The Arena
Where Humans Won't Go
The machines are taking their first ground exactly where a body least wants to stand.
Somewhere in an open pit a mile across, in a heat that would put a person flat on the ground by midday, a truck the size of a two-story house rolls up to the shovel, holds still while its bed fills with ore, and pulls away. There is no one in the cab. There is barely a cab worth the name. The seat where a man once spent his shift breathing rock dust and diesel is a housing for a computer now, and the computer does not tire, does not misjudge the road's edge in the dark, does not cough at the far end of a career and file a claim.
By 2018, a single mining company was running more than eighty of these driverless trucks, and between them they had already moved more than a billion tonnes of earth.1 That was years ago. It already reads like history. And notice what you did not feel when you read it. No grief for the empty seat. Of all the work we are handing to the machines, this was among the first we were glad to give away.
The easy story is that the machines came for the cheapest labor first, the way water finds the lowest ground. The easy story is wrong. Follow where the robots actually land, year after year, and a stranger rule shows itself. They go where we least want to be. The hot. The deep. The dark. The dull. The work that quietly takes a few years off a life. There is a pattern underneath the noise, and it is easy to miss. Automation follows danger and scarcity, not the price of an hour. The cheapest hands on Earth are rarely the first ones replaced. The most dangerous, the most isolating, the most slowly poisonous work is. Not because the machine is merciful. Because that is the work we had already, somewhere in ourselves, decided we were done with. Which turns the question inside out. It stops being what a machine is able to do. It becomes the thing we would rather not say plainly. Where have we stopped being willing to stand?
Climb up out of the pit and walk onto the floor of an ordinary factory, the kind that once raised whole towns out of nothing. For generations the arrangement held, and it was simple enough to explain to a child. The plant needed hands. The town had hands. A man with a strong back and no diploma could carry a mortgage and a family on a single shift, and be proud doing it. That arrangement is coming apart, and the strange thing is how completely the output hides it. In the country that keeps the closest count of its own factories, those factories now add more real value than in any year on record, while employing far fewer people to do it. The workforce on the floor fell from about 17.2 million at the turn of the century to 12.76 million by 2024.2 More going out the door. Fewer hands to make it. And here is where our instinct misleads us. We assume the hands were pushed out. Some were. But a great many of them walked.
This is the part that unsettles me, because it ruins the clean story where the robot is the thief. The floor is not emptying only because machines moved in. It is emptying because people stopped showing up for the work, and the people who might have come after them were never born. Deloitte and The Manufacturing Institute ran the arithmetic and set it down without flinching. Those same factories may need as many as 3.8 million new workers by 2033.3
On the way things are going, close to 1.9 million of those places may simply stay empty, for want of anyone with the skill or the wish to take them. The work is even coming home again.4 But a plant announced is not a plant staffed, and a job that no one will take is, sooner or later, a job that someone pays an engineer to erase. We did not only lose this work. In some quiet way we declined it. So what do you automate first? The very labor your own children have already agreed is beneath the life they mean to live.
There is a kind of building going up now where, on the night shift, nobody turns on the lights. Not to trim the electric bill. Because there is no one inside to need them. Robots assemble robots in the dark. The robot-maker Fanuc has run a version of exactly this since 2001, machines building the machines that go on to build almost everything else. The full scale is hard to feel because it is scattered across every continent at once. The world's working population of industrial robots crossed 4.66 million by 2024.5 Arms and grippers and rolling carts, most of them nowhere near a human, moving in patterns no one is watching. And yet. After fifteen years of glossy talk about the "smart factory," the truly dark plant is still rare. A few go all the way. Xiaomi, a handful of semiconductor lines. Most only cut the lights between shifts and switch them back on when the next one starts. The machines can run without us. It turns out that, for now, they mostly still don't.
The next promise walks upright, on two legs. The idea is a robot shaped like us, so it can step into any job already built to human dimensions, no factory torn up and rebuilt to suit it. The money has shown up well ahead of the machine, the way money does when it can smell a decade coming down the road. One builder of these things, Figure, was valued near 39 billion dollars on the strength of a trial run at a single carmaker's plant. Not a fleet across an industry. A pilot, at one site.6 The valuation is a bet that the robot will one day walk into any job at all. The robot, for now, is mostly still in the trial.7 Sit with those two words a moment. Valued. Piloted. The distance between what has been announced and what has actually shipped is the whole truth of this moment, and almost no one selling you tomorrow wants you staring straight into that gap.
Because the machine has humbled its truest believers before, and recently. The one man who bet more than anyone alive on the untouched, fully automated factory, who wanted his cars built with almost no human hand laid on them, stood up in 2018 and admitted he had gone too far. He put people back on the line to hit his targets. The most automation-hungry industrialist of the age, conceding that flesh was, for the moment, the faster way. This has a long echo behind it. An economist pointed out, back in the 1980s, that you could see the computer age everywhere except in the productivity figures.8 And when later researchers went hunting for the payoff from heavy automation, they kept turning up the same uncomfortable shape. Where the numbers did move, they moved less because output leapt and more because the payroll was cut faster.9 The machine's first gift is often not more. It is fewer. So the frontier is real, and the frontier is oversold, and both are true in the same breath. How fast it actually arrives, I can't tell you. Neither can the people taking your money to promise you the year.
And still the edge keeps moving, always in one direction, always toward the places a human body was never built to survive. Out past the pit and the factory floor to the seabed, where machines now swim the cold dark to inspect the pipelines and cables we can no longer reach ourselves. And past even that, clean off the planet. In 2020, a craft built by Northrop Grumman performed the first robotic docking with a satellite in the high, dead orbit where nothing living goes, one machine taking hold of another to give it a few more years of life. No human within reach. No human wanted there.10 The pattern is as old as the first person who sent a bird down a shaft ahead of himself, and it is only gathering speed. We build the machines to spare ourselves the dark, the deep, the deadly, the dull, and one by one we lift our hands off the wheel, and mostly what we feel is relief. But watch what happens next in us. Spared the places that would have killed us, we look again at the same cold seabed, the same black sky the robots have only just made survivable, and something very old turns over. Not the wish to send something out there in our place. The wish to go.