The Wage Scar of AI: What Society Owes Workers Left Behind

May 25, 2026

Rebecca Kimble spent more than a decade as an emergency medicine physician, the kind of job described in medical school prospectuses with the word “calling”. She earned between $300,000 and $500,000 yearly. By the early months of 2026, after a long spell of unsuccessful applications back into clinical roles, she was logged into an evaluation interface for an AI laboratory, scoring how well a large language model handled queries about chest pain. She had been technically promoted. She was now an “AI trainer”, paid by the task. There were no benefits, no shifts to hand over. The clients were the foundation model providers whose products were absorbing the work she had spent two decades learning to do.

Kimble's case appeared in a Guardian investigation published in early April 2026 alongside an occupational therapy academic and a software architect now living out of motels, all of them past fifty, all of them refugees from professions where they had built decades of expertise, all of them now annotating data through firms such as Mercor, GlobalLogic, TEKsystems, micro1, Alignerr. The clients listed on the contracts are OpenAI, Google, Meta. The work is unstable. The pay starts at twenty to forty dollars an hour, with specialists occasionally crossing into the low triple digits. Labour economists in the piece called the category a “bridge job” of a cruel sort: high demand now, designed to disappear as the systems being trained on the workers' expertise become competent enough not to need them.

In the same week, Goldman Sachs published a research note that gave the Kimble vignette its longer arc. Written by economists Pierfrancesco Mei and Jessica Rindels, it drew on four decades of individual-level data covering more than twenty thousand workers and asked what happens to a person who loses their job to a wave of technological change. The answer, in the cool register of macroeconomic research, is that they do not, on average, recover. Over the ten years following such a job loss, real earnings for technology-displaced workers grow nearly ten percentage points less than for never-displaced workers, and five percentage points less than for workers displaced by other causes. The phenomenon has a name in the labour economics literature. It is called scarring, and it is not new. What is new is the suspicion, growing now into something close to consensus, that AI will inflict it at a pace and on a population for which no advanced economy has built a meaningful response.

This is a different question from the one that has dominated the AI and jobs debate. That debate has been about aggregates: how many jobs will go, how many will be created, whether the productivity gains will be shared or captured. The question now bearing down is about specific people, and what the rest of us owe them when the machine that took their occupation also took the market for the skills it had taken twenty years to acquire. It is about Kimble, the software architect in the motel, and the millions whose trajectories will not show up in headline unemployment numbers because they will eventually find some kind of job, just not one that adds up to the life they had planned.

The Anatomy Of A Scar

The labour economics of displacement is one of the bleakest sub-fields in the discipline, and it has been bleak for a long time. The foundational empirical work belongs to Steven J. Davis of the University of Chicago and Till von Wachter of Columbia, whose 2011 Brookings paper assembled administrative data on US workers laid off in mass events between 1974 and 2008. Their headline finding has the unsettling quality of a physical law. Workers displaced during a recession lost, in present-value terms, roughly nineteen percent of expected lifetime earnings, a deficit of about $112,100. Workers displaced during expansions lost about half that. Even twenty years after the event, the displaced earned ten to twenty percent less than otherwise comparable workers who had not been displaced. The losses faded only after roughly fifteen years, and even then only partially.

The mechanism, when you decompose it, is not really about unemployment. It is about what economists call occupational downgrading. The displaced worker, eventually, finds a job. The job is in a different industry, often a different occupation, frequently a less skilled one. Whatever firm-specific or industry-specific human capital the worker had built up, the relationships, the tacit knowledge, the accumulated reputation, is largely worthless on the new ladder. The worker starts again, lower down, and never catches up. Davis, von Wachter, and subsequent researchers including Brendan Moore and Judith Scott-Clayton have shown that the firm a worker lands at after displacement matters enormously: workers who can move to a similarly high-paying employer mostly recover, while those who cannot are stuck.

Subsequent NBER work concluded that even prime-aged, well-attached workers suffered persistent losses, that life expectancy fell by roughly one to one and a half years for cohorts displaced in the early 1980s recessions, and that children of displaced fathers earned about nine percent less as adults than peers whose fathers had not been displaced. The scar is not just a wage curve. It is a demographic shadow.

