Study on the Impact of Artificial Intelligence on India's Labor Market: A Causal Analysis Based on Task Exposure

Basic Study Information

• Original Author: Sukin11
• Source: GitHub (Link: https://github.com/Sukin11/AI-Impact-Task-Based-Approach)
• Publication Date: May 26, 2026

Core Content

This study uses the continuous difference-in-differences method to analyze the impact of AI exposure on employment growth, labor productivity, and labor quality using data from 27 industries in India from 2009 to 2022. Key findings include: Industries with high AI exposure have faster employment growth but lower labor productivity and quality; effects have a 2-5 year lag; significant industry heterogeneity exists. The study fills the gap in research on AI's impact on labor markets in developing countries.

Global Controversies and Research Gaps

• Optimists: AI improves productivity and creates new jobs to boost employment growth
• Pessimists: AI replaces labor leading to unemployment and inequality
• Gap: Existing studies mostly focus on developed countries, with insufficient attention to developing countries (e.g., India)

Importance of India's Context

As the world's second most populous country and an emerging economy, India's labor market response to AI shocks is reference-worthy. The study aims to answer:

1. How does AI exposure affect employment growth?
2. What impact does AI have on labor productivity?
3. How does AI change labor quality?
4. Are there lagged effects?
5. What about industry heterogeneity?

Core Method: Continuous Difference-in-Differences (Continuous DiD)

• Suitable for scenarios where AI exposure is a continuous variable; compares labor market changes through cross-industry differences
• Key assumption: In the absence of AI shocks, trends of high/low exposure industries are parallel

Measurement of AI Exposure

Uses the task-based index by Felten et al. (2021):

• Based on job tasks rather than occupational classification, accurately captures substitution potential
• Maps to KLEMS classification for easy cross-country comparison

Data Sources

• Labor Market: India KLEMS Database
• Industries: 27 manufacturing and service industries
• Time: 2009-2022 (2014 as the starting point of AI diffusion)

Overall Effects

• Employment Growth: Positive and significant (β=1.681, p<0.05); industries with high AI exposure grow faster
• Labor Productivity: Negative and significant (β=-4.409, p<0.05); caused by short-term adjustment costs or skill mismatch
• Labor Quality: Negative and significant (β=-0.481, p<0.05); medium-skill jobs are replaced

Lagged Effects

• Effects not significant from 2014-2018; emerged after 2019 (2-5 year lag)
• Parallel trend test passed (coefficients not significant before 2014)

Industry Heterogeneity

• High Exposure: Information technology, finance, professional services
• Medium Exposure: Manufacturing, wholesale and retail
• Low Exposure: Agriculture, construction, personal services

Skill-Biased Technological Change (SBTC)

AI replaces medium-skill routine jobs, creates high-skill (AI development) and low-skill (manual service) jobs, leading to labor market hollowing

Productivity Paradox

Negative productivity effects stem from:

• Short-term adjustment costs of technology adoption
• Lagging organizational changes
• Learning curve effects
• Difficulty quantifying quality improvements

Job Creation Exceeds Substitution

In India's context, new jobs (AI services, data annotation) and transformation needs of traditional industries offset substitution effects; demographic dividend provides sufficient labor

Academic Contributions

1. Supplements evidence on AI's impact in developing countries
2. 1三年 span captures the full process of technology diffusion
3. Multi-dimensional assessment of employment, productivity, and quality
4. Strict causal inference ensures credibility

Policy Implications

1. Reform education system to strengthen AI-related skill training
2. Establish social security mechanisms to support affected workers
3. Guide industries to transform towards AI-complementary directions
4. Balance innovation and social protection in regulatory frameworks

Limitations

1. Industry-level data cannot reflect firm/individual heterogeneity
2. Reliance on existing indices may miss some AI applications
3. Causal mechanisms need more verification
4. Applicability of India's experience to other developing countries needs testing

Reproducibility Materials

• Stata Code: final_results.do (regression analysis)
• Python Code: results.ipynb (visualization)
• Dataset: All data used in analysis
• Output: plots/ (charts), results/ (regression results)

Analysis Steps

1. Descriptive statistics: Basic data characteristics
2. Regression analysis: Benchmark model and robustness tests
3. Event study: Visualization of dynamic effects
4. Heterogeneity analysis: Industry group regression