Universal Basic Income and Alternatives

Every October since 1982, the state of Alaska has sent its residents a check. Not for working. Not for filing a clever tax return. Just for being there. In 2025 the check was $1,000, deposited starting October 2 into the accounts of nearly every man, woman, and child in the state (Alaska Department of Revenue, 2025). The money comes from the Alaska Permanent Fund, a pool of oil wealth that the state invests and pays out from every year. The logic behind it is almost childishly simple: the oil under Alaska belongs to Alaskans. Companies pump it, sell it, and profit from it — and the owners of the resource get a cut. Nobody in Juneau calls this socialism. Republicans and Democrats have defended the dividend for more than four decades, and a politician who proposed abolishing it would be committing a form of career suicide.

Now hold that image in your mind and change one word. Replace oil with intelligence. AI companies extract value from data scraped off the open internet, from computing infrastructure built partly on public research, from the accumulated written and visual output of billions of people. They generate enormous profits. And in the newest version of a very old idea, everyone — the displaced call-center worker, the paralegal whose tasks a model now performs, the family whose town lost its largest employer — receives a share. That, stripped to its essence, is the case for a universal income in the age of AI: give people money unconditionally, so that machine-generated wealth does not flow only to those who own the machines.

For most of the last decade this was a thought experiment argued in op-eds and think tanks. In June 2026 it became a bill. Senator Bernie Sanders introduced the American AI Sovereign Wealth Fund Act, which would impose a one-time 50 percent tax — paid in stock, not cash — on the largest AI companies, place those shares in a Treasury trust worth an estimated $7 trillion, and pay out 5 percent of the fund each year, beginning with roughly $1,000 per American (Sanders, 2026). It is Alaska's oil dividend, transposed almost note for note onto the intelligence economy. It is also, by the admission of nearly everyone including its author, unlikely to pass the current Congress. But the fact that a mechanism this specific reached the Senate floor at all tells us the debate has moved. The question is no longer whether machine wealth should be shared. It is how, to whom, and at what cost.

Why the old safety net does not fit

The renewed interest in a universal income rests on a claim about scale. Previous waves of automation displaced workers in identifiable pockets — weavers, switchboard operators, bank tellers — while leaving most of the economy untouched. AI's reach looks broader. McKinsey has estimated that generative AI could automate activities absorbing up to 70 percent of the time employees currently spend on the job, and crucially it reaches into work long assumed to be safe: legal analysis, medical triage, software engineering, marketing copy, financial modeling (McKinsey, 2023). Whether that ceiling is ever reached is genuinely uncertain — it is a forecast, not a measured fact. But even a fraction of it would strain the machinery built to catch people when they fall.

That machinery was designed for a different kind of shock. Unemployment insurance assumes joblessness is temporary — a bridge between one job and the next. Retraining programs assume there is something to retrain for, and that a fifty-five-year-old logistics coordinator can plausibly become a data analyst. Traditional welfare assumes poverty is an individual misfortune, an exception against a backdrop of general employment. Each of these systems shares a hidden premise: that the labor market, given time, will reabsorb the people it spits out. Strip away that premise — imagine displacement that is persistent, broad, and faster than institutions can react — and the safety net starts to look like a fishing net held up against a flood. It was never built for water at that volume.

A universal income drops the premise entirely. Instead of assuming people will return to work, it decouples income from employment altogether. Give everyone a floor beneath which they cannot fall, and let them decide what to do above it — look for work, retrain, care for a parent, start something, or simply survive with dignity while the economy sorts itself out. The goal is not to engineer any particular outcome. It is to prevent mass destitution in a world where the labor market can no longer promise a place to everyone who needs one.

What the experiments actually show

It would be easy to assume this is all speculation. It is not. More than 160 basic-income pilots and cash-transfer trials have been run over the past four decades, at least 38 of them in Europe, North America, and Asia since 2015 (Universal Basic Income Pilots, 2025). None has been permanent, and none has been truly universal — they are samples, not systems — but together they form the largest body of evidence we have, and the patterns in it are surprisingly consistent.

