国际劳工组织-ChatGPT等生成式AI能增强就业，不会取代岗位（英）-2023.8-55页_1mb

Summary of "Generative AI and Jobs: A global analysis of potential effects on job quantity and quality"

Authors: Paweł Gmyrek, Janine Berg, David Bescond (ILO Research Department)
Date: August 2023 (ILO Working Paper No. 96)

Abstract

This study analyzes the potential exposure of global occupations to Generative AI, particularly GPT models, and its implications for job quantity and quality. Using GPT-4 to estimate task-level automation and augmentation potential, the research finds that while many tasks could be affected by GenAI, augmenting work (automating routine tasks while preserving others) is predominant over full automation. Clerical work is highly exposed, but impacts vary by income, gender, and geography.

Key Findings

Occupation Exposure

• High Exposure: Clerical support work (24% of tasks highly exposed, 58% medium exposure). Other groups have low exposure (1–4% highly exposed tasks) due to the nature of tasks (e.g., routine vs. cognitive).
• Augmentation vs. Automation: GenAI primarily aids tasks, with augmentation (not full automation) being the most likely outcome, affecting 10–13% of employment globally depending on income.
• Income Disparities: High-income countries have higher exposure (up to 5.5% employment potentially automated), while low-income countries have lower exposure (0.4%). The digital divide constrains GenAI use in lower-income countries.
• Gender Impacts: Automation disproportionately affects women's employment (up to 7.9% in high-income countries vs. 2.9% for men), exacerbating gender inequalities in labor markets.

Methodology and Scope

• Based on ISCO-08 standard occupations and tasks, using GPT-4 to score exposure.
• Focuses on potential exposure as an "upper bound" due to technological optimism in models, without accounting for real-world barriers like infrastructure or skills gaps.
• Provides global estimates using ILO employment data, highlighting uncertainties in country-level coverage.

Policy Recommendations

• Manage Transitions: Strengthen social dialogue and workplace consultation to address job losses and ensure fair transitions.
• Job Quality: Regulate GenAI's workplace integration to prevent adverse effects on autonomy and working conditions, and promote decent work.
• Address Digital Divide: Invest in infrastructure and skills training in low-income countries to leverage GenAI's benefits equitably.
• Support Affected Workers: Implement policies for retraining and social protection to mitigate negative impacts of technological change.

Conclusion

Generative AI represents a significant shift toward augmenting work, with substantial variations across occupations, genders, and income levels. While not leading to mass job losses in the short term, its adoption necessitates proactive policies to foster orderly, fair, and inclusive labor market transitions, addressing potential inequalities and harnessing technology for sustainable development. The report underscores that GenAI's societal impacts depend largely on how it is governed and deployed.

Copyright: Licensed under Creative Commons Attribution 4.0 International.