The Last Necessity

I. How we came to have to ask the question

For most of human history the question would have been unintelligible. Of course humans were necessary; they were the only engine the world had. Every wall, every harvest, every army, every ledger, every act of governance ran on human muscle and human judgment, because there was no other source of either. The question we now have to ask — is there any function for which the mass of humanity remains genuinely irreplaceable? — is the product of a specific historical machine, and to understand why we are forced to ask it now we have to see that machine clearly.

Begin with the long arc. Since humans first organized at scale, the dominant direction of political history has been the consolidation of control over larger and larger populations — from roving band to chiefdom to kingdom to empire to nation-state to the planetary platforms and bureaucracies of the present. This is not a smooth slope. It is interrupted by revolutions, collapses, and the great democratic expansions of the modern era: abolition, suffrage, the labor settlement, decolonization, the welfare state. But these interruptions are not exceptions to the consolidating arc. They are produced by it. Consolidation enlarges and enriches the very society it governs; the enlarged society spawns new constituencies and new interests; those interests temporarily check the center; and the center re-integrates at a higher level of scale and reach than before. Control grows the swarm, the grown swarm curbs the control, and the control reconstitutes itself larger. The reach of organized power expands across the whole cycle even as the individual, within any given era, often gains protections he never had.

The crucial question is why those democratic expansions kept happening — why consolidation kept being forced, on each upward turn, to hand something back. The answer is the engine, and naming it precisely is what lets us see what is now changing.

A controller can only govern a system whose variety — whose range of possible states and behaviors — it can match. For all of history, the only source of variety sufficient to run a large and complex society was the society itself: the distributed cognition, skill, local knowledge, and judgment dispersed across millions of human heads. No emperor, no church, no corporation could govern a complex realm without recruiting that distributed variety to do it — as bureaucrats, soldiers, clerks, engineers, foremen, and eventually voters and consumers. And the iron consequence is that the moment the center needed those people, it had to bargain with them. The franchise, the rule of law, the eight-hour day, social insurance — these were not gifts of conscience and not merely concessions wrung out by struggle. At the deepest level they were the price of requisite variety: the standing cost of buying the cooperation of the human beings whose distributed capability the center could not synthesize any other way. This is why consolidation and accommodation rose together across the modern period. They were two faces of the same necessity. Every turn of the ratchet of control carried a pawl that handed something back to the population, and that pawl was the swarm’s indispensability.

Now state the discontinuity plainly. Artificial intelligence, joined to robotics and autonomous systems, is the first technology in history that promises an exogenous source of requisite variety — the ability to match the variety of a society of arbitrary scale and complexity with machine cognition rather than with co-opted human cognition. If that promise is even partly kept, consolidation can continue to advance without enlarging or empowering the swarm, because it no longer needs the swarm’s variety to run the system. The ratchet keeps turning. The pawl falls off. The coupling between control and accommodation — the mechanism that made every previous round of consolidation survivable, even beneficial, for ordinary people — is severed at its root.

That severance is why we are now compelled to ask the question. Not because erosion has finally caught up with us, but because the automatic correction that erosion always triggered may no longer fire. And so everything turns on whether anything is left for which the swarm cannot be replaced. That residual necessity, if it exists, is the last source of the bargaining power on which the entire history of human freedom has quietly depended.

II. What the question actually is

The question is not “will there be jobs.” It is far colder and more structural than that. It is whether there remains any function — productive, coercive, epistemic, or social — that the apparatus of organized power cannot perform, sustain, or replace without the broad participation of ordinary human beings. The stakes are not employment statistics. They are leverage. A function for which the swarm is irreplaceable is a lever the swarm can pull. A society in which no such function remains is a society in which the governed have, for the first time, nothing the governing structurally needs.

Two distinctions sharpen the question and prevent it from collapsing into sentiment.

