A World Without Enough Jobs: What Happens Next?

Every major technological change has come with warnings about disappearing jobs. Agricultural machinery reduced the number of people needed to grow food, factories transformed manufacturing and computers automated large amounts of clerical work. The internet created industries earlier generations couldn't have imagined. Through all of it, people kept working, because new occupations emerged, existing jobs evolved and the economy adapted.

That's a reassuring story, but it's an incomplete one. Technological progress hasn't benefited everyone equally. Some communities lost industries that had supported them for generations, some workers never regained their previous earnings and others had to move, retrain or accept less secure work. The fact that an economy eventually creates new opportunities doesn't mean they arrive in the right places, at the right time or for the people who need them.

AI adds a further uncertainty. Earlier technologies mostly replaced particular physical or routine tasks, whereas AI can increasingly perform cognitive work such as writing, analysis, research, coding and coordination. That doesn't mean every job involving those activities will disappear, since most jobs combine many responsibilities and automating a task isn't the same as eliminating the person who does it.

But if AI can take on a growing share of both routine and skilled work, it's reasonable to ask whether new demand for human labour will keep emerging fast enough. It might. New industries could create millions of jobs, lower prices could increase demand and entirely new services could become viable. The transition could also prove harder than previous ones. History shows that economies can adapt to technological change, but it doesn't guarantee that everyone benefits or that the adjustment is easy.

It's worth taking seriously a possibility that's often dismissed before it's properly examined: that AI could eventually allow businesses to produce much of what society needs with substantially fewer hours of human labour, across a large part of the economy rather than in one industry. This could play out in several ways. Some businesses might reduce hours while keeping people on full pay, others might cut prices and attract more customers, some might expand into new markets and create different kinds of work, and others might reduce their workforce and keep more of the gains. The outcome would depend on business decisions, economic conditions, government policy and how quickly the technology develops.

We shouldn't assume widespread unemployment is inevitable. Many jobs depend on physical presence, relationships, responsibility and judgement, AI adoption is uneven, and rising productivity can create new demand as well as reduce the labour needed for existing work. But we shouldn't assume new jobs will automatically appear in sufficient numbers either. Telling someone who has spent twenty years in a profession that suddenly needs far fewer people that technology will create opportunities somewhere else in the economy offers little comfort when their mortgage is due next month.

The evidence so far points in both directions. The International Labour Organization estimates that around one in four workers worldwide is in an occupation with some exposure to generative AI, but expects most of those jobs to be transformed rather than eliminated.^[1] At the same time, US payroll data analysed by Stanford's Digital Economy Lab shows employment among 22- to 25-year-olds in the most AI-exposed occupations running about 19% below where it would otherwise have been, with no comparable gap for older workers.^[6] Neither finding shows mass job losses. The second does suggest that if AI reduces the demand for labour, the effect may show up first as fewer people being hired rather than as large-scale redundancies, which is easier to miss and harder to respond to.

There's also a wider economic problem. Imagine a supermarket where AI manages much of the ordering, logistics and administration, waste falls and operating costs drop, so prices come down while the business stays profitable. That's progress. Now imagine that many of its suppliers, along with local offices, retailers and service providers, have also reduced their workforces. Some former employees find new jobs, others work fewer hours or earn less, and some can't find work at all. Those households start cutting back, buying fewer non-essentials, cancelling subscriptions and postponing holidays and home improvements. Businesses have become better at producing goods and services, but some of their customers now have less to spend.

This is one of the central questions a highly automated economy would raise. Producing more doesn't automatically mean people can afford more. Lower prices, new industries and government support could all help sustain spending. But if most income continues to come from employment while employment becomes less available, society would need other ways to make sure people can participate in the economy. The danger would be less a shortage of goods and services than a shortage of the income people need to buy them.

In the first article in this series I looked at how the gains from AI are shared within individual businesses. The same question applies to whole countries. Imagine two countries adopting similarly advanced AI and becoming similarly productive. In the first, most of the gains flow to business owners, investors and the people who own valuable technology, so profits and asset values rise while many households see little improvement in their financial security. In the second, more of the benefit is shared through higher wages, shorter hours without lost income, better public services and support for people who can't find enough work. The technology is the same, but the outcomes for citizens are very different.

AI doesn't decide how the wealth it helps create is distributed. Businesses, governments and institutions do. If ownership of valuable AI systems and the businesses built on them becomes highly concentrated, the financial rewards are likely to become concentrated too. There are ways to respond. Tax policy can influence how gains are shared, public investment can spread the benefits, pension funds and other investment vehicles can give more people a stake in productive businesses, and competition policy can help prevent excessive concentration of market power. None of these guarantees a fair outcome, but together they show that the distribution of prosperity is a matter of policy rather than a technical consequence of automation.

Universal basic income, or UBI, comes up in almost every discussion about what happens if AI reduces the need for human work. The principle is simple: every eligible person receives a regular cash payment regardless of whether they're employed. They can still work and earn more, but their entire financial security no longer depends on finding a job. Imagine a payment of £1,000 a month to every adult in the UK. For someone in secure employment it might provide breathing room. For someone moving between contracts it could provide a financial foundation. For someone caring for a relative, it could recognise that valuable contributions to society don't always come with a salary.

