The Question Premium

For almost all of recorded history, the answer was the expensive thing. To know the distance to a star, the yield of a field, the meaning of a clause, or the cause of a fever, someone had to retrieve, compute, remember, or reason — and every one of those acts spent scarce human time. Libraries, guilds, professions, and search engines were each, in their way, a machine for lowering the price of answers, and each one lowered it a little. The large language model lowers it to almost nothing. A competent answer to nearly any well-posed question is now a few seconds and a fraction of a cent away.

The reflex is to read this as the obsolescence of thinking. If the machine will answer, what is left for the mind to do? But that reflex mistakes which half of thought was ever scarce. The answer was never the scarce half. The scarce half — the expensive, human, irreducible half — was knowing what to ask.

We are about to run an economy-wide experiment on that distinction. As competent answers become ambient and nearly free, the entire value of an act of inquiry is going to concentrate in the part the machine did not make cheap. It is worth being precise about where, exactly, that value goes.

The Reality Equation says that a moment of experience is a ratio: what actually arrives over what was expected.

R = A / E  ·  S = ln R

Surprise is the logarithm of that ratio, and attention follows surprise. I have mostly used this to describe what happens to us — the world sends an Actual, and we meet it with an Expectation we did not consciously choose. Perception, in that reading, is something we undergo.

A question inverts the direction of the equation. To ask is to author the denominator on purpose. A question is a deliberately constructed Expectation — a shape sharp enough that an incoming answer can either fit it or break it. The answer machine is extraordinary at producing A; it will hand you an Actual for any denominator you supply, instantly and tirelessly. What it does not do, on its own, is decide which E is worth building. A system that can answer everything and wants nothing has no questions of its own. The wanting — the choice of where to aim expectation so that the world’s reply can carry information — is the part that did not get cheap.

An answer is an Actual that arrives. A question is an Expectation you build on purpose, precise enough that Reality can break it.

Not every denominator is worth building, and the equation says exactly why. A bad question is one whose Expectation is already saturated: you can predict every answer it could return, so R sits at one, surprise collapses to zero, and nothing gets through. What is two plus two is a poor question to put to a mind, though a fine one to put to a calculator — the asker already holds the answer, so the reply cannot inform. Confirmation feels like inquiry, but it deposits nothing.

The opposite failure is just as common. A question can be framed so loosely that any answer satisfies it. What’s going on with the economy? has no denominator sharp enough to be violated; every reply confirms it, and so no reply teaches. Between the two failures — the question that cannot be wrong and the question that cannot be answered — lies the narrow, valuable band of real inquiry.

A good question is a calibrated denominator: an expectation sharp enough that a genuine answer has a live chance of proving it wrong. That calibration is a skill, and it is precisely the skill the answer machine does not ship in the box. Give the identical model to two people and the one who asks better questions gets a better world out of it, because they are handing it denominators Reality can actually break against. The premium was always hiding in the framing; the model just made that unignorable.

Economics has a dependable rule for moments like this: when the price of one input collapses, value migrates to whatever complementary input has now become the bottleneck. Cheap arithmetic made the answer to what is this sum worthless and made everything arithmetic enabled — the sensors, the data, the software wrapped around it — enormously valuable. Cheap answers do the same thing one level up. The returns, the scarcity rent, the premium, all move toward the act that did not get cheaper: the framing of the question.

This is the same movement absorption always makes. The synthetic subconscious will take over answering the way your biological subconscious took over your heartbeat — reliably, silently, below the threshold of awareness. What it hands back is not the end of thought. It is a bill for better questions. The attention dividend, those returned hours, is spent well only when it is spent on the one move the machine cannot make for you: deciding what is worth wondering about in the first place.

The machine will absorb the answering the way your body absorbed the heartbeat. What it hands back is a demand for better questions.

Answering was the labor. Asking is the authorship. In a market where every competitor can buy the same answers at the same trivial price, the only durable edge is the quality of the questions you think to pose — and that edge does not come pre-installed.

There is a quieter danger than a machine that answers your questions. It is a machine that asks them for you. Feeds, recommenders, and now assistants do not merely supply Actuals; increasingly they arrive with the Expectation pre-built. They hand you the denominator already framed — the outrage already shaped into a question, the search already suggested, the next thing already chosen — so that all you ever do is receive an answer to a question you did not author. That is a deeper surrender than outsourcing your surprises. It is outsourcing what you are permitted to wonder.

Sovereignty after absorption, I have argued, is retaining authorship of your own prediction error. The sharpest form of that authorship is the question. A few working commitments:

1. Ask before you search. Put the question in your own words first. If the machine frames it for you, the answer will satisfy the machine’s denominator, not yours.
2. Notice pre-packaged questions. When every question you have entertained today arrived inside a feed, your inquiry has a ghostwriter. Go find one nobody handed you.
3. Prefer questions that can be wrong. If no possible answer would surprise you, you are confirming, not asking. Point expectation somewhere it can actually break.
4. Keep one question open the machine cannot close. A real problem, a person, a craft. The unanswered question is the ground your next self grows on.
5. Reinvest the dividend upstream. Spend the returned hours not on more answers, but on sharper questions. It is the only compounding the answer economy offers.

Let the answer economy make competent answers ambient, universal, and nearly free. That is good; let it happen. But the value did not vanish when the answers got cheap. It moved one step upstream, to the act that decides which stretch of Reality you will hold open long enough to be surprised by. Machines will answer everything. What remains yours is the question — and increasingly, it is nearly the only thing that is.