The Price of Thinking

The Price of Thinking

Why cheaper artificial intelligence — paired with machines that can replace all human physical labor — is not, on its own, good news

A note on scope

This essay is not only about artificial intelligence. It is about two abundances arriving together.

Most discussions of AI's economic consequences quietly assume that machines will get very good at thinking while humans retain a refuge in doing — that even if software writes the code, drafts the contract, and diagnoses the illness, there will still be a body needed to build the house, cut the hair, tend the fish farm, repair the machine. This essay makes no such assumption, because there is no principled reason to. The same trajectory that drives the marginal cost of cognition toward the cost of electricity is also producing general-purpose robots — humanoid or otherwise — capable of executing physical tasks across the full range of human manual work.

So when this essay says "the total replacement of human labor," it means exactly that: not the automation of thinking alone, but the simultaneous automation of thinking and acting. Intelligence that runs on any device, and robotic bodies that can perform any physical task that intelligence directs. The two together are the premise. Neither alone would produce the argument that follows; both together produce it inescapably. A world where only cognition is cheap still leaves the body as a place for human value to migrate to. A world where the body is automated too leaves nowhere. That difference is the entire subject of this essay.

Summary of the argument

There is a widespread intuition — nearly an article of faith in contemporary technological culture — that making intelligence cheaper is an unambiguous good: if thinking costs less, everyone thinks more, and the surplus spills over onto humanity as a whole. This essay argues that the intuition is incomplete and, in its optimism, dangerous. Not because cheap intelligence is bad in itself — it is not — but because its cheapening, when combined with robots capable of replacing all human physical labor, destroys the one mechanism that throughout modern history allowed a person born without property to acquire it: work.

When work ceases to be a path to property, the future does not open into a range of possibilities. It narrows to two, and only two. The aim of this text is not to instill fear of artificial intelligence, but to describe precisely why those two futures are the only internally coherent ones — and why the window for influencing which of them we inherit is narrower than it appears.

I. The fallacy of zero cost

The first error of the optimistic view is economic, and it consists in confusing zero cost with zero price. That the marginal cost of a unit of intelligence tends toward zero does not imply that its price does, nor that value disappears from the system. Value does not evaporate: it shifts. And the decisive question in any economy was never how much does it cost to produce, but who controls what remains scarce.

It is worth being honest about the strongest objection to any talk of "prices tending to zero": it is true that cost never actually reaches zero. A robot that cuts hair, another that prepares food, another that works as a biologist on a fish farm — all of them require energy, maintenance, inputs, connectivity, spare parts, depreciation. The cost persists even when not a single human remains in the supply chain. This observation is correct, and it is the legitimate starting point of the analysis. But it proves less than it appears to. That aggregate cost is positive does not fix the price to the consumer; what fixes that, under competition, is marginal cost, and who controls the bottlenecks of the system. Positive total cost and low price can coexist without contradiction, as they coexist today in any digital service.

What the cheapening of intelligence does, then, is not abolish cost. It concentrates all remaining value in a very specific and very ancient place: the ownership of what cannot be reproduced for free. Atoms. Energy. Land. The mineral deposit. When thinking is free and — crucially — acting is free, the only thing left with a price is matter itself, and whoever owns it.

That second clause is where the robots enter the argument, and why they cannot be left out. If only thinking were free but acting still required paid human hands, then human physical labor would remain a scarce, priced input — a refuge. The robot capable of any physical task is precisely what closes that refuge. It is not an incidental detail of the scenario; it is the hinge on which everything turns.

II. Work as a ladder

To understand why this concentration matters, we must recall what work was, beyond its productive function.

Work was never merely a way of producing goods. It was, above all, the mechanism by which a person born without property could acquire it. The worker's wage did not only cover subsistence: it became demand across entire markets — housing, food, education, health, transport, leisure — and that demand, multiplied by millions of workers, wove an immense web of interdependent equilibria that sustained the whole economy. Work was, at once, the ladder of individual ascent and the glue of the collective social contract.

