Dull, Dirty, Dangerous? Working Conditions and the Limits of Occupational AI Exposure

TL;DR

The study investigates how occupational characteristics like dullness, dirtiness, and danger correlate with AI exposure across 754 U.S. occupations using O*NET data and two AI exposure measures. After adjusting for factors like physical demands and wages, the link between poor working conditions and AI exposure weakens, with confidence intervals including zero for one measure and remaining negative for the other.

Why It Matters

Understanding how AI exposure varies with job quality helps guide equitable automation policies. The findings caution against assuming AI will automatically alleviate undesirable work without considering worker preferences and job-specific contexts. This informs better design of automation strategies and targeted interventions.

Key Findings

• Higher AI exposure correlates with lower scores on dull, dirty, and dangerous work conditions using both LLM-based and AIOE measures.
• Adjusting for physical demands, education, experience, and wages diminishes the observed association, with the LLM measure's confidence interval including zero.
• Separating repetition from restricted discretion reveals that the combined 'dull' score captures different aspects depending on how items are grouped.