The Permanent Underclass: Artificial Intelligence and the Future of Work

By Matthew Parish

Sunday 23 August 2026

One of the darker possibilities accompanying the artificial intelligence revolution is that advanced economies may be approaching something historically unusual: the creation of a substantial class of people who are not temporarily unemployed, inadequately educated or victims of an economic downturn, but who are simply no longer required by the productive economy. They may remain citizens, consumers and voters, yet the economic system may have progressively less need of their labour. If that possibility materialises, then the consequences will reach considerably further than unemployment statistics. They will concern the distribution of property, political power, social dignity and ultimately the stability of democratic societies themselves.

Technological unemployment is hardly a new anxiety. Every major economic transformation has destroyed occupations. The mechanisation of agriculture eliminated enormous quantities of agricultural labour. Industrial machinery displaced craftsmen. The motor car destroyed trades associated with horses while creating entirely new industries surrounding roads, petroleum and automobile manufacture. Computers eliminated armies of clerks and typists, yet simultaneously produced occupations in software, telecommunications and information technology that scarcely existed a generation earlier.

For this reason economists have traditionally treated predictions of permanent technological unemployment with scepticism. Their scepticism has considerable historical justification. Human wants appear effectively unlimited. When productivity makes existing goods cheaper, people spend their additional wealth upon other things, thereby creating new industries and new employment. The nineteenth-century agricultural labourer could hardly have imagined that his descendants might earn their living designing websites, advising companies upon intellectual property or producing television programmes. Economic history repeatedly demonstrates humanity’s extraordinary capacity to invent new things for other human beings to do.

Artificial intelligence nevertheless presents a potentially different problem.

Previous waves of automation primarily substituted machinery for physical labour or routine intellectual processes. The conventional answer to displacement was therefore education. A factory worker whose job disappeared might acquire qualifications and become an administrator, technician or professional. The implicit bargain underlying post-war Western society was that education permitted people to climb away from those occupations most vulnerable to automation.

Generative artificial intelligence attacks precisely this refuge. Increasingly sophisticated systems can write, translate, summarise, conduct research, prepare computer code, analyse documents, produce images, undertake elementary accounting and perform significant portions of professional work. The people potentially exposed therefore include not merely factory workers and supermarket cashiers but lawyers, accountants, programmers, graphic designers, journalists, administrators and financial analysts. The machines are beginning to compete not principally with human muscles but with human cognition.

This does not mean that these professions will disappear. Technology usually changes occupations before eliminating them. A lawyer equipped with artificial intelligence may become dramatically more productive rather than redundant. The same may be true of doctors, engineers, academics and accountants. Yet therein lies the problem. If ten lawyers assisted by artificial intelligence can perform work previously requiring thirty, the legal profession need not disappear for twenty livelihoods to become unnecessary.

The effects may first become apparent at the bottom of professional hierarchies. Many professions traditionally operate through apprenticeship in everything but name. Junior employees undertake comparatively routine work while learning the judgement required for more sophisticated responsibilities. A young lawyer reviews documents. A junior programmer fixes elementary problems. A trainee accountant reconciles figures. A beginning journalist prepares straightforward reports. These tasks may be precisely those most readily undertaken by artificial intelligence.

Removing them creates an awkward paradox. A society may continue to require experienced professionals while gradually abolishing the occupations through which people become experienced professionals. The economic ladder is not necessarily kicked away at its summit. Its lowest rungs disappear.

The result could be a peculiar labour market in which established experts using artificial intelligence enjoy extraordinary productivity while younger people encounter increasingly formidable barriers to entering their professions. Companies would have compelling incentives to behave this way. An artificial intelligence system does not demand a salary increase, take maternity leave, become ill, resign for a competitor or require years of supervision before becoming competent. Even where its output requires checking, one experienced employee supervising machines may prove cheaper than several junior employees learning their trade.

This transformation would have profound consequences for inequality because artificial intelligence is ultimately capital. Someone owns the computers, models, intellectual property, electricity infrastructure and companies through which artificial intelligence operates. If the relative importance of labour declines while that of capital increases, then an increasing proportion of economic returns will naturally flow towards the owners of capital.

There is nothing particularly mysterious about this. If a factory previously requiring one thousand employees can produce twice as much with one hundred employees, enormous additional value has been created. But the distribution of that value is not predetermined. It might appear as higher wages, lower prices, greater corporate profits, taxation revenues or some combination of all four. Technology determines what can be produced. Political economy determines who receives the proceeds.

This distinction matters because the bleakest predictions about a permanent underclass sometimes confuse technological possibility with social inevitability. Artificial intelligence does not itself possess a theory of property rights. Algorithms do not determine rates of taxation, inheritance law, educational policy or welfare systems. Human institutions determine those things. A highly automated society could theoretically distribute the fruits of extraordinary productivity quite broadly. Equally, it could allow ownership to become concentrated amongst a remarkably small group.

Universal basic income is frequently proposed as an answer. If machines generate immense wealth while employing relatively few people, the state might tax some portion of that wealth and distribute an income sufficient for everyone to live comfortably. Economically this is not necessarily absurd. A sufficiently productive automated economy might be able to provide high material living standards while requiring comparatively little human labour.

Yet this solution addresses consumption more readily than status.

