Universal Basic Income: Will It Set Us Free?

The people building the systems most likely to replace human work are also among the most consistent advocates for paying people who do not work.

In 2016, Y Combinator, then led by Sam Altman, announced a pilot that would hand families in Oakland between $1,000 and $2,000 a month [1]. Dario Amodei, the chief executive of Anthropic, has written that if displacement runs deep enough, it will be necessary to go "beyond mere incentive programs to long-term income support for a significant fraction of the labor force," financed through taxes on the relevant companies or a higher capital gains tax [2]. Bill Gates published an essay in August arguing that we should tax AI tokens and robots [3]. Senator Bernie Sanders has proposed a one-time fifty percent tax, paid in stock rather than cash, to put the public on the boards of the largest AI companies [4].

There is an easy way to read that pattern, and most people reach for it straight away. The reading is guilt, or public relations, or some combination of the two. The people automating the jobs feel bad about what is coming, or they want to be seen to feel bad about it, and a monthly payment is what they are offering in place of slowing down.

I do not think that reading survives much scrutiny, and it lets everyone involved off far too easily. These are serious people making a serious argument in public with their names attached to it, and some of them have been making it for the better part of a decade, well before it became a fashionable thing to say. The more useful approach is to assume the proposal is sincere, take it at face value, and then work through what it would actually do once it existed. That is where it becomes interesting, because a policy can be offered in complete good faith and still produce an outcome that nobody in the room intended or wanted.

So the question I want to work through is this: when an economy stops needing people to work, what problem does a basic income solve, and whose problem is it?

The strongest case, stated properly

The argument I want to make does not hold unless the evidence on the other side is dealt with honestly, so that comes first.

The usual objection to giving people money is that they will stop working. That objection has now been tested repeatedly, in different countries, at different income levels, and it keeps coming back weaker than its advocates expect.

Finland ran the cleanest national version. Two thousand unemployed people received 560 euros a month, tax free and unconditional, through 2017 and 2018. Recipients worked about six days more than the control group over the measured year. On wellbeing the gap was wider and clearer: they reported more life satisfaction, and less mental strain, depression, sadness, and loneliness than the people who got nothing [5]. Stockton, California gave 125 residents $500 a month for two years with no conditions attached. Full-time employment among recipients went up, not down, and they were measurably less anxious and less depressed [6]. The largest study of the idea ever run, GiveDirectly's long-term trial in rural Kenya, found reductions in hours of wage work offset by increases in self-employment, which netted out to no effect on total household labor supply at all [7].

Three countries, three quite different designs, and broadly the same result. The caricature of the idle recipient does not survive contact with the evidence, and it has now failed to survive it enough times that continuing to lead with it is a choice rather than an oversight. What the studies actually show is that people given a reliable floor mostly keep working. Where hours did fall, as they did in the American trial, the reduction ran to something like one or two hours a week, which is a modest adjustment rather than a withdrawal from working life [20].

There is a moral case underneath the empirical one, and it is stronger than its critics usually admit. A basic income treats a person as a citizen with a claim rather than a client under supervision. It does not require anyone to prove they are deserving, submit to a caseworker, or perform the rituals of need that most benefit systems demand. These systems were built on the assumption that human labor would remain necessary, which is exactly the assumption now in question. Something will have to replace them.

And the money has to come from somewhere. If these systems were trained on the collective written output of humanity, on our books, our code, our images, our conversations, and the entire public record we posted online long before any of it was raw material, then the public has a real claim on what they produce. That claim does not evaporate because the claim is hard to price.

I find that argument persuasive, and it is the reason this piece is not a case against helping people. It is a case about what specifically we should be asking for, and about how much of the debate is currently spent arguing over the wrong variable.

The mechanism nobody names

Consumer spending is 68.0 percent of American gross domestic product as of the second quarter of 2026 [8]. That single number governs more of this debate than anything said in it.

An economy structured that way runs on a loop. People work, companies pay wages, wages become household spending, spending becomes corporate revenue, revenue becomes profit, and profit accrues to whoever owns the companies. Every link depends on the one before it.

