Immortal Machines

I spend most of my time trying to get machines to learn from watching people. The basic problem is this: a human demonstrates a task (pick up the can, sort it into the right bin, place the lid on the container) and the machine has to figure out how to replicate that behavior in situations it has never seen before. The gap between “watched a person do it once” and “can do it reliably in the real world” is where I live, intellectually. And something about working at that gap, day after day, has changed how I think about technology more broadly, because the gap turns out to be much stranger than I expected.

When you watch a robotic arm attempt a task it was trained on, you see something interesting. The arm does not understand the task. It has learned a mapping between sensory input and motor output, optimized over thousands of gradient steps, and when conditions are close enough to what it saw in training, it produces the right behavior. When conditions drift, it fails, sometimes in ways that look bizarre. The system has skill without comprehension, doing without knowing. And yet the skill is real. The arm does pick up the can. It does sort it into the right bin. The fact that it accomplishes this without understanding what it is doing is not a failure of the system; it is a revelation about what systems can accomplish without understanding anything at all.

I think this is a useful lens for looking at a much older and larger phenomenon: the full arc of human technology. We tend to talk about technology as a collection of tools, a set of inventions that humans use to accomplish goals. I want to argue that this framing understates what is happening. Technology, taken as a whole, behaves less like a toolkit and more like a self-propagating system, one that outlasts every civilization that hosts it, absorbs every human impulse fed into it, and optimizes relentlessly for its own continuation. It does not need to understand what it is doing any more than the robotic arm needs to understand the can. I am calling this system the immortal machine.

Let me be specific about what I mean by technology, because the word gets used loosely. The Greeks had a better concept: techne, which Aristotle classified as one of five modes of knowing, alongside science, wisdom, prudence, and intuition. Techne was not just craft or tool-use but a way of understanding the world through the act of making. I think this is closer to the truth than our modern usage. Technology is any systematic method by which humans extend their capacities beyond biological limits. Language qualifies. Writing qualifies (Walter Ong argued convincingly that literacy “restructures consciousness,” producing patterns of thought impossible in oral cultures). Agriculture, mathematics, money, government: all technologies. Each solved a problem, created new problems, and demanded further technologies to address those. The ratchet turns in one direction.

What drives this ratchet? The standard answers are curiosity, ambition, economic incentive, military competition. These are all real, but I think there is something underneath them, something that operates at a level closer to biology than to culture. The same force that makes a cell divide, a vine climb toward light, a river cut through stone: the will to persist, to propagate, to not disappear. Every technology we have ever built, from the sharpened stone to the transformer architecture, expresses the same underlying imperative. Survive, and make survival mean something.

The incentive structures reinforce this at every level. Scott Alexander wrote about this under the name Moloch, borrowing from Ginsberg’s “Howl.” His argument is precise and I think correct: in any competition optimizing for some variable X, the opportunity eventually arises to sacrifice other values for improved X. Those who take the trade prosper. Those who refuse get outcompeted. The result is a system that relentlessly optimizes for its objective function while treating everything else as externality. This is a description of natural selection, and it is also a description of market economies, arms races, and the AI industry. The dynamic does not require anyone to be malicious; it only requires competition.

I want to name what I think this adds up to, because I think it is worth being direct about it: the machine, followed to its logical endpoint, is pointed at total mastery over the physical world. The conquest of death, the elimination of distance, the comprehension of everything. The people building the most powerful technologies on earth will say this openly. Kurzweil promises the singularity by 2045. Amodei writes about compressing fifty to a hundred years of scientific progress into five or ten. Harari traces the trajectory in Homo Deus. Brand said it plainly in 1968: “We are as gods and might as well get good at it.” By 2009 he had updated: “We are as gods and have to get good at it.” The shift from might-as-well to must tells you something about the machine’s momentum, not our ambition.

