Technological displacement leaves workers with long-lasting scars, with larger shocks during recessions

Overview

This report studies the short-term and long-term effects of technological disruption on U.S. workers and applies historical technological substitution experience to assess labor market risks that AI may bring. Goldman Sachs uses U.S. national longitudinal surveys and occupation-level employment data to identify occupations with higher technological shock exposure since the 1980s and traces workers’ long-term outcomes on unemployment, re-employment, income, wealth, housing, and household formation.

Core views

The report reaches four core conclusions: First, workers displaced by technology face more difficult re-employment in the short term, with longer job search times, larger income losses after re-employment, and frequent occupational downgrading. Second, technological substitution effects are persistent, with incomplete income recovery even ten years later, along with elevated future unemployment risk, slower wealth accumulation, delayed home purchases, and delayed family formation. Third, impacts differ significantly across workers; younger, higher-educated, urban, and shorter-tenure workers adapt more effectively, and retraining can also help reduce losses. Fourth, economic recessions significantly amplify technological substitution costs because firms are more likely to cut routine jobs, making it harder for workers to secure stable re-employment in a shrinking and more competitive labor market.

Analysis framework

The report first uses employment growth and routine task intensity across more than 300 occupations in the American Community Survey to identify technology-impacted occupations, then uses the National Longitudinal Survey to track complete career histories for over 20,000 individuals. Occupations are defined as technology-shock occupations if their employment growth is in the bottom quartile of each decade and their routine task intensity is in the higher range, and matching methods based on propensity scores are used to compare subsequent outcomes for technology-displaced workers, other unemployed workers, and never-unemployed workers.

Methodology notes

• Occupational exposure identificationRoutine task intensity and employment growth screening

  Definition of technology impact occupations

  The report defines occupations in the bottom quartile of employment growth each decade and with high routine task intensity as technology-impact occupations to measure automation and technological substitution risk.

• Long-term labor market trackingNational Longitudinal Surveys

  Individual-level career history tracking

  The study uses the U.S. Bureau of Labor Statistics-supported National Longitudinal Survey, covering birth cohorts from the 1950s-1960s and the 1980s, to track career and economic outcomes of over 20,000 U.S. individuals.

• Causal comparison methodPropensity score matching

  Matching comparable workers

  The report matches workers by age, gender, race, marital status, education, income, net wealth, industry, job tenure, unemployment year, and urban/rural location to estimate differences between technology-substituted unemployment and other unemployment or non-unemployment states.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• U.S. labor market

  Direct research subject

  Strengths

  Younger, higher-educated, urban workers have stronger occupational mobility and skill upgrading ability; historical evidence shows this group more easily adapts to technological change.

  Weaknesses

  Jobs with high routine task intensity, low complexity, and high skill specificity are more prone to automation replacement, with slower income recovery after unemployment and higher repeat-unemployment risk.

  Comparison

  Compared with other unemployed and never-unemployed workers, technology-displaced unemployed workers show weaker long-term performance in income, job stability, wealth accumulation, and household formation.

  Risks

  If AI diffusion is faster than expected or coincides with a recession, unemployment rates may rise further, unemployment spells may lengthen, and income losses may persist longer.

• AI and information and communication technology

  Potential shock source and path for skill complementarity

  Strengths

  Can raise productivity and actual income, and create opportunities for workers with analytical capability, abstract-task ability, and ICT-complementary skills.

  Weaknesses

  It can reduce the value of some existing occupational skills, forcing displaced workers in re-employment to move into more routine, lower-complexity positions.

  Comparison

  Workers who undergo retraining are more likely to move into jobs with higher abstract-task content, higher complexity, and stronger complementarity with ICT.

  Risks

  If training and occupational transition mechanisms are insufficient, the AI dividend may become more uneven relative to workers’ transition costs.

Key data

• Potential AI replacement scopeMay affect 6–7% of U.S. workers over the next decadeGoldman Sachs’ global macro team previously estimated that AI could raise unemployment rates by up to about 0.5 percentage points above trend over a ten-year transition period.
• Re-employment time gapAbout one more monthCompared with workers losing jobs in stable or expanding occupations, those displaced by technology take longer to find new employment.
• Re-employment income lossOver 3%Technology-displaced workers experience materially larger real income losses after re-employment than workers displaced from stable occupations.
• Ten-year income scarNearly 10 percentage points lower than never-unemployed, about 5 percentage points lower than other unemployedThe real income of technology-displaced workers has not fully recovered even ten years after unemployment.
• Retraining effectCumulative real wage growth about 2 percentage points higher, re-unemployment probability about 10 percentage points lowerThe sample is defined as technology-displaced unemployed workers who undertook retraining for at least four weeks within three years after unemployment.
• Additional recession shockUnemployment duration increases by about 3 weeks, and subsequent unemployment and labor-force exit probabilities each rise by about 5 percentage pointsEconomic recessions amplify the long-term negative effects of technology-related unemployment.

Impact & implications

At macro and policy levels, AI is not simply a productivity story; its distributional effects and transition costs may persist for many years. If AI adoption accelerates or coincides with an economic downturn cycle, unemployment, income loss, and labor-force exit risks may become more concentrated. The report also notes that adaptation capacity is not evenly distributed: younger, higher-educated, urban workers are more likely to absorb shocks through job changes and skill upgrades, while routine, lower-complexity, highly occupation-specific jobs are more vulnerable. Retraining, occupational transition channels, and counter-cyclical labor-market policies may become important tools to mitigate AI’s impact.

Risks

• If AI adoption proceeds in a more concentrated and faster pattern, unemployment could rise more sharply in the short term than in a gradual adoption scenario.
• During recessions, firms are more likely to cut routine positions, amplifying income and employment losses from technology-related unemployment.
• Occupational downgrading may push workers into more routine, lower-complexity jobs, increasing the risk of being replaced again in the future.
• Workers with longer job tenure, more specialized skills, lower educational attainment, or non-urban location may face larger long-term income losses.
• If retraining coverage is insufficient or quality is poor, workers may struggle to move into jobs complementary to AI and ICT.

What to watch

• The speed of AI adoption in the United States and whether it is concentrated in particular white-collar or routine occupations.
• Future business-cycle shifts, especially whether AI-related layoffs coincide with recessions or demand slowdowns.
• Coverage, duration, quality, and alignment of retraining programs with demand for higher-complexity jobs.
• Employment transition outcomes for young graduates, lower-education workers, non-urban workers, and long-tenure workers.
• Long-term indicators such as re-employment wages, unemployment duration, labor-force exit rates, homeownership, and wealth accumulation.