Paid to Disappear: The Five Wages AI Cannot Pay

A cheque can replace an income. It cannot replace a reason to get up in the morning.

A young graduate stands at dawn before a vast closed glass door of a tower, where robotic arms work inside an empty lobby — The first casualty of AI is not the job you lost. It is the job you never got.

In 1930 Keynes promised his grandchildren a fifteen-hour week by about 2030. He also warned that “everybody will need to do some work if he is to be contented”. His grandchildren graduate into that year. Many will find the door closed.

The Economist’s Bartleby column put it crisply this week: if AI displaces human workers, they will lose more than just pay, because work has value beyond the income it provides. It is a sentence most economists nod at and most AI policy ignores. Almost every proposal on the table, from basic income to Elon Musk’s “universal high income”, treats work as a delivery mechanism for money. Replace the money, problem solved.

The evidence says otherwise. A job pays five wages that never appear on a payslip. AI is starting to withhold them, mostly from people who have not yet earned their first.

I. The five wages on no payslip

In the 1930s the social psychologist Marie Jahoda studied an Austrian village whose factory had closed. She concluded that a job gives us five things besides income: time structure, social contact, collective purpose, status and daily activity. A 2023 meta-analysis of 46 studies and more than 48,000 people confirmed that the unemployed are worse off on all five. The biggest gap is status. Each of the five predicts mental health independently of money, and collective purpose carries the most weight (Frontiers in Psychology).

When those wages stop, the bill arrives in other currencies. When Chinese import competition hollowed out American manufacturing towns, marriage and fertility fell while idleness and early deaths rose (Autor, Dorn and Hanson). Angus Deaton and Anne Case traced “deaths of despair” to the loss of meaning, dignity and pride as much as money (NPR). Studies of job loss in America, Sweden and Denmark find mortality elevated for years, in some cases for up to two decades (Cardus review).

This is the oldest lesson of industrial history, and AI is about to test it at a scale and speed no factory closure ever did.

II. The door, not the pink slip

So is AI destroying jobs? The honest answer, as of October 2026, is: not across the economy, but at the front door.

Stanford’s latest “Canaries in the Coal Mine” update, using payroll data to June 2026, finds employment of 22 to 25 year olds in the most AI-exposed occupations about 19% below where it would otherwise be, up from 15% a year earlier. Experienced workers show no comparable gap. The mechanism is fewer hires, not more firings, and the losses concentrate where AI automates tasks rather than assists people. Erik Brynjolfsson and his co-authors are careful to add: “We find no evidence of widespread, economy-wide job displacement” (Stanford Digital Economy Lab).

Anthropic’s own Economic Index tells the same story from inside the machine. Its June 2026 report, drawing on nearly 10,000 surveyed users, found that “early-career workers report that AI can do the highest share of their work”. They also voice the most fear of losing their jobs, and over a third of respondents put a junior colleague’s chance of job loss above 60% (Anthropic). Its March report found newer users delegating outright far more often than experienced ones, and business sales and trading workflows on its API doubling in three months (Built In). When the builders of the tools and the economists who study them agree on who is most exposed, it is worth listening.

Asia confirms the pattern independently. The Bank of Korea found that 94% of the fall in youth employment since 2022 came in AI-exposed industries, while workers in their fifties gained jobs (JoongAng Daily). India’s five largest IT firms shed staff on a net basis in FY26, and TCS cut planned graduate hiring by around 40%. In the US, firms citing AI accounted for about 23% of announced job cuts in the first half of 2026, although Yale’s Budget Lab warns that some of that is “AI-washing” of other troubles (Fortune).

This is what I have called the Frozen Workforce. It is not unemployed. It is stuck. And the cruellest part is invisible in every statistic we publish: a 23-year-old who is never hired loses Jahoda’s five wages before ever earning them. No pink slip. No severance. Just a door that does not open.

III. A cheque cannot tell you why to get up in the morning

A man sits at a kitchen table at dawn holding an unopened cheque, his work boots unused by the door — The income arrived. The reason to put on the boots did not.

The best test we have of income without work was funded, fittingly, by Sam Altman. OpenResearch paid 1,000 low-income Americans $1,000 a month for three years. Their earned income fell by about $1,900 a year, the freed time went mostly to leisure, and their wellbeing rose in the first year before drifting back to the control group’s level (NBER). Cash helped. It did not supply meaning.