Human beings do not work solely because they require money. Employment supplies structure, social relationships, achievement, hierarchy, identity and the sense — sometimes justified and sometimes not — that one’s existence matters to people beyond one’s immediate family. Asking somebody what he does remains one of the most common questions upon meeting him because occupation functions as shorthand for social identity.

A society in which millions of people receive adequate incomes but believe themselves economically unnecessary might therefore be considerably less harmonious than economic models suggest. Material comfort does not automatically produce dignity. Indeed a population permanently maintained through transfers from an extraordinarily wealthy technological elite might come to resent precisely the people upon whom its material security depends.

The political consequences could be severe. Liberal democracy has flourished most successfully alongside large middle classes whose members possess property, occupations and expectations of advancement. They have something to lose from political disorder and something to gain from institutional stability. A society divided between an exceptionally wealthy technological aristocracy and a much larger economically dependent population would possess a very different political character.

The wealthy might increasingly withdraw into private systems of education, healthcare, security and residential life. The economically marginalised might conclude that existing institutions offer them little prospect of advancement. Politics could consequently become an increasingly bitter contest over redistribution rather than a mechanism for enlarging shared prosperity. Populism of both left and right would find fertile territory in such circumstances.

There is also the geographical dimension. Technological wealth tends to cluster. A relatively small number of cities and regions attract capital, highly skilled workers and successful companies. Property prices rise accordingly, pushing ordinary workers towards distant suburbs or entirely different regions. A society can therefore become divided not merely economically but physically — prosperous technological enclaves surrounded by places increasingly detached from the principal sources of wealth creation.

Yet excessive pessimism remains unwarranted. Artificial intelligence may create categories of employment that are currently difficult to imagine. More importantly, it may make existing human abilities dramatically more valuable. A single entrepreneur may be able to establish a company with capabilities previously requiring hundreds of employees. Small businesses may acquire sophisticated legal, marketing and technical assistance at negligible cost. Individuals in poorer countries may gain access to education and expertise formerly available only in wealthy capitals. Artificial intelligence could therefore decentralise economic power as readily as it might concentrate it.

Much depends upon whether artificial intelligence develops principally as a substitute for human beings or as an amplifier of their abilities. The distinction will rarely be absolute. Calculators did not eliminate mathematicians. Spreadsheets did not eliminate accountants. Search engines did not eliminate researchers. Technologies that dramatically reduce the cost of particular intellectual activities often increase demand for the broader activities of which those tasks form part.

Education will nevertheless have to change. Teaching people primarily to perform tasks that machines can execute more quickly and accurately is unlikely to remain a sustainable model. Greater emphasis may need to fall upon judgement, interpersonal skills, entrepreneurship, creativity and the ability to supervise technological systems. There is a certain irony here. As machines become more intellectually capable, distinctly human qualities may acquire greater economic importance.

Governments will also face difficult choices about taxation and ownership. Taxing automation itself may discourage precisely the productivity improvements from which society could benefit. Yet allowing the returns from automation to accumulate indefinitely within a tiny ownership class could produce intolerable political tensions. Policymakers will need to distinguish between encouraging technological development and permitting the resulting wealth to become permanently hereditary.

Broader ownership may therefore prove more important than ever. Pension funds, employee shareholding, sovereign investment funds and other mechanisms through which ordinary citizens acquire interests in productive capital may provide part of the answer. If artificial intelligence causes capital to capture an increasing share of national income, then ensuring that citizens own capital becomes an obvious alternative to endlessly taxing one class to subsidise another.

The objective should not necessarily be to preserve every existing job. That would amount to preserving inefficiency for its own sake. Nobody seriously proposes restoring armies of telephone operators or agricultural labourers merely because technological progress eliminated their occupations. The purpose of economic development is partly to release people from work that machines can perform better.

The challenge is instead to preserve pathways towards independence, accomplishment and social participation. An economy in which people work fewer hours while enjoying greater prosperity might represent an extraordinary achievement. An economy in which millions of people are permanently excluded from meaningful economic participation while a small ownership class accumulates unprecedented wealth would represent something considerably less attractive. Both futures are compatible with extremely advanced artificial intelligence.

This is why the question of a permanent underclass ultimately concerns politics and institutions more than technology. Machines may transform the economic value of particular forms of human labour, but they cannot determine what society subsequently does about it. They cannot decide how property is distributed, how children are educated, how citizens acquire capital or what obligations accompany extraordinary wealth.

There is consequently no reason to assume either utopia or catastrophe. The technological revolution now unfolding may prove as economically transformative as industrialisation itself. If so, its benefits could eventually be immense. But transitions between economic orders are rarely painless, and societies that assume displaced people will somehow find their way into occupations that have not yet been invented may discover that historical analogies provide insufficient reassurance.

The gravest danger is therefore not mass unemployment tomorrow morning. It is a gradual narrowing of opportunity — fewer entry-level positions, weaker bargaining power, declining social mobility and increasing concentration of productive assets — proceeding slowly enough that each individual development appears tolerable until their cumulative effect becomes difficult to reverse.

A permanent underclass would not ultimately be something artificial intelligence created. It would be something human societies permitted to emerge around it. The decisive question of the artificial intelligence age may therefore prove not whether machines become capable of doing most of the things human beings presently do. It may be whether our political and economic institutions remain imaginative enough to ensure that, when they can, human beings continue to have both a stake and a purpose in the extraordinary prosperity those machines may create.