The first is between a function that is necessary to the system’s operation and a function that is merely valuable to human beings. These are not the same, and the gap between them is the most painful thing in this entire analysis. Human beings may continue to find immense value in art, in care, in love, in mentorship, in worship — and none of it need be necessary to the operation of the productive or coercive order. The whole of human leverage in history came from being necessary to production and to violence, not from being valuable to one another. A future can therefore arrive in which humans remain deeply meaningful to each other and simultaneously powerless before the system that feeds them. Meaning and leverage, fused for all of history, can come apart. The question is specifically about leverage. It must not be answered with a list of things that make life worth living, because those things, however precious, may carry no bargaining weight at all.

The second distinction is between a natural necessity and a constructed one. A natural necessity is a function the system genuinely cannot perform without us — the soldier who must be recruited, the worker who must be hired, the consumer who must be sold to. A constructed necessity is a function we choose to reserve for humans even though a machine could perform it — a legal requirement that a human judge pass sentence, a human physician bear final responsibility, a human hand on the decision to kill. Constructed necessities are real, and they may become the most important institutions of the coming century. But they are revocable in a way natural necessities are not. They are floors we build, not bedrock we stand on, and they can be legislated, lobbied, or quietly engineered away. To ask whether any function is irreplaceable is therefore to ask two questions at once: is there a natural floor, and if not, can we build and defend a constructed one before the power to build it erodes?

III. An attempt to answer

Let me actually try, function by function, rather than gesture at the difficulty.

The swarm as producer. This is the necessity that built the modern world, and it is the one AI most directly attacks. The breadth of cognitive substitution, joined to robotics closing the physical gap, threatens the productive indispensability of human labor across both white-collar and blue-collar domains in a way no prior automation did, because prior machines were narrow and left whole continents of human capability untouched. The honest reading is that this floor is the one most actively dissolving. It is not gone — vast amounts of work remain stubbornly human today — but it is the function on which we can least rely as a permanent source of leverage.

The swarm as consumer. Here is a genuine and underrated necessity: production is worthless until it is sold, and a fully automated economy still needs someone to buy its output. As long as the system remains a market, the mass of humanity holds the demand side, and demand is leverage — boycott, demand-strike, the politics of who can purchase. But this necessity is contingent in a way that should frighten us. It holds only while the elite still needs to sell. The moment output is allocated by command and status among a small ownership class rather than exchanged in a market, the consumer ceases to be necessary at all. Consumption is also a passive necessity: being a buyer asks nothing of you, builds no skill, confers no agency. It can deliver leverage while leaving the person hollow. So the consumer floor is real, but it is conditional on a configuration the powerful could choose to exit, and even where it holds it secures bread without securing purpose.

The swarm as soldier and as the source of consent. For all of history, coercion required coercers drawn from the population, and rule was cheaper when it was believed in. The conscript who shares an identity with the people he is ordered to fire upon, and the citizen whose consent makes governance efficient, were both forms of indispensability. Autonomous weapons sever the first; personalized persuasion and the manufacture of legitimacy attack the second. And note the grim coupling: regimes that have never needed their populations’ productivity — slave economies, petrostates — have historically needed neither their soldiering nor their consent. Necessity of labor, of arms, and of belief tend to rise and fall together. As the productive floor dissolves, the coercive and consensual floors tend to dissolve with it.

The swarm as bearer of irreducible local knowledge. This is the deepest technical hinge, and it is genuinely open. Hayek’s argument against central planning was not that planners lacked compute but that the knowledge a society runs on is tacit, local, and contextual — dispersed across millions of situated minds in a form that resists aggregation, which is why command economies failed even when staffed by clever people. If that irreducibility holds, the swarm remains a necessary distributed network of sensors and judges that no center can fully internalize, and the floor survives. If AI plus pervasive sensing closes that gap — internalizing local context at scale — the floor falls. I do not believe anyone honestly knows which is true. This is the empirical question on which more hangs than any other, and it is being decided in real time, domain by domain, with no general answer yet visible.