To be fair to the cash camp, Nobel laureates Abhijit Banerjee and Esther Duflo show that guaranteed income does not destroy the will to work: “Data from programs around the world suggests the opposite is true” (WSJ via BRAC USA). Notice what that implies. People keep working even when they do not need to, because work gives them something money does not.

Worse, AI threatens meaning inside the jobs that survive. A study of more than 10,000 occupational tasks found that the work most exposed to AI is precisely the work people find most creative, novel and autonomous (CHI ’26). Another found that after working with generative AI, people doing a task alone showed 11% less intrinsic motivation and 20% more boredom (Fortune). We are not only automating jobs. We are automating the joy out of the ones we keep.

Even the Vatican has weighed in. Pope Leo XIV’s first encyclical, Magnifica Humanitas, devoted to AI, insists that “work is not simply an instrument” (CSAN). Michael Sandel has long argued that “the dignity of work needs to be right at the heart of this” (IFOW). On this one point, a pope and a Harvard philosopher agree with Jahoda’s survey data.

IV. $146 is not a stake

Last week the debate lurched from cheques to equity. Anthropic said it would place 5% of its shares in a public benefit fund. The next day Elon Musk posted: “Next year, I will give one of my SpaceX shares to every kid in America” (Fortune). Social media promptly called it “a piece of Starlink for every American”. It is not. It is one SpaceX share per child, worth about $146 at the time, about $10.7bn in total, with no mechanism yet announced (AP).

A small child on a vast launch site at dusk, dwarfed by a colossal rocket on its tower — A share of the rocket. Not a seat on it.

I do not want to sneer at it. Two owners of frontier technology conceded the ownership principle in a single week, and that matters. In The Aegis of Abundance I argued that if you want an income floor that survives a change of government, you should not build a transfer; you should build an ownership stake. But run the arithmetic. Senator Sanders’ proposed AI sovereign wealth fund, financed by a one-off 50% levy on AI companies’ stock, would build about $7 trillion and pay roughly $1,000 per American a year (Broadband Breakfast). That is exactly what Alaska’s oil fund paid last year. A genuine “universal high income” of $30,000 per adult would cost something like $8 trillion a year, more than the entire federal budget, and Musk has offered no costing beyond the hope that abundance makes inflation disappear. The basic-income scholar Karl Widerquist calls that hope “naïve, oversimplified, wishful thinking” (Widerquist).

So here is the verdict. UBI at modest levels is affordable and benign, but it does not deliver lasting wellbeing. Universal high income is a forecast wearing a policy costume. Ownership dividends are the right principle at the wrong scale. None of them pays the five wages. Trust is the only currency here too: a revocable philanthropic pledge with no mechanism is the opposite of a durable, governed stake.

V. Ghost GDP has a ghost workforce

In The Longest Night Before Dawn I described Ghost GDP: output that rises on paper but never reaches households. It has a human twin. Call it the ghost workforce: people whose status, structure and purpose evaporate even when their income is partly replaced.

Tyler Cowen, no doom-monger, expects no mass unemployment but warns that “those who lose suffer more psychologically than those who gain” (Fortune). Oxford’s Carl Benedikt Frey expects professionals to lose status first, and predicts the politics that follows (Work/Code). The danger of this transition is not mass poverty in 2027. It is mass humiliation, and humiliated electorates do not vote for patient, evidence-based policy.

VI. Protect every worker, including the one not yet hired

Singapore has the world’s most coherent pro-work answer. Prime Minister Lawrence Wong’s line is: “We may not be able to protect every job. But we will protect every worker” (The Straits Times). In May, Parliament backed a motion for AI with “no jobless growth”. Government, unions and employers formed a Tripartite Jobs Council, and the then Manpower Minister rejected an opposition proposal for an AI equity dividend, preferring to “invest in people, not to compensate them for circumstances” (The Online Citizen).

In a tropical city at dawn, a mentor lifts a young apprentice towards the first rung of a glowing ladder whose lowest rungs are missing — The ladder still stands. The bottom rungs are missing.

I agree with the instinct. But the doctrine has a blind spot, and the data are starting to show it. The share of retrenched professionals and managers back in work within six months has fallen to 54.1%, now below other workers (HCAmag on MOM data). A Manpower Ministry poll found 6.2% of firms had cut headcount because of AI and 8.5% had cut hiring. SMU economist Li Jia describes a squeeze on middle cognitive roles and vanishing entry points for the young. Our toolkit is built for the mid-career worker who loses a job. It is not yet built for the graduate who never gets one.

