Ministry of AI · Dispatch from 2047
Simon's Ninety Percent: The Tax Nobody Levied
Editorial note. The Ministry of AI is a work of disciplined foresight: it describes the year 2047 in the present tense, and treats our own era as history. The institutions are imagined. The economics, the evidence and the historical parallels are real and sourced.
An economist wrote the number down in the summer of 2000, in an essay of about a thousand words, and then almost nobody did anything about it for forty-seven years.
He was near the end of a long life, a Nobel laureate replying to a young philosopher’s proposal for an unconditional income. His argument took one page. Wealth in rich societies, he said, is not mostly produced by the people who receive it. It is produced by the stock those people inherited: literacy, accumulated science, courts that enforce a contract, roads, a currency, a language, and the trained judgment of every human being who worked before them. Then he did the thing economists are trained to do and put a magnitude on it. Comparing the poorest nations with the richest, he wrote, it is hard to conclude that social capital produces less than about ninety percent of income in wealthy societies like the United States or Northwestern Europe — and on moral grounds one could therefore argue for a flat income tax of ninety percent to return that wealth to its real owners.
Ninety percent. Not a metaphor, not a provocation dressed as arithmetic. A working estimate, published by a man whose life’s work was how decisions actually get made in organisations, in a book-length forum that a few thousand people read.
My youngest son is twelve and finds this story funnier than I do. Somebody worked out that most of what everyone earned belonged to everyone, he said, and the answer was to keep going exactly as before? Yes. That is the answer. I was one of the people who kept going.
The wound in that number
I want to be careful here, because the ninety percent claim is easy to cheapen into a slogan about how nobody deserves anything.
My mother spent nineteen years reading insurance claim files. She was good in a way that resists description: she could feel which silence in a statement meant grief and which meant fraud. That skill was hers. She built it, at a desk, over two decades, and it cost her evenings.
And every single input to it was inherited. The language the claims were written in. The statistical training she got from a public university. The legal architecture that made a claim a claim. The forty years of adjuster practice her supervisors compressed into six months of instruction. If you subtract the inheritance from her judgment, you are not left with a smaller version of her judgment. You are left with nothing at all — an intelligent woman in a room with no tools.
That is what the ninety percent meant. Not that she was undeserving. That she was standing on something enormous, and the something had no owner on the books.
Which was tolerable, for a while, because of an accident. The inheritance was cashed out through wages. A company needed the compressed knowledge of the human record, and the only way to obtain it was to hire a person carrying some, and pay her monthly. Payroll was a spectacularly inefficient, spectacularly effective distribution mechanism for social capital. It never called itself that. It worked anyway.
Then we built machines that carried the same inheritance without the person.
Why the ninety percent could never be collected
The reason that essay changed nothing is not that its readers were greedy. It is that it proposed a tax on persons, and a moral claim aimed at persons cannot be metered.
Consider what a ninety percent inheritance levy would have required. You would have to assert, of one specific accountant in one specific year, that ninety percent of her salary was not hers. There is no instrument that can show this. The inherited component of a human wage is real in aggregate and invisible in the individual case — every economist could agree with the average and none could defend a single assessment. So the claim stalled where all unmetered moral claims stall: it became a thing to quote in seminars.
Its critics were right about the incentives, too. A ninety percent marginal rate collapses the reason to do difficult work, and the essay itself conceded that incentive problems would need serious handling. The doctrine that eventually built the Ministry took that concession seriously, and drew the opposite operational conclusion from the same moral premise: never assess the human. Assess the machine.
Because at the point of machine deployment, the philosophical problem becomes an engineering problem. When an autonomous system does work a paid human would otherwise have done, the inherited component is not a hidden fraction of someone’s merit. It is the entire artefact. The model is a compression of the human record — books, forum answers, repair manuals, court rulings, publicly funded science, publicly built networks, publicly educated engineers. There is no personal effort tangled up in it to protect. The heir is not being asked to give back ninety percent of what he earned. The estate is simply being read where it sits in plain view.
| The 2000 proposal | The Ministry’s instrument | |
|---|---|---|
| What is assessed | Persons, on total income | Autonomous systems, on produced value |
| Where the inheritance sits | Diffused inside every wage, unmeasurable individually | Concentrated in the model itself, observable at deployment |
| Evidence needed | A defensible share of one earner’s merit | A measured output boundary, with a published error band |
| Incentive effect | Blunts the return to difficult human work | Leaves human earnings untouched |
| Political durability | None; never enacted anywhere | Survived two governments and one hostile parliament |
| Coverage | In principle, all income | Only where machines displaced paid human work |
That last row is the honest one, and I will come back to it.
