AI Labor Tracker / Foundations

Who pays for people when work shrinks?

Cash pilots, robot taxes, sovereign funds, shorter weeks and retraining all have real evidence behind them now, and none of them is free.

There is no single answer to who pays for people if AI shrinks the amount of paid work. Anyone selling you one is selling something.

But there is a lot more evidence than there was five years ago. Cash pilots have run their full course. Economists have done the tax math.

A few countries and one American state have already built the plumbing. So this piece does something simple. It lays out the main options, what each one costs, and what the data says about whether it works.

First, why ask now.

Workers’ share of the output of American businesses fell to 52.8% in the second quarter of 2026, the lowest reading since the series began in 1947, according to the Bureau of Labor Statistics. We covered what that means in our opening piece. Unemployment, meanwhile, is a fairly normal 4.2%, per the September jobs report.

So this is not a crisis yet. It is a slow shift in who gets paid.

That shift matters for the government’s books. Individual income and payroll taxes produce roughly 85% of federal tax receipts, and both fall mainly on wages, note Elena Patel of Brookings and Aparna Mathur of the American Enterprise Institute in a new Brookings analysis. If more income goes to machines and less to people, the government collects less exactly when it might need more.

The economists don’t agree, and that’s useful

Before the options, the forecasts. They span a wide range.

On the cautious end, MIT’s Daron Acemoglu estimates that AI’s effect on total factor productivity will be “nontrivial but modest,” no more than about 0.66% over ten years, in his paper on the macroeconomics of AI. Tyler Cowen of George Mason is more upbeat but still gradual. “I’ve gone on record as suggesting that AI will boost economic growth rates by half a percentage point a year,” he wrote on Marginal Revolution, adding that it is “very much a guess.”

MIT’s David Autor sees a possible upside for workers specifically. “The unique opportunity that AI offers humanity is to push back against the process started by computerization — to extend the relevance, reach and value of human expertise to a larger set of workers,” he wrote in Noema.

On the other end, the University of Virginia’s Anton Korinek takes seriously the case where machines can do almost everything. “If there are significant wage declines, avoiding mass misery will require other ways of distributing income than labor markets, whether via sufficiently well-distributed capital ownership or via benefits,” he and Megan Juelfs wrote in an NBER paper.

That spread is the point. A good policy has to work if Cowen is right and still hold up if Korinek is right.

Option one: just send people money

This is the idea everyone has an opinion about. The good news is we no longer have to guess.

The largest US study, run by the nonprofit OpenResearch, gave 1,000 people in Illinois and Texas $1,000 a month for three years, while 2,000 people in a control group got $50 a month. Recipients worked about 1.3 fewer hours a week, their employment rate was about 2 percentage points lower, and their earned income fell about $1,500 a year relative to the control group, according to OpenResearch’s findings and the accompanying NBER paper.

So yes, people worked a little less. Not a lot less. Most of the money went to basic needs, and recipients mostly did not quit.

Other programs found smaller effects or none.

Table 1. Selected cash programs and their effect on work. Data file: who-pays-for-people-cash-pilots.csv. Data: real. Sources: OpenResearch; SEED; Finnish Government; GiveDirectly; Jones and Marinescu; Alaska Department of Revenue.

The Kenya results are the most interesting for a post-labor world. In villages promised a basic income for 12 years, people did not stop working. They started more businesses.

About two years in, total enterprise revenues in those communities were up 41% and net revenues 52%, according to the study by Abhijit Banerjee, Tavneet Suri and colleagues. Security seems to have made people take more risks, not fewer.

Now the hard number: the bill.

There were about 269.8 million American adults in mid-2025, per the Census Bureau. Paying each of them $1,000 a month would cost about $3.24 trillion a year. The entire federal government spent about $7.4 trillion in fiscal 2026 and took in about $5.4 trillion, according to the Congressional Budget Office.

That is 60% of current federal revenue, before anything else gets cut or raised. Some of it would be offset by replacing existing programs.

Most of it would not be. Anyone who says a full basic income pays for itself hasn’t done the arithmetic.

Option two: tax the machines

If work is moving from people to software, why not tax the software?

There’s a strong argument that the tax code already tilts the field the wrong way. Acemoglu, Andrea Manera and Pascual Restrepo estimated that effective taxes on labor in the US run 25.5% to 33.5%. Effective taxes on software and equipment were about 10% in the 2010s and about 5% after the 2017 tax law, down from about 20% in 2000, in Brookings Papers on Economic Activity.

In other words, before anyone writes a robot tax, the system already gives robots a tax break that workers don’t get. Their argument is that this pushes companies to automate even when a person would be just as productive.

Stanford’s Erik Brynjolfsson makes a related point about which kind of AI we build. “When AI is focused on augmenting humans rather than mimicking them, then humans retain the power to insist on a share of the value created,” he wrote in “The Turing Trap”.

Congress has started to move. In August, Rep. Greg Casar of Texas introduced the AI Tax and Work Protection Act, H.R. 10044, which would put an excise tax on foundation-model token usage and fund a new jobs program at the Labor Department, according to the bill text.

