AI and Work: Five Scenarios, Ordered by Severity

What could happen to European societies, including the one I live in, if current AI impact trends continue under different labor-market scenarios? Everyone has an opinion, but what do economists say?

This post organizes economists’ views on AI and work into five scenarios, ordered by severity rather than likelihood. For each scenario, it summarizes the expected effects on labor markets and society, the probability implied by the current mainstream economic consensus, the published indicators that would show whether the scenario is emerging, and the policy responses economists discuss to preserve social stability.

How scenarios are ordered

The scenarios below are ordered by severity, not by probability.

For each scenario, I include a qualitative probability estimate based on the current mainstream economic consensus. These probabilities should not be read as precise forecasts. They reflect what economists and institutions currently appear most willing to defend based on published evidence.

There is an important limitation: economists tend to assign lower probability to extreme scenarios unless there is strong empirical evidence already visible in labor-market data. This makes sense methodologically, but it may also cause the profession to underweight fast-moving technological transitions.

My own view is more severe than the current consensus. I think society may be moving gradually toward the post-labor scenario. However, the purpose of this post is not to present my view as consensus, but to separate:

  1. what economists currently regard as likely
  2. what they regard as possible but uncertain
  3. what society should track if the severe scenarios start becoming real

Main economists and sources

This post draws mainly on work by Daron Acemoglu, Pascual Restrepo, David Autor, Erik Brynjolfsson, Danielle Li, Lindsey Raymond, Anton Korinek, Joseph Stiglitz, Daniel Susskind, Carl Benedikt Frey, and Michael Osborne.

Important sources include:

Useful talks, interviews, and audio:

Scenario 5: Severe post-labor economy

Severity: highest | Probability: low

This is the most severe scenario. AI systems perform most economically valuable cognitive tasks and increasingly coordinate physical production through robotics and automation.

Human labor still exists, but wages no longer distribute income broadly. The traditional link between skill, employment, and income weakens substantially.

In this scenario, the key economic issue is ownership of productive capital. If AI capital produces most output, income flows to whoever owns the models, compute, data, platforms, and firms.

A video with Anton Korinek is also available on the Harvard Business Review (HBR) website.

What economists say

This scenario is discussed less in standard near-term labor economics and more in work on transformative AI, especially by Anton Korinek.

Anton Korinek and Joseph Stiglitz analyze cases where AI is strongly labor-saving. In these models, wages can fall and income shifts toward capital owners. Their conclusion is that redistribution becomes central if society becomes richer while labor income weakens. See Artificial Intelligence and Its Implications for Income Distribution and Unemployment.

Anton Korinek also works directly on transformative AI scenarios, where AI may become capable of performing a very large share of economically valuable tasks. In this literature, the main economic question becomes whether human labor remains a scarce input, or whether capital and AI systems become the dominant productive factors. See The Economics of Transformative AI.

Daniel Susskind argues that long-term technological unemployment should be taken seriously. His contribution is broader than income policy. Work also provides status, identity, structure, and social recognition. If work becomes less central, society must address both income distribution and the social role currently played by employment. See A World Without Work.

What would show this scenario is emerging

Relevant indicators:

  • sustained fall in labor share of income
  • weak wage growth despite strong productivity growth
  • declining employment in high-skill occupations
  • falling returns to education
  • rising corporate profit share
  • rising wealth concentration
  • lower headcount-to-revenue ratios across many sectors
  • rising dependence on transfers
  • falling tax revenue from labor
  • broad loss of bargaining power across occupations

The key warning sign would be strong GDP or productivity growth combined with stagnant or falling median labor income.

For Europe, this should be tracked through Eurostat labor-market data, national statistical offices, OECD productivity and wage data, and distributional national accounts where available. Relevant European indicators include labor share, unemployment, youth unemployment, occupational employment, enterprise AI adoption, tax composition, and household income distribution. Eurostat’s AI enterprise-use data is especially relevant because it tracks how quickly firms actually adopt AI technologies, rather than only what AI could theoretically do.

Policy response

In this scenario, ordinary retraining is insufficient.

The main policy tools become:

  • universal basic income
  • social dividends
  • public wealth funds
  • public ownership stakes in AI infrastructure
  • capital taxation
  • wealth taxation
  • taxation of monopoly rents
  • public provision of essential services
  • shorter working time
  • democratic oversight of key AI systems
  • new institutions for social status outside employment

The central problem is that wages no longer perform their historical role as the main distribution mechanism.

For Europe, this would likely imply much larger debates about welfare-state funding, taxation of capital and corporate profits, EU-level industrial policy, public AI infrastructure, and whether European societies can preserve social cohesion if employment becomes a weaker basis for income distribution.