Finland's experiment remains the most scrutinized. Two thousand unemployed people received €560 a month, no strings attached. They reported better health, less stress, and more trust in institutions than the control group — but they were no more likely to find jobs. Critics seized on the employment result: if a basic income does not put people back to work, what exactly is it buying? Supporters answered that reduced stress and better health are not consolation prizes; they are the point, and they carry real economic value that does not show up in a monthly employment statistic (World Economic Forum, 2019).

The most rigorous American evidence arrived in 2024 from OpenResearch, the study funded by OpenAI's Sam Altman. It gave 1,000 people in Texas and Illinois $1,000 a month for three years and compared them against 2,000 people receiving $50 (OpenResearch, 2024). The findings were more sobering than either camp expected. Recipients spent the extra money mostly on rent, food, and transportation — spending rose about $310 a month — and a thousand dollars was enough to lift nearly every single-person household out of poverty. But the striking early gains in mental health and reduced stress largely faded by the second and third years, and recipients worked 1.3 to 1.4 hours less per week than the control group. They did not drop out. They eased off. Whether that reduction is a problem or a feature — a parent trading an evening shift for time with a child, a worker declining a punishing second job — depends entirely on what you think work is for.

Kenya tells a different story, because the context is different. GiveDirectly's ongoing experiment, the largest ever run, pays about 23,000 people across 195 villages, some of them for a full twelve years (GiveDirectly, 2023). In a poor rural economy, cash did not sap the will to work; it financed it. Recipients started businesses, invested, and grew local commerce. The most counterintuitive result was that the villagers who received a single lump sum did better at building enterprises than those promised a steady monthly stream — the lump gave them enough capital to buy a cow, a motorbike, a market stall, where the trickle did not.

Put the trials side by side and a modest, honest conclusion emerges. A basic income does not reliably push people back into jobs, but it does not trigger the mass exodus from work that critics fear either. Its most consistent effects are humanitarian: less financial anxiety, better health, less destitution. In poorer settings it can actively stimulate economic activity. Whether those effects are worth the cost is not a question the pilots can answer — because the cost, at national scale, is enormous, and the behavior of a permanent program may look nothing like the behavior of a three-year one.

The Alaska model and its AI heirs

Of every real-world precedent, the Alaska Permanent Fund remains the most instructive, because it has done the one thing no pilot has: survive. Established in 1976 after the Prudhoe Bay oil boom, the fund banks a share of the state's oil royalties, invests the proceeds, and distributes a slice to every resident each year. The dividend swings with the market — a few hundred dollars in lean years, over two thousand in fat ones — but the institution itself has proven nearly indestructible. It has outlasted recessions, oil-price collapses, and repeated attempts to raid it, under governments of both parties.

Its durability comes from a single design choice that has almost nothing to do with economics. Alaskans do not experience the dividend as welfare. They experience it as a return on something they own. The oil is theirs; the check is their share of the proceeds. That framing — ownership, not charity — is what lets the program cut clean across the ideological lines that shred most redistribution schemes. You do not have to believe in a generous state to believe you should be paid for your own oil.

This is the insight the new wave of AI-dividend proposals is trying to borrow. If AI's value is extracted from communal inputs — data generated by all of us, research funded by the public, infrastructure built on shared foundations — then perhaps the returns should flow back the same way Alaska's oil money does. The mechanisms proposed vary. Some economists have floated a tax on AI model parameters, each parameter standing in for information distilled from public data, with rates rising as systems grow more capable. OpenAI itself, in an April 2025 policy paper, proposed a public wealth fund giving every citizen a stake in AI-driven growth. And the Sanders bill puts the sharpest possible point on it: a one-time 50 percent stock levy on AI firms with more than $200 million in annual AI revenue, the shares held in a public trust, 5 percent paid out yearly, and the whole thing governed by a new Independent Commission for Democratic AI whose seven members would be Senate-confirmed (Sanders, 2026).

