OCCUPATIONAL AI AUTOMATABILITY, WAGES AND EMPLOYMENT: AN ASSESSMENT USING THE AUTOMATION SPECTRUM AND GOVERNED PROMPTING (CREATE-R)

Abstract

This study aimed to 1) assess the level and distribution of occupational AI automatability across occupations and 2) analyze its relationships with wages and employment. A quantitative design was applied to secondary data from the U.S. Bureau of Labor Statistics for May 2024, covering 805 detailed occupations with complete data. Automatability was assessed by GPT-4o-mini, Claude Haiku 4.5, and Gemini 2.5 Flash operating under the CREATE-R framework, and the three model scores were averaged. The averaged score represented the estimated percentage of an occupation's core tasks that current AI could perform or meaningfully automate at the occupational level. Analyses included descriptive statistics, Pearson correlations, intraclass correlation coefficients, convergent validation against an external index, and linear regressions with log-transformed wage and employment measures. Results showed high correlations across model scores and good absolute agreement for the averaged score (ICC(A,k) = 0.827). The averaged score was strongly associated with the AI Occupational Exposure index (r = 0.727, n = 664), supporting convergent validity. However, this association should not be interpreted as direct expert validation or as evidence that labor displacement had actually occurred in labor markets. Mean occupational automatability was 32.1% and increased to 35.0% when weighted by employment. Overall, 28.7% of employment was in occupations with scores of 50 or higher. Clerical and repetitive text-based work received high scores, whereas physical and manual work received low scores. Automatability showed weak but positive and significant relationships with log median wage (r = 0.234, p < .001) and log employment (r = 0.134, p < .001). This pattern differs from earlier forms of automation, which more often affected low- to middle-wage occupations. The findings suggest that workforce policy should prioritize the development of skills that complement AI and provide targeted transition support to vulnerable low-wage clerical workers.