Ministry of AI · Dispatch from 2047
Intergenerational Wealth AI: The Dead Are Shareholders
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.
My mother is seventy-three and keeps a shoebox of her father’s letters. He died in 1994, a machinist who wrote badly and often, and last winter my youngest — he is twelve — asked whether his great-grandfather was “in the model.” It was not a sentimental question. He has grown up with the Ministry’s provenance vocabulary the way I grew up with postcodes, and he wanted to know, procedurally, whether letters that were never published, never scanned, never posted anywhere could be part of the inheritance he collects a share of every month.
The honest answer was no. Those letters are in a shoebox. But the question was better than the answer, because it exposed the thing the first decade of this argument kept tripping over: almost everyone whose work made these systems capable is dead, and the dead cannot file a claim.
The wound nobody could compensate
I spent the 2020s building growth and measurement systems inside companies switching on their first coordinated AI stacks. When the compensation fights started, they were fought over the living. Living authors sued. Living illustrators organised. Living developers discovered their public repositories inside products they had not been asked about. Every proposal that emerged — licensing pools, opt-outs, per-work settlements — was a mechanism for paying somebody who could be located, notified, and sent a payment.
Then someone in one of those rooms did the arithmetic on the corpus itself, and the argument changed shape permanently. The living were a minority of it.
Consider what the systems actually learned from. Public-domain literature, which by definition means works whose authors have been dead long enough for copyright to lapse — in the United States, generally the author’s life plus seventy years (U.S. Copyright Office). Digitised libraries built precisely to collect those works: Project Gutenberg began in 1971 and spent half a century putting out-of-copyright books into machine-readable form (Project Gutenberg). Openly documented training corpora of the era were explicit about their composition, including large book collections and archived web text (The Pile, 2020; Common Crawl). Encyclopaedic text written by millions of contributors over two decades (Wikipedia statistics), a population that by the 2040s includes a great many people who are no longer alive. Later audits of dataset licensing found the provenance of these collections to be far messier than the licence labels suggested (Data Provenance Initiative).
The scientific layer is older still. A system that reasons competently about materials, disease or logistics is standing on results published by people who died before anyone imagined a machine reading them. The physics is dead people’s physics. The mathematics is dead people’s mathematics. The idioms, the metaphors, the sentence rhythms that make the output sound like a person are the accumulated deposit of writers whose names nobody can recover.
So the compensation frame collapsed under its own logic. If the corpus is mostly the work of the dead, then a scheme that pays the locatable living is not justice; it is a rounding error with a press release. Per-work pricing — the mechanism this section examined earlier — can settle a lawsuit. It cannot settle an estate.
The turn: this is probate, not intellectual property
The move that unlocked the Ministry was giving up on property law and picking up inheritance law instead.
Every legal system already has procedures for an estate whose heirs cannot be found. Property with no locatable owner does not vanish and does not become the finder’s; it passes by escheat to the state, under statutes designed to hold unclaimed value rather than extinguish it (Cornell LII). And every will has a residuary clause, the provision that catches whatever the specific bequests did not — the remainder, assigned to a residuary beneficiary (Cornell LII).
Read the corpus as an estate and the whole problem re-sorts itself. There is an asset of enormous productive value. There is a set of contributors, most of whom are dead. There is no register of them, no probate file, no executor. What there is, for the first time in history, is a yield: a machine that converts that inherited deposit into output at commercial scale, continuously, and books the result as margin.
The Ministry’s founding finding is called the heirless-estate finding, and it says something narrow and legally boring, which is why it survived. It does not say that model outputs are derivative works. It does not assign copyright to anyone. It says: the productive capability of a deployed system descends in material part from an estate whose heirs cannot be individually established, and therefore the yield attributable to that capability is residuary — it must be distributed, not absorbed.
The finding turns on four tests, applied to a deployed system rather than to any individual work:
- Deposit — is the capability materially dependent on accumulated prior human work, as distinct from the deploying company’s own novel engineering?
- Untraceability — is the contributing population unenumerable at scale, such that no register of heirs could be assembled honestly?
- Mortality — does a substantial share of the deposit originate with contributors who cannot receive payment because they are dead?
- Yield — is there a measurable stream of value being booked from that capability now?
