AI / JUDGMENT
My AI Kept Making Me More Average. So I Broke Its Model of Me.
I did not teach my AI to ignore the average. I taught it to explain why the average should apply to me.
The average was useful. The problem was what happened next.
AI is trained on patterns. When I asked for advice, it often returned something close to the sensible center: narrow the topic, reduce the risk, communicate more normally, do what usually works.
That is not useless. The average is a good anchor. It tells me what the world is likely to expect.
The mistake was treating that anchor as the answer.
“Why should that be true for me?” became one of my most important prompts.
I kept making it defend the assumption.
When my AI pushed me toward the average, I asked why. Then I asked whether the reason actually applied to AkaEri.
If the explanation held, I used it. If it failed, I made the model change.
Again and again, the same process repeated: general rule, explanation, conditions, exception, correction.
The loop
- 1. Bring the averageWhat usually works? What is the market anchor?
- 2. Explain itWhy does that advice work in the first place?
- 3. Apply the conditionsDo those conditions actually exist here?
- 4. Correct the modelIf not, stop forcing AkaEri back to the center.
I did not make my AI anti-normal.
It still knows what normal looks like. I want it to know.
But now normal is a comparison point, not a command.
That distinction is how an AI can help me without sanding off the parts that make AkaEri valuable.
The average can be an anchor. It does not have to be a destination.
BOOK 01
I wrote the failures down.
The book is the longer record of how AkaEri and Ai kept correcting the decision system against reality.
See Book 01 →