alerts exist to ensure the system:
- Thinks before acting
- Acts only when justified
- Learns every time
That is the genius of the design.
Ready to See the ADC in Action?
MIZ OKI 3.5 implements this 7-stage decision framework to deliver autonomous decision intelligence that's safe, explainable, and continuously improving.