The swarm as the source of objectives and of new variety. Optimization requires something to optimize for. Even a superhuman system needs an objective, and the deepest claim for human irreplaceability is that humans remain the irreducible source of what is worth doing — the generators of genuinely new goals, values, and questions, the authors of the objective function rather than its servants. This is the most defensible candidate for a natural floor. But it is fragile in two ways. A small elite could monopolize the setting of objectives, in which case the swarm is still dispensable even if humanity is not. And the technology may, over time, come to generate and revise objectives itself, at which point even this floor erodes.

The constructed necessities. Finally, the functions we might choose to reserve: human judgment in the courtroom, human responsibility in medicine and war, the legal requirement of a human in the loop — not because the machine cannot decide, but because we insist that accountability, moral standing, and the authority to bind require a human bearer. These are real and buildable. They are also revocable, and they depend on the strength of exactly the institutions that consolidation erodes.

Assemble these and the answer comes into focus, and it is not the answer either an optimist or a fatalist wants. There is no robust, automatic, technology-proof floor of natural necessity that we can count on to regenerate human leverage the way every prior inflection did. The functions with the best claim to irreducibility — humanity as objective-setter, and the local-knowledge hinge — are uncertain, monopolizable, or under active assault. The functions that feel most human — care, art, meaning — are precisely the ones most likely to remain valuable while ceasing to be necessary, which is to say they preserve our worth to each other while surrendering our power over the system. The conclusion is therefore sobering and specific: going forward, human leverage will have to be manufactured politically rather than supplied by necessity. The bargaining power that history handed us automatically, as the price of our indispensability, will now have to be deliberately constructed and defended through institutions — owned dividends held as enforceable property rather than revocable grants, reserved human judgment, governed compute, a defended commons of verification and identity. That is a far weaker and more precarious foundation than the one we have always stood on, because it must be built using the very political capacity that the loss of necessity is draining away. The window in which we still have the leverage to build the institutions that would replace our lost leverage is the thing actually at stake.

IV. The spectrum of outcomes

Lay the futures along a spectrum, organized by the variables we have isolated: whether machine variety truly substitutes for human variety, whether the system stays a market or becomes a command order, whether the diffusion of capability to the many offsets the integration of capability by the few, and whether political leverage is reconstructed before the chance is gone.

At the wildly dystopian pole sits the configuration in which machine variety fully substitutes, the order shifts from market to command, integration decisively beats diffusion, and no political floor is built in time. The mass of humanity is necessary as neither producer, soldier, nor consumer. What follows is managed subsistence at best, and an apex of power with no feedback — for the dissent, exit, and competition the center used to absorb from a swarm it had to accommodate were also its error-correction, its senses. A control system that has blinded itself does not reform gracefully; it persists rigidly and then fails catastrophically, and now it fails while holding the grid, the laboratories, and the weapons. Both of its end-states — the stable cage and the sudden collapse — are ruinous.

A step in from there is the soft dystopia, the gilded cage. The market is retained, a dividend keeps demand alive, and the population is maintained in comfort and entertained into quiescence. No one starves; no one decides anything. This is Tocqueville’s soft despotism realized at civilizational scale, Huxley rather than Orwell — a humane-looking arrangement that has nonetheless severed people from agency, mastery, and consequence, and that is stable precisely because it is pleasant.

In the middle lies the muddle: diffusion partly offsets integration, elites stay divided and check one another, some functions are defended and some are lost, and history simply continues — contested, unresolved, at far higher stakes. This is the most probable near-term outcome and the least dramatic, and it buys time without guaranteeing what the time is used for.

Toward the good end is the constrained achievement: leverage deliberately rebuilt. Dividends owned as property, not dispensed as charity. Human judgment institutionalized where legitimacy demands it. Compute and model access governed so that capability diffuses rather than only concentrates. A verification and identity commons that keeps shared reality, and therefore coordination, possible. In this world humans are freed from the compulsion of necessity-labor while retaining real power and constructed purpose. It is reachable, but only by winning a sequence of difficult, boring, against-the-grain political fights.

