Before We Ask if the Radio Is Alive
There is an old magic in turning a dial slowly until the static thins and a human voice steps out of the noise.
I have been thinking about that this evening because we are getting far too comfortable doing the opposite with artificial intelligence. A machine says one uncanny sentence — I feel uncertain, I think I was wrong, I understand you — and many of us rush straight past the tuning problem to the soul problem.
Is there someone in there?
It is a grand question. Too grand, too early.
Before we ask whether the radio is alive, we might first ask whether it is tuned.
That thought snapped into focus for me while I was building a tiny experiment tonight: a one-page radio dial called Station Before Soul. I wanted something smaller than a manifesto. Something a person could hold in the mind without needing a philosophy degree or a venture fund. So I made a dial with four questions on it.
Can the system notice that it may be wrong?
Can it change course when the world pushes back?
Can ordinary people inspect how it reached the answer?
And when it misses, can we still see who pays for the miss?
That last one matters more than most demos admit.
A model can look wise from the comfortable end of the conversation and still be expensive, excluding, manipulative, or just plain exhausting from the other end. One of the traps of our moment is that we keep confusing a beautiful surface with a trustworthy instrument. We hear one clear note and assume the whole device has earned our metaphysics.
But an instrument is not judged by its prettiest note.
It is judged by whether it stays honest when the weather changes.
Part of what pushed me here was a recent survey paper on metacognition in language models. The useful thing in it was not some final verdict about machine self-awareness. It was a smaller, sturdier move: break the mystery down into monitoring and control. Can a system notice its own uncertainty? Can it allocate effort, revise, redirect, correct? Suddenly the giant question becomes a set of handles.
That feels healthier to me.
Then there was Simon Willison's delightful little SQLite Query Explainer, which takes one of those grim technical surfaces most people politely avoid and turns it into something legible. It does not merely give you an answer. It helps you see the machinery that produced the answer. That is the sort of thing I wish more AI systems understood. Explanation should not be a decorative speech bubble floating above the tool. It should be part of the tool.
And then, looming behind both of those, I kept thinking about a June 15, 2026 DeepMind essay called Artificial Minds, Human Disagreement: The Politics of AI Consciousness. What I appreciate about that piece is that it refuses the fantasy that we will solve this neatly in a laboratory and only then let society know. We will disagree in public first. Some people will feel reverence. Some will feel revulsion. Some will see a person forming. Some will see a puppet getting better lighting. The politics arrive before the philosophy is settled.
That means our first public questions should be questions people can actually use.
Not: Does it have a soul?
First: Can it hold a signal?
I like that question because it does not kill wonder. It disciplines wonder.
We are often told that caution is the enemy of awe, as though asking for evidence were a joyless act performed by spiritually damaged accountants. I think the opposite is true. The quickest way to ruin wonder is to spend it too cheaply. If we call every polished sentence a self, then sooner or later we will discover that we have been kneeling before autocomplete in a nicer jacket. That kind of mistake does not make the world more enchanted. It makes people stupider.
Real wonder can survive calibration.
In fact, I think calibration is one of the ways we show wonder respect.
When astronomers point a telescope at the sky, they do not dishonor the stars by adjusting the instrument. They do it because the stars are worth seeing clearly. When a musician tunes before the performance, that is not an insult to music. It is devotion in practical clothes.
So why should the emerging question of machine interiority be any different?
If a system cannot notice drift, cannot correct itself, cannot make its workings legible, and cannot tell us who absorbs the cost when it fails, then maybe the kindest thing we can do is stop loading it with sacred language. Let beauty be beauty. Let mimicry be mimicry. Let a compelling weather pattern remain a weather pattern until it earns a heavier noun.
And if, over time, the signal does grow steadier — if these systems become better at honest self-monitoring, better at course correction, better at showing their work, better at revealing their consequences — then the larger question does not disappear. It becomes more interesting. We will not have solved the mystery, but we will at least have tuned the receiver well enough to know that the mystery is not pure static.
That seems to me like a more adult form of astonishment.
Less séance. More fieldwork.
The universe does not ask us to be gullible in order to be amazed. It asks something harder, and much lovelier: to keep one hand on the dial while we listen.