This report describes a method for producing O*NET’s basic interest ratings—detailed indicators of the types of work activities people are likely to enjoy in each occupation. The project introduced a hybrid AI–expert approach. Large language models (LLMs) were used to generate initial ratings across all occupations, and a smaller set of human expert ratings was collected to validate and improve the results. The AI methods were then refined and equated to better align with expert judgment and applied at scale.
Key takeaways:
- The hybrid approach is efficient and scalable, enabling coverage of all occupations with less manual effort.
- AI-generated ratings are highly reliable and generally align well with expert evaluations.
- Initial AI results showed a slight tendency to overestimate, which was reduced through refinement and equating.
- The final method produces high-quality, expert-aligned ratings suitable for use in O*NET.
Overall, this work demonstrates that combining AI with targeted expert input can modernize and streamline how occupational data are developed and maintained.