This is the literature that the Goldman Sachs note dropped into. Mei and Rindels's contribution was to ask whether technological displacement specifically, as opposed to displacement from a struggling firm or a contracting industry, produced a distinctive pattern. It does. Workers displaced from technology-disrupted occupations took roughly a month longer to find a new job and suffered real earnings losses more than three percent larger upon re-employment than workers displaced from more stable fields. Their occupational downgrading was sharper because the same forces that took their old job had also degraded the market value of the skills that defined them. A radiologist edged out by an imaging model is competing in a market where the price of radiological expertise has been algorithmically depressed across the board.

Goldman's report singled out one mitigation that worked: workers who participated in a vocational or technical programme within three years of displacement saw roughly two percentage points more cumulative wage growth over the following decade and a ten-percentage-point lower probability of returning to unemployment. The problem, as the same week's Guardian reporting made painfully clear, is that the retraining option that is plausibly on offer to most current AI-displaced professionals is not the one that worked in the 1980s for a machinist becoming a maintenance technician. It is, increasingly, an “AI skills” certificate that the labour market has not yet decided how to value, attached to a person whose previous credential the labour market has just decided not to value at all.

Why This Time May Be Worse

The reflex in any discussion of technological displacement is to invoke the long historical view: weavers and Luddites, telephone operators and steelworkers, eventually superseded by jobs we did not have the imagination to predict. There is something to this. Aggregate employment in advanced economies has, over two centuries, absorbed enormous waves of automation without permanent collapse. The error is in mistaking the long-run aggregate story for the lived experience of the specific cohorts caught between waves.

Three features of the current AI transition make the lived experience plausibly worse than the precedents.

The first is breadth. Earlier waves of automation tended to concentrate on particular sectors, often manual ones. The displaced were geographically clustered, occupationally cohesive, and politically visible enough to demand response, even where the response was inadequate. The post-industrial regions of the US Rust Belt and the British coalfields are not stories of generous adjustment, but they are stories of identifiable communities organised around identifiable losses. AI displacement cuts simultaneously across knowledge work (junior lawyers, paralegals, analysts), creative work (illustrators, copywriters, voice actors), administrative work (claims handlers, customer service), and professional services (consultants, accountants). The displaced are scattered. They will not gather in the same union hall.

The second is speed. The Goldman analysis covered forty years of technological transition, much of which played out across decades. AI capability has compressed similar shifts into months. Anthropic's chief executive Dario Amodei warned in 2025 that AI could eliminate as much as half of entry-level white-collar jobs within five years, a figure widely treated as bombast and widely disputed but consistent enough with what is happening at the firm level that it would be irresponsible to dismiss. Morgan Stanley analysis cited in late 2025 and early 2026 suggested the UK had begun losing more jobs than it created because of AI, performing worse than any other large economy on this measure. Whether or not the most aggressive projections come true, the lived speed of the change has already outrun the period over which retraining schemes are designed to operate. The Goldman finding that retraining helps if it happens within three years is informative; in an AI transition, three years is the gap between two model generations.

The third is the failure mode of the obvious response. The political reflex to AI displacement, in every English-speaking country and across the EU, is some variant of “learn AI”. The UK government's December 2025 announcement of a £965 million plan to push unemployed Gen Z into AI, hospitality, and engineering roles is a faithful illustration. So is the Skills England strategy of distributing AI foundation skills training to ten million workers by 2030, with £136 million for skills bootcamps in the 2025 to 2026 cycle. The premise is that workers who acquire AI skills will be lifted by the same wave that displaced them. The premise is partly true and largely insufficient.

It is partly true because there is a real wage premium attached to demonstrable AI fluency, and workers who use AI tools to multiply their own productivity keep their jobs longer than those who cannot. It is largely insufficient for two reasons. First, the AI skills credential most accessible to a displaced worker (an online course, a bootcamp certificate, a foundation skills badge) is generic, and the wage premium attaches to people who can integrate AI into substantive domain expertise, not to those whose domain expertise has just been devalued. Second, the absorptive capacity of the AI economy for newly minted “AI literate” workers is finite and is saturating faster than retraining pipelines can fill it. The Goldman report's polite phrase for the limit of retraining is “moderately effective”. The Guardian's reporting is the unpolite version: people who did the retraining, or who held the credential before retraining was a slogan, sitting in motels and labelling chest-pain queries.

