section 03 of 11

This Is Not the Industrial Revolution

The first thing anyone says when confronted with the scale of AI disruption is: we have been here before. The industrial revolution destroyed entire categories of work, and we adapted. New industries emerged. New jobs were created. The economy restructured and, in the long run, everyone was better off. This is the most repeated reassurance in every boardroom, every policy discussion, every op-ed about the future of work. It is also, if you actually look at what happened, an argument for panic rather than comfort.

The industrial revolution “worked out” in the sense that, roughly a century and a half later, the descendants of displaced workers lived in relative prosperity. What happened in between was not adaptation. It was catastrophe. Sixty to eighty years of child labor, sixteen-hour workdays, urban squalor, cholera epidemics, mass displacement from countryside to factory slum, political upheaval, the rise of radical ideologies, and — downstream of all of it — two world wars that killed a hundred million people. The weavers who lost their livelihoods to power looms did not “adapt.” They suffered and died, and their grandchildren’s grandchildren eventually built a world that worked. Calling this a success story is survivorship bias applied across a century-plus timescale.

But even granting the optimistic reading — even accepting that the long arc eventually bent toward prosperity — the industrial revolution had a structural advantage that this transition does not.

The industrial revolution replaced muscle. AI replaces judgment.

When machines took over physical labor, humans still had something to offer that no machine could match: the ability to think, decide, communicate, and adapt. The displaced weaver could become a factory supervisor, a clerk, a teacher, an administrator. Not easily, not painlessly, but the path existed because these roles required a capability — cognition — that remained exclusively human. The entire modern economy is built on this premise. The shift from manufacturing to services, from manual labor to knowledge work, from making things to managing information — all of it rests on the assumption that human thinking is the scarce, valuable input that machines cannot replicate.

AI attacks this assumption directly. Not at the margins, not in narrow domains, but across the full spectrum of cognitive work. Writing, analysis, translation, coding, diagnosis, strategy, design, legal reasoning, financial modeling — the list of domains where AI performs at or above human level is growing not by the year but by the quarter. The standard response is that humans will move to “higher-level” tasks: creativity, complex judgment, emotional intelligence, strategic thinking. But these are not a stable refuge. They are the next frontier of AI capability, and the timeline is months, not decades. The escape hatch keeps moving upward, and the ladder is running out of rungs.

This creates a dynamic that has no precedent in economic history. Previous automation waves followed a pattern: a machine automates a task, the human moves to an adjacent task that requires some capability the machine lacks. The pattern worked because machines had categorical limitations. A power loom could not think. An assembly robot could not improvise. The boundaries were hard, and humans could always retreat behind them.

AI does not have categorical limitations in this sense. It has current limitations — things it does poorly today — that are eroding on a timeline measured in months. The human worker who retrains for a new role may find that by the time the retraining is complete, the new role is also within reach of the next model. This is not job transformation. It is a race where the finish line moves toward you faster than you can run.

And the institutions that would need to manage this transition cannot keep pace — not because of policy failures or political cowardice, but for a reason that runs deeper than either. Human beings are running on cognitive hardware that was optimized, over roughly three hundred thousand years of evolution, for small-group survival on the African savanna. Short time horizons, because planning past a season didn’t improve reproductive fitness. Tribal loyalty over abstract cooperation, because your tribe was your survival unit. Status competition over collective action, because hierarchy determined access to mates and resources. Loss aversion over rational calculation, because in a scarce environment, the cost of losing outweighed the benefit of gaining.

This is not loose metaphor. It is evolutionary mismatch: a set of instincts tuned for one environment, now running in a world nothing like the one that shaped them. The traits that kept a small band alive on the savanna — discount the distant future, trust the in-group, chase status, fear loss more than you value gain — are the same traits that make eight billion strangers unable to coordinate across decades. Every institution we have built — democracy, markets, international law, regulatory agencies — is a workaround for that mismatch, a scaffold that lets humans cooperate past the scale their instincts were built for. And the workarounds are failing, because the problems they now face exceed their design parameters. Climate change requires century-scale cooperation across every nation on Earth. We have spent fifty years failing to achieve it. AI governance requires the same kind of cooperation, on a faster timeline, with higher stakes.

The industrial revolution unfolded over generations. Institutions had time — ugly, insufficient, blood-soaked time — to adapt. Labor unions formed. Public education systems were built. Safety regulations emerged. Social insurance was invented. None of it happened fast enough to prevent decades of misery, but it happened.

AI capability is advancing on a timeline of months to years. The institutional capacity to respond operates on a timeline of decades. This is not a gap that effort or urgency can close, because the bottleneck is not will. The bottleneck is the glorified lizard brain running the show — the one that evolved to track threats in tall grass and compete for status around a campfire, now confronted with a problem that requires coordinated global action sustained over decades.

The industrial revolution is not a precedent for what is coming. It is a warning — and an inadequate one, because the thing that saved us last time, the one structural advantage we had, was that human cognition remained more valuable than anything a machine could do. That advantage is dissolving. The safety net had a floor, and the floor was us. What happens when the floor disappears?

Surely physical labor, at least, is safe. The body is harder to replace than the mind. You cannot automate a plumber with a chatbot. Can you?

going deeper · optional

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

This section rests on three claims. Two are solid. One needs a correction that the essay used to get wrong.