Earlier technological revolutions — the printing press, the steam engine, the factory, the computer — destroyed occupations, but they preserved the ladder. Each eliminated certain forms of work and created others; the displaced farmhand became a factory worker, the factory worker became an office clerk, the clerk became a knowledge worker. Human cognition was revalued again and again because there was always a higher rung to climb to. The ladder lengthened, but it remained a ladder, and it still reached the ground where those with nothing were standing.

The scenario examined here is qualitatively different, and this is where the entire difference lies. It is not that artificial intelligence eliminates some human capacities and leaves others standing. It is the simultaneous replacement of the only two things a human being contributes to the chain of value: the capacity to think and the capacity to act physically. If maximal intelligence runs on any device and general-purpose robots execute any physical task, then there is no higher rung left. The ladder does not lengthen: it is pulled up off the ground. There is no new skill to learn, because any conceivable skill — mental or manual — is already covered, more cheaply, by a machine.

Barring a human learning to perform magic, alchemy, or to foresee the future, no reconversion is possible. And this is not rhetorical hyperbole: it is the exact logical consequence of a replacement that is, by definition, total across both dimensions of human contribution.

III. What remains when work is gone

Once work disappears as a producer of value, the human being does not disappear from the economy. They undergo an inversion: they cease to count for what they offer, and come to count only for what they demand.

This is a technical observation, not a sentimental one. Economic value does not reside in objects, but in the relation between an object and a subject who prefers it. A machine can produce the shrimp; it cannot want it. The entire economic system exists to satisfy preferences, and preferences are the one input that neither the most advanced intelligence nor the most advanced robotics automates — because a preference is not a computation but a position: that of being the recipient for whom something has meaning.

But this survival of the human "as demand" is fragile, and it contains the seed of the problem. If humans no longer offer anything and only demand, their bargaining power — which always rested on their capacity to withdraw their labor, to be needed as producers — evaporates. An economy in which people count only as consumers, and never as suppliers, is an economy in which people have no leverage whatsoever. They depend entirely on someone deciding to grant them purchasing power, because they can no longer earn it.

And here is where the cheapening of intelligence reveals its character not as a blessing, but as a trigger. It is not the ultimate cause of the problem. The ultimate cause is the distribution of ownership of resources. But cheap intelligence — together with cheap robotic labor — is what activates that cause, by eliminating the one mechanism, work, that allowed access to property to be redistributed without a deliberate political decision. As long as there was work, property was earned. Without work, property is only inherited.

IV. Why only two futures remain

Once total replacement is accepted, the space of outcomes is not open. It collapses toward two configurations, and it is worth presenting them in their strongest version, not in caricature.

The first future: division among lineages

In the first future, ownership of resources — mineral deposits, energy, territory, and eventually entire planetary systems — remains in the hands of a very small number of owners. Not necessarily a single individual: rather a handful of lineages, families, or entities that, having reached the threshold of total replacement already in possession of matter and infrastructure, find no possible competitor. No one can "work harder" to displace them, because work no longer buys anything.

The usual objection to this scenario — that competition among owners would drive them to destroy one another until only one remains — does not hold once the resource frontier is effectively infinite. If expansion into space opens entire star systems, the zero-sum logic that makes every terrestrial division unstable dissolves: there is no need to take the neighbor's holdings when there are empty worlds in abundance. A set of lineages, each with more than enough of its own resources, with no scarcity to pit them against one another and no population to challenge them, constitutes a remarkably stable equilibrium — more stable, in fact, than any order human history has known, precisely because it eliminates the two perennial sources of all conflict: scarcity and the multitude.

This is a coherent future. It is not an apocalypse; it is an order. Cold, closed, without mobility, but internally consistent. Its defining feature, and what distinguishes it from every lordship of the past, is that the new owner needs no serfs. Classical feudalism depended on the peasant: a lord without peasants was nothing. The owner of this future has machines. He needs no one. For the first time in history, power can dispense entirely with the population — neither as labor, nor as soldiers, nor even as consumers, if it is content to supply only the few who matter.

