Universal Basic Income and Automation

Basic income is the standard policy answer to automation, and the standard supporting evidence is a set of trials. Those trials are good research. Almost none of them was designed to answer the question they get cited for, and the mismatch is specific enough to be worth spelling out.

What UBI is, exactly

A universal basic income is a cash payment that is universal (everyone gets it), unconditional (no work or job-search requirement), individual (not household-assessed), and regular. Each adjective excludes something that is often called UBI and is not:

The trials, and what they test

The evidence base is real and larger than people assume. The US and Canadian negative income tax experiments of the 1960s and 1970s were large randomised trials, including the Manitoba Mincome project. Finland ran a nationwide randomised experiment giving unconditional payments to unemployed recipients in 2017 and 2018. Stockton, California ran a guaranteed income pilot with a control group. GiveDirectly has run a long-horizon randomised universal basic income study in Kenyan villages with arms of different duration, which is the closest existing design to a permanent programme. Alaska’s dividend has been studied as a natural experiment in aggregate labour supply.

What this literature is built to measure, and measures well: labour supply responses among recipients, effects on health and reported wellbeing, how recipients spend money, and effects on children in recipient households. These are the outcomes the designs are powered for, and the general shape of the findings — that recipients do not broadly stop working, and that health and wellbeing measures move favourably — is reported consistently enough across studies to be worth taking seriously.

This page does not quote effect sizes. Read the individual studies for those; each defines its outcome differently and the numbers do not transfer between them.

The external-validity gap

Here is the part that matters for the automation question. Five properties of these designs make them poor evidence about a permanent, universal programme in an economy where labour demand has fallen.

• Partial equilibrium. When a few thousand people receive money, prices, wages and rents do not move. When everyone does, they may — and where the payment goes in the presence of inelastic supply, housing above all, is one of the central disagreements about UBI. No trial can observe this, because observing it requires the programme to be universal.
• Time limitation. A payment known to last two years supports different decisions than a permanent one. Leaving a job to retrain is rational under a permanent floor and reckless under a temporary one, so trials systematically understate the labour-supply response of a real programme. This is a known limitation, stated in the studies themselves.
• Funding is external. Trials are financed by researchers or philanthropy. A real programme is financed by taxation, and the incidence of that taxation has its own effects on employment and investment. The trials measure the transfer and not the tax, which is half of the policy.
• Labour demand still exists. Every trial ran in an economy where recipients could get a job. The automation scenario is one where they largely cannot, so the mechanism most trials actually document — that cash does not much discourage work — is measuring the wrong margin for that scenario.
• Selection and scale. Pilot participants are recruited and consent. Effects at the level of a city or a country include general-equilibrium and political dynamics with no analogue at pilot scale.

None of this makes the trials useless. It makes them evidence for a narrower claim than they are usually cited for: unconditional cash does not appear to produce the behavioural collapse its critics predicted. That is a genuine finding and it refutes a real objection. It is not evidence that a universal permanent programme is affordable, that it would not be capitalised into rents, or that it would work in an economy without wage labour.

The financing identity

The arithmetic is worth doing yourself with current figures for your own country rather than accepting anyone’s summary, and the identity is simple enough to do on paper:

gross cost   = payment per person x eligible population
net cost     = gross cost
             - existing transfers replaced
             - payments recovered through the tax system
share of GDP = net cost / GDP

Two lines do most of the work in any argument about affordability. What existing programmes are replaced determines whether the poorest are better or worse off, since replacing targeted benefits with a flat payment moves money from the neediest to the median. And how much is recovered in tax determines whether “universal” is a statement about who receives money or about who nets it: a universal payment funded by progressive taxation is arithmetically similar to a means-tested one, and differs mainly in administration, stigma and take-up.

Automation-specific funding proposals exist and are worth evaluating on their own terms: taxing capital or automation directly, on the argument that current tax codes favour capital over labour; sovereign wealth funds on the Alaska model, funded from a share of returns; and data or compute levies. Each has an incidence question — who ultimately bears it — that is not answered by naming it.

The questions money does not answer

Even granting that a basic income is affordable and works as designed, it addresses one variable. Three others come up in the automation literature and are not addressed by a payment.

Meaning and structure. Employment supplies routine, social contact and identity as well as income. The evidence on involuntary job loss finds wellbeing effects that are not fully explained by lost earnings. Whether a society without much wage labour reconstructs those goods elsewhere is a genuinely open question, and the job-guarantee camp makes it their central argument.

Bargaining power. A transfer replaces income and not leverage. The argument in gradual disempowerment is that a population whose labour is not needed has less influence over the institutions paying it, and that this does not depend on the payment being small.

Political durability. A programme that must be re-legislated is only as stable as the coalition behind it. Whether that coalition is stronger when a benefit is universal — everyone has a stake — or weaker when its recipients have no economic leverage is contested among political scientists, and it is a question about institutions rather than about economics.

Whether to prefer basic income, a job guarantee, expanded targeted transfers or something else is a value judgement about what matters: income, work, autonomy or power. The empirical literature constrains the options; it does not choose between them.