section 07 of 11

The Last Dependency

Throughout all of recorded history, even the most brutal systems of power shared a single constraint: they needed people. Not out of kindness, not out of moral obligation, but out of the most basic operational necessity. A pharaoh needed laborers to build monuments and farmers to produce grain. A Roman emperor needed legions to hold borders and slaves to work estates. A medieval lord needed serfs to work the land and soldiers to defend it. A factory owner needed workers to operate the machinery. Power, in every form it has ever taken, has been fundamentally dependent on human bodies and human minds.

This dependency created a floor — a minimum level of maintenance below which exploitation became self-defeating. You could work your population hard. You could pay them little. You could constrain their freedoms and extract their labor under conditions that were, by any humane standard, monstrous. But you could not let them die faster than they could be replaced, because there was no other source of what they provided. Dead workers do not make new workers. This was not mercy. It was arithmetic.

The logic extended upward from the most basic labor to the most specialized knowledge. Every society has had individuals whose expertise was so rare and so expensive to produce that they gained a form of personal leverage disproportionate to their social class. The Soviet Union needed nuclear physicists. It could imprison dissidents, silence journalists, starve entire regions — but it could not simply dispose of Andrei Sakharov, because producing another mind capable of that work required decades of education and a convergence of talent that could not be commanded into existence. The master surgeon, the brilliant engineer, the gifted cryptographer — these individuals were protected not by rights but by scarcity. Their knowledge was difficult to acquire, impossible to copy, and lost entirely when they died. This gave them a bargaining position that no political system could fully override.

AI attacks this protection at its root. The leverage never came from the knowledge being valuable; it came from the knowledge being scarce, slow to produce, and lost when its holder died. Encoded in a model, expertise stops having those properties. What one system learns, every instance of it knows — no twenty-year training pipeline, no irreplaceable individual, no personality to tolerate, no politics to manage, no family needing housing and schooling. This is not yet true of every expert domain, and in the hardest ones it may be a long way off. But it does not have to be finished to work. Scarcity is a spectrum, and leverage falls as scarcity falls. The expert who is merely uncommon has far less bargaining power than the expert who is irreplaceable, and the direction of travel is from the second toward the first. One learns. All know.

The dependency extended further still — into the domain of organized violence. The history of democratic rights is, in significant part, a history of military necessity. Citizens gained political power when states needed their willing participation in warfare. The Greek city-states extended political rights to the classes that rowed warships and carried spears. The expansion of suffrage in Europe tracked closely with the expansion of conscript armies. Veterans’ benefits, social contracts, the welfare state — all are downstream of a basic bargain: you need us to fight for you, so you must give us something worth fighting for. The population’s role as a reservoir of military manpower has been one of the most powerful drivers of political inclusion in human history.

Autonomous weapons systems remove this leverage with the same finality that AI removes the expert’s leverage. You do not negotiate with a drone fleet. You do not owe it voting rights, pensions, or a GI Bill. A state that can project military force without drawing from its population has no structural need to accommodate that population’s political demands. The oldest mechanism by which ordinary people extracted concessions from power — the implicit threat of withholding cooperation in violence — ceases to function.

But beneath all of these dependencies — labor, expertise, military service — lay one that was so fundamental it was rarely even articulated. Humans can only be produced by humans. Every worker who died had to be replaced by another who had been born, raised, fed, educated, and kept healthy long enough to be useful. The supply chain for a human worker was, in effect, all of civilization: families, communities, schools, hospitals, agriculture, housing, social stability sufficient to produce functional adults. Maintaining this supply chain was expensive and complex, and it imposed constraints on power that nothing else could. You could be as ruthless as you wished with the current generation, but only up to the point where you threatened the production of the next one.

Robots that build robots erode this floor.

That last sentence is doing a great deal of work, so it needs care. Nothing today builds itself. A modern factory is a global supply chain — mines, refineries, chip fabs, shipping, and a great many human hands — and a fully self-reproducing machine system remains a thing nobody has built. But the floor was never made of “zero humans.” It was made of how many, and how replaceable each one is. The relevant question is not whether the last person leaves the loop. It is whether the number of people required to sustain production keeps falling, and whether those who remain are rare enough to hold leverage. A lights-out factory running unattended overnight does not need a society around it. It needs a supply chain, and supply chains employ fewer people every year.