Automation cuts the first link. If the work is done by systems that do not draw a salary, wages stop reaching households, and the loop breaks at its narrowest point. That is a hardship problem for the people who lost the jobs, and it is simultaneously a demand problem for the companies that automated them, because a customer who earns nothing buys nothing.

A transfer repairs that break, and it repairs it rather elegantly. Money reaches households from the state instead of from an employer, households spend it much as they spent wages, spending becomes corporate revenue, revenue becomes profit, and profit accrues to exactly the same owners as before. The loop closes again, and everything downstream of the household continues exactly as it did. The same companies collect the same revenue, the same profit lands with the same people, and the only thing that has really changed is where the money enters at the bottom of the circuit. What used to arrive because somebody needed your work now arrives because a policy says that it should.

Martin Ford, one of the more thoughtful advocates for basic income, has been explicit that this is the point. He describes it as a way "to adapt capitalism to the future," and warns that without it, mass joblessness produces a deflationary spiral that drags the wealthy down alongside everyone else [9]. There is nothing concealed about the reasoning, and Ford is not being cynical in laying it out. He is arguing, correctly in my view, that the owners of automated production have a direct and self-interested stake in their former workers continuing to have money to spend.

That reframes who the policy is actually for. From where the recipient stands, the payment is support. From where the system stands, it is maintenance of demand: a way of keeping consumption running after the wage engine that used to fund it has been switched off.

I want to give this arrangement a name, because it does not have one and the absence of a name is doing real work in the debate. Call it managed dependency: a settlement in which people are no longer needed as workers, no longer powerful as citizens, and own no meaningful share of the productive systems around them, but are still carefully maintained as consumers, because the system requires somebody to buy what it produces. Life inside that arrangement is not necessarily miserable, and that is part of why it is difficult to argue against. The money still arrives every month, the groceries still get bought, and from the inside it can feel a great deal like being looked after. What has quietly gone missing is any point in the whole circuit where an ordinary person's participation is actually required, which is the same thing as saying there is no longer any point at which their objection would have to be answered.

Managed dependency does not require a conspiracy, and nobody has to intend it. It only requires each institution to keep doing what it is already built to do. Companies automate because it lowers their costs, investors fund the automation because it raises their returns, and governments tolerate the disruption because they want the growth, the tax revenue, and the strategic advantage that come with it. Then, when the public becomes anxious about where all of this is heading, those same institutions offer payments, credits, and dividends to keep things calm. Every one of those steps is rational on its own terms, and each of them is defensible when you examine it in isolation. The destination is not somewhere anyone sat down and chose to go, which is precisely why nobody involved will feel responsible for having arrived there.

Three tests the current proposals fail

If a proposal is going to be the answer to the largest labor disruption in a century, it should survive some obvious questions.

One: can it actually be paid for?

The best-known American proposal, Andrew Yang's Freedom Dividend, would pay $1,000 a month to roughly 236 million adult citizens. The Tax Foundation put the cost at $2.8 trillion a year [10].

Against that sits the industry expected to fund it. OpenAI's annualized revenue run rate passed $40 billion in August 2026, roughly double where it closed the previous year [21]. Anthropic told the world its run rate had crossed $47 billion when it announced a $65 billion funding round in May [22]. Those are the two largest American labs, and those are revenue figures rather than profits. Taking every dollar of both companies' entire revenue, leaving nothing at all for salaries, chips, or electricity, would cover something under four percent of the annual bill for a single year.

The gap runs to a factor of roughly thirty, and it is worth being precise about what that does and does not prove. It does not prove the idea is permanently unaffordable, because these companies are growing at a rate that makes any static comparison look foolish within about a year. What it establishes is that the money to do this does not currently exist, which means every proposal on the table is a promise about a future balance sheet rather than a plan against a present one.

There is a serious objection to that arithmetic, and it is the one I would raise myself. Revenue is the wrong measure, because the real wealth here is in the valuations. Anthropic's May round priced the company at $965 billion [22]. OpenAI was valued at $852 billion in March, has filed confidentially for a listing, and is reported to be targeting a trillion dollars or more when it goes public [23]. Two companies, nearly two trillion dollars of paper, and more of it created every quarter. Surely somewhere in that is the money for a dividend.