The tradeoffs are real, and the Brookings authors lay them out well. A token tax is easy to explain but hard to define, since tokens vary in size across models and don’t map cleanly onto computing power. It would also tax the uses of AI that make workers more productive, not just the ones that replace them.

An automation tax runs into the problem of proving which job a machine displaced. And a training-compute tax could push model training offshore.

A tax aimed squarely at AI is easier to sell than broad reform of capital taxes, they concede. “That advantage is real, but it’s an advantage in politics rather than economics.”

Option three: own the machines

The third option skips the tax fight. Instead of taxing capital, give people a piece of it.

America already has a working model. The Alaska Permanent Fund, built from oil revenue, was worth $91.9 billion at the end of June 2026, according to the Alaska Permanent Fund Corporation.

It paid every eligible resident $1,200 this year, per the state Department of Revenue. Economists Damon Jones and Ioana Marinescu found the dividend did not reduce employment, while part-time work rose 1.8 percentage points, in American Economic Journal: Economic Policy.

Scaling that idea to AI is what OpenAI CEO Sam Altman proposed in his essay “Moore’s Law for Everything.” He suggested an American Equity Fund paid for by taxing large companies 2.5% of their market value each year, in shares, plus 2.5% of the value of privately held land. Under his assumptions, each adult would get about $13,500 a year within a decade. Those assumptions include market values doubling, which is a big if.

There’s also a bottom-up version. About 15.1 million Americans participate in 6,609 employee stock ownership plans, according to the National Center for Employee Ownership, based on 2023 filings.

Worker ownership won’t cover everyone. But it means a company that automates also enriches the people it employs.

The ownership route has one big advantage. If the machines do all the work, the people who own the machines are fine. The question is just how many people that is.

Option four: share the work

Keynes thought we’d solve this by working less, and some firms are trying.

In the UK’s 2022 four-day week pilot, 61 organizations with about 2,900 workers cut hours with no cut in pay. Afterward, 56 of them kept the policy, 18 permanently. Revenue rose 1.4% on average over the trial, and the number of staff leaving fell 57%, according to the Autonomy Institute.

A larger study of 2,896 employees across 141 organizations, published in Nature Human Behaviour, found the four-day week reduced burnout and improved job satisfaction and mental and physical health, compared with workers at control companies.

The caveat is selection. Firms that volunteer for these trials are the ones that expect it to work.

A four-day week at a law firm is a very different thing from a four-day week on a hospital ward. Still, sharing the remaining work is cheaper for taxpayers than any other option on this list.

Option five: retrain people

Retraining has a bad reputation, and some of that is earned. The good news is that the evidence has gotten more specific about what works.

The best results come from sectoral programs, which train people for specific in-demand jobs with real employers involved. In randomized trials, programs such as Year Up, Project QUEST and WorkAdvance raised earnings by 14% to 39% in the year after training, and the gains largely lasted, according to Lawrence Katz and coauthors in the Journal of Labor Economics.

The older federal model is weaker. Benjamin Hyman found that Trade Adjustment Assistance retraining raised displaced workers’ cumulative earnings by about $50,000 over ten years, but the yearly advantage faded until it disappeared by year ten, according to his Census Bureau working paper. Helpful, not permanent.

Lesson: training works when it points people at a real job. It’s not a strategy on its own if the jobs aren’t there.

Putting it together

In my opinion, the smart way to see these options is as a sequence.

Retraining and shared work are what you do while there are still jobs to move into. They are cheaper, the evidence is decent, and they preserve the part of work that pays more than money: being needed. That dignity matters, and it’s why I don’t think a check alone is a plan.

Fixing the tax tilt toward machines is the next step, and it should be the least partisan one. It doesn’t require a new tax. It requires not subsidizing automation more than we subsidize people.

Ownership and cash are the backstop. If Korinek’s scenario arrives, we’ll need them, and it would be far better to have the plumbing built before then.

Alaska shows a dividend can run for decades without wrecking the work ethic. Kenya suggests a guaranteed floor can make people more willing to start something.

What I’d do

Start small and build the pipes. Fund sectoral training that is tied to employers, because the trial evidence is strong.

Close the gap between how we tax workers and how we tax equipment and software. And set up a modest public or private dividend that can grow if AI’s gains concentrate, rather than waiting to invent one in a crisis.

Workers shouldn’t wait for Washington. Sign up for training that is tied to a real employer while the jobs are there, and keep your skills relevant to the work that’s growing.

Don’t bet everything on any single forecast, Acemoglu’s or Korinek’s. Build a system that works in both. That’s my recommendation.