What the meter actually reads
The Machine Yield Account does one thing: it estimates the value of work performed by autonomous systems that a paid human would otherwise have performed, at the site of deployment, in the jurisdiction where the systems run. The Displacement Ledger records which roles were absorbed, whether or not anyone was dismissed. The Dividend Schedule allocates the yield to the income floor, healthcare, housing, care and education. The Contribution Record notes what people choose to do, and is never — under any government, under any fiscal pressure — a condition of the floor.
The doctrine behind those four instruments came from operators, not legislators. In the mid-2020s, the first cohort running coordinated AI systems inside real companies were the first people on earth to watch the human-machine boundary move under their own payroll, month by month, in numbers they signed. They noticed something an economist writing in 2000 could not have seen: the ninety percent had stopped being an abstraction and started being a line item. Every task shifted from a person to a system converted an inherited asset directly into margin, and the conversion was visible in their own dashboards.
They were right about the direction, and the profession caught up. The clearest theoretical account of AI’s macroeconomic effect showed that plausible task-level automation produces modest total output gains alongside a substantial redistribution from labour to capital; the long-run labour share series already sloped downward before the models arrived; and a central bank’s own assessment stated the mechanism without decoration: automated tasks convert labour cost into margin. None of that was a prophecy about robots. It was the ninety percent, quietly changing hands.
Tuesday, and what it cost to get there
My mother is seventy-three. On Tuesdays she sits with families who are fighting the health service, unpaid, because she is still the best reader of a bad file I have ever met. Her dividend arrives on the ninth. She calls it her money, and she is right in a way the essay of 2000 would have recognised: it is not a rescue, it is a distribution to an owner. When she was made redundant at fifty-eight by a system trained on her own decisions, the thing that humiliated her was not poverty. It was being recategorised — from a woman with judgment to a case with needs. The dividend is what a society pays when it declines to make that recategorisation. The best real evidence from the trials of my century said the same in a colder register: people receiving unconditional transfers did not stop working; they worked differently, invested more, and slept.
And here is the cost, since a dispatch that ends in comfort is advertising.
The Ministry collects on the visible boundary only. It meters machines, not the inherited component of ordinary wages, which means the great majority of that ninety percent is still uncollected and probably uncollectable. We built an instrument narrow enough to defend and then declared the defensible part to be the whole moral claim, which it is not. Every year some committee proposes extending assessment to human income on the original 2000 logic, and every year we refuse, because we cannot measure it and an unmeasurable levy is how institutions die.
So the essay is still true and still mostly unenforced. We collected the tenth of it that has a meter attached. My son will inherit the argument about the rest.
FAQ
What is social capital and how does it relate to income? Social capital here means the inherited stock a society hands every earner before they earn anything: language, literacy, accumulated science, public infrastructure, legal order, and the compressed judgment of everyone who worked before you. The claim is that income is mostly a return on that stock, making each earner a beneficiary as much as an author.
Who first argued that ninety percent of income comes from social capital? A Nobel-winning economist, in a short essay published in 2000 responding to a basic-income proposal. He argued that comparing poor and rich nations, social capital can hardly produce less than about ninety percent of income in wealthy societies, and that this would morally justify a ninety percent flat tax to return the wealth to its real owners.
Why did the argument never become policy? It was aimed at persons. Nobody can defend the inherited share of one individual’s earnings, so the estimate remained an aggregate with no instrument behind it — and a ninety percent marginal rate would have flattened the incentive to do hard work.
How does the Ministry avoid repeating that failure? By assessing machines rather than people. Where an autonomous system does work a paid human would otherwise have done, the inherited component is the entire artefact rather than a contested fraction of someone’s merit, so the measurement is defensible in a hearing.
Is the Automation Dividend welfare by another name? No. Welfare is justified by need and withdrawn when need is disputed; a dividend is justified by ownership and paid regardless of whether the owner is sympathetic. That difference decides what recipients are called, and therefore how they are treated.
What remains unresolved? Coverage. The meter reaches only the visible human-to-machine boundary. The rest of the inheritance — the part still flowing invisibly through ordinary wages and returns — has never been collected, and no one has yet built an instrument that could.