The trouble is that a valuation is not a sum of money, and this distinction turns out to matter more than almost anything else in the debate. A trillion-dollar valuation is a share price multiplied by a share count. There is no vault. The company itself receives cash only on the new shares it actually sells at listing, reportedly somewhere in the tens of billions for OpenAI, and that cash goes to the company to buy computing power rather than to the public. The rest of that trillion is a number attached to shares that other people already hold.

So the question becomes which specific mechanism is supposed to convert that number into monthly payments, and there are only three candidates. The first is to tax the capital gains when shares are sold, which produces a large one-off receipt from a one-off event, while a basic income is a permanent annual obligation. The second is to tax profits, which runs into the awkward fact that these companies are consuming capital rather than generating it; a business raising $65 billion in a single round is not a business sitting on distributable earnings. The third is for the public to hold the shares themselves, which is the only one of the three that survives contact with the arithmetic, and it is the one that requires giving up ownership rather than writing a cheque.

Then there is what an initial public offering actually is, which is where this gets uncomfortable. A listing is the moment private holders convert paper into cash by selling to the public: to pension funds, to retirement accounts, to index funds, and to retail investors. The public will be offered ownership of these machines. It will be offered at full market price, funded out of the public's own savings, at a valuation that already reflects the expected profits from displacing the public's own labor. Working people will be invited to buy back, at retail, a share of the thing partly built out of the wages they are no longer being paid.

That is close to the exact inverse of the Alaskan arrangement, where residents were given a stake in the resource rather than sold one. And it changes how the advocacy should be read, though not in the direction of accusing anyone of lying. A commitment that cannot be honored at present scale, and that will not fall due for years, costs the person making it very little today. What it buys is immediate and quite valuable: a reputation for having thought seriously about the people the technology will displace, and a reason for the rest of us to wait and see rather than to object now, during the window when ownership is actually being distributed.

The Alaskan comparison makes the scale problem concrete in a way the trillions do not. Alaska has operated the only durable unconditional citizen dividend in the United States for over four decades, funded by oil and paid out of a permanent invested fund, and the 2025 payment was $1,000 [11]. Not a month. For the year. That is what a mature, well-managed, genuinely funded citizen dividend delivers in practice, and it sits at roughly one percent of what the AI proposals discuss in passing. The distance between those two figures is the distance between a policy that exists and a policy that is being described.

Two: where does the money actually end up?

Suppose the funding problem were solved tomorrow. The money still has to land somewhere, and it is worth following where.

Gabrielle Fack studied a French housing benefit expansion as a natural experiment and found that each additional euro of benefit raised the rent paid by new claimants by 78 cents [12]. The subsidy was intended for tenants. Most of it reached landlords, because in a market where supply is fixed, a transfer to buyers becomes a price increase.

The largest recent American trial found the same shape. A thousand people received $1,000 a month for three years, against a control group receiving $50. Spending rose by at least $300 a month, concentrated in housing, food, and vehicles. And the effect on participants' net worth was near zero, because whatever assets they gained were offset by the debt they took on [13].

That last finding is the one that matters most, and it deserves stating slowly. A thousand dollars a month for three years comes to $36,000 per person, which is a genuinely significant sum for a low-income household. At the end of it, measured against the control group, those households owned no more than when they started. The money reached them, improved their lives while it lasted, and then moved on to the landlords, the grocers, the auto lenders, and the utilities. It passed through the household rather than settling there, and what a household actually accumulates over time is the only thing that eventually turns into security.

This is where Bill Gates's proposal belongs, and it belongs here precisely because it is the most carefully reasoned plan on the table rather than the least. Gates is in no sense a cheerleader for any of this. He writes that we are not preparing for what is coming, that he sees no evidence leaders are confronting the challenge adequately, and that there is at present no plan of any kind. He diagnoses the mechanism driving substitution more sharply than most of his critics: hire a person and you pay payroll taxes on their earnings, buy a robot and you usually write it off immediately as a business expense, so "the tax system nudges you toward replacing people with machines." His remedy is a tax on AI tokens and robots, which would "slow the rush away from human labor a little and raise money for retraining and a stronger safety net" [3].

