I agree that building an interactive model of the supply chain’s labor intensity, and how it has evolved over time would be really impactful piece of work. A few resources that I would take a look at:
This report from Goldman Sachs from 2023 was a good early first pass at estimating the share of tasks within occupations potentially impacted by AI. The actual data tables they referenced aren’t (to my knowledge) publicly available, but their methodology should be replicable with some effort. This McKinsey report published later that year uses a similar methodology and considers the historical trajectory back to 2016.
If you want to going back further in time, the BLS published a helpful guide for mapping O*NET data going back to 1998, which tracks tasks and required skills with occupations, though there are some limitations here. The BLS also released this new data product about skills last year that I haven’t had a chance to explore thoroughly yet.
In addition to uncertainty about future AI capabilities, there could be considerable variation in how important skills are within each occupation. If AI only partially automates or de-skills an occupation, the extent to which other skills remain a bottleneck is an important question where estimates may be imprecise.
Forward projections may be more helpful to do by industry. The BLS helpfully maintains industry-occupation matrices, but this adds another layer of complexity to the analysis.
I agree that building an interactive model of the supply chain’s labor intensity, and how it has evolved over time would be really impactful piece of work. A few resources that I would take a look at:
This report from Goldman Sachs from 2023 was a good early first pass at estimating the share of tasks within occupations potentially impacted by AI. The actual data tables they referenced aren’t (to my knowledge) publicly available, but their methodology should be replicable with some effort. This McKinsey report published later that year uses a similar methodology and considers the historical trajectory back to 2016.
If you want to going back further in time, the BLS published a helpful guide for mapping O*NET data going back to 1998, which tracks tasks and required skills with occupations, though there are some limitations here. The BLS also released this new data product about skills last year that I haven’t had a chance to explore thoroughly yet.
I’m currently working on making the historic occupational data I used for this analysis of occupational churn going back to 1870 publicly available, hopefully by the end of this month.
Some limitations to be aware of:
In addition to uncertainty about future AI capabilities, there could be considerable variation in how important skills are within each occupation. If AI only partially automates or de-skills an occupation, the extent to which other skills remain a bottleneck is an important question where estimates may be imprecise.
Forward projections may be more helpful to do by industry. The BLS helpfully maintains industry-occupation matrices, but this adds another layer of complexity to the analysis.