Introduction

This paper explores the defining economic challenge of our time: the prospect of artificial intelligence rendering human labour uncompetitive. Our aim is to set out clearly the challenges posed by the Economic Singularity, and we invite thoughtful commentary, debate, and – most importantly – proposals for solutions.

The paper combines analytical argument with storytelling. This may feel an awkward combination to some readers, but we believe it is often the best way to explore the future.

Section 1 argues that the Economic Singularity is coming, and that we need to plan for it.

Section 2 describes one possible scenario, in which the transition is rapid, and on the whole successful. Of course we do not claim that this is the only possible scenario, but we do think it is realistic, and framing challenges with a story can make them seem more real. Section 3 continues the story, and explores the challenges the transition raises.

Section 4 poses some of the outstanding questions. These questions deserve the broadest possible responses, from economists and technologists to policymakers, ethicists, and ordinary citizens whose lives will be shaped by the decisions we make.

The Economic Singularity is not a distant hypothetical; it is approaching, and we are unprepared.

1. The Inevitability of the Economic Singularity

1.1 Automation

Automation has been happening for centuries, but until recently, it has mostly been mechanisation. This caused technological unemployment for horses, but not for humans, who have cognitive abilities as well as muscle power. Automation makes production processes more efficient and creates wealth, which in turn creates new jobs. This will continue, but as AIs take on more of the cognitive work, the new jobs will eventually be taken by AIs and not by humans.

Unless we stop developing increasingly capable AI systems, or unless there exists some inherent “silicon ceiling” beyond which machine intelligence cannot improve, we are heading towards an economy in which there are no jobs for the great majority of humans. This is the Economic Singularity: the moment when, for every task a human might perform for money, there is a machine – whether software, hardware, or some combination – which can do the job cheaper, better, and faster. The shift might be gradual, but we think that economic forces are likely to make it sudden.

There will still be plenty of work for humans, but no paid work – i.e., no jobs. Humans do a great deal of unpaid work, carrying out domestic chores, caring for family and friends, and pursuing hobbies and sports. But our labour will be economically uncompetitive, so there will be no jobs for humans unless they are subsidised.

1.2 Closing the gap

This is not a prediction about the distant future. Each year, AI systems demonstrate new capabilities that were recently thought to be uniquely human: legal analysis, medical diagnosis, creative writing, software engineering, scientific research, and even emotional support. The trajectory is clear.

Today, in mid-2026, the world´s largest economies are close to full employment. AIs are making some of us much more productive, but they are not yet taking over entire jobs at scale. They are still unreliable. This is partly because their brilliance and speed is undermined by their hallucinations, but this, while not a solved problem, is a shrinking one. The main reason why AIs are not yet reliable enough to take our jobs is that they lack context: they do not understand the world around them. They are geniuses in some respects, and fools in others. This has been called their jagged edge, and another term for it is a lack of common sense.

It would be a profound and dangerous folly to rely on this shortcoming lasting forever. It may be solved by the brute force of the scaling laws. Or by world models – AIs that are not trained by language alone, but by interacting with the real world, and simulations of it. Or it may require a third AI big bang, following the arrival of deep learning in 2012 and transformer technology in 2017. One way or another, AIs will acquire common sense.

1.3 Timing and planning

The most important open questions are when the transition will happen, and how we can survive it. The people running the largest AI firms think it could happen this decade. Some of them have a track record of being over-optimistic, so maybe it will be sometime in the 2030s.

As for how we could survive it, we currently have no plan. Until very recently, most members of the profession which might be expected to prepare such plans – economists – have been in denial. They insisted that automation did not cause lasting technological unemployment in the past, and therefore it will not in the future. This assumes that the future will be like the past, but if that was true we would not be able to fly. They also complain, correctly, that there is no data to support the technological unemployment thesis. This is because the transition has not started yet. In 1908 there was no data about international air travel.

The arrival of large language models, their astonishing capability and their dizzying rate of progress, has started to weaken the resolve of economists to deny that the Economic Singularity is coming. But only to a degree: a new academic paperⁱ forecasts a permanent decline in labour force participation by 2050, but only by 13%.

2. The Economic Singularity arrives: a scenario

2.1 Full automation

It is 2033. Like most people you know, you make your living by guiding AI agents. Working within agreed parameters, you choose which problems they solve. You monitor their performance. You correct their assumptions when they drift. You shepherd them. You enjoy your work.

Your job description changed four times in the last year, as upgraded AI agents took over aspects of the work you were doing. The previous year it changed twice, and the year before that just once. Something similar has happened to everyone you know.

There is a common belief that the Economic Singularity will arrive soon, and recently there have been rumours from inside the leading AI development companies about a new type of model that has as much common sense as an adult human. They have become known as Human-level Agentic Labour, or HALs for short, after the AI in “2001”. There is a race between these companies to be the first to release this new technology, but it is widely assumed that they will all get there about the same time. Key staff have been moving between them, cross-pollinating their best ideas, driven by excruciatingly large sums of money.