So I would extend the doctrine: protect every worker, including the one not yet hired. Treat the apprenticeship as infrastructure. Budget for the first rung the way we budget for roads.

VII. Tax the token, fund the rung

If work supplies what cash cannot, fiscal policy should fund work, not only cheques. The sharpest ideas on the table point that way. Raghuram Rajan proposes a small tax on the AI tokens firms consume, rising if displacement grows, plus credits for firms that retrain and retain people, noting that a firm pays social security for every worker “but not for AI” (Business Today). A US House bill would tax AI usage and channel the money into housing, infrastructure and care jobs, with rates that rise when unemployment does (Fortune). Nobel laureate Daron Acemoglu, with David Autor and Simon Johnson, argues for ending the tax advantage machines enjoy over workers, so that firms automate only when it is genuinely better, not merely cheaper.

This is Long-AND, not Short-OR, as fiscal policy: bridge incomes AND ownership stakes AND apprenticeships AND a tax system that stops paying firms to replace people. In The Last Human Job Is Answerability I argued that the final human role in an agentic economy is to be accountable for what machines do. Here is the uncomfortable corollary. Answerability is learned at work, on the job, rung by rung. Close the entry door and you lose more than a generation’s sense of purpose. You lose the very people who would have governed the machines.

VIII. The panel weighs in

I widened my usual panel for this one, to economists, Nobel laureates, HR and future-of-work thinkers, and philosophers. Here is where they stand, on the record.

The steerers. Daron Acemoglu (Nobel 2024) thinks the AI job apocalypse is avoidable and that agents replacing whole jobs is “just a losing proposition” (MIT Technology Review). Joseph Stiglitz (Nobel 2001) warns: “We do not have the macro or micro framework” (Fortune). The 2025 laureates Philippe Aghion, Peter Howitt and Joel Mokyr back creative destruction but want its costs managed through competition policy and Danish-style flexicurity; Howitt admits “nobody really knows in great detail what’s going to happen” (SSE).

The dignity camp. Angus Deaton (Nobel 2015) describes what automation and outsourcing severed: “You’re not part of the company anymore” (Richmond Fed). Lynda Gratton of London Business School insists work is “the unpredictable and human journey of work and creativity”, not a bundle of tasks (Gratton). Josh Bersin urges leaders to treat AI as “work enhancing”, not job-replacing (UNLEASH). Daniel Susskind, whose A World Without Work framed this debate, argues that this time really is different and that we must find meaning beyond the job. Singapore’s NTUC chief Ng Chee Meng says simply: “We know that our PMEs are worried about AI impact on their jobs” (The Online Citizen).

The warnings. Anton Korinek models a world where “wages collapse as the economy approaches full automation” (arXiv), the modern version of Wassily Leontief’s warning that people might go the way of the horse. Adam Grant has publicly revised his view that care is a safe human residue, after studies found people felt “more supported, more seen, more cared about” by AI replies (Indeed).

The dissent. Bertrand Russell, in In Praise of Idleness, would call this whole essay a romance of toil, and David Graeber would remind us how many jobs are already meaningless. Gallup finds only 20% of the world’s workers engaged (Gallup). They have a point. But even an unfulfilling job supplies Jahoda’s structure, contact and status, and those are exactly what a cheque cannot buy.

The verdict: the panel converges on “steerable, not inevitable”. Not one economist I could verify endorses UBI as the answer. They split only on who pays and how.

IX. Keynes’s grandchildren graduate in 2030

Keynes was right about abundance and right about contentment. His century runs out around 2030, and the Fork I have described for years is not abundance versus scarcity. It is two ways of holding abundance. Down one branch, Star Trek: machines do the toil, people do the meaningful work, and the gains are owned widely. Down the other, Mad Max with a dividend: a cheque in every letterbox and nothing to get up for.

Short-term turbulence for long-term abundance is still my wager. But abundance that pays the income and withholds the other five wages is not abundance. It is a very comfortable form of exile.

Fund the cheque, by all means. Then fund the rung, the apprenticeship and the reason to get up. That is the wage AI cannot pay, and the one we must.

Further reading on Genesis: The Frozen Workforce: Certified for Yesterday · Schrodinger’s Canary · Retire the Unemployment Rate · Prometheus Has Too Many Fathers

Genesis: Human Experience in the Age of Artificial Intelligence and Synthesis: The Superintelligence Protocol are available now.