AI and the Looming Permanent Underclass: How Automation Could Deepen Inequality

One of the darker possibilities accompanying the artificial intelligence revolution is that advanced economies may be approaching something historically jarring: a permanent underclass of workers who, rather than simply transitioning between jobs, become structurally excluded from stable, well-paid employment. This shift goes beyond the familiar fear of robots taking over factory floors; it targets the cognitive and administrative tasks that have long anchored the middle class.

A Structural Threat to Work

The core argument is not merely about job displacement in the short term. AI’s capability to learn, analyse, and generate content at superhuman speed threatens to reshape labour demand so profoundly that some workers could face persistently lower wages, weaker bargaining power, and dwindling pathways back into the workforce. Unlike past technology waves that automated physical labour, today’s algorithms are encroaching on routine cognitive work—data entry, scheduling, basic legal and financial analysis, customer service, and even aspects of software development.

The OECD’s ongoing analysis of employment trends highlights that the disruption is not uniformly distributed. Where earlier automation polarised the labour market into high-skill and low-skill jobs, AI is now gnawing at the middle, potentially hollowing out the very roles that once offered a ladder into the middle class.

The Hollowing Out of Middle-Skill Work

The long-run “hollowing out” effect is especially pronounced for roles that combine routine cognitive tasks with administrative coordination. Clerical workers, paralegals, bookkeepers, and call-centre agents are among those whose daily workflows overlap significantly with what large language models can now perform. Even some professional domains, such as junior-level medical diagnosis or architectural drafting, face partial automation. The result could be a labour market resembling an hourglass: a concentration of high-end technical and creative jobs at the top, a mass of insecure, low-paid service jobs at the bottom, and a narrowing tube in between.

This pattern is not purely hypothetical. Studies cited by the International Labour Organization suggest that advanced economies with deep digital infrastructure, large service sectors, and knowledge-intensive industries are especially exposed. When businesses can deploy AI rapidly across office networks, the speed of job reconfiguration can outpace the ability of workers to retrain or relocate.

Why Advanced Economies Are on the Front Line

Advanced economies are uniquely vulnerable precisely because of their strengths. High labour costs make automation financially attractive. Widespread broadband and cloud platforms mean that an AI tool can be rolled out across a multinational firm almost overnight. Furthermore, the knowledge-work content of these economies—financial services, legal advice, marketing, software—overlaps heavily with the capabilities of generative AI. This sets up a situation where the very drivers of past prosperity become accelerants for workforce disruption.

Who Wins from AI-Driven Productivity?

A critical question is who captures the productivity gains. Evidence from earlier technological leaps suggests that the spoils often flow upward: to shareholders, to the firms deploying the AI, and to the highly skilled workers who design, fine-tune, and interpret the systems. The World Economic Forum’s Future of Jobs Report projects a net increase in demand for AI specialists, data analysts, and engineers, while roles in administration, customer service, and repetitive analysis are expected to decline sharply. Unless deliberately managed, this concentration of gains risks widening wealth disparities and shrinking the tax base that funds social safety nets.

Policy Questions in Urgent Need of Answers

The prospect of a permanent underclass forces a rethink of labour-market institutions built for a different era. Key policy challenges include:

• Retraining effectiveness: Can mid-career workers realistically acquire new skills faster than AI evolves, and who pays?
• Wage insurance and unemployment support: Should governments step in to top up incomes when displaced workers are forced into lower-paying jobs?
• Education redesign: From early schooling to university, curricula may need to prioritise uniquely human skills—critical thinking, emotional intelligence, creativity—over content knowledge that AI can replicate.
• Labour protections: Can existing employment law cope with a gig-like reorganisation of cognitive work, where tasks are broken into micro-contracts and mediated by platforms?

Is a Permanent Underclass Inevitable?

Whether “permanent underclass” is a realistic outcome or a warning label remains debatable. Historical parallels with earlier waves of automation—the mechanisation of agriculture, the decline of manufacturing—show that economies eventually generate new types of employment, often in entirely new sectors. However, those transitions were not painless; they required decades of adjustment, large-scale migration, and the creation of a social safety net that only emerged after widespread hardship. The worry today is that the pace of change could outstrip the capacity of institutions to adapt, turning what used to be temporary dislocation into a lasting fault line. Ultimately, the future of work under AI will be shaped less by the technology’s raw capability and more by the choices societies make about distribution, training, and the social contract.