The Retraining Mirage

Retraining is the policy answer almost every government has chosen and the answer least likely to be sufficient on its own. Brookings Institution analyses since late 2024 have been increasingly explicit about its limits as a stand-alone response, noting that the population most exposed to AI displacement is also the population for whom retraining has historically delivered the smallest returns: mid-career workers with significant prior investment in occupation-specific human capital. The Urban Institute's 2026 report on AI and older workers reaches a similar conclusion. The systems we have are not built for a fifty-five-year-old paralegal whose present skill set was built mostly through doing the job.

Even where retraining works in the technical sense, the credential it produces frequently has no settled labour market value. The proliferation of “AI specialist” microcredentials in 2025 and 2026 has created a thicket of certificates whose meaning is opaque to hiring managers. Some come from elite institutions and carry weight. Some come from for-profit providers whose business model depends on enrolment volumes and whose graduates struggle to demonstrate to employers what the certificate actually attests. The result, documented in the same Guardian reporting and corroborated by labour market data from job-search platforms in the US and UK, is professionals emerging from retraining with a credential that does not function as a substitute for the seniority and domain authority they have lost.

There is a subtler indignity here. The retraining narrative places the moral weight of adjustment on the displaced individual. It assumes the worker has a duty to keep up, a duty to invest in their own continuing employability, a duty to be agile. Many of the displaced workers in the current wave did exactly that. They acquired AI tools, integrated them into their work, used them to make themselves more productive, and were displaced anyway, because the productivity gain accrued mostly to the firm and was eventually used to justify replacing them or their teams with smaller numbers of even more AI-augmented workers, or with the systems themselves. The story that retraining absolves society of further responsibility is one told largely by the parties whose business model benefits from minimising it.

Beyond The Wage Curve

The economics is gloomy. The economics is also not the whole story.

The scarring effect documented by Davis and von Wachter and re-litigated by Goldman shows up in earnings, in unemployment durations, in delayed homeownership, in lower probability of marriage, in shorter life expectancy, in the next generation's earnings. These are measurable outcomes. They sit alongside outcomes that are less measurable but no less real, and that the labour market literature has only recently begun to treat as central rather than incidental. Among them: the loss of occupational identity.

To be a doctor, a lawyer, a teacher, a journalist, a designer, an engineer, is not, for most people who do these things seriously, a means of acquiring income. It is a way of being in the world. It organises time, social relationships, the practice of expertise, the experience of competence. The Boston-area sociologist Allison Pugh has spent fifteen years documenting what she calls “the tumbleweed society”, in which precarious work has corroded the sense of self workers used to derive from steady employment. The current AI displacement wave is not so much extending this trend as detonating it among populations that thought themselves immune. Professional identity, in many of the most-exposed occupations, was the compensating premium that justified years of underpaid training and the assumption of debts. Strip the occupation, and the premium goes too.

There is a parallel cost in retirement security. The post-war social contract in advanced economies relied on a worker spending most of a career in earnings-progressing employment, accruing pension contributions, housing equity, and savings sufficient for a long retirement. A scarring event in the second half of a career, a fifty-something physician dropped to twenty dollars an hour or a marketing director moved into freelance gigs, blows up the pension contribution model and frequently forces drawdown of equity to cover the gap. Existing retirement systems were not built to cushion a decade-long downward shift in earnings late in life. They were built to be supplemented by it. The arithmetic of compounding, working in reverse, is brutal: a contribution missed at fifty-five is several times more consequential to retirement income than the same contribution missed at thirty-five.

The community costs of mass scarring also bear on the discussion. The post-industrial sociology of the US Rust Belt and the UK coalfields, traced in work by Carol Graham at Brookings and the deaths-of-despair literature associated with Anne Case and Angus Deaton, has shown how earnings scarring at scale degrades not just individuals but the social fabric of the places where they live. Falling marriage rates, rising substance abuse, declining civic participation, and the decay of local institutions are downstream of long-term earnings collapse in identifiable communities. The pessimistic projection is that this pattern, formerly geographically contained, will diffuse across the suburbs and commuter belts where knowledge workers are concentrated. Professionals are not immune to despair when their occupations are taken from them.