The obvious problem is cost. The UK has roughly 55 million adults, so paying each of them £12,000 a year would cost in the region of £660 billion before any changes to tax or existing benefits.^[7] That's an illustration rather than a costed policy, and the net cost could be very different depending on how taxes and benefits were redesigned around it. But the scale shows why UBI isn't simply a matter of deciding that everyone deserves financial security. Someone has to fund it, and there are hard questions about whether a universal payment is the best use of public money. Would it replace existing benefits? Would people with disabilities or very high housing costs still receive extra support? Would a flat payment be enough in expensive areas, and how much would taxes need to change to pay for it?

Real-world experiments offer some evidence, but less than people sometimes assume. Finland's basic income experiment paid 2,000 unemployed people €560 a month, with no conditions, for two years from 2017.^[4] The effect on employment was small, but recipients reported greater life satisfaction, less mental strain and more confidence in their financial situation. That's useful evidence about how people respond to greater income security. A temporary trial involving a small group of unemployed people can't tell us what would happen if a whole country introduced a permanent payment for everyone, though. UBI is one possible response to less dependable employment, but it isn't the only one, and it isn't automatically the best.

Imagine two people who lose their jobs when their employer introduces AI. The first has substantial savings, investments and a partner with a secure income. The second has no savings, rents privately and supports two children. They've experienced the same event, but the consequences are completely different. A universal payment would treat them identically, while a more targeted system could give far more support to the second household. That raises a genuine choice between providing everyone with the same financial foundation and concentrating resources on the people who need help most.

Both approaches have merit. Universal schemes are simpler and avoid some of the stigma and complexity of means-tested support. Targeted schemes direct more money towards people facing the greatest hardship, but they bring eligibility rules, administrative burdens and situations where people lose support as their earnings rise. Support also doesn't have to come entirely as cash. Affordable housing, accessible healthcare, reliable transport and childcare all reduce the income a household needs for a reasonable standard of living, and governments could also strengthen support for people changing careers or moving into work that remains in demand.

The right mix would depend on what actually happens to employment. A short period of disruption in particular industries calls for a different response from a lasting, economy-wide fall in demand for labour. In either case, though, the underlying issue is the same: if employment becomes less reliable, the systems that provide financial security may need to depend less on employment too.

Suppose a company that used to employ 500 people introduces AI and automation over several years and now produces the same output with 300. It's still profitable and more productive, but it's paying 200 fewer salaries. That matters for public finances, because governments raise a large share of their revenue through taxes linked to employment, alongside taxes on profits, spending and property. If economic output becomes less closely tied to paid work, the balance of taxation may need to shift. AI sharpens the question, because much of the value may sit with the companies that own the models and the computing infrastructure they run on, often in a different country from the work being replaced.

Some people have proposed taxing robots or automated systems directly, on the basis that businesses benefiting from replacing human labour should help support the people affected. The idea runs into difficulties immediately, starting with what counts as a robot. An industrial machine might, but what about an AI recruitment system, or a spreadsheet that automates a calculation? A tax on automation could also discourage investment that would make businesses more productive and raise living standards. Economists who have modelled the question, such as Guerreiro, Rebelo and Teles, have found that the case for a robot tax is limited and largely temporary, strongest while workers whose jobs are being automated are still in the workforce.^[5]

Other proposals focus on taxing profits, capital gains or certain forms of economic rent more effectively, or on public investment funds that give citizens a broader stake in productive assets. All of these involve trade-offs, because businesses respond to taxes, investment can move elsewhere and poorly designed policies have unintended consequences. There's no simple formula that turns automation into a guaranteed source of public revenue. But if AI changes where economic value is created and who receives it, governments will need to reconsider how the costs of running a society are shared, with the aim of ensuring that technological progress continues to help fund the society that makes it possible rather than punishing it.

Imagine a different version of the future, in which your housing is secure, you can afford food, healthcare, transport and the occasional holiday, and you no longer need to work forty hours a week to maintain that standard of living. What would you do with your time? Some people would keep working because they enjoy their profession, like solving problems and value being part of something. Others would work fewer hours and spend more time with family, volunteering, travelling or pursuing interests they've put off for years. Some would start businesses without needing them to provide a full-time income straight away.

Others would struggle, because work provides more than money. It gives structure to the week, social connection, and a sense of achievement and belonging, and I explored in the fourth article how closely many people's identities are tied to what they do. Removing the financial need to work wouldn't automatically produce a fulfilling life.

At the same time, paid employment has never been the only meaningful way to contribute. Caring for relatives, supporting neighbours, volunteering, creating art and mentoring young people all create real value without a payslip, and much of the work that keeps communities functioning has always been unpaid. A society with less paid employment would need to get much better at recognising that contribution. A world with less work could offer more freedom, but freedom and purpose aren't the same thing, and the second doesn't follow automatically from the first.

There's a danger in discussing a highly automated future as though everyone will arrive there together. Some people already work in occupations where AI is changing expectations, while others work in industries where adoption is much slower. Some have savings, property, investments and professional networks that help them adapt, while others live from one payday to the next. For one household, losing a job might mean postponing a holiday. For another, it could mean falling behind on the rent.

That's why the pace of change matters so much. If employment patterns shift gradually, people and institutions have time to adapt. If disruption hits particular industries or communities quickly, the consequences could be severe, and the people most exposed are often the least able to protect themselves. Governments shouldn't try to prevent every technological change, but they do need to consider who bears the costs while society waits for the benefits. A transition can be economically successful in the long run and still cause avoidable hardship along the way, and the promise of a better future is little comfort to someone who can't get through the present.