What this threat is
The history of automation is, in large part, a history of machines doing physical and repetitive work that humans used to do. The loom displaced hand weavers. The tractor reduced agricultural labor. The assembly line reorganized factory work. In each case, the disruption was real and painful for those directly affected, but the economy eventually produced new kinds of work to replace what was lost. The standard argument for why AI will follow the same pattern is that it, too, will displace certain tasks while creating new demand for human skills in areas machines can't handle. That argument isn't necessarily wrong, but it's built on an assumption that's now being challenged: that cognitive work, professional judgment, and creative output are safely in the human column.
What makes AI different from earlier automation waves is that it's moving up the skill ladder rather than across it. Previous waves tended to automate routine physical tasks, which pushed workers toward non-routine physical work and, especially, toward cognitive and interpersonal roles. AI doesn't respect that boundary. It can write, reason, research, diagnose, design, code, and analyze at a level that competes with entry-level to mid-level professional work in a wide range of fields. The categories of work most exposed right now include data processing, legal and financial analysis, content production, customer service, software development, and certain medical and diagnostic functions. These aren't low-skill jobs. Many of them represent the first rungs on professional career ladders that allowed previous generations to build economic stability.
It's important to distinguish between task displacement and job destruction, because the two have different implications. AI doesn't usually eliminate an entire job at once. It automates specific tasks within that job, which means the remaining work gets reorganized. A lawyer using AI to handle routine contract review doesn't stop being a lawyer, but their firm may need fewer lawyers to handle the same caseload. Over time, those efficiency gains can translate into fewer positions being filled, without any single dramatic layoff event. This makes the labor market effect slower and harder to see in aggregate data, but not necessarily less severe for the individuals affected.
The polarization effect is something labor economists have documented across earlier automation waves and expect to intensify with AI. When middle-skill jobs get automated, the labor market tends to bifurcate: demand grows at the high end for workers who can direct and augment AI systems, and at the low end for jobs requiring physical presence or genuine human interaction that AI still can't replicate well. Workers displaced from the middle tend to end up competing for lower-skill, lower-wage work rather than moving up into the high-demand roles, because the skills required for those roles are genuinely different and the retraining pathways are slow, expensive, and uncertain. That bifurcation tends to worsen income inequality even if total employment remains high.
Why it matters
Labor market disruption that happens gradually gets absorbed. Disruption that happens faster than institutions can respond creates crises. The question with AI isn't just whether it will displace work but whether it'll displace it faster than retraining programs, social safety nets, and new job creation can compensate. There are reasons to worry that the answer is yes. AI capability is developing and deploying at a pace that doesn't give institutions much lead time, and the institutions that manage labor market transitions, community colleges, workforce development agencies, unemployment insurance systems, were mostly built for a slower world.
The political consequences of rapid labor market disruption are well documented and don't require much speculation. When large populations experience economic dislocation that the mainstream political system doesn't visibly address, they tend toward political instability. This isn't a partisan observation; it's a pattern that plays out across different countries and political contexts. The rise of populist movements in the aftermath of earlier automation waves and deindustrialization isn't coincidental. If AI-driven displacement is as broad as some projections suggest, and if governments don't have credible responses ready, the political fallout could be significant.
The inequality dimension compounds the political problem. The economic gains from AI won't be distributed evenly. They'll flow primarily to the organizations that own or control AI systems and to the workers who can effectively collaborate with those systems. Both of those groups skew toward the already-wealthy end of the income distribution. If AI simultaneously reduces demand for middle-skill labor and concentrates productivity gains at the top, it's a mechanism for accelerating the inequality trends that most developed economies have been managing imperfectly for decades. A society where a small group captures most of the productivity gains from a transformative technology while a large group bears the transition costs is one with serious social stability problems.
Where things stand today
The empirical evidence on AI's labor market effects is still limited, because the most capable AI systems have only been widely deployed for a few years. What we have is a patchwork: some sectors showing early displacement effects, some studies finding productivity gains that could theoretically fund new job creation, and a great deal of uncertainty about how the picture will look at scale over the next decade. What's clear is that AI is already affecting employment in some categories of knowledge work, particularly in tasks that involve drafting, searching, summarizing, and classifying. The debate among economists is about magnitude and timing rather than direction.
Policy responses are scattered and mostly insufficient. Retraining programs have a poor track record; people who lose jobs to automation in their forties or fifties rarely successfully transition into the technical roles that AI creates. Social safety net reforms, including discussions about things like universal basic income or expanded unemployment insurance, are being debated in various countries but haven't been implemented at scale. Some jurisdictions are exploring robot taxes or automation levies as a way to fund transition support, but none of these are operational in major economies. The EU AI Act includes requirements around automated decision-making that affects employment, which at least creates some transparency and accountability obligations for AI systems making or influencing hiring, performance management, and dismissal decisions.
The employer perspective is genuinely complex. Many organizations using AI to handle tasks that would previously have required human labor aren't doing so as part of a calculated workforce reduction plan. They're responding to competitive pressure, cost constraints, and the availability of tools that make certain work faster and cheaper. The decision to adopt AI is often made by people who aren't thinking primarily about employment effects, and the downstream consequences play out through attrition, slower hiring, and reorganized roles rather than explicit layoffs. This distributed decision-making makes the aggregate employment effect harder to measure and harder to address through any single policy lever.
How Better Societies helps
Summit: The AI and labor question spans economic policy, technology governance, education systems, and social infrastructure in ways that no single sector can address alone. The Better Societies Summit creates space for policymakers, researchers, employers, and advocates to build shared understanding and coordinate on responses that actually match the scale of the problem.
Compliance: The EU AI Act includes specific requirements for AI systems that make or meaningfully influence decisions about employment, including hiring, performance assessment, promotion, and termination. These systems are classified as high-risk, meaning they require risk assessments, transparency disclosures, and meaningful human oversight. Our compliance programs help organizations deploying AI in HR and workforce contexts understand what the regulation requires and build the processes to meet it.
Accelerator: Some of the most important work being done on labor market transition sits outside traditional policy institutions. Founders are building platforms for skills recognition, portable credentials, adult learning tools, and social infrastructure for workers navigating career transitions. If you're building something that genuinely helps people adapt to a changing labor market, the Better Societies Accelerator connects you with the resources and community to move faster.