“It worked out” hides the cost. The stretch from roughly the 1780s to the mid-1800s in industrializing Britain was not a smooth upward glide. Urban life expectancy in the worst manufacturing cities fell below rural levels for a generation. Child labor, twelve-to-sixteen-hour days, cholera outbreaks tied to overcrowding, and periodic mass unemployment were the texture of the transition, not the exception. Real wages for many workers stagnated or fell for decades before the sustained rise began — the “Engels pause.” The point is not that industrialization failed. It succeeded, eventually, spectacularly. The point is that “eventually” was measured in generations, and the people who paid the cost were not the people who collected the benefit. Citing the outcome while skipping the interval is where the reassurance cheats.

Muscle and judgment are different targets. Every previous wave of automation replaced a physical or narrowly procedural capability and left a cognitive one standing to retreat into. The loom took the weaver’s hands; the weaver’s judgment was still worth something. This is the structural feature that made “we adapted” true — there was always a higher rung. The claim here is narrow and defensible: when the capability being automated is judgment itself, the retreat direction is no longer obviously open. You do not have to believe any particular timeline to accept the asymmetry.

The correction — read this if you came to argue about the “lizard brain.” Earlier drafts of this section said the coordination problem “is not a metaphor, it is neuroscience,” and leaned on the picture of a rational cortex bolted onto an ancient reptilian core. That picture — the triune brain, MacLean’s model — is not how modern neuroscience describes the brain. Brains did not evolve in tidy additive layers, and there is no discrete “lizard” module in charge. So the anatomy claim is dropped. What survives, and what the argument actually needs, is evolutionary mismatch: a psychology selected for small-group, short-horizon survival now operating in a global, long-horizon world. Steep discounting of the future, in-group preference, status competition, loss aversion — these are well-documented behavioral tendencies, and they are a poor fit for problems that require sustained cooperation among strangers at planetary scale. “Lizard brain” stays in the essay as a figure of speech, not a brain diagram. The floor it holds up is mismatch, and mismatch stands.

One honest concession is in the thinking-aloud note below.

— thinking aloud —

Here is the part I am least sure of, and I would rather say it than have you catch me hiding it.

The section slides from “our instincts are mismatched to this problem” toward “so we cannot solve it.” Those are not the same claim, and the second one is doing more work than I can strictly defend. Mismatch is a headwind, not a wall. We have occasionally coordinated against it. The hole in the ozone layer is the cleanest example: a genuinely global, long-horizon, invisible threat, and the Montreal Protocol actually worked — near-universal ratification, the CFC phase-out, the layer measurably recovering. That is exactly the kind of cooperation the section implies we are incapable of, and we managed it.

So why do I still think the pessimistic read is right for AI, and not just gloom for its own sake? Because the cases where we coordinate share features this one lacks. Ozone had a small number of chemicals, a small number of manufacturers, cheap substitutes already on the shelf, and no serious military or economic advantage to defecting. AI has the opposite profile on every axis — diffuse, dual-use, no substitute for “having the better model,” and a direct competitive and military payoff for whoever refuses to slow down. Mismatch is not destiny, but the escape hatches that worked before do not seem to be here.

That is a judgment call, not a proof. If you think the Montreal case generalizes further than I am giving it credit for, that is a real argument, and it is the one I would push on if I were you.

Objections 5 the strongest counters, answered
But the industrial revolution did work out. We live longer, richer, healthier lives than any pre-industrial human. Isn't that the whole point?
Yes — and the section doesn't deny it. The claim isn't that the transition never resolves. It's that resolution took generations, the cost fell on people who never saw the payoff, and it only worked because displaced humans still had cognition to sell after machines took muscle. Point to the destination all you like; the argument is about the road and about which capability is being automated this time.
The 'lizard brain' stuff is pop-neuroscience. The triune brain model is debunked. Doesn't that sink the whole section?
You're right about the model, and the section no longer rests on it — the anatomy claim was removed. 'Lizard brain' now stands only as a figure of speech. The load-bearing claim is evolutionary mismatch: instincts selected for small-group, short-horizon survival operating in a global, long-horizon world. That's ordinary behavioral science — discounting, in-group bias, loss aversion — and it doesn't require any layered-brain story to hold. If you can knock down mismatch itself, you've got something. Knocking down the metaphor doesn't touch it.
AI progress is going to plateau. 'Timelines of months' is marketing. Scaling laws are already bending.
Maybe. The section doesn't actually need exponential progress — it needs the response to be slower than the disruption. Even a decade-long rollout outpaces institutions that take decades to legislate, fund, and implement. Slow the AI timeline down by a factor of five and the mismatch between disruption speed and adaptation speed still holds. The 'months' framing is the aggressive version; the argument survives the conservative one.
Humans always invent new job categories nobody could have predicted. Nobody forecast 'app developer' in 1990.
Historically true, and it's the strongest optimistic case. The catch is that every unforeseen category we've ever invented still ran on human judgment — that was the common ingredient. The section's specific claim is about what happens when judgment itself is the thing being automated. New jobs may appear; the question is whether they require a capability that stays exclusively human, and that's exactly the assumption under pressure.
This is just Luddite panic in a nicer suit. Every generation thinks its technology is the one that ends work.
The Luddites were wrong — about muscle. The pattern of failed predictions is real and worth respecting. But 'people cried wolf before' is not itself an argument that there's no wolf; a turkey fed every morning for a thousand days has excellent evidence the farmer is friendly. The section's whole burden is to name the disanalogy — muscle then, judgment now. If that disanalogy is wrong, the Luddite charge lands. If it's right, prior false alarms are irrelevant.
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).