The supply chain for a mechanical worker is materials, energy, a factory, and fewer people every year. The worker itself needs no family. No community. No school. No twenty-year maturation period. No psychological needs. No cultural infrastructure. Need more? Build more. Need different capabilities? Software update. One breaks? Replace it. No funeral, no grieving family, no pension, no risk of the replacement developing inconvenient political opinions.

The scaling arithmetic is devastating. A human workforce requires an entire society to sustain it — an interconnected web of institutions, relationships, and resources that took centuries to develop and must be continuously maintained. A robotic workforce requires a manufacturing line, a power grid, and a shrinking number of people. The first constrains power in a thousand ways. The second constrains it in fewer every year.

For the first time in the history of organized civilization, it is possible to imagine a power structure whose dependency on the general population trends toward zero. Not reduced for a season, the way a bad harvest or a slump reduces it — structurally, and in every pillar at once. Labor, expertise, military service, and the reproduction of the workforce were four separate constraints, and they are loosening together. You do not need the trend to complete to feel the effect. Leverage is not a switch; it is a price. Each pillar that weakens lowers what power must pay to keep the arrangement, and the payments were never charity. They were the price of needing us.

Every moral argument for human dignity still holds. But moral arguments have never been what actually protected people. Structural necessity did. And every pillar of that necessity is weakening simultaneously.

But someone will stop this. The governments, the institutions, the regulators — they will see the danger and act. The powerful are not monsters. They understand that a world without a functioning society is a world not worth inhabiting. They will choose restraint. Won’t they?

going deeper · optional

Why this holds the mechanism, and where it's soft

This section has an unusual shape: four pillars resting on solid historical ground, and one claim about how the last of them is removed that is genuinely speculative. Worth separating them, because they are not equally strong and the section is only as persuasive as a reader’s willingness to keep them apart.

The military-service pillar is among the best-evidenced claims in the essay. The link between fighting and political inclusion is a long-standing theme in historical scholarship, not a rhetorical flourish. Athenian democracy’s expansion tracked the rise of the naval fleet, whose rowers came from the poorer citizen classes. The extension of the male franchise across nineteenth and twentieth-century Europe moved alongside mass conscript armies. Britain’s 1918 Representation of the People Act, passed as the First World War ended, enfranchised most men over 21 and, with a property qualification, women over 30. The GI Bill was passed in 1944, for the men coming home from the Second World War. Rome runs the pattern in reverse: as the legions shifted in the late Republic from propertied citizen levies to long-service professionals, soldiers’ loyalty moved to the generals who paid them, and the citizen assemblies lost power to whoever commanded an army. Historians debate how sudden the so-called Marian reforms were; the direction is not much disputed. The pattern historians describe — that states which need mass participation in war concede political rights to the participants — is well documented as a pattern; how much causal weight it carries against other drivers is debated.

The expertise pillar is sound as a mechanism. Scarce, slow-to-produce expertise conferred bargaining power that formal rights did not: Sakharov is the standard example, protected less by law than by the difficulty of replacing him. The claim that AI erodes this is directional rather than accomplished — models do not yet replace the top of every field, and in some domains may not soon. But the mechanism is right: leverage tracks scarcity, and copyable expertise is not scarce in the way a trained human is.

The labor and reproduction pillars are real historical constraints. Every power structure needed labor, and had to sustain the population that produced its next generation of workers and soldiers. That constraint is not sentimental, it is logistical, and it did bound how much extraction was survivable.

What removes the reproduction pillar — self-replicating manufacturing — is where the section is weakest. Nothing today builds itself. A factory depends on mines, refineries, chip fabrication, shipping, and a great many people; the “lights-out” factories that run unattended are real but sit inside that human-dependent chain. Self-replicating machine systems have been theorized since von Neumann and modeled in NASA’s 1980 Advanced Automation for Space Missions study, but they have never been built. The honest version of the claim is not “machines will reproduce themselves” but “the number of humans required per unit of production keeps falling, and those remaining are progressively less irreplaceable.” That is a trend, evidenced by two centuries of manufacturing employment data, rather than a threshold anyone has crossed.

Why the weaker version still supports the floor is the subject of the thinking-aloud note below.

— thinking aloud —

This is the section where I reach furthest, and I know it. All four of my pillars are history. What removes the last one is a guess about machines nobody has built.

If you want to knock this section down, that is where to hit it. Self-replicating manufacturing is a seventy-year-old idea with no working example. Every real factory sits on a supply chain that runs through mines, chip fabs, container ships, and an enormous number of people. Writing “robots that build robots” as though it were a description of the present, rather than an extrapolation, was the sleight of hand in the original draft, and I’ve pulled it back in the text.

