Stanford Digital Economy Lab

This is so cool. I have lived and breathed O*NET tasks and occupations over the last few years while I researched AI's job impacts during my time at Brookings. It is the best data we have... ....but oy, was it ready for a refresh. I love this TaskNet project and I hope it can be scaled, especially for the occupations that are likely to be impacted by AI first. Kudos to @mercor @onet_center @DigEconLab @andreas_h0wpt @robinsonaisaac @erikbryn @jennjwang @Diyi_Yang @stanfordnlp @StanfordAILab! One thing I would love to see emerge from this work is a clearer delineation of task bundles by seniority. One of the big limitations of O*NET is it treats occupations as a monolith: there is just one "lawyer" occupation, for ex. But obviously the daily tasks of a newly minted junior lawyer look almost nothing like those of a law firm partner. I think this is why the risks to early career roles were originally missed, until the Stanford "Canaries" paper showed it in the ADP data last year. Early studies at the time suggested that AI might be a great "leveler," boosting the inexperienced the most. After ChatGPT was launched, I was hearing something different when I interviewed workers, from Hollywood writers to insurance underwriters and illustrators. They all told me AI put junior roles most at risk. But it was hard to systematically analyze this experience-adjusted exposure risk from O*NET as only a tiny number of occupations split junior and senior roles into distinct jobs. (e.g., market research analyst vs marketing manager.) I ended up using my subjective judgement *within* occupations to rank tasks that required more or less experience. For instance, in this essay I wrote for Bloomberg two years ago on "How AI Could Break the Career Ladder," I used the example of O*NET's lawyer tasks and broke out what I thought were the junior vs senior tasks to illustrate the AI exposure gap: bloomberg.com/news/articles/…. Without better data, it was hard to do much more than that. The TaskNet approach of using AI to interview workers is a great way to fill in these gaps systematically and enable more granular insights. I am excited to see where this goes.