Rapid robot adoption in South Korea raises worker stress, depression and alcohol use within years, even as overall job satisfaction appears unchanged; declines in task enjoyment and perceived job meaningfulness point to technostress spillovers that call for targeted upskilling, job redesign and mental-health supports.

Main Finding

In South Korea—a highly robotized economy—increased regional exposure to industrial robots causally raises workers’ psychological distress (higher perceived stress and depression), worsens self-rated health, and increases alcohol consumption (notably among men). Overall job-satisfaction scores show little net change, but several job-quality dimensions (daily task enjoyment/meaningfulness, confidence in benefits, long-term commitment) decline in a staggered pattern consistent with an extension of technostress from digital tools to factory automation. The authors conclude that productivity gains from robotization may come with material costs to worker well-being unless paired with worker-centered supports.

Key Points

• The paper extends technostress theory (techno-overload, techno-invasion, techno-complexity, techno-insecurity, techno-uncertainty) from information technologies to industrial robots, arguing robots create performance norms and cognitive burdens that spill over beyond the shop floor.
• Primary psychological outcomes (from the Community Health Survey): perceived daily stress, depressive episodes (≥2 weeks), self-rated health, and monthly alcohol consumption (men).
• Workplace outcomes (from KLIPS): overall job satisfaction plus nine domain-specific satisfaction measures (compensation, employment stability, job content, environment, hours, career development, interpersonal relations, HR management, welfare).
• Main result pattern: rapid and persistent increases in stress/depression and drinking, deterioration in self-rated health; little change in aggregate job satisfaction but declines in specific meaningfulness/commitment-related domains.
• Heterogeneity: evidence of increased alcohol use among men highlighted; other subgroup patterns discussed but not fully summarized in the unedited manuscript excerpt.
• Robustness: effects are similar over 1-, 2-, and 3-year windows; results are instrumented and pass first-stage strength checks for the core instruments.

Data & Methods

• Data:
  • Community Health Survey (CHS): nationally representative repeated cross-sections with psychological and behavioral measures.
  • Korean Labor and Income Panel Study (KLIPS): longitudinal panel of urban households with detailed job-satisfaction components.
• Measure of robot exposure:
  • Region-level robot exposure computed using industry-level robot stocks (International Federation of Robotics) combined with baseline (year-2000) local industry employment shares (Bartik-style exposure measure).
• Identification strategy:
  • Two-way fixed-effects (region and year) long-difference / rolling-window design (1-, 2-, and 3-year intervals) to capture short- and medium-run effects while controlling for time-invariant regional heterogeneity and national trends.
  • Instrumental variables: Bartik-style instruments constructed from foreign industry-level robot adoption trends (notably from export-oriented economies such as Singapore, Taiwan, Germany, Japan) interacted with local baseline industry shares to isolate exogenous variation in domestic robot adoption.
  • Additional controls: lagged county-level log electricity sales and lagged log employment; province fixed effects; clustering of standard errors.
• Robustness & diagnostics:
  • Multiple instrument specifications (IV1–IV4), with some instruments showing strong Kleibergen–Paap F-statistics.
  • Sensitivity checks reported for alternative country sets in the Bartik instrument, different baseline years, and censoring approaches (e.g., Tobit for zero-robot regions).
• Strengths:
  • Combines health-focused repeated cross-sections with a longitudinal labor panel.
  • Uses an exogenous shock design via foreign-robot trends to mitigate endogeneity of local adoption.
• Limitations noted by authors / to be kept in mind:
  • Manuscript provided in unedited (pre-final) form—results pending final edits and peer review.
  • Identification rests on the exclusion assumption that foreign robot trends affect Korean worker well-being only through domestic robot adoption (potential channels via global demand or spillovers warrant careful consideration).
  • Findings are from Korea and may not generalize directly to less-automated or differently regulated labor markets.

Implications for AI Economics

• Welfare accounting: Standard productivity-focused assessments of automation should incorporate psychological and behavioral externalities. Well-being losses (stress, depression, increased substance use) constitute social costs that may offset measured gains unless mitigated.
• Labor-market modeling: Models of automation adoption ought to include non-pecuniary worker utility effects (technostress channels), dynamic adjustment costs (mental-health trajectories), and behavioral coping responses when computing net welfare, labor supply, and reservation wages.
• Policy design:
  • Complementary investments are required alongside automation: targeted upskilling and reskilling programs, job redesign to preserve meaningful tasks/autonomy, and accessible mental-health services.
  • Social insurance and income-smoothing mechanisms may need recalibration to reflect psychosocial risks, not only earnings risk.
  • Firm-level governance: monitoring and management practices that reduce techno-insecurity and techno-overload (e.g., participatory implementation, transparent evaluation systems, and workload safeguards) can mitigate adverse effects.
• Microeconomic incentives: When firms internalize long-term productivity losses due to degraded worker well-being (e.g., higher absenteeism/turnover), optimal automation adoption paths may differ from those predicted by short-run productivity metrics alone.
• Research agenda:
  • Quantify monetary equivalents of the mental-health impacts to integrate into cost–benefit analyses of automation.
  • Explore heterogeneity by occupation, task routineness, firm practices, and worker demographics to target interventions effectively.
  • Investigate long-run labor-market adjustments (reallocation, wage compensation) and whether initial well-being harms dissipate as workers adapt or are offset by policy/firms’ responses.

Note: the manuscript is an early-access, unedited version; treat the findings as provisional pending final publication.