Artificial intelligence: optional work or global inequality?

Monday 9 February 2026

The promises attached to artificial intelligence have begun to sound like the closing chapter of industrial history. Work will become optional. Incomes will be universal. Human beings will be liberated for art, care, discovery and play. The old injustices of class and geography will melt away in a warm bath of abundance.

It is an alluring story, and it is not wholly fantasy. Artificial intelligence is a general-purpose capability — like electricity or the internal combustion engine — that can be applied across the economy, raising productivity in places where inefficiency once looked permanent. But the leap from “higher productivity” to “greater equality” is not automatic. In fact the default trajectory of artificial intelligence, without deliberate political choices, looks more like an accelerant poured onto existing inequalities than a solvent that dissolves them.

To judge the entrepreneurs’ claims fairly, it helps to separate three questions that are often fused together.

First — will artificial intelligence produce enough wealth that society could afford “optional work” and a universal income?

Second — will that wealth be distributed broadly, including across poorer countries, or captured narrowly by owners of capital, data and intellectual property?

Third — if a broad distribution is possible in theory, is it plausible in practice given how power, institutions and international markets actually function?

On the first question, the optimistic camp has the strongest case. Artificial intelligence can reduce the cost of many forms of knowledge work — writing, translation, software coding, design iteration, legal drafting, accounting triage and customer support. That does not mean it will abolish these professions, but it can lower the labour required per unit of output, freeing resources for other uses. This is how past productivity waves worked. A society that could afford tractors did not have to keep most people on farms.

However there is a crucial caveat — productivity gains are not the same thing as a social dividend. The fact that a technology can, in principle, generate large surpluses does not tell us who receives them. The last four decades have already offered a warning. Many advanced economies experienced productivity growth alongside stagnant median wages, with the gap absorbed by higher profits, rising asset values and sharply unequal wealth accumulation. The mere presence of a powerful technology does not compel equal outcomes. Distribution is a political economy question, not a computing question.

On the second question — distribution — the empirical signs point towards divergence unless counteracted.

Start with labour markets. The International Labour Organization’s recent work on generative artificial intelligence (systems that produce text, images or code from prompts) finds that exposure is uneven across countries and between men and women. Globally, roughly one in four workers is in an occupation with some exposure to generative artificial intelligence, but the share is far higher in high-income countries than in low-income countries — and higher for women than for men. The ILO estimates around 34 per cent exposure in high-income countries compared with about 11 per cent in low-income countries. This matters because “exposure” is a proxy for where rapid transformation will occur first — where tasks can be reorganised, where wages can be pressured, and where new productivity gains can be captured.

Exposure alone does not prove harm. Some roles will be enhanced rather than replaced. Yet even enhancement can widen inequality if the gains flow to a limited slice of workers and owners. The OECD’s research on artificial intelligence and wage inequality underlines the ambiguity — artificial intelligence can compress pay within occupations by raising the output of weaker performers, but it can also produce selection effects and bargaining shifts that disadvantage those who cannot adapt. In other words artificial intelligence may reduce inequality in one narrow place whilst increasing it in the wider system — between occupations, between regions and between those with and without the complementary skills, credentials and networks that turn an artificial intelligence tool into income.

Now look at geography inside countries. The OECD has warned that generative artificial intelligence is likely to hit local job markets unevenly, exacerbating existing urban–rural divides and productivity gaps. This is not mysterious. Artificial intelligence benefits cluster where there is strong broadband, modern firms, venture finance, research universities and dense labour markets — the same places already advantaged by globalisation. A factory town struggling with weak investment does not automatically become a beneficiary because an algorithm exists somewhere else.

Then consider the global level — the divide between rich and poor countries. The infrastructure of frontier artificial intelligence is heavily concentrated: advanced semiconductors, specialised data centres, large-scale cloud services, and the scientific and managerial talent that can run them effectively. Even where software is accessible, the ability to deploy it at scale depends on electricity reliability, connectivity, capital, and institutions that can absorb new methods.

The World Trade Organization has explicitly warned that artificial intelligence could widen global inequality if developing countries do not gain access to its benefits and the infrastructure required to use it.  This is a familiar pattern. The gains from previous technology waves often accrued disproportionately to those who already had the ability to invest and organise production, whilst others became consumers of imported value rather than producers of it.

Against that background, the entrepreneurial promise of “global incomes” runs into a practical problem — there is no global treasury. There are national tax systems, national welfare states and fragile international mechanisms that are already struggling with debt, ageing populations and geopolitical rivalry. A universal income on a global scale would require not just wealth but sustained international political solidarity and enforcement capacity — the very things that have been fraying.

That leads to the third question — plausibility.

The optimistic narrative usually assumes one of two routes to equality.

