Distributing the AI Economy and the 1.6-Point Drop in Labour Share── As AI-driven automation pushes down the labour income share, three policy tools — robot taxes, basic income, and workforce redeployment — are being tested, but none alone can halt widening inequality

01The global labour income share fell 1.6 points in 20 years

The ILO's World Employment and Social Outlook, published in September 2024, shows that the global labour income share dropped from 53.9% in 2004 to 52.3% in 2024 — a 1.6-percentage-point decline. In purchasing-power-parity terms, that shift represents $2.4 trillion per year in income that moved from workers to capital owners. Roughly half of the decline occurred during the pandemic years of 2020–2022, but technology-driven automation has been a structural driver throughout.

Daron Acemoglu laid out the mechanism in his 2024 NBER paper "The Simple Macroeconomics of AI." In a task-based framework, automation expands the set of tasks performed by capital — machines, algorithms — and contracts labour's share of value added. Over the past 30 years, worker displacement has outpaced reinstatement, the creation of new labour-intensive tasks. Whether AI will widen this imbalance is the central distributional question.

02Twenty-seven percent of OECD jobs sit in high-risk occupations

The OECD's 2023 Employment Outlook found that 27% of jobs across member countries are in occupations at high risk of automation. Earlier waves of automation targeted routine manual and clerical work; AI extends the threat to non-routine cognitive tasks. Medicine, law, and finance — fields that require years of education and rely on accumulated experience — now face displacement risk as well.

The distributional effects run in two directions. First, if high-wage cognitive work is automated, pressure reaches into the middle and upper tiers of the income distribution. Second, if the productivity gap between workers who can use AI and those who cannot widens, wage dispersion will grow even within firms. The OECD also notes that the workers in the highest-risk occupations tend to be low-skilled, young, and male — a pattern that reinforces existing inequalities.

03Robot taxes aim to correct a structural asymmetry in the tax code

In an August 2026 essay, Bill Gates identified a concrete rationale for taxing automation. Employers pay payroll taxes when they hire people but can deduct the cost of equipment through depreciation. This asymmetry makes replacing a worker with a machine cheaper on paper than it should be. Gates proposed levying fees on both AI tokens — the computational units AI models consume to process requests — and physical robots, with the revenue funding retraining and transition support.

South Korea was the first country to act. In 2017, it reduced the tax deduction rate for automation investment by up to 2 percentage points — a de facto robot tax. In August 2026, the Korean legislature received bills that would charge employers directly for AI-driven layoffs, tying the levy to headcount outcomes rather than capital expenditure.

The European Parliament, by contrast, rejected a robot-tax proposal in 2017, and the International Federation of Robotics supported that decision. The core objection is that taxing automation would suppress productivity-enhancing investment and erode international competitiveness. Set the rate too high and companies move investment abroad; set it too low and the proceeds fail to fund meaningful redistribution.

04Basic-income experiments disproved the laziness hypothesis but face a scale problem

Universal basic income is a recurring proposal for cushioning AI-driven job loss. Experimental evidence is accumulating.

Finland ran a two-year trial in 2017–2018, paying 2,000 randomly selected unemployed people €560 per month with no conditions. Recipients reported higher life satisfaction and lower mental strain than the control group. They worked an average of 78 days during the measurement period — six more than the control group. The hypothesis that unconditional cash discourages work did not hold.

In Kenya, GiveDirectly has been running the world's largest UBI study since 2017, covering roughly 20,000 people across about 200 villages. The first results, published in December 2023, showed that households receiving a one-time lump sum of approximately $500 saw their incomes rise by 50% compared to the control group. Recipients started more businesses and earned more from them than even the group receiving 12-year monthly transfers. The fear of idleness was not borne out in either experiment.

Both experiments share a constraint: they cover thousands to tens of thousands of people, not entire national populations. Scaling Finland's €560 per month to the country's 4.4 million adults would cost roughly €30 billion a year — about 40% of Finland's 2023 revenue. Financing remains UBI's largest unresolved problem, and it is the main reason the idea is discussed alongside robot taxes.

05Challenger data show AI-cited US job cuts surging in 2026

To ground the distributional question in observed data, consider the pace of AI-related layoffs. According to Challenger, Gray & Christmas, the US outplacement firm, employers have cited AI in approximately 100,000 job-cut announcements since tracking began in 2023. Through May 2026, the cumulative figure stood at 87,714 — already exceeding the 54,836 for all of 2025. In May 2026 alone, 38,579 cuts named AI as the reason, the highest single-month total on record.

These figures reflect employer-stated reasons and do not prove direct causation. But the sharp increase in the frequency with which companies publicly cite AI as a reason for headcount reductions is itself a signal of labour-market pressure. In the pharmaceutical sector, AI-driven automation of document drafting, data entry, and analytical tasks is already reshaping the staffing profile of support functions.

06Pharma is spending $3.2 billion on reskilling but the training-deployment gap persists

The pharmaceutical industry is both subject to AI's labour effects and a testing ground for workforce redeployment. Gartner estimated that pharma reskilling expenditure would reach $3.2 billion by 2025, driven by projections that 50–60% of analytical tasks could be automated by 2030.

Yet a gap persists between training and deployment. A 2025 industry survey found that 73% of quality-unit leaders said AI oversight had become part of their function, but only 28% reported that their teams had received formal AI training. Training programmes that exist on paper but are not integrated into day-to-day work produce little practical effect.

Cross-industry hiring — bringing data scientists from finance, aerospace, or technology and giving them crash courses in pharmaceutical regulation — is spreading. But acquiring contextual knowledge of drug regulation takes years, not months. The ability to use AI and the ability to make safety and efficacy judgements about medicines are distinct capabilities, and developing both in the same person is the operative challenge.

07Distributional design requires moving taxation, social security, and workforce investment simultaneously

No single policy can govern the distribution of AI-generated value. A robot tax alone risks suppressing investment. UBI alone lacks a viable financing mechanism at national scale. Reskilling alone fails if there are no roles to redeploy people into. The three instruments must be combined.

In practical terms, the first step is to neutralise the tax-code asymmetry between payroll taxes and equipment depreciation, so that the choice between human and machine labour is not distorted by the tax system. The second is to extend existing social-security mechanisms — unemployment insurance, vocational-training allowances — to cover structural displacement driven by AI. The third is to link reskilling programmes to actual internal transfers, so that training leads to a job, not just a certificate.

For the pharmaceutical industry, the third challenge is the most immediate. Regulatory review, safety evaluation, and quality oversight are judgement-intensive tasks that resist automation. Securing pathways to redeploy people into these functions is the key to reconciling automation with employment.