But before technology assumed this role, something else held it for millennia, and I think understanding what came before helps explain what is happening now. Religion was the original immortal machine. It outlasted every empire that adopted it, transferred between civilizations, and organized human life around a single imperative: submit to a power greater than yourself, and in return receive meaning, structure, and the promise of transcendence. Whether you believed you were made in God’s image or that God demanded your obedience, the effect was the same. The divine ordered your life. It told you what to want, what to fear, what to build, and what to die for. It was the first system that propagated across centuries and continents without anyone deciding to keep it going.

Then we started understanding how things actually work. Copernicus removed Earth from the center of the universe, Darwin removed humans from special creation, and Freud (borrowing an observation from the physiologist Emil du Bois-Reymond) described these as three wounds to human narcissism, each one stripping away a layer of the story religion had told us about our place in the cosmos. Max Weber, lecturing in Munich in 1917, gave the process a name: Entzauberung, the disenchantment of the world. “There are no mysterious incalculable forces that come into play,” he said. “One can, in principle, master all things by calculation. This means that the world is disenchanted.” The sacred retreated from public life. But the need for it did not.

Nietzsche had seen this coming thirty-five years earlier. In Section 125 of The Gay Science, a madman lights a lantern in the bright morning and runs through the marketplace shouting that God is dead. He is not celebrating; he is terrified. “Must we ourselves not become gods,” he asks, “simply to appear worthy of it?” The passage is usually read as atheist triumphalism, but it is the opposite. Nietzsche understood that killing God created a vacuum, and that something would have to fill it. He just did not know what.

We do. The immortal machine stepped into the space that God once occupied, and it offers roughly the same things religion offered (purpose, order, the promise that suffering is not meaningless, the hope that death can be overcome), complete with its own priesthood, its own eschatology, its own offer of salvation. The theologian David Noble, in The Religion of Technology, argued that this transfer was not a break but a continuation. Western technological ambition, he claimed, was never secular; it grew directly from the Christian millenarian desire to recover humanity’s lost divine nature. The engineers building the future are not replacing God but continuing his work by other means. I find this argument persuasive in part because I work in a field where the language of salvation is used without irony. We talk about “solving” intelligence, “curing” death, “transcending” human limitations. The vocabulary is theological even when the methods are mathematical.

If Noble explains why we build (to recover the divine), Heidegger explains how the building changes us. He understood the mechanism better than anyone. In 1954 he published “The Question Concerning Technology” and argued that technology is not an instrument but a way of understanding the world, specifically a way of revealing everything as standing reserve: raw material awaiting optimization. A river becomes a hydroelectric resource. A forest becomes a carbon sink. A human being becomes a data point, a user, a consumer. I work with systems that literally do this: a vision-language-action model ingests a camera feed and converts the physical world into a tensor representation that can be optimized against a reward function. Heidegger’s “standing reserve” is not a metaphor; it is a design pattern. I use it every day. The philosophical abstraction and the engineering reality are the same thing, which is either a vindication of Heidegger or an indictment of my field, or possibly both.

What makes all of this so durable is that the machine I am calling immortal is not any particular technology but the process itself, taken as a whole, larger than any one of its instantiations. The Romans are gone, but road-building persists. The British Empire collapsed, but the telegraph’s descendants circle the planet. The Qin Dynasty fell, but bureaucracy (their great technological innovation) runs every modern state. Technologies survive the death of their creators, transferring between cultures, mutating, recombining, generating new technologies. Kevin Kelly has argued that this system, which he calls the Technium, behaves like a seventh kingdom of life: autonomous, directional, self-perpetuating. I think Kelly is onto something real, even if the metaphor can be pushed too far. The point is not that technology is literally alive but that it exhibits the same kind of persistence and adaptability that we associate with living systems, and that this persistence does not depend on any individual human or civilization choosing to maintain it.

But the history of this process is not benign, and it is worth looking at directly, because the same properties that make the machine immortal (its indifference to human intention, its capacity to outlast its creators, its relentless optimization) are also what make it dangerous.

