O*NET (Occupational Information Network) is a US Department of Labor database that catalogues hundreds of occupations with detailed descriptors: the tasks performed, skills required, knowledge areas, work activities, and work context. Researchers have used this rich task-level data to build automation scores — quantitative estimates of how automatable each occupation is.
The most influential early work was Frey and Osborne's 2013 paper "The Future of Employment," which used O*NET task data to classify 702 occupations by automation probability. Subsequent researchers (Autor, Levy, Murnane; Arntz, Gregory, Zierahn) refined these methods by decomposing occupations into individual tasks rather than treating the whole role as a unit — acknowledging that most jobs contain both automatable and non-automatable tasks.
The core methodology rates each O*NET work activity on dimensions like: requiring finger dexterity, social perceptiveness, originality, and manual dexterity. High scores on cognitive-routine and manual-routine activities correlate with high automation probability. High scores on social, creative, and unstructured physical activities correlate with low automation probability.
While O*NET scores are a useful research baseline, they have limitations for individual career assessment: they reflect average task distributions across all workers in a role (not your specific version of the job), and they were built before the LLM revolution dramatically expanded what AI can do to cognitive tasks.