Anthropic, Google and Mistral announce the release of their HALs on the same day. OpenAI, Baidu, and a couple of others announce a few days later. Adventurous firms deploy them, and it is true: these agents can execute any task that humans can, without supervision. Across numerous industry sectors, aggressive firms deploy them and gain an immediate and massive cost advantage by letting go most of their human employees, leaving a handful of people in post to set strategic direction. Their competitors know they must either follow suit quickly or go bust.

With remarkable speed, in a matter of months, humans are replaced in the majority of jobs in most companies in the developed world, and the same is happening slightly more slowly elsewhere. The companies deploying HALs are now generating most of the added value in the world´s economy, and although they are fiercely competitive with each other, they are accruing astonishing revenues and profits. Within a few months, they are earning a substantial proportion of global GDP.

2.2 Governance

Around the world, politicians and voters have been slow to acknowledge the coming change, but in the last few months it has become undeniable, and emergency plans have been made. The planning process was rushed, and the plans are incomplete and inconsistent. Some countries decided to ban the use of HALs, but as widely predicted, their economies quickly become uncompetitive, and their populations can foresee a future of increasing poverty and discontent. Almost all of these countries relent within a year.

Other countries argue for the nationalisation of “Big AI”, the companies deploying HALs. In practice, of course, this is only an option for the countries where those companies are domiciled, so it is only exercised twice. Elsewhere, the companies remain privately owned, but their governments quickly subject them to the most intrusive forms of regulation ever seen in capitalist economies. It turns out that the governance and behaviour of Big AI firms varies little whether they are nationalised or not.

The owner of one of the Big AI companies resists this intrusion, launching legal action and a massive political campaign. He is distressed to discover that he has virtually no supporters apart from the people he is paying. The power of oligarchs, in fragile democracies like the USA as well as kleptocracies like Russia, had been a source of great concern for many years. When push came to shove, presidents proved more powerful than plutocrats.

The days immediately after the launch of the first HALs are surreal. In cities and towns around the world, newly unemployed people spend their time in cafes and bars, and on the phone, talking to friends and strangers about what will happen next. There are occasional riots spawned by fear and frustration, but mostly the world is anxiously quiet, waiting to see how matters will unfold.

3. Meaning and Distribution

3.1 The nature of the challenge

It has been obvious to almost everyone since early 2032 that the Economic Singularity was coming, and both the mainstream media and social media have been consumed by discussion of how it should be handled.

Something of a panic had set in when people realised that almost everybody was going to become unemployed, and for a while the plaintive cry was “what will we do for meaning in a world with no jobs?” But as the event drew nearer, a consensus emerged that this was actually a lesser problem, and the much more pressing question was, “what will we all do for money?” The challenge, people realised, was how to distribute resources efficiently and equitably enough to enable everybody to live decent lives, and without chaos erupting. This was obviously a job for governments, but could politicians really be trusted to get it right?

The idea of a universal income dates back at least as far as Thomas More’s “Utopia”, published in 1516, and campaigns for a Universal Basic Income began in the mid-1980s. Numerous experiments proved that giving people money generally does them more good than harm, but economists pointed out that it was unaffordable in a world where goods and services have significant costs. On this occasion the economists were right. As John Kay wrote, “Either the level of basic income is unacceptably low, or the cost of providing it is unacceptably high.”

3.2 Lessons from Covid

Kay’s claim was supported by the experience of the Covid pandemic in 2020-23. The disease forced governments everywhere to close down swathes of their economies, requiring most of the population to remain at home to reduce transmission. This in turn obliged governments to channel money to the people no longer working, to prevent them from starving. This was the largest experiment in something like UBI that was ever undertaken, and it was successful. The 15 million or so excess deaths due to Covid were caused by the disease, not by starvation.

It was also ruinously expensive. Governments spent trillions of dollars on furlough schemes and direct payments, and national debts rose sharply. It was necessary, it worked, and it could not have been sustained for much longer. It took years and considerable pain to unwind those debts.

The other lesson from Covid was that sclerotic and incompetent as they often are, governments can turn on a dime when they really have to. It turns out that sometimes, humanity can perform well in a crisis. Not perfectly, but well enough.

3.3 Abundance

It is 2034. HALs are extraordinarily productive. They work around the clock, they require no salaries, and they continuously improve themselves. They are unhindered by ego or cognitive bias, they exchange vast amounts of information with each other flawlessly and instantaneously, and they forget nothing. They quickly generate immense wealth.

In all the countries where they are based, there is overwhelming agreement that Big AI must share their enormous rewards with the general population. Sometimes this is framed as an argument about restorative justice, based on the claim that the HALs destroyed human jobs, and were trained on data which many argue was stolen. More often it is framed as distributive justice, based on fairness, and pragmatic considerations.

Among those drawn to conspiracy theories, and those who believe that wealth inequality is immoral, rumours spread about the Machiavellian intentions of some leaders of the Big AI firms. These rumours are vigorously denied, and fade with the absence of evidence.