Scenario 4: Broad skilled-labor displacement

Severity: very high | Probability: low to medium

In this scenario, AI substantially reduces demand for skilled cognitive labor across many occupations.

Affected groups could include:

  • programmers
  • analysts
  • designers
  • lawyers
  • consultants
  • accountants
  • writers
  • translators
  • researchers
  • marketers
  • administrators
  • junior managers

This scenario is less extreme than the post-labor economy because human work remains important, but large parts of the professional middle class lose wage security.

What economists say

Daron Acemoglu and Pascual Restrepo provide the central task-based framework. In their model, automation creates a displacement effect when machines take over tasks previously performed by workers. New tasks can offset this through a reinstatement effect, but if new task creation is too weak, labor demand falls. See Artificial Intelligence, Automation and Work.

Daron Acemoglu also warns about “so-so automation”: automation that replaces workers but produces limited productivity gains. Firms may still adopt it because it reduces wage costs, even if the social gain is small. His later macroeconomic work is also skeptical of very large near-term productivity effects from AI unless AI affects economically large tasks at scale. See The Simple Macroeconomics of AI.

Anton Korinek and Joseph Stiglitz focus more directly on distribution. If AI is strongly labor-saving, income shifts from labor toward capital. Their policy conclusion is that capital taxation, redistribution, and social insurance become more important. See Artificial Intelligence and Its Implications for Income Distribution and Unemployment.

David Autor is more optimistic. He argues that AI could allow more workers to perform expert tasks, potentially rebuilding some middle-class work. This depends on AI being used as a decision-support and augmentation tool rather than mainly as a labor-replacement tool. See Applying AI to Rebuild Middle Class Jobs.

What would show this scenario is emerging

Relevant indicators:

  • declining employment in professional occupations
  • falling entry-level hiring in skilled fields
  • graduate underemployment
  • occupational downgrading
  • falling college wage premium
  • falling wage growth for skilled workers
  • fewer junior roles in law, software, finance, consulting, media, and analysis
  • rising productivity with flat skilled wages
  • higher profits with lower skilled headcount
  • reduced career mobility for younger workers

This scenario may first appear among young workers because they depend on entry-level tasks to build experience.

In Europe, the most relevant data would include Eurostat labor-force statistics, national graduate employment surveys, occupational wage data, vacancy data, and firm-level data on AI adoption. The current European baseline matters: Eurostat reported that 20.0 percent of EU enterprises with 10 or more employees used AI technologies in 2025, up from 13.5 percent in 2024 and 7.7 percent in 2021. That type of adoption data should be monitored together with employment and wage data by occupation.

Policy response

Economists move beyond general reskilling and toward distribution and bargaining institutions.

Policy tools include:

  • wage insurance
  • stronger unemployment insurance
  • public retraining tied to real labor demand
  • firm-funded transition support
  • capital and rent taxation
  • social dividends
  • shorter working time
  • public AI infrastructure
  • collective bargaining around AI deployment
  • employee ownership and profit-sharing
  • protection of entry-level career ladders

The social stability risk is high because skilled workers have made large investments in education and career formation. If those investments lose value, the political reaction may be severe.

For Europe, this scenario would put pressure on welfare states, unemployment insurance, professional education systems, and collective bargaining institutions. It would also raise questions about whether European societies can protect the professional middle class without blocking productivity growth.

Scenario 3: Significant sectoral displacement

Severity: high | Probability: medium

In this scenario, AI causes large job losses in specific sectors, but not across the whole economy.

Likely exposed areas include:

  • customer support
  • clerical administration
  • legal support
  • translation
  • marketing content
  • basic software development
  • accounting support
  • research assistance
  • media production
  • HR administration
  • back-office finance

The economy as a whole may continue to create jobs, but the affected sectors experience real displacement.

What economists say

This scenario fits the task-based literature of Daron Acemoglu, Pascual Restrepo, and David Autor. AI is most likely to displace labor where tasks are digital, repetitive, text-based, pattern-based, or easy to verify. The employment effect depends on whether demand expands enough to absorb displaced labor, and whether new tasks appear quickly enough. See Acemoglu and Restrepo, Artificial Intelligence, Automation and Work.

Daron Acemoglu’s “so-so automation” argument is relevant here. Firms may adopt AI to reduce headcount even when the productivity gains are modest. This can be privately profitable while still producing weak social gains. See The Simple Macroeconomics of AI.

The OpenAI and University of Pennsylvania exposure study by Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock is also relevant. It estimates that many workers have tasks exposed to large language models, but it does not predict direct job losses or adoption timelines. See GPTs are GPTs.

The older paper by Carl Benedikt Frey and Michael Osborne is useful historical context because it estimated automation risk at the occupation level. Later work has generally moved toward task-level exposure because occupations contain many different tasks. See The Future of Employment.

