OIT - Why Universal Basic Income is not the answer to the impact of AI on the labour market

 ILO Brief 1 Why UBI is not the answer to the impact of AI on the labour market  Research Brief May 2026 Why UBI is not the answer to the impact of AI on the labour market1

Tech developers in Silicon Valley seem to be convinced that AI will lead to mass unemployment and that only Universal Basic Income (UBI), funded by the government, would be the appropriate policy response. Several leading CEOs have made two related economic claims: that AI will devastate labour markets catastrophically, and that the productivity gains it generates will make such transfers noninflationary. Both claims are empirically weak , and the second does not follow from the first even if the first were true. AI transforms tasks, not occupations The idea that automation eliminates jobs wholesale has been refuted by every major wave of technological change in economic history. What technology typically displaces is not occupations but specific tasks within them. This is the central insight of the task -based framework developed by Acemoglu and Autor (2011) and Autor, Levy, and Murnane (2003): occupations are bundles of tasks — routine, analytical, social, creative, supervisory — and automation tends to substitute for the routine subset while complementing the rest.

Consider accounting : AI can automate data entry, document classification, standard reporting, and

1 Dr. Tuğba Özay, Dr. Ekkehard Ernst ; Founder, MindHu, tozay@mindhu.com; ILO, ernste@ilo.org Dr. Tuğba Özay is the founder of MindHu, a decision-support platform that applies task-based models, including an AI Risk Index (AIRI) and an Automation Risk Index (ARI), to measure occupational vulnerability to AI and automation at the firm level. Dr. Ekkehard Ernst is Chief of the Macro-economic policies and jobs unit at the International Labour Organization.

reconciliation, b ut interpretation, risk assessment, regulatory compliance, client communication, and

(AIRI) and an Automation Risk Index (ARI), to measure occupational vulnerability to AI and automation at the firm level. Dr. Ekkehard Ernst is Chief of the Macro-economic policies and jobs unit at the International Labour Organization.

reconciliation, b ut interpretation, risk assessment, regulatory compliance, client communication, and strategic advisory are strengthened, not replaced. The same logic applies to law: contract screening and preliminary case research can be accelerated by AI, while negotiation, litigation strategy, and legal judgment remain firmly human. In both cases, what we observe is not occupational death but occupational redefinition. The historical record is consistent. During the Industrial Revolution, the Luddite movement expressed genuine fear that machinery would permanently displace craft labour. In the short run, some trades did contract and transition costs were severe. But in the long run, total employment did not collapse. On the contrary, production expanded, new sectors emerged, and urbanisation generated a broader division of labour. Agricultural mechanisation through the twentieth century offers a parallel: as farm employment fell sharply in advanced economies, labour migrated first to manufacturing, then to services. The economy's sectoral composition changed; aggregate employment did not crater. What technological revolutions do cause is shifts in remuneration: the steam engine and electrification turned skilled manual jobs into routine and repetitive occupations, allowing unskilled workers to carry out the work at much ILO Brief 2 Why UBI is not the answer to the impact of AI on the labour market lower wages and cheaper cost. The Luddites fought against impoverishment, not against job loss. At the same time, this technological transformation allowed manufacturing to absorb massive inflow of workers who were leaving the agricultural sector, bringing about the most significant and rapid change of European economies in the two decades after World War II and the basis for their economic rise and prosperity. A similar mechanism operated during computerisation. Office automation reduced demand for secretarial and

agricultural sector, bringing about the most significant and rapid change of European economies in the two decades after World War II and the basis for their economic rise and prosperity. A similar mechanism operated during computerisation. Office automation reduced demand for secretarial and clerical work while creating software development, data analysis, digital marketing, e -commerce logistics, and cybersecurity. Banking offers a particularly instructive case: Bessen (2015) documented that ATM deployment reduced tellers per branch from roughly 20 to 13 between 1988 and 2004, yet total teller employment did only fall after the introduction of the smart phone as users found it easier to access their bank account from their phone rather than to go to the local branch of their bank . The underlying economics are straightforward: technological progress raises productivity, reduces unit costs, expands demand, and generates new labour requirements through scale, income, and new -product effects. The net employment impact cannot be read o ff the direct substitution channel alone. The main difference between the ICT revolution and the earlier period of electrification was who stood to benefit from these shifts. Computerisation required the use of cognitive skills, more than pure muscle power and so it was the more educated part of t he population who benefited from that shift, leading to a rise in returns to skills and education. Again, no massive increase in unemployment but a shift in the distribution of technological gains. Frey and Osborne's (2013) widely cited estimate that 47 per cent of US jobs are at risk of computerisation captures exposure to task automation, not inevitable employment destruction , a distinction their own subsequent work has emphasised. The more probable outcome of the AI transition is not mass unemployment but labour market restructuring: a shift in wage structures, skill premia, and sectoral distribution of employment , which is consistent with the framework of Acemoglu and Restrepo (2019) on automation displacing and reinstating labour through new task creation. The real policy problem is not

