Ministry of AI · Dispatch from 2047
Political Risk of Basic Income: The Fourth Year
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.
In the fourth year, the dividend did not fall because the argument was lost. It fell because a finance minister discovered that she could reduce it by one line in a transfer schedule, and that no law required her to explain herself first.
I remember the week clearly. My mother had been receiving the payment for three years by then, long enough to have stopped treating it as luck. She had rebuilt her Tuesdays around it — the unpaid hours she now spends sitting with families arguing with the health service began in that period — and when the notice came she did the thing that people who spent nineteen years reading insurance claims do. She read the schedule. Then she called me and asked whether this was allowed. I told her the truth, which was that it was, and that we had not thought hard enough about the difference between winning an argument and building something that survives a bad election.
Everyone who works on this eventually learns the same lesson in the same order. The design problem is hard and public. The survival problem is harder and nobody writes about it.
Programmes are rarely killed; payments are reduced
The historical record is unambiguous and it embarrassed us. Universal payments almost never get abolished. Abolition requires an elected official to tell every household in the jurisdiction that it is losing money, and that official generally does not remain an elected official. So opponents do not abolish. They reduce.
The clearest case is the oldest one. Alaska has paid a share of resource income to every resident since the early 1980s, and it has never been repealed. In 2016 the governor used a line-item veto to cut the transfer that funded it, and residents received $1,022 — roughly half of what the formula would have paid. Legislators sued. In 2017 the state supreme court ruled unanimously against them: the fund’s income could not be constitutionally dedicated, and the veto was a proper exercise of executive power. The programme was untouched. The cheque was halved.
And it kept moving. The fund’s own payment record shows the dividend at $3,284 in 2022, $1,312 in 2023, $1,702 in 2024 and $1,000 in 2025 — a fall of nearly fifty-nine percent in the last of those steps. Some of that is investment performance. Much of it is politics with an accounting vocabulary.
Set against that, the fate of the pilots. Ontario ran a basic income trial paying up to about seventeen thousand Canadian dollars a year to four thousand low-income people across three cities, designed to run three years. A new government cancelled it in 2018, roughly a year in, before the outcome data existed. The payment was not found wanting. It was never measured. Four thousand households is not a constituency; it is an anecdote with a budget line.
That contrast is the whole of the political theory. Alaska’s payment was cut and survived because everyone gets it. Ontario’s payment was ended without a fight because almost nobody did.
The veto surface
So the Ministry publishes something no programme published before it: the veto surface — the enumerated list of every point at which the dividend can be lawfully reduced without repealing anything.
There are five in our jurisdiction. The executive transfer approval. The annual appropriation vote. The smoothing reserve, which can be drained or hoarded. The indexation clause, which can be suspended. And the deferral provision, which allows a payment to be moved into the following fiscal year — technically not a cut, and functionally the cruellest of the five, because it is invisible in the year it happens.
Publishing this list was unpopular inside the institution. The objection was obvious: you are handing opponents a map. The answer, which I still believe, is that they already have the map. Only the households are missing it. A risk you have named can be argued about in advance of its use, and every one of these levers was pulled somewhere in the first fifteen years.
Alongside it sits the formula lock. The dividend is computed from measured yield by a published formula, and departing from that formula requires a reduction notice: a signed public document naming the official, the amount withheld, the reason, and the households affected, entered in the same register as the yield figures themselves. It stops nothing. It makes every cut attributable to a person with a name, which is a different kind of defence — the kind that works.
| Targeted payment | Universal dividend | |
|---|---|---|
| Who has a direct stake | The eligible minority | Every resident household |
| Typical failure mode | Cancellation, often before evidence exists | Reduction, formula suspension, deferral |
| Defended by | Advocacy organisations | Voters, in their own interest |
| Visibility of a cut | Low; affects people with little political weight | High; arrives in every household in the same month |
| Cost of the protection | — | Cannot be tilted toward greater need |
The constituency test
From that table comes the least idealistic rule we have. Before any change to the dividend’s design, we ask one question: does this change reduce the number of households with a direct financial stake in the payment?