Retraining and a stronger safety net are both worth having, and I would rather live in a country that had them than one that did not. What neither of them does is move a single share of ownership in the machines from the people who currently hold it to the people being displaced by it. Gates has correctly identified that the tax code is subsidising the substitution of capital for labor, and his answer is to make that substitution slightly more expensive and to cushion the people it lands on. The most carefully reasoned proposal available is still, in the end, a plan to compensate people for losing their position rather than a plan to give them a different one.

Three: can it be taken away?

The Kenya trial contains a finding that has been underdiscussed, and it is the most important one in this entire literature.

Recipients who were guaranteed twelve years of payments improved more than recipients given two years of payments on nearly every measure, even though both groups had received exactly the same amount of money at the point where they were compared. The cash was identical and the outcomes were not, which means the difference came from something other than the money itself. What the long-horizon group had was a horizon. They invested, they joined savings and lending groups, and they behaved like people who could make plans that extended past next year [7]. A separate arm of the study points the same way: recipients handed the money as a single lump sum started more enterprises than those given the same total in a stream of small monthly payments.

So what made the money work was not really its size. It was the certainty attached to it, and the recipient's ability to turn it into something that goes on producing after the payments stop.

What is actually on the table has neither property. A dividend funded by tax receipts is set by whoever holds office. It can be cut in a downturn, means-tested by the next administration, made conditional on behavior, or paid through a payment rail that can decline particular purchases. American benefit history is not reassuring on this point, and the leading US proposal already required recipients to give up food assistance and family support to receive it [10]. Every one of those design choices removes the property that the evidence says made the cash effective.

There is a live illustration of how quickly a commitment can be renegotiated once it becomes inconvenient, and it comes from the company that made the commitment.

In March 2019, OpenAI announced the structure it called a "capped-profit" company. Investors and employees could earn a capped return, and "any returns beyond that amount" were, in the company's own words, "owned by the original OpenAI Nonprofit entity." The announcement was specific about how large that overflow was expected to be: if the company succeeded, it anticipated generating "orders of magnitude more value" than it would owe the people who had invested in it or worked there [19]. That excess was the public's share, and it was written into the corporate form rather than promised in an essay.

On October 28, 2025, OpenAI completed a recapitalization. The nonprofit, now the OpenAI Foundation, holds conventional equity valued at roughly $130 billion, with "all stockholders participating proportionally in any increase in value" [14][15]. The Foundation's initial commitment is $25 billion, directed at health and AI resilience [14].

Reasonable people can argue that the new structure is the better one, and $130 billion is a serious endowment by any measure at all. What matters for this argument is the direction of travel rather than the merits. A mechanism that assigned the entire extraordinary upside to a nonprofit holding it on behalf of the public was replaced by an ordinary proportional stake, and it was replaced by the same company that had designed it, at the moment when it started to constrain fundraising. The promise lasted about six years, which is worth holding in mind when the next commitment of this kind is announced. It is the best evidence currently available on how durable a voluntary undertaking to share the proceeds of AI turns out to be once it meets a funding round.

The objection I owe you

Here is the strongest argument against everything above, and it deserves a real answer rather than a dismissal.

Money circulates, and it always has. Wages return to companies when workers spend them just as surely as a transfer would, which means the loop I described in the previous section is not really a description of basic income at all. It is a description of any economy that has ever functioned. Drawing a circle, pointing at money travelling around it, and calling the circle a trap is not analysis, and anyone with an economics background will say so within about a minute of reading it.

That objection is correct and I accept it entirely. Circulation is not the problem, and it never was. The question is what the circulation is evidence of, and the answer changes completely depending on why the money is moving.

Think about what a wage actually is. It is a price, paid because your effort is needed by somebody who cannot obtain it any other way, which is exactly why it can be negotiated and exactly why it can be withheld. Nearly every protection ordinary people won over the last century, the weekend, the eight-hour day, workplace safety law, and the right to organize, rests on that one piece of structural leverage. None of it was won by persuading owners to be more generous. It was won because the work stopped when the people doing it decided to stop, and because that consequence was expensive enough to bring somebody to the table.

A maintenance payment is a different instrument altogether, even when the sum involved is identical. It is a discretionary transfer to somebody whose participation is no longer needed for anything, which means the recipient has nothing available to withdraw. Nothing they do is load-bearing, so nothing they decline to do imposes a cost on anyone. The money moves in the same circle it always did, and the person receiving it stands in a completely different relationship to the people sending it.