The appeal of the Alaska framing is obvious; so is the reason it is harder to apply to AI than to oil. A barrel of oil is a countable thing pulled from a fixed patch of ground. "Value extracted by an AI model" is contestable, mobile, and easy to relocate to a friendlier tax jurisdiction. Alaska could tax its oil because the oil could not flee to Texas. A model's parameters, and the corporation that owns them, can. That is why even sympathetic analysts call the Sanders numbers optimistic — a 50 percent one-time tax on the value of firms that could restructure, offshore, or simply crater in valuation is not the same reliable spigot as a royalty on physical extraction (World Politics Review, 2026). The Alaska model proves that a resource dividend can be politically immortal. It does not prove that intelligence behaves like oil.

The menu of alternatives

A universal basic income is only one item on a longer menu, and its critics argue it is neither the cheapest nor the fairest. Four main rivals compete for the same problem, each making a different bet about what matters most — cost, targeting, work incentives, or administrative simplicity.

The Negative Income Tax, associated most closely with Milton Friedman, pays only those below a set income line, tapering the payment away as earnings climb until it disappears. It preserves the incentive to work — every extra dollar earned still leaves you ahead — while concentrating help where it is needed. The price is means-testing: the forms, the eligibility checks, the bureaucracy, and the quiet stigma of being sorted into the category of people who qualify.

Wage subsidies take the opposite tack, topping up low pay so that employment stays viable without asking workers to prove they are poor. Britain's furlough scheme during the pandemic showed such programs can be stood up fast and at scale. But they carry a fatal assumption for the AI era: that there are jobs to subsidize. When a technology eliminates positions rather than merely depressing their pay, a subsidy for wages has nothing left to attach to.

A job guarantee attacks the problem from a different angle. Rather than compensating people for the absence of work, it manufactures the work — the government becomes the employer of last resort, offering a living wage for jobs in care, maintenance, environmental restoration, and other areas of public need. The Levy Economics Institute estimated that guaranteeing jobs for 11 to 16 million people at $15 an hour plus benefits would cost somewhere between 0.8 and 2 percent of GDP net — less, its authors note, than the country spends on primary and secondary schooling (Levy Institute, 2018). Supporters prize it for preserving the psychological and civic scaffolding of employment. Skeptics warn about the machinery required to invent millions of genuinely useful jobs on demand, and the risk that guaranteed work curdles into make-work that dignifies no one.

The targeted AI dividend threads between universality and means-testing by paying workers in occupations most exposed to automation, expanding eligibility as AI spreads into new sectors. It scales with the actual damage. But "actual damage" has to be measured, sorted, and constantly updated — exactly the kind of administrative complexity that makes targeted programs expensive and error-prone in the real world.

The cost problem

Whatever its merits, a universal basic income runs first into arithmetic. Handing every American adult $1,000 a month costs roughly $3.1 trillion a year; extend it to every resident and the figure climbs toward $4 trillion (UBI Center, 2024). Against total federal individual income-tax revenue of about $3.5 trillion, the program is nearly as large as the entire income-tax base. There is no rounding error that makes that small.

How to pay for it splits along predictable lines. A value-added tax spreads the burden widely but lands hardest on the poor, who spend a larger share of what they earn — which risks making the whole scheme regressive, taking with one hand what it gives with the other. Taxes on wealth, capital gains, or financial transactions are more progressive but draw fierce lobbying and raise the specter of capital flight. AI-specific levies of the Sanders variety are the most poetically fitting — machine wealth funding the people machines displace — but they are also the least tested, easiest to avoid through relocation, and most dependent on international coordination that does not yet exist. The choice of funding is not a technicality. It determines who actually bears the cost, and therefore whether a universal income ends up progressive or regressive once every transfer and tax is netted out. A UBI funded by a wealth tax and a UBI funded by a sales tax are, in distributional terms, almost opposite policies wearing the same name.

There is also the question of what else the money could do. Every trillion routed into universal transfers is a trillion not spent on healthcare, housing, education, or research — investments that might yield larger gains in welfare and productivity over time. Whether a universal income beats targeted spending is a real empirical dispute, and the honest answer is that it depends on the details and the alternative on offer. The cost problem does not prove a universal income is unaffordable in principle. It proves that the revenue and trade-off questions must be answered convincingly before anyone writes the first check at scale.