What would show this scenario is emerging

Relevant indicators:

  • layoffs concentrated in exposed sectors
  • falling job openings in exposed occupations
  • weak wage growth in exposed sectors
  • longer unemployment duration after displacement
  • lower reemployment wages
  • high applicants-per-vacancy ratios
  • reduced junior hiring
  • productivity rising while headcount falls
  • regional unemployment where exposed sectors cluster

A strong warning sign would be sector-level productivity gains combined with falling employment and poor reemployment outcomes.

In Europe, this scenario should be monitored through sector-level employment statistics, vacancy data, job posting data, unemployment duration, and reemployment wages. Regional concentration matters because European political reactions often become stronger when labor-market stress is geographically concentrated.

Policy response

The standard policy package includes:

  • unemployment insurance
  • wage insurance
  • active labor-market policies
  • transition grants
  • occupational retraining
  • relocation support
  • regional support
  • firm-funded retraining
  • collective bargaining over automation
  • public employment options where regional displacement is severe

The social risk depends on concentration. Sectoral displacement can be politically destabilizing if it hits specific regions, age groups, or professional identities.

For Europe, this scenario would probably be handled through existing welfare-state institutions, but those institutions may need adjustment if displacement is faster, more white-collar, or more geographically clustered than previous labor-market shocks.

Scenario 2: Uneven disruption and labor-market polarization

Severity: medium | Probability: high

In this scenario, AI benefits some workers while hurting others.

Workers who use AI effectively become more productive. Workers whose tasks are easy to automate face weaker demand. Entry-level workers may be especially exposed because their early-career tasks are often simpler, more codifiable, and easier to supervise with AI.

The result is higher inequality and weaker career ladders, even if headline unemployment remains low.

What economists say

This scenario is close to the current mainstream institutional view.

The IMF argues that AI exposure is high in advanced economies, especially because they have many cognitive-intensive jobs. It also distinguishes between exposure and complementarity: some workers may benefit from AI, while others face lower demand or lower wages. See Gen-AI: Artificial Intelligence and the Future of Work.

The ILO similarly finds that many jobs are more likely to be augmented than fully automated, but clerical work is especially exposed. It also highlights gendered effects because clerical occupations are an important source of employment for women in many countries. See Generative AI and Jobs.

David Autor’s optimistic view fits partly here. He argues that AI could help less-specialized workers perform higher-value tasks, but this outcome depends on institutions, firm choices, and whether AI systems are designed to complement workers. See Applying AI to Rebuild Middle Class Jobs.

Daron Acemoglu is more cautious. His work emphasizes that productivity gains do not automatically become wage gains. If AI weakens labor demand or bargaining power, the benefits can accrue mainly to firms and capital owners. See The Simple Macroeconomics of AI.

What would show this scenario is emerging

Relevant indicators:

  • rising wage inequality
  • weaker wage growth in exposed occupations
  • declining entry-level hiring
  • reduced graduate employment quality
  • stronger wage growth for AI-complementary workers
  • increasing returns to firm-specific AI adoption
  • falling demand for routine office roles
  • more contract and freelance work
  • higher productivity without broad wage growth

This scenario can exist even when headline unemployment looks healthy.

In Europe, this should be tracked through wage dispersion, youth unemployment, graduate outcomes, occupational employment, job openings, and AI adoption by firm size and sector. Eurostat unemployment and youth unemployment data are useful, but they are insufficient on their own. A healthy headline unemployment rate can hide weaker entry-level hiring, reduced career mobility, and occupational downgrading.

Policy response

Economists usually recommend:

  • AI literacy
  • education reform
  • targeted retraining
  • wage insurance
  • stronger safety nets
  • support for job mobility
  • protection of entry-level training paths
  • incentives for augmentation rather than replacement
  • competition policy
  • monitoring of AI’s effect on young workers

The social risk is moderate but can become serious if younger educated workers lose access to stable careers.

For Europe, this scenario is especially relevant because many countries rely on the promise that education, welfare-state support, and regulated labor markets can produce broad social stability. If AI weakens entry-level paths into skilled work, the problem may appear first as frustration among younger workers rather than as mass unemployment.

Scenario 1: Productivity improvement with limited job loss

Severity: lowest | Probability: highest

This is the mildest scenario.

AI improves productivity in many jobs, but employment remains broadly stable. Workers use AI to write, summarize, code, translate, search, analyze, and automate routine parts of office work.

Firms become more productive, but most occupations remain intact.

What economists say

This scenario is supported by empirical evidence showing productivity gains from AI tools in specific workplaces.

Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied the introduction of a generative AI assistant in customer support. They found that access to the tool increased productivity, with especially large effects for novice and lower-skilled workers. This supports the augmentation view in at least some settings. See Generative AI at Work.

The OECD’s employment work is also relevant. It finds limited evidence so far that AI is producing broad job losses, while warning that automation potential remains high and that many jobs are exposed to AI and related automation technologies. See OECD Employment Outlook 2023.

This scenario is also compatible with David Autor’s view that AI may expand what workers can do, especially if used to support judgment rather than simply replace labor. See Applying AI to Rebuild Middle Class Jobs.

What would show this scenario is emerging

Relevant indicators:

  • stable unemployment
  • stable or rising employment-to-population ratio
  • rising productivity
  • wage growth broadly tracking productivity
  • low displacement in exposed occupations
  • rising AI adoption without major layoffs
  • increased output per worker
  • workers reporting improved productivity
  • limited change in labor share of income

The key signal is that AI adoption raises output without reducing employment or wages broadly.

In Europe, this scenario would be consistent with increasing enterprise AI adoption combined with stable employment, stable labor share, rising productivity, and wage growth that reaches ordinary workers. Eurostat reported that around one fifth of EU enterprises with 10 or more employees used AI technologies in 2025. If that number continues rising without broad labor-market deterioration, the mild scenario gains support.

Policy response

Policy can remain relatively moderate:

  • AI literacy
  • education updates
  • worker mobility support
  • diffusion of AI tools to smaller firms
  • competition policy
  • public-sector AI adoption
  • monitoring of exposed occupations

The risk of unrest is low if productivity gains are broadly shared.

For Europe, the main policy challenge would be to ensure that AI adoption is not limited to large firms and that productivity gains are shared through wages, public services, lower prices, or shorter working time.

Cross-scenario monitoring framework

A serious AI labor-risk dashboard should not rely on headline unemployment alone.

Employment quantity

Track:

  • unemployment rate
  • employment-to-population ratio
  • labor-force participation
  • layoffs and discharges
  • hires
  • job openings
  • unemployment duration

Employment quality

Track:

  • wage growth by occupation
  • reemployment wages
  • involuntary part-time work
  • contract and freelance share
  • benefits coverage
  • job security
  • underemployment

Distribution

Track:

  • labor share of income
  • capital share of income
  • corporate profit share
  • wealth concentration
  • wage inequality
  • college wage premium
  • income mobility

Occupational structure

Track:

  • entry-level hiring
  • graduate employment
  • junior roles in exposed professions
  • occupational downgrading
  • movement from professional work into lower-paid services

Firm behavior

Track:

  • headcount versus revenue
  • headcount versus profit
  • AI adoption
  • AI spending
  • layoffs following AI investment
  • market concentration
  • profit concentration in AI-intensive firms

Social and political stress

Track:

  • strike activity
  • unionization activity
  • protest activity
  • trust in institutions
  • support for extremist parties
  • household financial stress
  • youth unemployment
  • regional unemployment clusters

European data sources to use

For a European version of this monitoring framework, relevant sources include:

Summary table

Scenario Severity Mainstream probability Main risk Main policy response
5. Severe post-labor economy Highest Low, high impact Wages no longer distribute income broadly UBI, social dividends, public ownership, wealth funds
4. Broad skilled-labor displacement Very high Low to medium Skilled workers lose wage security Capital taxation, shorter work week, bargaining power
3. Significant sectoral displacement High Medium Concentrated layoffs in exposed sectors Wage insurance, transition policy, regional support
2. Uneven disruption and polarization Medium High Inequality and weaker career ladders Education, wage insurance, safety nets
1. Productivity improvement with limited job loss Lowest Highest Gains may still be uneven AI literacy, diffusion, monitoring

Conclusion

The current mainstream economic consensus puts the highest probability on the mild and medium scenarios. Economists generally see strong evidence for AI exposure, task change, and productivity effects, but weaker evidence so far for broad AI-driven unemployment.

The severe scenarios receive lower probability in the mainstream view because they require more assumptions: stronger AI capabilities, faster adoption, broad substitution across skilled occupations, weak creation of new tasks, and a sustained shift of income from labor to capital.

However, severity matters. A low-probability scenario can still deserve serious attention if the social cost is high.

If the world moves toward broad skilled-labor displacement or a post-labor economy, retraining will not be enough. Society would need to change how income is distributed. The key tools would be capital taxation, public wealth funds, social dividends, shorter working time, stronger labor bargaining power, competition policy, and public AI infrastructure.

The central question is whether AI productivity gains continue to reach people through wages.

If they do, the transition may be manageable.

If they increasingly flow to capital owners, society will need new institutions to avoid political instability and unrest.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.