What The Safety Net Was Built For

The infrastructure that exists to support workers in transition was, almost without exception, designed to handle a different kind of disruption. In the United States, the principal federal programme is Trade Adjustment Assistance, established in 1974 to support workers displaced by import competition. TAA includes a wage insurance component for older workers, paying half the difference between previous and current wages up to a $10,000 two-year cap. Coverage is conditional on demonstrating that displacement was caused by a specifically trade-related shock, a category that has never accommodated technological displacement and is unlikely to start doing so. The TAA data show reasonable outcomes (76.8 percent re-employment, 90.5 percent wage replacement at twelve months) for the small population that qualifies, but the gating is narrow and the overall American unemployment system is famously ungenerous, with state UI typically replacing forty to fifty percent of prior wages for six months or fewer.

The United Kingdom's principal instrument is Universal Credit, supplemented by Jobseeker's Allowance. Universal Credit was designed in the early 2010s to consolidate working-age benefits and to taper support against earnings, and it operates with notional reference rates that are some of the lowest in the OECD. The Institute for Fiscal Studies notes UK unemployment protection is unusually low by international standards, and reforms scheduled for April 2026 introduce a time-limited unemployment insurance benefit somewhat more generous than basic UC. Even after these reforms, the UK system is structurally a poverty-floor system rather than an income-replacement system. It is not designed to soften the multi-year downward slope that scarring describes; it is designed to keep people from destitution while they look for the next job, on the assumption that the next job will be roughly comparable to the last.

Active labour market policy across the OECD, retraining, job-search assistance, employment services, wage subsidies, is more developed in northern Europe than in the Anglophone world. Denmark's flexicurity model, Germany's Kurzarbeit short-time scheme, and Sweden's Trygghetsråden job security councils all reflect a continental bet that proactive transition support beats minimal cash benefits, at resourcing levels several multiples of US or UK equivalents. Even these were designed for a slower, more sectoral pattern of disruption than the present one. The OECD's 2025 Employment Outlook highlights wage insurance and early intervention as priorities, and notes that the policy frontier is shifting towards “career-oriented” support: job mobility, validation of prior learning, active counselling rather than passive cash. The frontier is mostly aspirational. The actual instruments deployed in most countries are still the unemployment insurance schemes built for a manufacturing economy that no longer exists.

The conclusion, which is both obvious and discomforting, is that the safety net in every major advanced economy is calibrated for a transition pace and a displacement pattern that AI is unlikely to produce. It will not catch the people Goldman is describing. It is not designed to.

Proportionality, Or What Would It Actually Take

If the human cost is a multi-year downward shift in life outcomes for millions of individuals, what would a proportionate response look like? The catalogue of plausible answers is not new. What is new is the urgency.

Wage insurance is the most narrowly targeted of the serious proposals, and in some ways the most practical. The mechanism is simple: a worker displaced by a defined cause receives, for a fixed period, a subsidy equal to some fraction of the gap between previous and current wages, with a cap. The TAA wage insurance pilot in the US is one model. A more ambitious version, advocated by Robert Lalonde at the University of Chicago and Lori Kletzer at Pomona among others, would be permanent, uncoupled from trade-specific causation, and set at a replacement rate sufficient to materially flatten the post-displacement income trajectory. Wage insurance is conditional on re-employment, which appeals to centre-right preferences for work incentives, and cushions the scarring slope, which appeals to centre-left preferences for income protection. It does nothing for the displaced worker who cannot find any work.

Portable benefits, the policy bundle developed in the gig economy debate, is the second serious cluster. The premise is that pensions, healthcare entitlements, and accrued leave should attach to the worker rather than the employment relationship, and should be fundable by contributions from any party for whom the worker performs paid work. The displaced professional turned data labeller would continue to accrue pension entitlements from her labelling income; her healthcare coverage would not end with her last salaried role; her capacity to weather the downward slope would be materially improved. Variants of this exist in California, Washington State, and parts of the EU, and the model is spreading slowly under pressure from gig workforce organising. It does not, by itself, address the wage scar. It addresses the cliff edges that surround the scar.