The work problem

Beyond money lies a deeper unease. If people are paid whether or not they work, will they stop? The pilots push back on the strongest version of this fear — in every trial, the overwhelming majority kept working, because people work for reasons that outlast financial necessity: purpose, structure, status, company, identity. But the OpenResearch result, with its 1.3-hour weekly dip and its fading well-being gains, is a caution against too-easy reassurance. People did work a little less. And every pilot shares a disqualifying feature when it comes to predicting a permanent program: it was temporary, and participants knew it. You behave differently toward money you know will stop than toward money you know never will. A three-year windfall is a cushion; a permanent floor is a foundation, and foundations reshape the buildings on top of them in ways no short trial can observe.

The most serious unknown is not about workers at all — it is about employers. If everyone has a guaranteed income floor, a firm might reason that it no longer needs to pay a living wage, since its workers can survive on the government's money plus a little more. In that scenario a basic income quietly mutates from a benefit for workers into a subsidy for low-wage employers, its value captured at the top of the pay stub rather than the bottom. Whether this happens turns on the structure of the labor market. Where employers compete hard for workers, wages should hold, because a firm that cuts pay loses its people to a rival. Where a few employers dominate a local labor market — the single hospital, the lone warehouse, the company town, a condition economists call monopsony — the floor could sag, and the dividend could leak upward into profits.

Here we reach the edge of what anyone actually knows. No pilot has been large or permanent enough to reveal how a universal floor would move the wage bargain across a whole economy, and detecting the effect afterward would be genuinely hard — wages move for a hundred reasons at once, and disentangling the fingerprint of a basic income from ordinary market churn is a problem we do not yet have the tools to solve cleanly. This is not a footnote. It is a hole near the center of the case for a universal income, and intellectual honesty requires naming it as one.

Timing, politics, and the shape of what comes

Even people convinced a universal income is the right destination disagree about when to leave. One camp argues that AI will not generate enough new fiscal capacity to fund a meaningful universal income until the productivity gains actually arrive — somewhere late this decade, depending on how fast capability compounds — and that moving too early means building an unaffordable program that discredits the whole idea when it buckles. The opposing camp answers that policy takes years to design, legislate, and implement, and that if mass displacement arrives before the architecture is standing, the result is not a managed transition but a humanitarian emergency. Waiting for certainty, in this view, is its own reckless bet.

The compromise favored by many who study the problem is to start narrow and grow. Begin with a modest targeted dividend for the most exposed workers, or an Alaska-style fund seeded small, paired with public investment in housing, healthcare, and education — building administrative muscle and political legitimacy without gambling the whole treasury up front. Notably, the Sanders bill is structured exactly this way: 5 percent of the fund paid out annually, directed first toward dividends and over time toward health, education, and housing (Sanders, 2026). Whether it passes or not, it sketches the incrementalist path that even skeptics tend to concede is the only realistic one.

Because politics, in the end, is the binding constraint. Support for guaranteed income is wide but shallow — high in the abstract, collapsing the moment a survey mentions the tax bill or attaches a work requirement. Pew found 45 percent of Americans backing a universal income in 2020; a State Policy Network poll had it down to 33 percent by late 2023, with more opposed than in favor (State Policy Network, 2023). Finland's experiment is the parable: it ended not because it failed — participants were measurably better off — but because political enthusiasm drained away as costs became visible and governments changed hands. Deep cultural convictions about the link between effort and reward proved sturdier than the data, and elected officials, reading the same polls, learned to keep their distance from anything that could be branded as paying people not to work.

