The demographic cliff: accelerating automation

Around the world, societies are heading towards a demographic cliff. In 1950, the global total fertility rate stood at roughly five children per woman. Today it is about 2.3 – barely above the 2.1 replacement threshold – and still falling. More than two-thirds of humanity now live in countries where fertility has dropped below replacement level. In South Korea, the total fertility rate has collapsed to 0.8, the lowest recorded on Earth. China’s is around 1.0. Across much of Europe, fertility now ranges from 1.0 to 1.6, with the EU average at a record low of 1.34 in 2024. Even the United States, long a demographic outlier among rich countries, has slipped to 1.6.

The consequences are already visible. China’s population fell for a fourth consecutive year in 2025, as deaths outpaced births by a widening margin. By 2050, China is projected to lose roughly 145 million people from its current 1.4 billion. Japan’s working-age population peaked in 1995 at 87 million and has since fallen by 16%; it is projected to shrink by a further 31% by 2060. Over the coming quarter century, 38 nations of more than one million people each will probably experience population declines, up from 21 in the past 25 years. By 2050, the share of people aged 65 and over in countries experiencing population declines is expected to nearly double, from roughly 17% to 31%.

This implies fewer workers to fill jobs, smaller tax bases to fund pensions and healthcare, and mounting pressure on economic growth. The question is whether technology can help. Specifically, can automation and artificial intelligence compensate for the workers that demography will fail to deliver? The historical record suggests they can and, in some cases, that a scarcity of workers is precisely what triggers the adoption of labour-saving machines in the first place.

When workers disappear, machines arrive

The idea that labour shortages spur mechanisation is not new. One striking example comes from the American South. In 1927, the Great Mississippi Flood displaced hundreds of thousands of agricultural workers from the Mississippi Delta. Many never returned. As the economic historians Richard Hornbeck and Suresh Naidu have shown, the counties that lost the most workers did not simply accept lower output. Instead, landowners adopted tractors and other mechanical equipment at sharply higher rates than counties where the labour force stayed intact. A sudden scarcity of hands made machines worthwhile.

A similar story played out in France after the First World War. The conflict killed or maimed a staggering share of the male working-age population. When the survivors came home, French agriculture and industry faced an acute shortage of labour. The response was a wave of mechanisation and labour-saving innovation that might otherwise have taken decades. Farms that had relied on manual labour and animal power invested in machinery, and employers across the economy turned to capital-intensive methods to compensate for the missing workers. War, for all its destruction, inadvertently accelerated technological adoption by removing the cheap labour that had made older methods viable.

These episodes share a common logic: when labour is abundant and cheap, employers have little reason to invest in expensive machines. When labour becomes scarce or costly, automation suddenly looks like a bargain. The economic incentives shift, and technology fills the gap.

Demography and robots in modern economies

The same logic applies to the demographic transitions unfolding today. In a series of influential papers, the economists Daron Acemoglu and Pascual Restrepo have demonstrated a robust link between population ageing and the adoption of industrial robots. Countries and regions that are ageing faster tend to install more robots per worker. The pattern holds across a range of settings, from the rapidly greying prefectures of Japan to the older industrial heartlands of Europe.

The mechanism is straightforward. As the working-age population shrinks relative to the overall population, wages for carrying out routine tasks tend to rise and firms struggle to fill positions. Robots become an increasingly attractive substitute. Acemoglu and Restrepo find that demographic pressure is one of the strongest predictors of robot adoption – stronger, in many specifications, than the differences in industrial composition or trade exposure. Japan is a case in point: its working-age population has been declining for three decades, it faces a projected shortfall of over 11 million workers by 2040, and it is one of the most robot-dense economies on the planet.

Yet there is an important caveat. Until very recently, the robots in question have been overwhelmingly confined to manufacturing. Industrial robots – the kind that weld, paint, assemble, and palletise – are superb at structured, repetitive tasks in controlled environments. A car factory or a semiconductor plant is the ideal habitat: fixed layouts, predictable inputs, minimal variation. These robots have transformed manufacturing productivity, but manufacturing accounts for a shrinking share of employment and output in most advanced economies. In the United States, it is roughly 8% of employment; in the UK, around 7%, and less than half of those workers toil on the factory floor. The vast bulk of the economy – and of the demographic challenge – lies in the service sector.

The constraint that is breaking

This is the constraint that is now beginning to break. For decades, the main limitation on robotic automation was not hardware per se, but perception and cognition. A factory robot can repeat a precise motion thousands of times, but it cannot recognise an unfamiliar object, navigate a cluttered room, or understand a spoken request. These are the capabilities required to automate services – cleaning an office, stacking supermarket shelves, assisting a patient, delivering a package – and they have been beyond the reach of traditional robotics.

Two developments are changing this. The first is the rapid advance of AI, and in particular of large language and vision models, which give machines something closer to general-purpose perception and reasoning. A robot guided by a modern AI system can identify objects it has never seen before, interpret ambiguous instructions, and adapt its behaviour to novel situations. The second is the emergence of humanoid and general-purpose robots designed specifically for unstructured environments: hospitals, shops, and eventually cities and homes. Companies including Tesla, Figure and several Chinese manufacturers are racing to produce humanoid robots that can walk, grasp, and manipulate objects in the kind of messy, variable settings that define the service economy.

Together, AI and next-generation robotics promise to extend automation into the sectors that have long resisted it. If a robot can clean a hotel room, sort packages in a logistics hub, assist a nurse with patient transfers, or restock shelves overnight, then the productivity gains from automation are no longer confined to the factory floor. They reach into healthcare, hospitality, retail, logistics, and care work – precisely the sectors where demographic pressure is felt most acutely, and where productivity growth has been slowest.

The potential ahead

Consider the numbers. The International Federation of Robotics reports that the global stock of industrial robots stands at roughly 4.7 million units – impressive, but serving a narrow slice of the economy. Goldman Sachs has estimated that humanoid robot shipments could reach 1.4 million units per year by 2035, precisely because the service economy is so vast and so labour-intensive. In ageing societies, the arithmetic is particularly compelling: fewer young people entering the workforce, more elderly people requiring care, and a widening gap that human labour alone cannot close. Japan, where the average farmer is 70 years old and the construction sector has 5.6 job openings for every applicant, illustrates how acute these shortages have already become.

None of this will happen overnight. Humanoid robots remain expensive, and their capabilities, while improving fast, are still limited compared with a human worker’s adaptability. Regulatory frameworks for robots operating alongside people in public spaces are only beginning to take shape. And the social and political implications of widespread service automation will be significant, requiring careful management.

But the direction of travel is clear. Demography is creating substantial demand. AI is supplying the capability. And the historical pattern – in which labour scarcity drives technological adoption – suggests that the societies ageing fastest may also be the ones that automate fastest. The demographic cliff will likely prove to be one of the most powerful spurs to the next wave of automation.