AI and Labor Markets: What We Know and Don't Know – Stanford Digital Economy Lab (digitaleconomy.stanford.edu)

Bharat Chandar’s state-of-knowledge piece synthesizes recent empirical work — including his “Canaries in the Coal Mine” paper — showing that early AI adoption has not produced large economy‑wide employment declines but appears to have a concentrated, meaningful impact on entry‑level workers. Multiple analyses using CPS, ADP and Revelio data find that employment fell most for 22–25‑year‑olds in AI‑exposed occupations (software development, customer service, clerical), with Chandar reporting a roughly 13% within‑firm drop in entry‑level hiring after the spread of LLMs. Corroborating studies in the US and UK support this pattern, while at least one Danish study finds no effect, highlighting geographic and measurement uncertainty. Technically, researchers used occupation‑level AI exposure indices (including an LLM‑usage measure from Anthropic) and ran robustness checks: excluding tech sectors, controlling for firm‑time effects, comparing college vs non‑college workers, and extending samples back to 2018. Validation is improving — e.g., Eloundou‑style exposure scores correlate strongly with Copilot usage — but causal identification remains the key open problem. Chandar calls for better firm‑level adoption data, quasi‑experimental or randomized adoption evidence (A/B tests), expanded public trackers, and integration of workflow‑level productivity studies to understand heterogeneous effects across ages, education levels, and occupations.