So the near-term future, in most wealthy democracies, is likely to be incremental rather than transformative: expanded unemployment insurance, extended retraining subsidies, targeted cash for the hardest-hit industries, scattered municipal pilots, perhaps a private initiative or two from tech firms nervous about the backlash they are helping to create. A genuine AI revenue fund is plausible on a longer horizon, and smaller countries with existing sovereign-wealth machinery may move first. But history offers a darker possibility too. The largest expansions of the social safety net — the New Deal, the postwar European welfare states — were not products of calm foresight. They were responses to catastrophe, built in the wreckage of depression and war. If AI-driven displacement outruns policy, societies may back into income support the same way: through crisis rather than design, at higher human cost and lower dignity than proactive planning would have required. Which road we take will depend less on economic analysis than on political will and public understanding — on whether we choose to build the boat before the flood, or after.

Summary

1. The safety net was built for a different shock. Unemployment insurance, retraining, and welfare all assume the labor market will reabsorb displaced workers. If AI makes displacement persistent and broad, that assumption fails — which is why a universal income, decoupling money from work, has returned to serious debate.

2. The evidence is real but limited. Over 160 pilots consistently show that cash transfers reduce financial stress and improve health without triggering mass withdrawal from work. But the 2024 OpenResearch study found well-being gains that faded and a modest drop in hours, and every trial has been short-term — leaving the behavior of a permanent, national program genuinely unknown.

3. Alaska works because of ownership, not generosity. The Permanent Fund's four-decade durability rests on framing the dividend as a return on a resource people own, not charity. AI-dividend proposals — most concretely Bernie Sanders's June 2026 bill for a $7 trillion fund paying ~$1,000 a year — try to copy that framing, but intelligence is harder to tax than oil because it can move.

4. No single alternative dominates. The Negative Income Tax, wage subsidies, a job guarantee, and targeted AI dividends each trade differently among cost, targeting, work incentives, and administrative burden. Wage subsidies fail if jobs vanish; a job guarantee preserves work but risks make-work; a targeted dividend tracks damage but is complex to run.

5. The hardest problems are practical, not conceptual. A universal income is fiscally enormous (roughly $3 trillion a year in the US), its funding choice determines whether it is progressive or regressive, its effect on employer wage-setting is a genuine unknown, and its political support collapses the moment costs are named. The likely near-term path is incremental — unless displacement arrives as a crisis and forces the issue, as depression and war once did.

The question underneath all of this — what a society owes its members when machines can do most of the economic work — will not be settled soon. But it is no longer merely academic. There is now a bill with a number on it, and the debate has moved from whether to how.

Sources

• Alaska's 2025 Permanent Fund dividend will be $1,000 | Alaska Beacon
• Department of Revenue Announces 2025 Permanent Fund Dividend Amount
• Alaska Permanent Fund Dividend
• Sanders Introduces Legislation to Create $7 Trillion AI Sovereign Wealth Fund | Senator Bernie Sanders
• Sovereign wealth fund tax on AI companies unveiled by Sanders | Roll Call
• An AI Sovereign Wealth Windfall? First, Find the Windfall | World Politics Review
• Millions Would Get 'Direct Payments' Under Bernie Sanders' Proposed AI Sovereign Wealth Fund | Newsweek
• Here's what a Sam Altman-backed basic income experiment found | CBS News
• UBI Study Backed by OpenAI's Sam Altman Bolsters Support for Basic Income | Bloomberg
• OpenAI Founder Sam Altman Gave Thousands Of People Free Money | Forbes
• Early findings from the world's largest UBI study | GiveDirectly
• Results are in from one of the world's biggest experiments in fighting poverty | NPR
• The Results of Finland's Basic Income Experiment | World Economic Forum
• Why Basic Income Failed in Finland | Jacobin
• Why Not UBI? | UBI Center
• Universal Basic Income May Sound Attractive But Would Likelier Increase Poverty | CBPP
• The Job Guarantee: Design, Jobs, and Implementation | Levy Economics Institute
• The Federal Job Guarantee | Center on Budget and Policy Priorities
• Alaska Is a Model for Taxing AI | Bloomberg Tax
• Universal Basic Income Pilots | Wikipedia
• Universal Basic Income (UBI) Pros, Cons, Debate | Britannica
• Universal Basic Income: Preparing for the AI Future? | Tax Project Institute
• An AI Capability Threshold for Rent-Funded UBI | arXiv

Last updated: 2026-07-15