Universal Basic Income and AI: What the Trials Show

The best evidence on basic income comes from trials, not forecasts, and it is more modest than either side of the debate suggests. In Finland and the United States, regular unconditional payments reduced work only slightly and improved how people felt about their lives, but no trial has yet tested an income paid to everyone, for good, in an economy reshaped by AI.

This article sets out what the main trials found, what they cannot tell us, and what that means for planning income support as AI changes work.

Why is basic income back on the agenda?

Forecasts of AI’s effect on jobs are large. The IMF estimates that almost 40% of global employment is exposed to AI and recommends social safety nets and retraining for workers at risk. A universal basic income, a regular payment to everyone with no conditions, is one of the safety nets proposed.

The data so far is calmer than the forecasts. The Budget Lab at Yale found no discernible disruption of the broader US labour market in the 33 months after ChatGPT’s release. That gives time to learn from trials before any decision is forced.

What did Finland’s basic income experiment find?

Finland ran a national randomised experiment in 2017 and 2018. According to the University of Helsinki’s summary, 560 euros a month was paid to 2,000 randomly selected unemployed people, replacing their unemployment benefit and continuing if they took up a job. Other unemployed people on the same benefit formed the control group.

• Work: in the first year there was no difference between the groups. In the second year the basic income group worked six days more, but that difference cannot be attributed to the basic income alone: a new sanction regime, the activation model, came in, and the researchers say its effects cannot be separated from those of the second year of the experiment.
• Wellbeing: the basic income group reported clearly higher wellbeing across many measures. There was no baseline survey, so the researchers could not measure the change during the experiment.

The lesson is mixed: a basic income did not raise employment, and it did not reduce it either.

What did the largest US study find?

OpenResearch’s Unconditional Cash Study is a randomised trial of 3,000 people in Illinois and Texas. From 2020 to 2023, 1,000 recipients received $1,000 a month and 2,000 people in the control group received $50 a month.

On average, recipients worked 1.3 fewer hours a week and were 2 percentage points less likely to be employed than the control group. The study also reports on health, parenting, agency and politics; its site sets out those results in full and is worth reading before drawing conclusions from one number.

What about a payment to everyone?

The closest working example is Alaska’s Permanent Fund Dividend, paid every year to eligible residents from the state’s oil wealth. The 2026 amount is $1,200, including a $200 energy relief payment. It is universal but far smaller than a living income.

The Stanford Basic Income Lab maps many more pilots around the world. Most are time-limited, targeted at a group, and smaller than a full income.

What can the trials not tell us?

Analysis:

• Scale. Paying a sample of people does not show what happens to prices, wages or taxes when everyone is paid.
• Duration. Two or three years is not a lifetime. People may plan differently when an income is permanent.
• Cause. The trials test a cash payment, not AI. If AI removes whole kinds of work, the choices people face will differ from those in a trial run in a normal labour market.
• Cost. No trial shows how a country would pay for a full income. We look at that in Can AI Pay for a Universal Basic Income?.

What should decision-makers take from the evidence?

1. Do not expect a collapse in work. The trials show small effects on employment, in both directions.
2. Count wellbeing as an outcome. Both studies found benefits that employment figures alone miss.
3. Design trials that answer the AI question. Pay a whole community, for longer, with a baseline survey, in places where AI is changing jobs.
4. Fix delivery early. How a payment reaches each person once is a hard problem in itself; see How Would an AI Dividend Reach Everyone?.
5. Keep watching the labour data. Decide on the strength of what is happening, not on deadlines in headlines; see Is There an 18-Month AI Jobs Cliff?.

Frequently asked questions

Does a basic income make people stop working?

Not in the trials so far, though it can reduce work slightly. In OpenResearch’s US study, recipients of $1,000 a month worked 1.3 fewer hours a week and were 2 percentage points less likely to be employed. In Finland’s experiment, the employment effects were small.

What did Finland’s basic income experiment find?

Finland paid 560 euros a month to 2,000 unemployed people in 2017 and 2018. There was no difference in employment in the first year; in the second, the basic income group worked six days more, which cannot be attributed to the basic income alone. Recipients reported clearly higher wellbeing.

Has any trial tested a truly universal income?

No large trial has paid everyone in a country. Finland’s experiment was limited to unemployed people, and US studies select a sample of participants. Alaska’s Permanent Fund Dividend is universal for residents but far smaller than a living income.

Why does AI make basic income a live question?

Because forecasts suggest AI could affect a large share of jobs. The IMF estimates that almost 40% of global employment is exposed to AI, and recommends social safety nets and retraining for workers at risk.

References

1. University of Helsinki (2020), The basic income experiment in Finland yields surprising results — Design and results of Finland’s 2017 to 2018 experiment: 560 euros a month to 2,000 unemployed people; small employment effects; higher wellbeing; no baseline survey.
2. OpenResearch, Unconditional Cash Study — 3,000 participants in Illinois and Texas; $1,000 a month for three years versus $50 for the control group; effects on work hours and employment.
3. Alaska Department of Revenue, Permanent Fund Dividend — A universal resident dividend; the 2026 amount is $1,200.
4. Stanford Basic Income Lab, Global Map of Basic Income Experiments — A database of basic and guaranteed income pilots around the world.
5. IMF (2024), Gen-AI: Artificial Intelligence and the Future of Work, Staff Discussion Note SDN/2024/001 — Almost 40% of global employment exposed to AI; safety nets and retraining are crucial.
6. The Budget Lab at Yale (2025), Evaluating the Impact of AI on the Labor Market: Current State of Affairs — No discernible disruption of the broader US labour market in the 33 months after ChatGPT’s release.