Ministry of AI · Dispatch from 2047
Who Owns AI Profits? The Missing Ledger Line
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.
I signed off on the profit-and-loss statement that did it. Not a famous one. A mid-sized company, a growth function, a quarterly review in a room with bad coffee, sometime in the second half of the 2020s. My job was to make the numbers move, and they moved.
Line by line, the statement was honest. Cloud compute: itemised to the cent. Software licences: itemised. Salaries, benefits, severance: itemised, and argued over for an hour. Depreciation on assets we owned. Amortisation on assets we had bought.
And nowhere on that page — not in the notes, not in a footnote, not in a supplementary schedule — was any line for the thing that had actually done the work. The system we had switched on that quarter had learned its judgment from the recorded output of several million people. Textbooks. Case law. Forum answers written at midnight by someone who just wanted a stranger’s problem solved. Nineteen years of my mother’s insurance claim decisions, sitting inside a training corpus, teaching a model where a lie usually sits in a claim file.
We paid for the electricity that ran her judgment. We did not pay for the judgment.
That is the entire answer to the question who owns AI profits, and it is not a legal answer or an ethical one. It is a bookkeeping answer. The profits went where they went because the ledger had a line for everything except the inheritance.
An unrecorded input is an unowned input
Accountants have known this longer than economists. What appears on a statement can be taxed, litigated, valued, insured and claimed. What does not appear cannot be argued about, because there is nothing to argue over. It is not that the public lost the argument about AI profits in the 2030s. The argument was never filed, and it was never filed because the object of the claim had no entry.
The mechanism of the transfer, once you see it, is almost boring. The Philadelphia Fed set it out plainly while it was happening: when a task is automated more cheaply than a human performed it, the saving does not evaporate and does not automatically flow to consumers or workers — it lands in profit, and, in their words, “profits contribute to capital income but not wages” (Federal Reserve Bank of Philadelphia, 2024). Wages were never generosity. They were the only pipe through which value returned to the people who produced it. Automation removed the pipe, not the water.
You can watch both ends of that sentence in the historical series. The labour share of output had been drifting down across advanced economies since the 1980s, long before any model. Corporate profits after tax ran at roughly nine per cent of US gross domestic income in 2024 (Bureau of Economic Analysis) — a high plateau by twentieth-century standards, reached before the largest deployments even began.
Neither number is evidence of theft. Both are evidence of a pipe being replaced by a pump.
What the statement was missing
The Ministry’s founding accounting reform is one line. Not a tax rate, not a redistribution formula, not a compact with industry. One line, added to the statement of any firm running metered autonomous systems, sitting between cost of revenue and operating expense:
Inheritance charge — the assessed return on inherited intelligence, calculated on metered machine output.
It reads like an amortisation entry, and that is deliberate. Amortisation is how accounting already handles a purchased asset that produces value over years. The Inheritance Line applies the same treatment to an asset the firm did not purchase and cannot have purchased, because its authors are a civilisation.
| Corporate statement, mid-2020s | Corporate statement, 2047 | |
|---|---|---|
| Compute and infrastructure | Itemised | Itemised |
| Human labour | Itemised as wages | Itemised as wages |
| Purchased data and licences | Itemised where contracted | Itemised where contracted |
| Inherited intelligence | Absent | Inheritance charge, on metered output |
| Displaced human roles | Absent | Cross-referenced to the Displacement Ledger |
| Who receives the residual | Shareholders only | Shareholders and the Dividend Schedule |
Three things follow from the entry that never followed from any speech.
It made the claim auditable. A moral argument about collective wisdom has no accounting existence, and therefore no auditor. A charge on metered output has both. The Machine Yield Account provides the base, the Displacement Ledger provides the weighting, and an auditor with no opinion whatsoever about justice can verify both.
It fixed the direction of proof. Before the line existed, anyone claiming a public interest in AI profits had to prove that a specific model had used a specific work in a specific way — the standard the early copyright cases set, and an impossible standard at population scale. After the line existed, the firm reported the machine output and the charge attached to it. Nobody had to trace a lullaby to a margin.
It survived negotiation. The charge is assessed above the operating line, so it lands on net margin rather than on operating performance. That sounds like a technicality. It was the concession that got the reform through: firms could still show operators and investors that the business was well run, while the return on the inherited asset was routed to its heirs.