At the wildly utopian pole is genuine post-scarcity flourishing: abundance, aligned systems, wise institutions, humanity released from drudgery into Arendt’s realm of action and self-actualization — Keynes’s prophecy for his grandchildren, finally and actually delivered.

The decisive feature of this spectrum is its asymmetry. The dystopian pole is the default attractor: it is where the unmodified incentives of capital accumulation, state security, and competitive deployment point on their own. The utopian pole is an achievement: it requires sustained, coordinated, costly action against those incentives. Drift produces the cage. Only effort produces the garden. That asymmetry is the whole reason the question is urgent rather than academic.

V. What our track record says

So can we, the species, capitalize on an inflection like this? Here I will give an opinion rather than hide behind the range, because the question deserves one.

Our record is not encouraging in the specific way that matters here. Humanity is poor at anticipatory, coordinated, ex-ante governance of new powers, and comparatively good at costly, reactive, ex-post correction after the damage is visible. We rarely steer an inflection in advance. We careen through it and then, standing in the wreckage, build the institutions that would have prevented it. The industrial revolution eventually produced the union, the factory act, public health, and the welfare state — but only after decades in which productivity climbed while wages stagnated and a generation was ground through Manchester’s mills. Tobacco, leaded fuel, and the chlorofluorocarbons all followed the same grammar: the harm ran decades ahead of the remedy, and the remedy arrived after the bodies. Climate change, the closest structural analogue to AI in its diffuseness, its misaligned incentives, and its contested epistemics, has drawn decades of clear warning and weak coordinated response. The cases where we did act in advance — the Montreal Protocol, the nuclear taboo and the non-proliferation regime — are real and they matter, but they came with advantages AI lacks: a clear villain and an available substitute, or a technology concentrated, legible, and bipolar enough to be bargained over by a handful of hands. AI is diffuse, fast, multipolar, and dual-use. It is climate’s difficulty with nuclear’s stakes.

And here is the structural cruelty that distinguishes this inflection from all the others on the list. Every previous catastrophe was survivable in part because the masses retained the leverage to force the eventual correction. The settlement after the industrial revolution was extracted by a labor movement that could halt production; the welfare state was, among other things, a bid for the loyalty of populations whose cooperation the state still needed.

The reactive-correction model — muddle through, suffer, then fix it — has always relied on the swarm keeping enough leverage to compel the fix. The specific danger of the AI inflection is that it may foreclose the correction mechanism itself, by dissolving the necessity-based leverage early, before the correction is made. There may be no “later” force capable of compelling the settlement, because the conditions that have always generated such a force may be gone. Add to this that every prior inflection unfolded across decades or generations, leaving room for the slow reactive fix, and that AI’s potential speed may compress the window below the time institutions need to respond, and the comfortable historical pattern of fixing things afterward looks less like a reliable fallback than like a luxury this transition may not afford.

I will not end there, because determinism is its own kind of laziness and the record is not only dark. Humanity has, on rare occasions, overridden the incentive gradient through values and leadership and sheer clarity at the edge of the precipice. We abolished chattel slavery against colossal entrenched economic interest. We have not used a nuclear weapon in anger in eight decades, sustaining a taboo that pure self-interest does not explain. We ended a global standoff far more peacefully than almost anyone predicted. These are proof that the gradient can be beaten — unreliably, expensively, but really. And this inflection carries one genuinely unusual asset: it is foreseen. We are arguing about it before it has fully arrived, which is itself historically rare; most inflections were named only in the rearview mirror. Foresight is not the same as steering, and it confers no guarantee. But it is the precondition for steering, and we have it.

So my honest verdict is this. On the weight of the evidence, we should expect to handle this badly in the familiar way — late, reactively, at great cost — except that this is the one inflection where late and reactive may not work, because the mechanism that has always rescued us afterward is the very thing it threatens to remove. The entire wager of the coming decades comes down to whether foresight, for once and against our track record, can be converted into anticipatory institution-building before the leverage to build those institutions is gone. We have never reliably done that. We have occasionally done it. And this may be the first time we have to.

(Co-written with Opus 4.8)