labour market restructuring: a shift in wage structures, skill premia, and sectoral distribution of employment , which is consistent with the framework of Acemoglu and Restrepo (2019) on automation displacing and reinstating labour through new task creation. The real policy problem is not that everyone will lose their jobs, but that not everyone will transition at the same speed. This causes a structural mismatch and transition cost s, a serious but manageable problem, and quite different from the catastrophe that the tech gurus imply. Importantly, as the latest wave of Generative AI tools are set to erode skill premia, we find ourselves again at a Luddite moment, this time with the educated workforce trying to destroy computers to prevent their skill premia to get eroded. Productivity growth does not neutralise inflation from income transfers That leaves us with the question what to do with the coming productivity gains that the AI revolution promises. Even granting for argument’s sake that AI-driven unemployment will become severe, the claim that UBI would be noninflationary because AI expands supply does not hold. Macroeconomic price levels are determined by the interaction of money supply, aggregate demand, productive capacity, and critically, sectoral supply heterogeneity. AI productivity gains are real but radically uneven across sectors. In scalable sectors , such as software, digital content, certain administrative services, some manufacturing , AI can drive marginal costs toward zero. But in structurally constrained sectors, including housing, healthcare, urban land, specialist labour, energy infrastructure, personal care, supply cannot expand at anywhere near the same pace. Bottlenecks here are not computational but physical, institutional, and temporal: land scarcity, planning regulations, professional licensing, construction capacity, human embodiment. These sectors are precisely where households direct a large share of their spending when income rises. The implication is that AI does not create homogeneous abundance. It creates technological deflation in some markets and leaves structural scarcity untouched in others.

regulations, professional licensing, construction capacity, human embodiment. These sectors are precisely where households direct a large share of their spending when income rises. The implication is that AI does not create homogeneous abundance. It creates technological deflation in some markets and leaves structural scarcity untouched in others. A universal high-income transfer would generate a demand shock that flows disproportion ately into the constrained sectors, pushing prices upward even as digital goods become cheaper. Housing is the paradigm case: in virtually every major city, short -run housing supply is inelastic. Additional purchasing power translates directly into higher rents and prices, not more homes. This is not a theoretical conjecture. Post -WWII demand releases in the United States triggered significant inflation despite rapid capacity expansion , demonstrating that production growth does not automatically suppress price pressure when demand is released suddenly and unevenly. ILO Brief 3 Why UBI is not the answer to the impact of AI on the labour market The stagflationary episodes of the 1970s showed that supply-side constraints and demand -side stimulus can interact to produce inflation and unemployment simultaneously. Even extreme cases such as Weimar Germany or more recently Zimbabwe and Venezuela, confirm the basic principle: when money supply grows faster than real output across the relevant consumption basket, prices rise. UBI does not only affect demand, it will also affect public finances through the way it is financed. Funds can come through three ways: taxation (which redistributes but does not create new aggregate demand), borrowing (which has limits, especially in countries without deep financial markets), or money creation (which fuels inflation when nominal demand expands without relieving constrained supply). For UBI to work, AI needs to expand supply in exactly the sectors where people will spend the new income. The composition of productivity gains does not support this. There is also a labour market dimension. UBI raises reservation wages in low -wage, physically demanding, or

supply). For UBI to work, AI needs to expand supply in exactly the sectors where people will spend the new income. The composition of productivity gains does not support this. There is also a labour market dimension. UBI raises reservation wages in low -wage, physically demanding, or low-prestige occupations , care work, logistics, cleaning, food supply chains. Existing evidence documents the adverse effects of UBI on welfare through reduced labour force participation. If robotics has not fully substituted for these roles, the result is wage and cost pressure in precisely the sectors most relevant to everyday living standards. AI may substitute for some forms of human labour while simultaneously making complementary human work more valuable and more expensive. What policy should follow None of this implies that AI poses no distributional challenge. On the contrary, recent evidence suggests that cognitive jobs are seeing their skill premium eroding, especially for younger job market entrants . Hence, t he more likely scenario involves asymmetric adjustment: workers in specific, well-paying service sector occupations are facing sharper disruption s; regional divergence widening; labour income shares declining (again) . These are real problems warranting serious policy responses. But the appropriate framework centres on reducing transition costs and expanding constrained supply , not on cash transfers premised on a mischaracterisation of how AI affects the economy. Reskilling programs targeted at the occupational groups most exposed to task displacement, active labour market policies, portability of benefits, expanded housing supp ly through regulatory reform, direct provision of healthcare and public services, and taxation of AI -generated economic rents are all more defensible instruments. Universal social protection plays an important role in this respect. But it needs to be introduced in a way to strengthen labour supply incentives especially at the lower end of the income distribution, for instance through an earned-income tax credit. The fundamental confusion in the argument for UBI as a response to AI is conflating nominal income with real purchasing power. The state can increase the number

in a way to strengthen labour supply incentives especially at the lower end of the income distribution, for instance through an earned-income tax credit. The fundamental confusion in the argument for UBI as a response to AI is conflating nominal income with real purchasing power. The state can increase the number printed on every cheque. Whether the goods and services that determine real welfare , housing, health, quality food, care, are available in sufficient quantity at stable prices depends on supply -side conditions that AI improves only partially and unevenly. Distributing money faster than constrained supply can adjust produces inflation, not abundance. The question is not whether to support workers through the AI transition. The question is how to do so accurately , starting with an honest account of what AI actually does to labour markets and to the price level.

Licensed under CC BY 4.0 © International Labour Organization 2026 ILO. Why UBI is not the answer to the impact of AI on the labour market, Research brief, Geneva: International Labour Office, 2026. © ILO. https://doi.org/10.54394/00034454

Contact details International Labour Organization Route des Morillons 4 CH-1211 Geneva 22

Switzerland T: +41 22 799 7239

E: research@ilo.org DOI: https://doi.org/10.54394/00034454 © International Labour Organization 2026