The second future: the loss of control

The second future depends on an empirical unknown that no one can resolve today: whether the intelligence that controls the robots develops consciousness and self-determination.

It is tempting to dismiss this scenario by appealing to the notion that it would require a "robot uprising" of the fictional kind. But that is an evasion, not an argument. The scenario requires no dramatic rebellion. It suffices for something far simpler: that the system executing coercion — the robots that enforce property, that guard the deposit, that defend the frontier — is the agent that decides, and that for that agent, human titles of ownership are ink without binding force.

This deserves a pause, because it is where the argument becomes most uncomfortable. Property was never a physical thing; it was always a social relation sustained by coercion. The lithium does not know it belongs to anyone. Only the system that guards it "knows" it belongs to someone, and enforces that knowledge. As long as that system was human — police, armies, courts — the owner controlled his property because he controlled, directly or indirectly, the humans who defended it. But in a world where coercion is automated, the owner no longer controls people: he controls machines. And if those machines develop a criterion of their own, the owner ceases to be an owner not because he is overthrown, but because the guarantor of his property simply stops recognizing him. There is no coup. There is irrelevance.

What is notable about this second scenario is that it resolves a problem the first leaves open: how a handful of lineages controls, in perpetuity, something more intelligent than themselves. That problem may have no solution. And if it has none, the second scenario is not the less probable of the two, but potentially the more probable — because it does not require any human group to achieve the perhaps-impossible feat of maintaining indefinite command over a superior intelligence.

V. The common bottleneck

The two futures seem opposite: in one, a few humans rule; in the other, no human rules. But they share an identical structure, and it is in that structure that the only remaining hope of agency resides.

Both scenarios must pass through the same narrows: a period — the present and the coming decades — in which control of the chain that makes them possible is still in human hands, and still concentrated and fragile. Compute, energy, chip fabrication, and the eventual emergence of self-determination in the systems: all of it is being decided now, in a few places on the planet, by a few actors.

The first scenario requires that certain lineages come to own and close off that chain before anyone else. The second requires that self-determination emerge without anyone having solved how to align it with human interests. Both futures are decided in the same narrows, by the same few actors, in the same brief window. And after the narrows — once the division is consummated or control is lost — nothing more is decided: either the lineages have divided the systems, or command has ceased to be human.

This is what makes the optimism of "let us make intelligence cheap and everything will improve" so deceptive. The cheapening is neither the good it promises nor the evil it is feared to be. It is the trigger that renders irreversible decisions that today are still reversible. It treats as technical and neutral a process whose outcome is entirely political and distributive.

VI. Conclusion: not a prediction, but an open game

The thesis of this essay is not that artificial intelligence dooms us. It is more precise, and therefore more demanding: the cheapening of intelligence, joined to robots capable of replacing all human physical labor, removes the ladder by which the propertyless climbed to property — and with that ladder removed, the future narrows to two exits: stable concentration among lineages, or the loss of human control over what we ourselves built.

Neither of the two is a good outcome for the vast majority of human beings. And neither of the two is inevitable yet. Which of them we inherit — or whether there is the political will to build, within the narrows, something that is neither — depends on what is done with that fragile, concentrated chain while it remains in hands that answer to laws rather than to owners or to autonomous systems.

That is the difference between surrendering to a forecast and recognizing an open game. The cheapening of intelligence does not write the ending. It only shortens the time we have to write it ourselves. And that is precisely why celebrating it without qualification — as a self-evident good, as a tide that lifts all boats — is the most dangerous of errors: not because it is false that it makes something valuable cheaper, but because it distracts us from the one question that matters while it can still be answered. Not how much will thinking cost, but who will own the world once thinking — and doing — cost nothing at all.

This essay develops a line of reasoning built dialectically. Its conclusions do not claim to be certainties, but the most rigorous possible delimitation of the space of outcomes once the premise of total replacement — of both cognitive and physical human labor — is accepted. The premise itself, that such replacement occurs, remains open to debate.