Sources

1. “Labor share at its lowest level, 52.8 percent, in second quarter 2026,” The Economics Daily, U.S. Bureau of Labor Statistics, Sep 2026. https://www.bls.gov/opub/ted/2026/labor-share-at-its-lowest-level-52-8-percent-in-second-quarter-2026.htm
2. “The Employment Situation, September 2026,” U.S. Bureau of Labor Statistics, Oct 2026. https://www.bls.gov/news.release/empsit.nr0.htm
3. Patel, E. and Mathur, A., “Taxing artificial intelligence: What’s on the table, and what’s at stake,” Brookings Institution, Oct 27, 2026. https://www.brookings.edu/articles/taxing-artificial-intelligence-whats-on-the-table-and-whats-at-stake/
4. “Monthly Budget Review: September 2026,” Congressional Budget Office, Oct 8, 2026. https://www.cbo.gov/publication/61985
5. “Vintage 2025 National and State Population Estimates,” U.S. Census Bureau, 2026. https://www.census.gov/newsroom/press-kits/2026/national-state-population-estimates.html
6. “Key Findings: Employment and Income,” OpenResearch, 2024. https://www.openresearchlab.org/findings/key-findings-employment-and-income
7. Vivalt, E. et al., “The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States,” NBER Working Paper 32719, 2024. https://www.nber.org/papers/w32719
8. “Employment,” Stockton Economic Empowerment Demonstration. https://www.stocktondemonstration.org/employment
9. “Results of the basic income experiment: small employment effects, better perceived economic security and mental wellbeing,” press release, Ministry of Social Affairs and Health and Kela (Finnish Government), May 6, 2020. https://valtioneuvosto.fi/en/-/1271139/perustulokokeilun-tulokset-tyollisyysvaikutukset-vahaisia-toimeentulo-ja-psyykkinen-terveys-koettiin-paremmaksi
10. Banerjee, A., Faye, M., Krueger, A., Niehaus, P. and Suri, T., “Universal Basic Income: Short-Term Results from a Long-Term Experiment in Kenya,” 2023. https://econweb.ucsd.edu/~pniehaus/papers/UBI_main_paper.pdf
11. “2023 UBI results,” GiveDirectly, 2023. https://www.givedirectly.org/2023-ubi-results
12. Jones, D. and Marinescu, I., “The Labor Market Impacts of Universal and Permanent Cash Transfers: Evidence from the Alaska Permanent Fund,” American Economic Journal: Economic Policy, 2022. https://www.aeaweb.org/articles?id=10.1257%2Fpol.20190299
13. “DOR Announces 2026 PFD Amount,” Alaska Department of Revenue, 2026. https://dor.alaska.gov/docs/default-source/press-releases/2026-004-dor-announces-2026-pfd-amount.pdf
14. “APFC Board Approves Audited FY26 Financials and Advances FY28 Budget Proposal,” Alaska Permanent Fund Corporation, 2026. https://apfc.org/fund-news/fy26-audit-and-budget-proposal/
15. Acemoglu, D., Manera, A. and Restrepo, P., “Does the US Tax Code Favor Automation?” Brookings Papers on Economic Activity, Spring 2020. https://www.brookings.edu/wp-content/uploads/2020/12/Acemoglu-FINAL-WEB.pdf
16. Acemoglu, D., “The Simple Macroeconomics of AI,” NBER Working Paper 32487, May 2024. https://www.nber.org/papers/w32487
17. “H.R. 10044, AI Tax and Work Protection Act,” U.S. Congress, introduced Aug 6, 2026. https://www.govinfo.gov/content/pkg/BILLS-119hr10044ih/html/BILLS-119hr10044ih.htm
18. Brynjolfsson, E., “The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence,” Daedalus, Spring 2022. https://arxiv.org/abs/2201.04200
19. Altman, S., “Moore’s Law for Everything,” Mar 2021. https://moores.samaltman.com/
20. “Employee Ownership by the Numbers,” National Center for Employee Ownership, 2026. https://www.nceo.org/research/employee-ownership-by-the-numbers
21. Korinek, A. and Juelfs, M., “Preparing for the (Non-Existent?) Future of Work,” NBER Working Paper 30172, 2022. https://www.nber.org/papers/w30172
22. Autor, D., “AI Could Actually Help Rebuild The Middle Class,” Noema, Feb 12, 2024. https://www.noemamag.com/how-ai-could-help-rebuild-the-middle-class
23. Cowen, T., “Why I think AI take-off is relatively slow,” Marginal Revolution, Feb 2025. https://marginalrevolution.com/marginalrevolution/2025/02/why-i-think-ai-take-off-is-relatively-slow.html
24. “The results are in: the UK’s four-day week pilot,” Autonomy Institute, Feb 2023. https://autonomy.work/portfolio/uk4dwpilotresults/
25. Fan, W., Schor, J. B. et al., “Work time reduction via a 4-day workweek finds improvements in workers’ well-being,” Nature Human Behaviour, 2025. https://www.nature.com/articles/s41562-025-02259-6
26. Katz, L., Roth, J., Hendra, R. and Schaberg, K., “Why Do Sectoral Employment Programs Work? Lessons from WorkAdvance,” Journal of Labor Economics, 2022. https://www.journals.uchicago.edu/doi/10.1086/717932
27. Hyman, B., “Can Displaced Labor Be Retrained? Evidence from Quasi-Random Assignment to Trade Adjustment Assistance,” Census Bureau Center for Economic Studies Working Paper 22-05, 2022. https://www.census.gov/library/working-papers/2022/adrm/CES-WP-22-05.html