If it does, it is a survival risk, however much fairer it looks on paper. Means-testing fails the test. A work condition fails it. Excluding high earners — the reform proposed in almost every election since I was forty — fails it decisively, because it converts an owner’s share into a transfer to the poor, and transfers to the poor are, on the historical evidence, the first thing cut and the last thing defended.
I dislike this rule. It means the payment cannot be made more generous to the people who need generosity most, and I have written elsewhere that this is a real and permanent limitation of ownership language rather than a temporary one. But it is the rule that kept the architecture alive through two hostile governments, and I would rather hold an imperfect instrument than an elegant memory.
There is a financing version of the same discipline, and Norway wrote it before anyone was thinking about machines. That country’s fund is spent under a fiscal rule limiting withdrawals to the fund’s expected real return, so the principal survives the government of the day. Our funding-source rule does the same work at a different point in the pipe: the Ministry pays only out of yield already measured for a closed period, never out of borrowing against a forecast. Both rules exist to make a good year non-spendable, because the political risk in a good year is precisely that someone spends it.
What survival costs
Here is the fourth year as it actually ended. The reduction stood. It was not reversed by a court, a protest or an argument. It was reversed eighteen months later by an election in which four separate candidates promised to restore the formula, because by then eleven million households had noticed a number change in their own account — and a payment everyone receives creates, over time, the only lobby that has ever reliably protected anything.
My mother got the difference back in a single deposit. She was not grateful. She was annoyed, in the specific way of someone who has been made to prove she was entitled to what she was already owed, and I think that annoyance is the healthiest political emotion the dividend has produced. Gratitude is what you feel toward a benefactor. Annoyance is what you feel toward an administrator who mishandled your property. The second one votes.
What it cost is written into everything above. Because the payment must stay equal to stay defended, it cannot bend toward need. Because it must be financed from measured yield, it falls in bad years — Alaska’s own record is the standing demonstration, and every household that built a rent payment on the higher number learned the lesson the way households always do. Because the smoothing reserve reduces that swing, it also creates a discretionary pot, which is a sixth veto surface we chose to accept.
Capacity is not contribution, and work that must be done to survive is not chosen work. That sentence is carved into a building now, which was supposed to make it permanent. It didn’t. What made it durable was duller: a published formula, a named list of the places it can be broken, and eleven million people who would notice.
FAQ
What is the real political risk to a basic income?
Reduction rather than repeal. Universal payments are almost never abolished, because abolition means telling every household it is losing money. They are cut, frozen, deferred and re-based instead, while the programme name survives untouched. Alaska’s 2016 line-item veto halved the dividend to $1,022 and was upheld unanimously by the state supreme court in 2017.
What is the veto surface?
The published list of every point at which the dividend can be lawfully reduced without repeal: executive transfer approval, appropriation vote, smoothing reserve, indexation clause, deferral provision. Publishing it hands opponents nothing they did not have and gives households the one thing they lacked.
Why do targeted pilots die so easily?
Because they pay too few people to defend themselves. Ontario’s pilot paid four thousand low-income households and was cancelled about a year into a three-year design, before any outcome data existed. Advocacy is not a constituency.
What is the constituency test?
Before any design change: does it reduce the number of households with a direct stake? If yes, it is a survival risk regardless of its fairness. Means-testing, work requirements and high-earner exclusions all fail it.
Does the formula lock actually prevent cuts?
No. It requires a signed reduction notice naming the official, the amount and the reason, filed in the public yield register. It converts an anonymous accounting adjustment into an attributable decision, which is the only sanction that has ever worked on this class of behaviour.
What is the unavoidable cost of designing for survival?
The properties that make the dividend hard to kill — equality, universality, a formula rather than discretion — are the same properties that make it hard to improve. It cannot be tilted toward greater need without weakening the coalition that protects it, and it falls in bad yield years. Targeted programmes still have to exist alongside it, and they remain as fragile as they have always been.