The Impact of AI on the Labour Market: Technical Annex

These studies are novel in that they are using ChatGPT as a form of the wisdom of the crowds – on the basis that the LLM was trained on a large volume of data that reflects the accumulated knowledge of the world’s population and provides a probabilistic assessment of the most likely outcome based on those data. [...] Our analysis of the “management analyst” and “project-management specialist” professions, both of which involve tasks closely related to those in the study, estimated that the use of AI would introduce a time saving of 24.7 per cent and 19.5 per cent respectively – very close to the empirical results from the study. [...] This process of experimentation eventually leading to deployment is reflected in the history of technological adoption which tells us that it can take a long time for the impact of new inventions to be fully felt and to appear in productivity and growth statistics.24 Firms need to first experiment to work out how to best reorganise their production to take advantage of new technologies. [...] But the remainder would 39 That is to say, the number of redundancies in a given year is in proportion to the share of the rollout that occurs in that year. [...] Observing past patterns in the data – specifically, comparing US productivity growth over time from the US Bureau of Labor Statistics and estimates of new task creation from Acemoglu and Restrepo (2019) – we find that in the past, increases in labour demand from the introduction of new tasks tend to occur around a decade after productivity increases, and that in the past a 1 per cent increase in p.