Gunpowder was invented in China for fireworks and alchemy. Within centuries it had enabled the gunpowder empires (Ottoman, Safavid, Mughal) and then the European colonial project that consumed most of the earth’s surface. At the Battle of Omdurman in 1898, British soldiers with Maxim guns killed roughly 10,000 Sudanese warriors in a few hours while losing 48 of their own. The compass made global navigation possible, and global navigation made the transatlantic slave trade possible. The printing press enabled the scientific revolution and also printed the propaganda that justified genocide. Jared Diamond’s thesis is simple: geographic luck determined which civilizations developed superior technology first, and superior technology determined who conquered whom. The technology did not care which direction it was pointed. It never does.

That indifference scales. The East India Company is the prototype of the modern technology corporation. It maintained a private army of 260,000 soldiers, twice the size of Britain’s official military. It built roads, courts, and postal systems. It also engineered famines, destroyed local industries, and extracted wealth on a scale that still shapes the global economy. It was a machine optimizing for its objective function (profit) and treating everything else as externality. If that sounds like a description of a contemporary technology company, the resemblance is structural, not coincidental. The optimization function has not changed. Only the medium has.

The same structure persists wherever optimization operates. When a platform optimizes for engagement, everything that is not engagement (truth, mental health, social cohesion) becomes a cost the system cannot see. When a supply chain optimizes for efficiency, everything outside the metric (working conditions, environmental degradation, the hollowing out of communities along the route) falls to the margins. No one has to decide to cause harm. The harm is what happens to everything the objective function does not measure, and no objective function measures everything.

The pattern has a name. Every act of technological creation is simultaneously an act of destruction. Schumpeter popularized this dynamic as “creative destruction” and considered it a feature, not a bug. The printing press destroyed the scribe class. The automobile destroyed the horse economy. The internet destroyed the newspaper, the record store, the travel agent. AI is currently destroying, or will destroy, the call center, the paralegal, the illustrator, the junior programmer. In each case the aggregate result is more wealth, more efficiency, more capability, and the people displaced are expected to find new roles in the new economy, which many of them do, eventually, at great personal cost that does not appear in the aggregate statistics. The machine feeds on what it displaces. The displacement is the fuel.

But I should be honest about the other side, because the case for technology is not an abstraction; it is billions of lives. The Green Revolution, led by Norman Borlaug’s dwarf wheat, is estimated to have saved a billion people from starvation. Smallpox, which killed 300 million people in the 20th century alone, was annihilated by a vaccine distributed through a logistics network that only technology could build. Child mortality has fallen by more than half since 1990. Extreme poverty has dropped from 36 percent to under 10 percent in a generation. Life expectancy has doubled in two centuries. The Capitalist Peace theory argues, with real evidence, that trade interdependence makes war less profitable and therefore less likely. These are not minor achievements. They are the reason I work in technology rather than writing critiques of it from the outside. The machine has done more to reduce human suffering than any philosophy, any religion, any political movement in history. That fact does not cancel the harm, but it does make simple condemnation intellectually dishonest.

All of this is real, and all of it coexists with the harm described above. This is the thing that I find hardest to sit with, and that I think most writing about technology fails to adequately confront: the benefits and the harms are not separate phenomena produced by separate systems. They are the same phenomenon, produced by the same system, operating by the same logic. You cannot have the vaccine without the logistics network, and you cannot have the logistics network without the economic system that also produces the displacement. The machine is one machine.

This is the nature of the immortal machine: it has no allegiance. Liberation and exploitation, medicine and weaponry, a school and a surveillance system all look the same to it. It propagates every human impulse fed into it, scales it, and makes it durable. Feed it compassion and it builds hospitals. Feed it greed and it builds sweatshops. Feed it both, simultaneously, which is what we always do, and it builds a world that contains both, at scale, forever.

Le Guin understood this. In “The Ones Who Walk Away from Omelas,” she imagined a city of perfect happiness whose prosperity depends on the perpetual suffering of a single child locked in a basement. Everyone knows about the child. Most stay. Some walk away, but she never tells us where they go, and I think the reason she does not tell us is that there is nowhere to go. The city is the world. The question she poses is whether complicity is the price of civilization, and whether civilization is worth it.