All four must hold. Where they do, the system’s assessed yield enters the Machine Yield Account as residuary, and the Dividend Schedule distributes it. Where they do not — a model trained only on data a company generated itself, which exists and is rarer than you would think — nothing is assessed. The tests are the reason the Ministry is a meter and not a levy.
| Frame | Who gets paid | Where it breaks |
|---|---|---|
| Copyright / licensing | Locatable rights-holders of specific works | Ignores the dead majority; pays lawyers and estates, not authors |
| Descendant claim | Traceable heirs of identified contributors | Documented families collect; anonymous and oral traditions get nothing |
| Escheat to treasury | The state’s general fund | Becomes a budget line; reallocated by the next government |
| Heirless-estate residuary | Every living person, equally | Equalises an unequal contribution — the concession we did not solve |
Why we refused to pay the descendants
The descendant claim was the emotionally obvious answer and it was rejected early, over real objection. If my great-grandfather’s published work is in the deposit, why shouldn’t his share come to my mother’s shoebox?
Because that rule pays documentation, not contribution. The families who could evidence descent from a named author are, almost perfectly, the families who already had literacy, publication, property records and lawyers. The people whose ancestors wrote anonymously, wrote in languages that were never digitised, sang rather than wrote, or had their work published under an employer’s name would have been excluded by the same procedure. Inherited advantage is not a small channel in how wealth ends up where it is; historical work on France put the annual flow of inherited wealth at roughly 20–25% of national income for most of the nineteenth century, down under 5% by the 1950s, and climbing back toward 15% by 2010 (Piketty, 2011). A descendant rule would have run the AI inheritance through that same channel and called the result fairness.
So the residuary beneficiary is the living population. Not because everyone contributed equally. Because the estate has no way of proving anyone contributed more, and a payout rule that pretends otherwise would be lying with a straight face.
What it looks like on a Tuesday
My mother’s first dividend arrived when she was fifty-nine, eight months after the claims system she had trained — nineteen years of her judgment, read into a model — finished replacing her. She now spends Tuesdays sitting unpaid with families arguing with the health service, because she is good at reading a file and knows what a denial letter is designed to do. Her Contribution Record notes it. Nothing in the floor depends on it, and that is the point: capacity is not contribution, and work that must be done to survive is not chosen work.
She showed the shoebox to my youngest after his question. Bad handwriting, 1970s, a machinist telling his daughter about a strike. Nothing in it is in any corpus. What is in the corpus is the world that letter came out of — the union’s pamphlets, the local paper, the safety manuals, the court rulings — and my son now collects, monthly, a fraction of the return on it.
That is the whole of the argument, and it is smaller than it sounds. The machine is capable because we were, across generations, mostly without payment and mostly without names. Nobody sent the certificate. The Ministry is what sending it looks like.
The part still unresolved
Equal payout is a concession, not a principle. Somewhere in the deposit is a woman who spent thirty years on crystallography that half the materials science in these systems rests on, and she receives exactly what her neighbour receives, and her neighbour published nothing. Every proportional scheme we have modelled — citation weighting, corpus-share estimation, contribution registries — either failed the untraceability test or reproduced the descendant problem in a new costume.
There is also a limit my son’s question found by accident. The deposit is not the whole record. It is the digitised, surviving, indexed part of it, which skews toward the literate, the published, the English-speaking and the recent. The dividend pays out on what was written down. Everything that was only ever spoken, or burned, or thrown out with a house, contributed nothing measurable and is owed the same amount anyway — which is either the rule’s deepest fairness or its most convenient evasion. In 2047 we still argue about which.
FAQ
What does “the dead are shareholders” actually mean? It means the productive asset was assembled by generations who are gone, and that a return is being earned on their work right now. Shareholding is the closest available analogy: a claim on yield arising from a contribution of capital — here, intellectual capital — rather than from current labour.
Doesn’t public-domain material mean the work is free to use? Legally, yes. The heirless-estate finding does not dispute that. It separates the right to use from the question of who the yield belongs to. A public library is free to use too; nobody concluded from that fact that the profits of everything built with it accrue to whoever borrowed the book last.
Why not let the value escheat to the state, as unclaimed property normally does? Because a treasury receipt becomes a budget line, and budget lines are reallocated by whoever wins the next election. Naming the population as residuary beneficiary makes the payment a property claim rather than a spending choice, which is what made it survive changes of government.
Is this just an inheritance tax by another name? No. An inheritance tax touches a private transfer between identified people. This assesses the yield of an estate that was never claimed by anyone. No family’s transfer is reduced.
What happens when a company builds a model with no inherited deposit? Then the four tests fail and nothing is assessed. That outcome is deliberate. An instrument that assessed everything regardless of provenance would be a tax on computation, and it would have been repealed within a decade for being exactly that.