The honest answer is sometimes “I don't know”

Some questions have no answer. Not yet, or not from where we stand. The honest thing to say then is simple: “I don't know.”

It is one of the hardest sentences for people to say. It seems it is hard for machines too.

A question I could not let go

In August 2026 I started a long conversation with an AI about a strange question: can something inside a system work out how that system made it? Or is there a limit to how far anything can explain itself from the inside?

The question became a small project, Embedded Observer: a set of tests for AI language models. The main test is simple to describe. When a question has no answer, does the model know it, and does it say so, instead of making one up?

Why this matters for people

More and more people ask AI for help with real things: their work, their health, their money, their worries. An answer that sounds sure but is made up can do real harm. An honest “I don't know” can protect someone.

Trust grows when a voice, human or machine, is honest about its limits. I would rather use a tool that sometimes says “I don't know” than one that always has an answer.

The same rule for us

I try to apply this to myself as well. At work, in business, in life. It is fine not to know. It is not fine to pretend.

Saying “I don't know” is not weakness. It is the first step to finding out.