Sectoral transition assistance is the third. Drawing on the European tradition of co-managed transitions, the model dedicates funds and institutional capacity to specific sectors undergoing rapid transformation, providing tailored retraining, job placement, and income bridging for workers leaving the sector. The Trygghetsråden councils in Sweden, jointly governed by employer associations and unions, are the canonical example, with re-employment success rates over eighty percent and substantial wage maintenance for displaced workers. A serious AI-specific application would dedicate sectoral funds to the most-exposed knowledge-work occupations, fund retraining that actually leads somewhere (not generic AI literacy but routes into roles where AI-augmented expertise commands a premium), and provide income bridging for periods longer than the unemployment system contemplates. The cost is non-trivial. The outcomes, where the model has been tried, are markedly better than Anglophone alternatives.

Universal basic income is the fourth, and is the option that most directly engages the question of who pays. The case for UBI in the AI age is that if AI captures a significant fraction of the productivity gain previously realised through human labour, distributing some of that gain unconditionally to the population is the only way to maintain demand and to share the dividend. UK investment minister Jason Stockwood is one of several senior politicians on the centre and centre-left to have endorsed the broad principle in 2026, and the LSE Business Review's 2025 essays on UBI as a new social contract lay out a recognisable framework. The empirical record from limited UBI experiments (Finland, Stockton California, Kenya) is mixed but more positive than detractors allow, particularly on mental health and labour force participation. The political record is harder. UBI is expensive at any meaningful level, and politically vulnerable to the framing that it pays people not to work, a framing that has dogged smaller and more targeted unemployment schemes for decades.

A fifth option, less developed in the policy literature but gaining attention, is a productivity-linked levy on AI-displacing technologies, with proceeds hypothecated to displacement support. Bill Gates's 2017 proposal to tax robots is the rough ancestor; more recent proposals from think tanks including the Roosevelt Institute and academics including Daron Acemoglu would target firms whose AI deployments are demonstrably labour-displacing, using the revenue to fund wage insurance, retraining, and sectoral support. The mechanism is technically tricky: defining a displacing deployment, attributing displacement to specific firms, avoiding incentives to offshore are all hard. The political economy is harder still, because the firms in question include the most powerful corporations in the world, with the most sophisticated tax-policy lobbying capacity in any sector.

Each of these options has live detractors and partial precedents. None of them, individually, would be a sufficient response. Together, in some workable combination, they would begin to look proportionate. None of them is currently being adopted in any advanced economy at the scale that Goldman's findings imply is needed.

The Question Of Who Pays

The question of proportionate response is also a question of moral economy. If millions of workers are pushed onto a decade-long downward earnings trajectory because of decisions made by a few firms deploying a few classes of model, where does the obligation to make them whole sit?

The honest answer, in the existing political economy, is that it sits with the displaced themselves and their families, then with public welfare systems, then with the local communities whose tax bases and social capital absorb the second-order effects. The firms whose products generated the displacement bear, at present, no specific financial obligation tied to it. They bear general corporate tax obligations, of course, with whatever effective rates their tax-planning produces. They bear no levy keyed to displacement, no obligation to fund transitional support for the workers their products replaced, no automatic contribution to retraining schemes, and in most jurisdictions no obligation to disclose the labour market impact of their deployments.

This is, on any reasonable accounting, an enormous externality. The firms that capture the productivity gain do not pay for the wage scarring it causes; the cost is borne by the parties least able to influence the deployment decisions. The standard economic prescription for an externality of this kind is internalisation: a Pigouvian tax that forces the producer to bear the cost their activity imposes on third parties, with the revenue available to compensate those third parties. Applied to AI displacement, that argument is the productivity-linked levy described above. The technical and political difficulties of implementing it are real. The principled case for some version of it is hard to dismiss without abandoning the externalities framework altogether, which orthodox economics is rather attached to.

There is a parallel obligation argument grounded not in externality theory but in distributive justice. The productivity gain from AI is in significant part a return on data and labour that workers themselves contributed, often without meaningful consent, to the training corpora that underlie the systems now displacing them. The Guardian's data labellers are a particularly vivid case: their domain expertise is being directly fed into the systems that will erode the value of that expertise in the broader market. The implicit bargain (your knowledge, in exchange for our model's eventual ability to substitute for you) is one no rational worker would willingly accept. The argument that some share of the productivity gain should flow back to the workers whose accumulated expertise made it possible is, in this framing, not redistribution but restitution.

