This report examines how the impact of artificial intelligence (AI) on work can be systematically measured and indexed within the O*NET System. Drawing on a review of 19 major studies published in recent years, the authors analyze the different methods researchers have used to assess AI’s impact on work, including measures of AI exposure, automation potential, augmentation potential, and real-world AI usage. A key finding is that most existing research relies heavily on O*NET data and typically evaluates AI’s influence on specific job tasks, worker knowledge and skills, or job vacancy information before aggregating those results to the occupational level. The report introduces a framework for organizing these approaches, helping distinguish between studies that assess AI based on estimating its capabilities to augment or automate work and those that examine how AI tools are actually being used in workplaces.
Building on this review, the report argues that current AI impact research often overlooks nearly a century of established job analysis and job performance research. Many existing approaches focus narrowly on tasks, potentially overstating AI’s overall effect on occupations by not considering modern perspectives of job performance such as contextual and adaptive performance behaviors. The authors also identify limitations in relying solely on expert judgments, job postings, or usage data as indicators of AI impact. Instead, they advocate for a more comprehensive, job-analysis-based perspective that considers both well-established determinants of job performance and the multi-faceted nature of job performance.
To address these gaps, the report recommends that O*NET consider developing a suite of potentially 16 AI impact indices organized around modern models of job performance. These indices would measure AI’s ability to exhibit required levels of knowledge or skill to perform work effectively or augment or automate various elements of task and contextual job performance. The proposed approach emphasizes transparency, scalability, and regular updating by leveraging large language models, O*NET’s existing occupational framework, and potentially expanded use of job vacancy data to produce consistent and sustainable measures of AI impact. The result would be a richer, more nuanced set of indicators that could help policymakers, workforce development professionals, employers, researchers, and workers better understand how AI is reshaping skills and occupations across the economy.
