Source: X / @TheHumanoidHub
AI isn’t just another industrial revolution
For two centuries, technological change has followed a reassuring pattern. Machines displaced some tasks, but new industries and occupations appeared quickly enough to absorb displaced workers. Steam power replaced muscle; electricity streamlined factories; computers automated routine calculations. Each wave raised productivity, lifted living standards and ultimately created more work, not less.
Artificial intelligence may prove a different beast. Modern systems can already summarise legal documents, diagnose illnesses, write software and increasingly orchestrate physical activity. The International Monetary Fund estimates that in advanced economies around 60% of jobs are exposed to AI, compared with about 40% in emerging markets and 26% in low-income countries. That is a much broader impact on cognitive work than previous waves of automation (IMF Staff Discussion Note).
Evidence from labour markets is beginning to bear this out. In Britain, the Chartered Institute of Personnel and Development reports that one in six (17%) employers expects AI to shrink their workforce over the next year, with junior roles most at risk. In large private-sector firms, one in four (26%) expects headcount to fall because of AI (CIPD press release).
Corporate restructuring tells a similar story. Amazon recently announced it would eliminate around 14,000 corporate jobs—about 4% of its corporate workforce—while at the same time accelerating investment in AI and cloud infrastructure. Company leaders have framed this as part of a pivot towards a more AI-centric model of growth, not just a cyclical cost-cutting exercise (Associated Press coverage).
Unlike earlier technologies, AI does not merely mechanise routine labour; it automates cognition itself. Because the marginal cost of software labour is close to zero, substitution could spread much more rapidly than new jobs appear. The comforting narrative that “technology always creates more work eventually” looks less secure than it once did.
The decoupling of work and income
The most plausible future is not one of mass unemployment, but one of weaker bargaining power for labour. Firms that can scale production without proportionately expanding payrolls will see profits rise while wages stagnate. That dynamic—rising output, flat incomes—has already appeared in several advanced economies and could intensify as AI matures.
If a smaller share of national income flows to workers and a growing share to owners of chips, cloud platforms and data centres—the industrial base of the machine age—the traditional route to prosperity becomes unstable. For generations, work has been the primary mechanism for distributing income and conferring social status. But as AI takes over more economically valuable tasks, that link begins to fray: the economy may hum with abundance while household finances feel increasingly constrained.
Economic growth alone will not solve this. Even if AI delivers extraordinary productivity gains, societies will struggle if those gains accrue narrowly. When broad prosperity decouples from broad labour participation, the political consensus underpinning market economies weakens. Citizens tolerate rapid technological change when they believe they will share in the dividends. Without that confidence, resentment grows.
The real danger is therefore less a jobs apocalypse than a legitimacy crisis: an economy that visibly needs fewer workers cannot indefinitely base social status and basic security on paid employment alone.
A dividend society?
Governments will soon face an uncomfortable design problem: how to distribute income in an economy that may not need full employment to generate abundance. One answer gaining traction is the idea of a dividend society—a system in which citizens receive a modest, predictable share of the gains generated by an automated economy.
This need not mirror the more utopian visions of universal basic income. A dividend could be modest and phased in gradually. It might be combined with active labour-market policies, retraining programmes and incentives for education, care and community work. The aim is not to replace work as a source of meaning and status, but to ensure that income security does not collapse simply because fewer workers are needed.
Where such a dividend comes from will depend on how countries govern AI infrastructure. Compute clusters, specialised chips and hyperscale data centres are already chokepoints where economic rents accumulate. If these become toll roads, owned by a handful of firms, the distribution of gains will skew sharply. If they are regulated more like common carriers—open, with reasonable returns and some form of rent capture—part of the surplus can be recycled into public benefit.
Timing matters. Once the architecture of the machine economy hardens—once control over chips, models and energy grids concentrates—reshaping the system will become much harder. Waiting until wages are visibly decoupled from output is likely to mean acting only after political damage is done.
Artificial intelligence may ultimately deliver remarkable abundance. But without institutions that share the gains beyond those who own the algorithms and hardware, the machine age will widen inequality and strain democratic stability. A carefully designed dividend society is not an act of generosity. It is a pragmatic attempt to ensure that AI strengthens, rather than fractures, the social contract.