Artificial intelligence is global. But its potential impacts vary enormously between economies. How can all governments make the most of technological change, regardless of their stage of development? Era Dabla Norris of the International Monetary Fund argues that one of the main ways in which governments translate productivity into shared prosperity will be through policy.
Every major technological shift eventually becomes a policy test, and AI has now reached that point. Firms are experimenting and public institutions are adopting. AI capabilities are moving faster than labour market institutions, tax systems and safeguards can adjust.
The labour market and macroeconomic effects are still taking shape. We do not yet know how quickly adoption will spread, which tasks and jobs will change the most, how workers will adjust or how gains will be distributed across firms, workers and countries. That is why policy matters.
The path ahead will depend on three forces: capabilities, diffusion and readiness. Capabilities set the frontier. Diffusion determines where AI is actually used. And readiness determines whether countries can capture the productivity gains and manage any disruption.
The scale of potential labour market exposure is large
Our research estimates that almost 40 per cent of global employment is exposed to AI, with higher exposure in advanced economies. But exposure simply captures the different ways in which AI can change tasks within jobs: by automating or displacing them, by complementing them or by reorganising them.
Adoption determines where those changes take hold. That depends on whether firms can integrate AI into workflows. It requires usable data, skilled workers, financing and management capable of redesigning tasks and processes. Energy, grid capacity and data centre infrastructure also shape the pace.
These factors, which I call “preparedness”, are already beginning to show up in actual adoption patterns. Economies with stronger digital infrastructure, skills and labour market foundations are adopting AI more intensively. Others face slower diffusion and more uncertain productivity gains.
AI could widen gaps between countries. So closing these adoption gaps will be critical.
Three policy issues for labour markets in the age of AI
I see three critical policy issues. The first is skills and training. The labour market evidence shows that a transition is already under way. Our new work on job postings shows that about one in ten vacancies in advanced economies now requires at least one new skill. In emerging market economies, the share is about half as large.
Labour markets are already beginning to reprice skills. Jobs requiring “new skills” tend to pay more. Demand is strongest in professional, technical and managerial roles, especially in IT.
But priorities vary dramatically across countries and sectors. In some countries, the immediate issue is supply: whether education, retraining and mobility systems can help workers acquire the skills that firms are beginning to demand. In others, the issue is demand: whether firms are creating enough opportunities for workers to use those skills. Some countries may face both challenges.
The training evidence sharpens the concern
Among workers in the most AI-exposed occupations, our forthcoming research shows that most took part in some form of organised training over the past year, well ahead of the least exposed occupations. But participation is not the same as access. A sizeable minority received no training, and many more wanted more than they were able to take, most often because of work commitments.
That is a warning signal. Workplace change is concentrated among the most exposed workers, yet employer-paid training does not reliably accompany it. If training does not keep pace with changing jobs, AI could widen labour market divides. The policy question is where these gaps matter most. AI may also require stronger social protection and welfare systems to cushion potential labour market disruption.
Adjustment takes time, especially when workers need to retrain or find jobs that use new skills. AI-related disruption or displacement may appear as job losses, but it may also show up as lower earnings, fewer hours, weaker bargaining power or movement into lower-paid work.
This puts pressure on social protection systems that are already uneven in reach. Only 17 per cent of unemployed people globally receive cash unemployment benefits. Coverage exceeds about half of unemployed workers in high-income economies and is close to zero in low-income countries. Most countries would enter a potentially large AI-related disruption with limited automatic stabilisers.
Earlier episodes of automation in the United States offer some important lessons. In states that were more exposed to robot adoption, more generous unemployment insurance was associated with wage losses that were about two-thirds smaller. Importantly, it helped cushion workers without preventing labour markets from adjusting. Social assistance also helped to cushion the effects on poverty.
This brings the design choices into focus. How broad should coverage be if employment relationships become less stable? How generous should support be if disruption is prolonged? Can benefits become more portable as workers move across jobs, contracts and sectors? Is there a role for wage insurance where workers return to employment at lower pay? And how can countries with high informality and limited administrative capacity broaden their social safety nets while managing the fiscal costs?
Political economy matters too. If the gains from AI are concentrated, while the adjustment falls on workers with weaker protection, broadening the gains from AI becomes much harder.
Tax systems will need to be fit for purpose in an age of AI
AI raises the stakes for fiscal policy. If it lifts productivity and growth, it can strengthen revenues. If it disrupts labour markets, governments may face greater demands for training, income support and adjustment assistance.
The tax challenge is to preserve revenue, fairness and resilience as AI changes where income arises, who receives it and how concentrated the gains may become.
There are two distinct tax challenges.
The first concerns the taxation of AI-producing firms: chips, data centres, frontier models, software, cloud services and data. This ecosystem is concentrated, research-intensive and heavily reliant on intangible assets. It can generate rents that are mobile and difficult to locate, raising important questions about corporate taxation, profit shifting, tax competition, taxing rights and international co-operation.
The second concerns AI adoption across the wider economy. As AI spreads beyond the firms that produce it, the fiscal issues broaden. In labour markets, the issues are jobs, wages and possible transition spending. In product markets, the issues are concentration and rents outside the AI sector itself.
Distribution brings us to the balance between labour and capital. We know that labour’s share of income has been declining across many countries in recent decades. If AI adds further pressure, tax systems that rely heavily on labour income and social contributions could come under strain.
What matters is both the tax base and the incidence of the tax: what is taxed, where income arises and who ultimately bears the burden. In the age of AI, tax policy will be one of the main ways in which governments translate productivity into shared prosperity.
For governments, the agenda is clear
Prepare workers and firms, protect those facing disruption, share the gains through tax systems that remain fair and resilient, and keep adapting policy as AI capabilities evolve.
That work has to start now. The priority is to confront the hard issues early, before gaps in skills, social protection and taxation become harder to close.
How governments respond will shape whether AI becomes a source of stronger productivity and broader opportunity, or of deeper divides among workers, firms and countries. That is the policy test.
Watch Era’s speech at LSE’s Global Forum for AI and the Social Sciences.
This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.
The post The policy tests facing governments in the age of AI first appeared on LSE Business Review.
