Standing in the ruins of a simulated city, Morpheus outlines the history of the world: “We have only bits and pieces of information, but what we know for certain is that at some point in the early twenty-first century, all of mankind was united in celebration. We marvelled at our own magnificence as we gave birth to A.I.” Unsurprisingly, a war followed, and humanity blacked out the sun to starve the machines. The machines answered by turning humans into living batteries and harvesting them for energy. The Matrix, Morpheus explains, exists to keep the crop docile.
Back in the real world, a quarter of the way into the twenty-first century, our nightmare runs the other way. We do not fear becoming indispensable to the machines. We fear becoming useless to them. And the warnings come from the people building them. “Probably none of us will have a job,” Elon Musk predicted in 2024. OpenAI’s Sam Altman expects “entire classes of jobs” to go away. Anthropic’s Dario Amodei has warned that half of entry-level white-collar jobs could vanish within five years. The IMF reckons almost 40 per cent of global employment is exposed to AI. Everyone is counting the jobs AI will eliminate. Almost no one sees the ones it has already created and depends on.
A few weeks ago, a job advert landed in my LinkedIn feed. A company I had never heard of, recruiting on behalf of one of the major AI labs, was looking for experienced university lecturers to help train its latest language model. We are told AI is coming for our jobs. Here it was, offering me one. I had to know why. I took the red pill.
The mirror swallowed me faster than it swallowed Neo: my acceptance arrived fifteen minutes after I submitted my CV. Then I woke in the pod — a group chat with thousands of other recruits (academics, doctors, lawyers, businesspeople and consultants), all as disoriented as I was, all asking each other what exactly we had signed up for. The rescue came in the form of an induction session, which finally explained what the machine needed from us. Then came the training and the screening tests. Neo had kung fu uploaded into his brain. We were being tested on whether our skills were worth downloading.
Having passed the tests, I was deemed worthy and allowed to work. My first tasks involved evaluating the layouts of AI-generated documents. The machine’s most common mistake was failing to understand that a heading belongs at the top — not at the bottom, not in a corner, and not in tiny font behind a table or chart. I used to think marking first-year undergraduate essays was the dullest way an academic could spend an afternoon. How little I knew. What surprised me the most was that thousands of highly qualified professionals had not been hired to tell the machine what was right — there are only so many correct ways to place a heading. We were paid to spend days eliminating the infinite number of wrong ones. That, it turns out, is much of what human expertise is: recognising the countless ways of being wrong. And that is what the machine was harvesting.
I lasted a week. The work was mentally draining: hour after hour of the same judgement, applied to the same mistakes, made by a machine that never got tired. Cypher was right: ignorance is bliss. I could not manage it. Others stayed, and I understand why. The work is steady, the hours are flexible and the pay is good.
And there are a great many others, “fields… endless fields.” Mercor, one of the firms that recruits doctors, lawyers, bankers and academics to train AI models, says its experts earn an average of $95 an hour, and that it pays them more than $2 million a day. Surge AI, whose clients include Google, Meta, Microsoft and Anthropic, earned $1.2 billion in 2024 from a workforce of more than a million gig workers across some 50 countries. Meta paid $14.3 billion for a 49 per cent stake in Scale AI, another supplier of human-trained data. This is not a cottage industry. It is a multibillion-dollar supply chain, and its raw material is human minds.
We have been told that AI is coming for white-collar jobs first, with the rest to follow. But if it needs thousands of highly paid experts to teach it where a heading does not belong, unemployment may not be our worst nightmare. The more likely future is not one in which AI works instead of us, but one in which we work for it: not as its consumers, but as its suppliers. In the Matrix, humans lie in their pods, wired in, dreaming they are free, while the machines harvest what they need. We are awake, and paid, and we log in by choice. But the arrangement is not so different. The machine still needs the crop. It has already acquired a taste for what grows in our heads.
Rumour has it the Wachowskis first imagined humans as processors, not batteries. If so, they should have trusted their first idea.