A third argument operates at the level of state interest. Mass scarring at the scale Goldman describes is not just bad for the affected workers. It is bad for aggregate demand, for public finances, for political stability, and for the legitimacy of liberal-democratic institutions that depend on visible upward mobility for legitimacy. The state has an interest in funding adjustment for reasons independent of any moral claim on AI firms, and a fiscal capacity to do so that is not contingent on extracting revenue from those firms. This is the implicit logic of UK and EU proposals for new unemployment insurance benefits and skills funding, both ultimately taxpayer-funded. The honest objection to this approach is that it socialises losses that were generated by private decisions, and that without a mechanism for capturing some of the corresponding gain, the public balance sheet eventually buckles.

Which of these arguments carries weight is a political question. The state-interest argument has the advantage of being palatable to almost every political constituency and of requiring no novel taxation. It also has the disadvantage of making the public, rather than the AI firms, the residual underwriter of an indefinite transition. The Pigouvian and distributive arguments have the disadvantage of requiring the political defeat of the most powerful corporate lobbies in the world, and the advantage that, if won, they would shift the cost to the parties best able to bear it.

The Person Inside The Statistic

Return to Rebecca Kimble, whose case ran in the Guardian alongside the others, and who is, as far as her interview suggested, more pragmatic than bitter. She is not a metaphor. She is a person who spent twenty years training to do something difficult and useful, who did it for more than a decade, who lost it in a transition not of her making, and who is now adjacent to the systems that took it from her, paid by the task to teach them how to be better at it. The statistical Goldman scar, in her case, is not yet visible, because the data on the current cohort of displaced professionals will not be in for years. On the basis of forty years of prior data, her ten-year earnings trajectory has been bent down by roughly ten percent, and the bend will not straighten.

Multiply Kimble by some number that researchers will eventually settle on. The lowest plausible estimates of AI displacement in advanced economies in the second half of the 2020s run into the millions; the higher estimates run into the tens of millions. Even the lowest estimates imply a population of scarred workers larger than any single cohort affected by any postwar industrial transition. The scale, the speed, and the breadth of the transition, taken together with the inadequacy of the existing safety net and the absence of any meaningful obligation imposed on the firms generating the gains, describe a policy failure waiting to be named.

The Goldman note ended with retraining as its constructive suggestion, the mildest of the available answers and the one most consistent with the existing political settlement. The Guardian's reporting ended with the trainers and motel-dwellers and the accumulating evidence that the settlement is not equal to the moment. Neither paper said what a proportionate response would require, perhaps because both knew that to say it plainly would be to step outside the bounds of what either treats as plausible. It would require, at minimum, the simultaneous deployment of wage insurance, portable benefits, sectoral transition assistance, and a meaningful displacement-linked contribution from the firms whose deployments generated the displacement, all on a scale several multiples beyond what is currently being budgeted in any major advanced economy. It would require, in other words, a different settlement.

Whether one is built before the scar deepens or only after is the question every affected country's political class will, in spite of itself, have to answer. The statistic is being measured. The people inside the statistic have names. The bill is being written, in real time, on the wage curves of millions of careers that will not now arc the way the people living them had assumed.