That is the reason the size of the payment is the wrong thing to be arguing about. A generous payment inside that structure remains a payment that can be reduced whenever the budget tightens or the politics shift, and the person receiving it has no lever to pull when that happens. Income of that kind determines how comfortable your life is. Ownership determines whether anybody is obliged to ask you first.

What a version that passes would look like

None of this makes the benevolent version impossible. What it does is turn the benevolent version into a design problem rather than a moral one, and design problems have the considerable advantage of having answers.

Alaska's fund is instructive for a reason that has nothing to do with the size of its dividend. The state did not simply take the oil money and hand it out, which is what most governments would have done and what most of the AI proposals amount to. It converted a finite resource into a permanent invested asset base, and it pays residents out of what that asset earns rather than out of the resource itself. Norway did the same thing at larger scale, and its Government Pension Fund Global now holds, on average, 1.5 percent of every listed company in the world [16]. Both are cases of the public holding capital rather than receiving compensation from whoever does.

That distinction is the whole test, and it can be stated as a set of properties. A version that increases ordinary people's freedom is unconditional, protected by law rather than granted at the discretion of an administration, large enough to allow refusal, paid in money that cannot be restricted by category of purchase, and paired with genuine ownership of productive assets. A version that increases the system's leverage is small, conditional, revocable, administered through a rail that can approve or decline particular transactions, and unaccompanied by any transfer of ownership whatsoever.

The proposals with the most promise are the ones that treat ownership as the point. Sanders's stock tax is aggressive, and it is the only major proposal that asks for voting shares and board representation rather than a payment stream [4]. Amodei's mention of universal capital accounts belongs in the same category, and it is notably the part of his argument that gets quoted least [2]. Both would give the public a position rather than an allowance.

The concentration has already run further than the debate usually acknowledges. The top one percent of American households hold 50.2 percent of all corporate equities and mutual fund shares, and 31.6 percent of total net worth [17][18]. A dividend distributed across the population while ownership stays where it is does not correct that. It formalizes it, and adds a monthly reminder of who is doing the distributing.

What to actually ask for

Gates closes his essay with a line considerably more radical than any of the policies he proposes around it. The questions this technology raises, he writes, are "too consequential to leave to a small group of technologists" [3]. He is right about that, and it is the single most important sentence in the piece. His own proposals then leave ownership of the technology precisely where that small group already holds it, and the sharpest question he asks, which is who gets to decide what we reserve for humans, he explicitly declines to answer.

That gap between the quality of the diagnosis and the modesty of the remedy is where this entire debate currently sits, and it is not unique to Gates. It runs through nearly every serious proposal on the table.

Whenever a basic income comes up, the argument that follows is almost always about the number. Whether $1,000 is enough, whether it can be afforded, whether it should be $500 or $2,000, whether it should taper as other income rises. Those arguments will go on for years and they are largely beside the point, because a number is the most easily revised thing in any policy. It can be adjusted downward by any future government facing a difficult budget, and a commitment made voluntarily can be renegotiated by the party who made it, as one of them already has been.

The question that survives all of that is narrower, and it is considerably harder to answer with a check. When the machines are doing the work, what is it that you actually own?

That is the thing to ask for, and to keep asking for. Everything else is an allowance.

---

Asim Javed writes The Consent Report, a weekly briefing on how AI is being deployed and who is deciding. Subscribe at theconsentreport.substack.com.

---

References

[1] Sadowski, J. "Why Silicon Valley is embracing universal basic income." The Guardian, June 22, 2016. https://www.theguardian.com/technology/2016/jun/22/silicon-valley-universal-basic-income-y-combinator (opens in a new tab)

[2] Amodei, D. "Policy on the AI Exponential." darioamodei.com. https://darioamodei.com/post/policy-on-the-ai-exponential (opens in a new tab)

[3] Gates, B. "An Epochal Shift: The turbulent AI era is here. The choices we make now are critical." Gates Notes, August 26, 2026. https://www.gatesnotes.com/a-turbulent-ai-era-and-critical-choices-to-make (opens in a new tab)

