Trend AnalysisEconomics & FinanceExperimental Design

AI and Labor Markets: Between the 92 Million Jobs Lost and the 170 Million Created

The World Economic Forum's Future of Jobs Report 2025 projects that by 2030, AI and automation will displace approximately 92 million jobs while creating roughly 170 million new onesโ€”a net gain of ...

By Sean K.S. Shin
This blog summarizes research trends based on published paper abstracts. Specific numbers or findings may contain inaccuracies. For scholarly rigor, always consult the original papers cited in each post.

The World Economic Forum's Future of Jobs Report 2025 projects that by 2030, AI and automation will displace approximately 92 million jobs while creating roughly 170 million new onesโ€”a net gain of 78 million. This framing is reassuring in aggregate but potentially misleading in distribution. The jobs displaced and the jobs created will not go to the same people, in the same places, with the same skills. The research on this transition is accumulating rapidly, and the picture is more textured than the headline numbers suggest.

The Research Landscape

Racial and Structural Dimensions of Displacement

Broady, Booth-Bell, Barr, and Meeks (2025) provide one of the more empirically grounded analyses, examining how automation and AI affected U.S. workers between 2019 and 2022. The COVID-19 pandemic served as a natural experiment: employers accelerated automation adoption during labor shortages, creating a compressed timeline of displacement effects. The key finding is distributional: Black and Latino or Hispanic workers are overrepresented in occupations most susceptible to automationโ€”roles involving routine physical and cognitive tasks. Workers in occupations with lower automation susceptibility earned higher wages and experienced greater employment stability during the pandemic.

This is not a new finding, but the pandemic data makes it more concrete. The structural vulnerability is not about technology per se but about the pre-existing occupational segregation that places certain demographic groups in the path of automation.

Generative AI and White-Collar Exposure

Jiang (2025) shifts the focus from routine manual tasks to white-collar occupations, examining how generative AI tools (ChatGPT, large language models) are affecting wage dynamics among knowledge workers. The study finds that occupations with higher exposure to generative AI experienced greater wage dispersion in 2024โ€”not uniform wage decline, but a widening gap between workers who can leverage AI tools productively and those whose tasks are substituted by them.

This patternโ€”divergence rather than uniform displacementโ€”is consistent with what economists call "skill-biased technological change," but with a twist: the relevant skill is not traditional education level but the ability to work effectively with AI systems, a competency that existing educational credentials do not reliably signal.

Displacement or Complementarity?

Zakerinia, Chen, and Srinivasan (2025) at Harvard Business School frame the core question directly: is generative AI a substitute for or a complement to human labor? Their analysis, which has attracted early academic attention finds that the answer depends heavily on the task level within occupations. Within the same job title, some tasks are displaced (routine drafting, data summarization, initial coding) while others are augmented (strategic decision-making, client interaction, creative problem-solving). The net effect on employment depends on which tasks dominate the occupation's time allocation.

Higher-Order Skills and Demand Shifts

Gulati, Marchetti, and Puranam (2025) examine how generative AI adoption reshapes not just tasks but the skills that employers demand. Using job posting data, they find that as organizations adopt GenAI tools, they shift demand toward higher-order skillsโ€”critical thinking, complex problem-solving, and interpersonal communication. This is consistent with a complementarity view of AI, but it creates a transition problem: workers whose comparative advantage lies in the routine tasks being automated must acquire new skills to remain competitive, and the timeline for this reskilling may be shorter than historical technological transitions have allowed.

Critical Analysis: Claims and Evidence

<
ClaimSourceAssessment
AI displacement disproportionately affects minority workersBroady et al., 2025Supported โ€” U.S. occupational data confirms structural vulnerability
GenAI creates wage divergence within white-collar occupationsJiang, 2025Plausible โ€” early evidence, needs longer time series
Same occupation can experience both displacement and complementarityZakerinia et al., 2025Supported โ€” task-level analysis is well-designed
AI adoption shifts employer demand toward higher-order skillsGulati et al., 2025Supported โ€” job posting data is informative but may lag actual practice
Net job creation will exceed net displacement by 2030WEF Future of Jobs Report, 2025Uncertain โ€” aggregate projection with wide confidence intervals

Open Questions

  • Reskilling timelines: How quickly can displaced workers acquire the skills needed for emerging occupations? Historical evidence from manufacturing automation suggests the answer is "not quickly enough" without significant institutional support.
  • Policy mechanisms: What mix of education reform, income support, and labor market regulation is most effective for managing the transition? The evidence base is thin because the current AI transition has limited historical precedent at this speed.
  • Developing economies: Most research focuses on the U.S. and Europe. The implications for developing economiesโ€”where labor-intensive manufacturing is a pathway to middle-income statusโ€”may be more severe and less studied.
  • Measurement: Standard labor statistics (unemployment rate, labor force participation) may not capture the quality of transition outcomes. A worker who moves from a displaced $60,000/year manufacturing job to a $30,000/year service job is employed but not equivalently situated.
  • Feedback effects: If AI increases productivity and lowers costs, the resulting economic growth could create demand for labor in ways that static models do not capture. The magnitude of this effect is an empirical question that current projections cannot answer.
  • Where This Stands

    The labor market impact of AI is neither the catastrophe that pessimists fear nor the seamless transition that optimists predict. The evidence points to a distributional story: who benefits and who bears the cost of the transition depends on pre-existing structural factorsโ€”race, geography, education, occupationโ€”that are amenable to policy intervention but resistant to market self-correction.

    Explore related work through ORAA ResearchBrain.

    References (5)

    [1] Broady, K. E., Booth-Bell, D., Barr, A., & Meeks, A. G. (2025). Automation, artificial intelligence, and job displacement in the U.S., 2019โ€“22. Labor History, 66(2).
    [2] Jiang, H. (2025). Generative AI Exposure and Wage Disparities among White-Collar Occupations: Evidence from the U.S. Labor Market in 2024. Advances in Economics, Management and Political Sciences, 15(1), 854.
    [3] Zakerinia, S., Chen, W., & Srinivasan, S. (2025). Displacement or Complementarity? The Labor Market Impact of Generative AI. Harvard Business School Working Paper 25-039.
    [4] Gulati, P., Marchetti, A., & Puranam, P. (2025). Generative AI Adoption and Higher Order Skills. SSRN Working Paper.
    [5] Okur, F., & ร–zdemir, E. (2025). Artificial intelligence, automation and employment dynamics: empirical evidence from G7 economies. Journal of Economic Studies.

    Explore this topic deeper

    Search 290M+ papers, detect research gaps, and find what hasn't been studied yet.

    Click to remove unwanted keywords

    Search 7 keywords โ†’