Predicting Job Automation: What have we observed?

Abstract

This research considers the ability to predict job automation based on two models. The first is a job model developed by Frey and Osborn and published in 2017. With 12000+ citations, that article appears to be the most highly cited academic article on predicting job automation. The second is a job automation model developed by Sampson and published in 2021. Coincidentally, both models were developed using the same U.S. Department of Labor database called O*Net, although using different data from different years. We use historical and current O*Net data to see how each model does in predicting observed changes in job automation over a wide range of jobs. A surprising finding is a negative correlation between degrees of automation for various jobs and changes in the degree of automation over the subsequent decade. This analysis leads to interesting theories about how job automation can be predicted, including an AI explanation.