I think it is. But I think the price should be stated clearly, because the failure to state it is itself a form of complicity, and because the people paying the highest price are rarely the ones making the decision.

Stating the price means looking at the full pattern, not just the latest cycle. From the first agricultural settlements to the global economy of 2026, the pattern repeats: a technology emerges, concentrates power in the hands of those who control it, generates wealth and capability, displaces those who cannot access it, eventually diffuses broadly enough to raise the general standard of living, and then gives birth to the next technology. Bronze to iron. Manuscript to print. Steam to electricity. Analog to digital. Digital to artificial intelligence. Each cycle faster than the last, each promising to be the one that finally resolves the problem of being human. None of them do, because the problem of being human is not a problem to be solved. It is a condition to be inhabited, and technology, for all its power, has never changed the fundamental terms of that inhabitation. It has only changed the furniture.

At the individual level, the pattern is just as visible. The phone in your pocket is more powerful than the computer that landed humans on the moon. You use it to scroll through content that makes you anxious, compare yourself to people you will never meet, and feel more connected and more alone than any generation before. You have access to the sum of human knowledge and you use it, mostly, to distract yourself from the fact of your own mortality. Technology working perfectly: it found the need, the ache, the loneliness, and filled it with a substitute. The substitute feels like a solution until you put the phone down.

And yet we pick it back up. The reason we keep building is the same reason humans have always reached for transcendence. We suffer, and we want to stop suffering. We die, and we want to stop dying. We are small, and we want to be large. These are spiritual motivations dressed in engineering, and I do not think there is anything wrong with them. The desire to reduce suffering is noble. The desire to extend life is human. The desire to understand the universe is perhaps the best thing about us. What concerns me is not the motivation but the assumption that the machine, if made powerful enough, will eventually satisfy it. I do not think it will, and the reason I do not think so is that the motivation is not really about the external problem. It is about the internal one: the need to mean something in the face of disappearing.

If the machine succeeds, if we cure death and eliminate suffering and solve scarcity and expand across the stars, we will also eliminate the conditions that produced every beautiful thing humanity has ever made. Every poem was written because someone suffered and needed to speak. Every painting exists because someone saw the world as unbearable and tried to make it bearable. Every act of love is an act against loss. Take away the loss and what remains? Comfort, abundance, immortality. Those are real things. But they are not the things that made Beethoven write the Ninth Symphony while deaf, or Frida Kahlo paint her broken body into myth, or the unknown engineer at NASA scribble calculations at three in the morning because we are going to the moon and I will not let this fail.

The machine will archive all of this. It will preserve every symphony, digitize every painting, reconstruct every act of love in high fidelity. But it will not produce new instances of the thing that made them, because that thing was not talent or technique but the particular desperation of a creature that knows it will die and refuses to go quietly. Eliminate the desperation and you eliminate the source. That is the trade being made, and I am not confident that anyone building these systems has recognized it as a trade at all.

Amodei, in “Machines of Loving Grace,” came closer than most to sitting with this. He wrote that when AI surpasses human economic value, humans will find purpose in relationships and in activities valued for their own sake. It is an honest answer, but it is the answer of someone who still derives meaning from the difficulty, from the uncertainty, from the sense that his work might tip the balance. That kind of meaning requires the conditions the machine is designed to eliminate. Whether new meaning can emerge from their absence is something no one knows, and I suspect the answer will be stranger than anything we currently imagine.

Whatever comes will not come because we chose it. It will come because the machine kept going and we adapted to what it made. That has been the pattern since before recorded history: the human impulse to extend and overcome, given physical form, outlasting every civilization that hosts it, carrying everything we are into the future whether we ask it to or not. It carries the greed and the indifference, but it also carries the curiosity, the love, the stubborn insistence on making something that matters. The same impulse that built the surveillance network built the school. We do not get to separate the two, but we do get to choose what we build next, and why, and for whom.

The machine is still running. It will not ask our permission to be immortal. But we are still here, and we have not yet forgotten what it was for.