Home » Year » 2021 » Predicting the Work Task Replacement Effects of the Adoption of Machine Learning Technology

Predicting the Work Task Replacement Effects of the Adoption of Machine Learning Technology

by Shreya Hurli

This paper develops a methodology to attempt to predict which tasks in the workforce will be resistant to the replacement of labor by machine learning technology in the near future given current technology and technology adoption trends. Tasks are individual activities completed as parts of a job. Prior research in the field suggests that characteristics of tasks (non-roteness, creativity, analysis/cognitive work) that make them harder for machine learning technology to complete are good predictors of whether those tasks will be resistant to replacement in the workforce. This study utilizes O*NET (Occupational Information Network) task description and education data from October 2015 to August 2020 and Bureau of Labor Statistics salary data to use task characteristics to predict tasks’ resistance to replacement. Normalized scores, average salaries, and average worker education levels are calculated to quantify the relative presence or absence of non-roteness, creativity, and cognitive work in a task. This paper then uses the calculated scores, salary, and education data, as well as a number of interaction terms as inputs to a support vector machine (SVM) model to predict which tasks will be resistant to decline in their shares of workplace tasks weighted by the jobs under which the tasks fall. Using task characteristics, the SVM predicts that just approximately 39% of tasks are likely to be resistant to replacement. These tasks tend to be highly non-deterministic (very non-rote, analytical/cognitive, and/or creative) in nature.

View Thesis

Advisors: Professor David Berger, Professor Michelle Connolly | JEL Codes: J23, J24, O33


Undergraduate Program Assistant
Matthew Eggleston

Director of the Honors Program
Michelle P. Connolly