Popular AI exposure scores are useful but incomplete—they don't capture how AI adoption actually happens, who really benefits or loses, or how impacts change over time.
This paper examines AI exposure scores—metrics measuring how much AI can assist with job tasks—that have become central to workforce policy debates. The authors show these static scores have significant limitations when applied to real-world policy questions, and identify a critical gap: researchers keep using outdated scores while ignoring newer methodological improvements.