The owners of two of the Big AI firms decide to transfer the ownership of their companies to their governments. They become folk heroes as a result. One of them tried to make the transfer to the United Nations rather than their national government. This proved impractical in the short term, but there are discussions about revisiting the idea in the future.

An economy of abundance is developing fast, in which the cost of all the goods and services required for a good standard of living are falling close to zero. An element of cost remains, as raw materials, land, and energy are not free, although thanks to the accelerating shift towards renewable energy sources, it is envisaged that energy will soon be too cheap to meter. There is general agreement that a zero-cost economy is not a desirable outcome, however, as prices are the best way to regulate demand and avoid damaging profligacy.

3.4 Distribution

Even given the full compliance of the leaders of Big AI, the need to distribute resources from them to the rest of the population has raised enormous challenges. Around the world, different solutions have been tried. Some have succeeded and some have failed – partly through honest and understandable mistakes, partly through stupidity, and partly through greed and corruption. There have been severe economic and political dislocations as a result.

The major challenges have included:

How to allocate resources between the individuals within countries. One of the drawbacks of the idea of Universal Basic Income was the explicit assumption that everybody needs the same resources, which is patently untrue. A healthy adult single male has different requirements to a sick mother or a severely disabled teenager. Our needs vary over time and between geographies and communities. Preserving justice and the appearance of justice is a knotty problem.

How to determine who makes and executes these decisions. How much to rely on AI.

How to allocate resources between countries. This inevitably requires transfers from countries which host Big AI firms to those which do not. In the years running up to the Economic Singularity the concept of AI sovereignty became fashionable, and many countries became determined to develop their own Big AI firms, either individually, or collectively, in groups like the EU and the GCC. For most countries, however, this was simply not an option.

How to overcome political opposition to radical distribution. Even though the alternative was almost certainly turmoil and massive death tolls, there was fierce resistance to central planning in some circles. These people argued strongly to retain some of the virtues of the market, and to institute Fully Automated Luxury Capitalism rather than Fully Automated Luxury Communism.

How to avoid corruption and graft. Not just at the outset, but forever.

3.5 Meaning

For some people, their job defines their lives. It not only gives them structure, including a reason to get out of bed in the morning, but it gives them a sense of purpose, and a way to define themselves. When asked at a party what they do, they feel better about themselves if they can honestly reply “I’m a doctor” (or journalist, or carpenter) rather than “I’m unemployed”.

These people are a minority. Gallup runs polls every few years asking how engaged people are in their jobs, and across time and geography, the result is consistent. Only around 15% of people are truly engaged by their jobs. The rest do their jobs to put food on the table and keep a roof over their heads. Being released from wage slavery would be a release for most people.

For most people, meaning does not come from jobs, but from family and friends, perhaps religions, or hobbies and sports. Most people do not suddenly lose their meaning when they go on holiday, nor when they retire, if they have the means to do comfortably. Do we really want to live in a world where the meaning in your life comes from being an Amazon delivery driver, or an actuarial assistant?

None of this means that freedom is easy. With much more time on our hands, many of us will be tempted to over-indulge in vices like drink and drugs, and other forms mindless recreation. A whole new range of support services will probably be needed to help us all live rich, fulfilling lives. But we will be starting from a much better place.

3.6 Transition

Even if we can design a viable economic system for a post-labour world – one that provides every human being with a high standard of living – we will still face the enormous challenge of getting from here to there. The transition from our current economy to a radically different one will be fraught with disruption, resistance, and unintended consequences.

A viable plan must address not only the destination but the journey. It must account for the messy, uneven, politically constrained reality of moving from an economy built on human labour to one that is not.

4. Questions

Assuming that full automation is coming, here are some of the questions and challenges it raises.

1. When will it arrive?

2. Can we establish and monitor any metric that will improve our ability to forecast the date of the Economic Singularity?

3. Is the scenario of rapid transition described in sections 2 and 3 the most likely one? If not, why, and what alternative scenario is more likely?

4. Who should determine which values the resource distribution systems optimise for, and who should build and operate the systems?

5. How much of a role should AIs play in designing the new structures and processes that will be needed?

6. Will radical economic transfers prove politically unacceptable in some places?

7. Are international transfers politically feasible? What consequences would follow?

8. If full automation entails more centralised planning, as seems inevitable, how can we avoid corruption and graft arriving with it?

9. How should financial systems adapt to the new circumstances?

10. What happens to democracy if most citizens are economically dependent on transfers from a small number of firms? Can voters retain their power when the productive economy doesn’t need them?

11. How should education systems change before and during the transition, when many of the skills being taught may become worthless before students graduate?

12. How can we draw the attention of policy makers, economists, and the general public to this problem?

The stakes in the Economic Singularity could not be higher. If we get it right, we could usher in an era of unprecedented prosperity and human flourishing. If we get it wrong – or if we fail to plan at all – the consequences will be catastrophic. The time to begin this conversation is now.

i. https://static1.squarespace.com/static/635693acf15a3e2a14a56a4a/t/69cbb9d509ada447b6d9013f/1774959061185/forecasting-the-economic-effects-of-ai.pdf