Korean industries more exposed to AI saw steadily shrinking workweeks after 2022, with the largest reductions by 2025; the pattern is consistent with AI diffusion rather than pre-existing trends.
Artificial Intelligence Exposure and Working Hours: Evidence From South Korea
openalex
quasi_experimental
medium evidence
7/10 relevance
Summary only
summary available; pdf_status=paywall
DOI
Source
PDF
In Korea, industries with higher AI exposure experienced progressively larger declines in weekly working hours after 2022, with the negative association growing through 2025 and no pre-trend differences detected.
ABSTRACT This study examines whether working‐hour trends differed across Korean industries with varying levels of AI exposure following the diffusion of AI technologies that accelerated in 2022. While a growing body of research has examined the effects of AI on productivity, employment and wages, relatively little attention has been paid to working hours. To address this gap, this study constructs a Korean AI Industry Exposure Index. Using an exposure‐based event‐study framework, the analysis estimates differential working‐hour trends across industries between 2020 and 2025. The results show that industries with higher levels of AI exposure experienced larger declines in weekly working hours in 2023, 2024 and 2025. Moreover, the magnitude of the negative association grows over time, reaching its largest value in 2025. In contrast, no statistically significant differences are observed in 2020 and 2021, consistent with the parallel‐trends assumption. Additional analyses reveal no statistically significant heterogeneity by employment type, flexible working arrangements, labour union membership or part‐time employment status. Overall, the findings suggest that the diffusion of AI technologies after 2022 was associated with progressively larger reductions in working hours in highly AI‐exposed industries.
Summary
Main Finding
Industries in Korea with higher exposure to AI experienced progressively larger declines in weekly working hours after the 2022 acceleration in AI diffusion, with statistically significant reductions appearing in 2023–2025 and the largest negative association observed in 2025. There were no pre‑treatment differences in 2020–2021, supporting parallel trends.
Key Points
- The study constructs a Korean AI Industry Exposure Index to rank industries by their degree of AI exposure.
- Uses an exposure‑based event‑study framework to compare working‑hour trends across industries from 2020 to 2025.
- No statistically significant differences in working hours across exposure levels in 2020–2021 (consistent with parallel trends).
- Significant declines in weekly working hours for more AI‑exposed industries begin in 2023 and strengthen in 2024–2025, peaking in 2025.
- Additional heterogeneity checks find no significant variation in the estimated effect by employment type, flexible working arrangements, labour union membership, or part‑time status.
Data & Methods
- Outcome: weekly working hours at the industry level (period 2020–2025).
- Treatment measure: a constructed Korean AI Industry Exposure Index indicating each industry's level of exposure to AI diffusion.
- Empirical approach: exposure‑based event‑study design that estimates differential trends in working hours by industry exposure over time, testing pre‑trend validity (no differences in 2020–2021).
- Robustness/heterogeneity: analyses conducted to test whether the working‑hours effect differs by employment type, flexible work arrangements, union membership, or part‑time status (no significant heterogeneity found).
- (Note: details on data sources, index construction algorithms, covariates, and econometric specifications are described in the full paper.)
Implications for AI Economics
- Labor margins: Beyond employment and wages, AI diffusion appears to operate through working‑hour adjustments — firms or workers in more AI‑exposed industries are shortening weekly hours as AI adoption spreads.
- Mechanisms to consider: task automation/substitution reducing time needed per output, productivity gains enabling shorter hours, or changes in demand/composition of tasks that shorten schedules.
- Policy relevance: monitoring hours is important for assessing welfare impacts of AI; policies could include work‑sharing, taxation and benefit adjustments, and retraining programs to manage transitions even where headcount effects are muted.
- Generality and research needs: results highlight a potentially broad effect across worker groups (no heterogeneity found), but causal pathways, index construction, and applicability to other countries warrant further investigation. Future work should link hours changes to productivity, wage dynamics, and worker welfare.
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — The paper uses a plausible quasi-experimental event-study design and tests for no pre-trend differences, which supports a causal interpretation, but identification depends on the validity of the AI exposure index, potential omitted time-varying industry shocks, and aggregate/industry-level measurement that may mask within-industry confounding; no exogenous instrument or random assignment is used.
Methods Rigormedium — Approach appears rigorous in using event-study dynamics, pre-trend checks, and heterogeneity analyses, but likely lacks an exogenous source of variation in AI diffusion, may be sensitive to index construction and industry composition changes, and relies on observational industry-level panel data rather than firm- or worker-level causal leverage.
SamplePanel of Korean industries observed 2020–2025 (industry-level or industry-aggregated worker data on weekly working hours), with industry AI exposure measured via a constructed AI Industry Exposure Index and analysis covering differential trends across industries before and after 2022 AI diffusion.
Themeslabor_markets adoption
IdentificationExposure-based event-study / difference-in-differences: constructs an AI Industry Exposure Index and compares working-hour trends across industries with higher vs lower AI exposure before (2020–2021) and after the accelerated diffusion (post-2022), relying on the parallel-trends assumption and industry fixed effects to identify differential changes.
GeneralizabilityFindings are specific to Korea and the 2020–2025 period (post-2022 AI diffusion) and may not generalize to other countries or earlier/later periods., Industry-level exposure may not reflect firm- or worker-level AI adoption, limiting extrapolation to individual worker outcomes., Results pertain to weekly working hours only and do not directly inform effects on wages, employment, or productivity., Validity depends on the AI exposure index construction; different exposure measures or AI definitions could change results., Potential confounding from concurrent macro or sectoral shocks (e.g., pandemic recovery or supply-chain changes) may limit external applicability.
Claims (6)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| This study constructs a Korean AI Industry Exposure Index. Automation Exposure | null_result | AI exposure (Korean AI Industry Exposure Index) |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Industries with higher levels of AI exposure experienced larger declines in weekly working hours in 2023, 2024 and 2025. Employment | negative | weekly working hours |
Reading fidelity
high
Study strength
medium
|
not reported
|
| The magnitude of the negative association between AI exposure and weekly working hours grows over time, reaching its largest value in 2025. Employment | negative | weekly working hours |
Reading fidelity
high
Study strength
medium
|
not reported
|
| No statistically significant differences in working-hour trends by AI exposure are observed in 2020 and 2021, consistent with the parallel-trends assumption. Employment | null_result | weekly working hours (pre-treatment differences) |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Additional analyses reveal no statistically significant heterogeneity in the AI exposure — working-hours relationship by employment type, flexible working arrangements, labour union membership or part-time employment status. Employment | null_result | weekly working hours (heterogeneity of effects) |
Reading fidelity
high
Study strength
medium
|
not reported
|
| The diffusion of AI technologies accelerated in 2022. Adoption Rate | positive | pace of AI diffusion/adoption |
Reading fidelity
medium
Study strength
speculative
|
not reported
|