Automation is often promoted as a way to relieve dull, dirty and dangerous work. Using 21 O*NET items, I examined AI exposure across 754 U.S. occupations. Higher exposure correlates with lower condition scores under both a large language model (LLM) metric and the AI Occupational Exposure (AIOE) index, whether occupations are weighted equally or by employment. After adjusting for physical demands, required education, training, experience and wages, the overall association weakens; the adjusted confidence interval includes zero for LLM exposure but remains negative for AIOE. Separating repetition from restricted discretion changes the meaning conveyed by the combined dullness score. These comparisons demonstrate that item groupings and occupational characteristics shape how exposure estimates are interpreted. Occupational profiles provide a starting point for selecting contrasting research settings. Achieving desired automation or realized relief requires evidence about workers' preferences and specific interventions.
Review