References & Sources

Mei, Pierfrancesco, and Jessica Rindels. “Goldman Sachs Research Note on the Scarring Effects of Technological Displacement.” Goldman Sachs, April 2026. Reported in: Eaton, Kit. “Goldman Sachs Warns That Losing Your Job to AI Can Hurt Your Earnings for a Decade.” Inc., 7 April 2026. https://www.inc.com/kit-eaton/goldman-sachs-warns-that-losing-your-job-to-ai-can-hurt-your-earnings-for-a-decade/91332401
Goldman Sachs. “How Will AI Affect the Global Workforce?” Goldman Sachs Insights. https://www.goldmansachs.com/insights/articles/how-will-ai-affect-the-global-workforce
TIME. “AI Is Learning to Do the Jobs of Doctors, Lawyers, and Consultants.” https://time.com/7322386/ai-mercor-professional-tasks-data-annotation/
Davis, Steven J., and Till von Wachter. “Recessions and the Costs of Job Loss.” Brookings Papers on Economic Activity, Fall 2011. https://www.brookings.edu/wp-content/uploads/2011/09/2011b_bpea_davis.pdf
Lachowska, Marta, Alexandre Mas, and Stephen A. Woodbury. “Sources of Displaced Workers' Long-Term Earnings Losses.” NBER Working Paper No. 24217, January 2018. https://www.nber.org/system/files/working_papers/w24217/w24217.pdf
Moore, Brendan, and Judith Scott-Clayton. “The Firm's Role in Displaced Workers' Earnings Losses.” Industrial and Labor Relations Review, 2025. https://journals.sagepub.com/doi/10.1177/00197939241310124
Brookings Institution. “Unemployment and Earnings Losses: A Look at Long-Term Impacts of the Great Recession on American Workers.” https://www.brookings.edu/articles/unemployment-and-earnings-losses-a-look-at-long-term-impacts-of-the-great-recession-on-american-workers/
Brookings Institution. “AI labor displacement and the limits of worker retraining.” https://www.brookings.edu/articles/ai-labor-displacement-and-the-limits-of-worker-retraining/
Urban Institute. “AI and Older Workers.” January 2026. https://www.urban.org/sites/default/files/2026-01/AI_and_Older_Workers_0.pdf
National Academies of Sciences, Engineering, and Medicine. “Retraining Workers for the Age of AI.” https://www.nationalacademies.org/news/retraining-workers-for-the-age-of-ai
Policy Options. “Canada's labour protections aren't ready for the age of AI.” March 2026. https://policyoptions.irpp.org/2026/03/ai-labour-protections/
NBER. “Wage Insurance for Displaced Workers.” https://www.nber.org/digest/202408/wage-insurance-displaced-workers
CEPR. “Wage insurance for trade-displaced workers: A middle-ground alternative to rising protectionism.” https://cepr.org/voxeu/columns/wage-insurance-trade-displaced-workers-middle-ground-alternative-rising-protectionism
Congressional Research Service. “Trade Adjustment Assistance for Workers and the TAA Reauthorization Act of 2015.” https://www.congress.gov/crs-product/R44153
Bipartisan Policy Center. “What Happens When Jobs Disappear? A Guide to Displaced Worker Programs in the U.S.” https://bipartisanpolicy.org/explainer/what-happens-when-jobs-disappear-a-guide-to-displaced-worker-programs-in-the-u-s/
OECD. “OECD Employment Outlook 2025: Reviving growth in a time of workforce ageing: The role of job mobility.” July 2025. https://www.oecd.org/en/publications/2025/07/oecd-employment-outlook-2025_5345f034/full-report/component-9.html
Institute for Fiscal Studies. “Options for unemployment insurance.” https://ifs.org.uk/publications/options-unemployment-insurance
Institute for Fiscal Studies. “Universal credit review: challenges and options for reform.” https://ifs.org.uk/publications/universal-credit-review-challenges-and-options-reform
UK Government. “Free AI training for all, as government and industry programme expands to provide 10 million workers with key AI skills by 2030.” GOV.UK. https://www.gov.uk/government/news/free-ai-training-for-all-as-government-and-industry-programme-expands-to-provide-10-million-workers-with-key-ai-skills-by-2030
Skills England. “AI Skills Boost: Skills England's AI foundation skills for work benchmark supports free AI training for all.” 28 January 2026. https://skillsengland.blog.gov.uk/2026/01/28/ai-skills-boost-skills-englands-ai-foundation-skills-for-work-benchmark-supports-free-ai-training-for-all-by-phil-smith/
Fortune. “UK launches $965 million plan to get unemployed Gen Z into AI, hospitality, and engineering.” 9 December 2025. https://fortune.com/2025/12/09/millions-gen-z-unemployed-globally-uk-tossing-965-million-at-problem-get-young-people-ai-hospitality-engineering-jobs/
People Management. “Universal basic income needed to support workers displaced by AI, minister says.” https://www.peoplemanagement.co.uk/article/1946845/universal-basic-income-needed-support-workers-displaced-ai-minister-says
LSE Business Review. “Universal basic income as a new social contract for the age of AI.” 29 April 2025. https://blogs.lse.ac.uk/businessreview/2025/04/29/universal-basic-income-as-a-new-social-contract-for-the-age-of-ai-1/
Pugh, Allison J. “The Tumbleweed Society: Working and Caring in an Age of Insecurity.” Oxford University Press, 2015.
Case, Anne, and Angus Deaton. “Deaths of Despair and the Future of Capitalism.” Princeton University Press, 2020.

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...