The first route is a market utopia — artificial intelligence makes everything so cheap that poverty becomes impossible. If food, energy, housing, education and healthcare approach near-zero marginal cost, then even unequal money incomes might not matter.

The second route is a social dividend — artificial intelligence produces vast profits, and governments tax those profits to fund universal incomes and public services, making work increasingly optional.

Both routes face hard constraints.

The “near-zero cost” vision bumps into the reality that the most important goods are not infinitely scalable software. Housing sits on land — and land is scarce in places where people want to live. Health care depends on human attention, trust and complex systems. Education at its best is relational. Energy requires physical infrastructure and geopolitical supply chains. Even if artificial intelligence reduces some costs, it does not abolish the scarcity that underpins many of the most unequalising markets.

The “social dividend” route collides with power. To tax a concentrated, mobile and politically influential sector at the level required to fund large universal incomes is not technically difficult. It is politically difficult. If the owners of artificial intelligence capital can shift profits across borders, lobby regulators, and shape public narratives, then the gap between what is economically possible and what is politically enacted can be vast.

There is also a deeper structural reason why artificial intelligence tends towards inequality — it exhibits “scale effects”. The best models often benefit from more data, more computing power, more specialised labour and more distribution channels. That can produce winner-takes-most dynamics. A single firm’s system can serve hundreds of millions of people at low per-user cost, crushing competitors and concentrating rents. Even when open-source alternatives exist, the frontier often remains expensive and centralised.

This is why the debate about “optional work” needs more realism. Work is not only a way to earn money — it is a way societies allocate status, structure time and maintain social contribution. Removing the necessity of work without replacing its social role can produce alienation and resentment. It can also produce a bifurcated world — an elite designing systems, owning systems and governing systems, and a large population offered transfers whilst being excluded from influence. That is not equality. It is a different kind of hierarchy — softer in material terms perhaps, but potentially sharper in dignity and power.

None of this means the pessimists must be right. It means the equality-promoting outcomes are conditional — and the conditions are demanding.

A more credible path to artificial intelligence reducing inequality would require at least five pillars.

First, competition policy with teeth — to prevent the extraction of monopoly rents from essential artificial intelligence services, and to keep markets open to smaller firms, including outside the richest countries.

Second, labour institutions that translate productivity into pay — collective bargaining where appropriate, minimum standards, portable benefits, and active labour market policies that help workers move into tasks that remain stubbornly human: care, repair, supervision, teaching, safety, conflict resolution and skilled trades.

Third, public investment in the complements — broadband, electricity resilience, education, and the capacity of public services to procure and deploy artificial intelligence safely. Without these, poorer regions and poorer countries are not empowered by artificial intelligence — they are merely exposed to it.

Fourth, taxation that tracks value creation — not only corporate tax, which is easy to avoid, but taxes on economic rents, on extreme wealth, and on the extraction of value from data and platforms. Without a fiscal mechanism, “social dividends” remain marketing copy.

Fifth, international support that is not paternalistic charity but development in the older, serious sense — technology transfer, financing for infrastructure, and rules that allow poorer countries to build productive capacity rather than being locked into dependency. The alternative is a world where artificial intelligence increases global output whilst the distribution of that output becomes more skewed.

In this context, the most honest answer to the entrepreneurs’ claims is that they are not impossible, but they are not self-fulfilling. They require politics, institutions and, above all, a willingness by those who benefit first to share power rather than merely surplus.

Recent warnings from senior international figures underline that the stakes are not abstract. At Davos in January 2026, the Managing Director of the IMF described artificial intelligence as a “tsunami” likely to affect a large share of jobs, with younger workers particularly exposed and with risks that productivity gains will not translate into broad wage gains.  Whether one agrees with the metaphor or not, it captures the speed of change — and the danger that societies will react after the distributional settlement has already been written.

So will artificial intelligence entrench global inequality? If left to the default settings of contemporary capitalism — concentrated ownership, weak labour bargaining, fragmented regulation and international rivalry — that is the more likely outcome. Artificial intelligence will raise the returns to capital and scarce skills faster than it raises the earnings of those with little bargaining power. It will reward the places with infrastructure and institutions, and bypass those without them. It will create dazzling new forms of wealth and then ask the rest of society to applaud the spectacle.

But the same technology can, under different rules, become a tool of equalisation — not because it magically dissolves scarcity, but because it can be harnessed to expand capability: better health triage, cheaper access to knowledge, faster translation across languages, more efficient public administration, and productivity gains in smaller firms that have long been excluded from advanced methods. The ILO’s findings on uneven exposure across income groups should be read as a call to action — not a prophecy. 

The decisive question then is not whether artificial intelligence can produce abundance. It can. The question is whether societies, and the international community, will choose to convert that abundance into shared security, shared dignity and shared opportunity — or whether they will allow a new machinery of wealth to harden into an old pattern of inequality, only faster, and with a more persuasive sales pitch.