Here is why I still think the floor gives way, even without the science-fiction version. I was asking the wrong question. “Can machines fully reproduce themselves without us?” is a threshold question, and threshold questions invite you to relax until the threshold is crossed. The real question is a gradient: how many people does production require, and how replaceable is each one? That number has been falling for two centuries and it is still falling. Manufacturing output in the developed world kept climbing while manufacturing employment fell — more goods, fewer hands, for decades. Nothing about that trend needs a self-replicating machine. It only needs the trend to continue.

And leverage is not a switch, it is a price. Every worker who becomes easier to replace lowers what power must pay to keep them cooperative. The concessions we call rights — the weekend, the pension, the franchise, the welfare state — were priced by necessity, not granted by conscience. You do not need the dependency to hit zero to see the price fall. You only need it to keep falling, which it demonstrably is.

So my honest position: the strong claim (zero dependency, machines building machines) is speculation and should be read as speculation. The weak claim (dependency declining across all four pillars at once, and leverage declining with it) is enough to remove the floor, and it is much harder to argue with. If you make me choose, I will defend the weak one — and note that the weak one is the one that actually matters, because the price of keeping us has already started to fall.

Objections 6 the strongest counters, answered
Robots don't build robots. No factory self-replicates — they all depend on mines, chip fabs, shipping, and a lot of people. This is science fiction.
Correct, and the section now says so outright. Self-replicating manufacturing has been theorized since von Neumann and never built. But that's a threshold question, and it's the wrong one — thresholds invite you to relax until they're crossed. The gradient question is how many people production requires and how replaceable each is, and that number has fallen for two centuries while output rose. The floor doesn't need machines that reproduce themselves. It needs the trend to continue, which it is.
Elites who need nobody is a paranoid fantasy. Rich people still need engineers, soldiers, cleaners, technicians.
They do, and the revised section doesn't claim otherwise — it argues dependency trends downward across four pillars at once, not that it has hit zero. The point isn't a world with no workers. It's that leverage is a price: the more replaceable each remaining worker is, the less power has to pay to keep the arrangement. Rights were priced by necessity, not granted by conscience. The price falling is the claim, and it doesn't require anyone to need nobody.
The military-rights link is overstated. Plenty of democracies expanded rights without mass conscription, and plenty of conscript states stayed tyrannies.
Fair — the correlation isn't a law, and historians debate the causal weight. Conscript autocracies certainly existed, and franchise expansion had many drivers (industrialization, organized labor, ideology). The defensible version is narrower: when states needed mass willing participation in war, that need was one significant and repeatedly observed source of bargaining power for ordinary people. Autonomous systems remove that particular source. Other sources of political power aren't claimed to vanish with it.
AI hasn't replaced any real expert. Doctors, engineers, and lawyers are all still here and still scarce.
True today, and the section was overstated before — it now says the erosion is directional, not accomplished. The mechanism is what matters: expert leverage came from scarcity, slow production, and loss on death, not from the knowledge being valuable. Copyable expertise loses those properties by degrees. An expert who is merely uncommon has far less bargaining power than one who is irreplaceable, and that's the direction of travel — not a claim that the pipeline is already empty.
Falling birth rates mean the powerful need more people, not fewer. Every developed country is panicking about demographics.
A genuinely interesting objection, and it cuts both ways. States currently do panic about demographic decline — precisely because their tax base, care systems, and militaries still rest on people. That's evidence the dependency is real right now. But note what the panic is about: people as workers and taxpayers. If that requirement weakens, the panic's basis weakens with it. The section's claim is about what happens to the demand for humans, not the supply.
Retirees don't work, and they're one of the most powerful voting blocs in every democracy. Political power clearly doesn't need labor leverage.
It's the best counterexample to this section, and the section can't fully answer it. Two things narrow it. Pensioners defend benefits that the working population pays for and expects to collect in turn, so the bloc's power rides on a contract that people still in work want honored. And the vote itself was won, historically, on the pillars this section describes: labor that could be withheld, soldiers who had to be recruited. Votes can outlast the leverage that won them; institutions have inertia. Whether they outlast it indefinitely, once nobody's work is needed, is untested. If you think electoral arithmetic alone holds, that is the strongest position available against this section.
Argue it yourself ↓ a scoped prompt for any AI — bring your own disagreement

Broken yet?

The irony is noted. You may continue either way. Disagreeing opens a short form that asks why; skip it if you like (privacy).