[4] Sanders, B. "The Public Should Own Half of the Big A.I. Companies." Office of Senator Bernie Sanders. https://www.sanders.senate.gov/op-eds/the-public-should-own-half-of-the-big-a-i-companies/ (opens in a new tab)

[5] VATT Institute for Economic Research. "Results of the basic income experiment: small employment effects, better perceived economic security and mental wellbeing." https://vatt.fi/en/-/results-of-the-basic-income-experiment-small-employment-effects-better-perceived-economic-security-and-mental-wellbeing (opens in a new tab)

[6] Stockton Economic Empowerment Demonstration (SEED). https://www.stocktondemonstration.org/ (opens in a new tab)

[7] GiveDirectly. "Early findings from the world's largest UBI study." https://www.givedirectly.org/2023-ubi-results (opens in a new tab)

[8] Federal Reserve Bank of St. Louis, FRED. "Shares of gross domestic product: Personal consumption expenditures (DPCERE1Q156NBEA)." Q2 2026: 68.0 percent. https://fred.stlouisfed.org/series/DPCERE1Q156NBEA (opens in a new tab)

[9] Ford, M., interviewed in "Are robots taking our jobs?" Phys.org, June 2015. https://phys.org/news/2015-06-martin-ford-robots-job.html (opens in a new tab)

[10] Tax Foundation. "Andrew Yang's Value-Added Tax and Universal Basic Income." https://taxfoundation.org/blog/andrew-yang-value-added-tax-universal-basic-income/ (opens in a new tab)

[11] State of Alaska, Permanent Fund Dividend Division. 2025 dividend amount: $1,000. https://pfd.alaska.gov/ (opens in a new tab)

[12] Fack, G. "Are housing benefit an effective way to redistribute income? Evidence from a natural experiment in France." Labour Economics, Vol. 13, Issue 6, December 2006, pp. 747-771. https://ideas.repec.org/a/eee/labeco/v13y2006i6p747-771.html (opens in a new tab)

[13] Bartik, A. W., Rhodes, E., Broockman, D. E., Krause, P. K., Miller, S., and Vivalt, E. "The Impact of Unconditional Cash Transfers on Consumption and Household Balance Sheets: Experimental Evidence from Two US States." NBER Working Paper 32784. https://www.nber.org/papers/w32784 (opens in a new tab)

[14] Taylor, B. "Built to benefit everyone." OpenAI, October 28, 2025. https://openai.com/index/built-to-benefit-everyone/ (opens in a new tab)

[15] OpenAI. "Our Structure." https://openai.com/our-structure/ (opens in a new tab)

[16] Norges Bank Investment Management. "The fund." https://www.nbim.no/en/ (opens in a new tab)

[17] Federal Reserve Bank of St. Louis, FRED. "Share of Corporate Equities and Mutual Fund Shares Held by the Top 1% (WFRBST01122)." Q1 2026: 50.2 percent. https://fred.stlouisfed.org/series/WFRBST01122 (opens in a new tab)

[18] Federal Reserve Bank of St. Louis, FRED. "Share of Net Worth Held by the Top 1% (WFRBST01134)." Q1 2026: 31.6 percent. https://fred.stlouisfed.org/series/WFRBST01134 (opens in a new tab)

[19] OpenAI. "OpenAI LP." March 11, 2019. https://openai.com/index/openai-lp/ (opens in a new tab)

[20] Vivalt, E., Rhodes, E., Bartik, A. W., Broockman, D. E., Krause, P., and Miller, S. "The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States." NBER Working Paper 32719. https://www.nber.org/papers/w32719 (opens in a new tab)

[21] Bloomberg. "OpenAI Revenue Run Rate Tops $40 Billion, Doubling From 2025." August 13, 2026. https://www.bloomberg.com/news/articles/2026-08-13/openai-s-revenue-run-rate-tops-40-billion-ahead-of-ipo (opens in a new tab)

[22] Anthropic. "Anthropic raises $65B in Series H funding at $965B post-money valuation." May 28, 2026. https://www.anthropic.com/news/series-h (opens in a new tab)

[23] CNBC. "OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employees." August 19, 2026. https://www.cnbc.com/2026/08/19/open-ai-ipo-timing-2027-friar.html (opens in a new tab)