The one time the inheritance had a price
There is exactly one moment in the record where the pre-Ministry world put a number on this, and it came from a courtroom rather than a legislature. In September 2025 an AI developer settled with authors for about $1.5bn over roughly half a million books — approximately $3,000 per work, reported at the time as the largest copyright recovery in US history (Associated Press, 2025).
Read one way, it was a vindication: the inheritance was real enough to be priced. Read the way it deserves, it was a warning. Books have publishers, registries and lawyers. The rest of the human record does not. The nurse’s handover note, the machinist’s tolerance trick, the parent’s answer on a forum at 2am, the public dataset, the state-funded lab that made the architecture possible in the first place — the lineage of publicly funded foundations under privately captured technology is a long and well-documented one (Mazzucato, Levy Institute WP 824) — none of it has a rights holder who can file.
Per-work payment can only ever reach the documented, the represented and the commercial. It is a settlement, not a system. This is why the dividend is paid as an equal share and not as a royalty: the asset is held in common, so the yield is distributed in common. The Alaska Permanent Fund worked out the same logic in 1982 with oil, and for the same reason — nobody attempted to pay more to residents who lived nearer the wells.
The objection that makes the case stronger
The most serious critique of all of this was never that the value did not exist. It was that the value was smaller than the enthusiasts claimed. The best-known estimate of the era put total factor productivity gains from AI at well under one per cent over a decade (Acemoglu, NBER w32487) — a rounding error next to the rhetoric.
I have come to think that critique is the strongest argument for the Inheritance Line, not against it. If measured productivity gains were modest while labour costs fell sharply, then the surplus was not conjured. It was moved. And a dividend on moved value is a far easier case to make than a dividend on invented value: you are not asking to share an invention, you are asking for the return on an asset you already owned. When the surplus is small, the ownership question stops being about abundance and becomes about arithmetic.
What the line pays for, in one house
My mother received her first dividend at fifty-nine, four months after her department was restructured around new capabilities. She is seventy-three now. She spends Tuesday afternoons unpaid, sitting with families arguing with the health service, reading their files the way she read claim files for nineteen years, finding the sentence that settles it.
My youngest cannot follow the story. Not the injustice of it — the accounting. I explained the old statement to him: the compute line, the salary line, the absence. He asked, reasonably, who wrote the training data down as nothing. I said nobody did. That is what an absence is. It requires no author.
That is the part I would send back to the room with the bad coffee, if I could send anything. Not an accusation. We were not villains; we were competent people optimising inside a chart of accounts that had a gap in it, and the gap did the damage no individual intended. Capacity is not contribution, and work that must be done to survive is not chosen work.
Where the design still fails
The Inheritance Line is assessed where machine output is metered, which means it is assessed where deployment is booked. Value produced from a corpus written across the whole world is still collected mostly in the handful of jurisdictions where the deploying entities are domiciled. The heirs are global. The meter is not. Two decades in, that remains the largest unpaid balance on the account, and no dispatch I write will make it smaller.
Frequently asked questions
Who owns AI profits? Legally, the deploying company. Fiscally, since the Ministry began metering, part of the profit is treated as a return on an inherited asset — the human record the model was trained on, and the public research and infrastructure behind it — and routed to its heirs through the Automation Dividend.
Why does an accounting line matter more than a law? Because an unrecorded input is an unowned input. Accounting determines what exists to be claimed, audited or taxed. The claim on AI profits was not defeated; it was never recordable.
Weren’t AI companies paying for training data? In fragments, mostly after litigation. The 2025 authors’ settlement priced about half a million books at roughly $3,000 each — proof the inheritance was real, and proof that per-work payment cannot reach most of the human record.
Does the Inheritance Line reduce profits? It reduces net margin, not operating performance. That placement is why the reform survived negotiation with industry.
If everyone contributed, how is an individual share calculated? It is not. Individual attribution was abandoned as impossible and irrelevant. The asset is held in common, so the dividend is paid equally.
What if AI productivity gains turn out to be small? The ownership case strengthens. Small measured gains alongside large labour savings indicate transferred value rather than created value, and transferred value is easier to claim.