This desk review toolkit brings together recent methodologies used to analyse AI’s effects on employment, wages, and productivity. Shift from Occupation-Level to Task-Level Analysis: Recent methodologies increasingly focus on tasks rather than entire occupations, recognising the heterogeneity within jobs. For example, Felten et al. (2023) and Eloundou et al. (2023) use AI exposure indices to measure task-specific impacts.Integration of Advanced Technologies: Methods now incorporate Natural Language Processing (NLP) (e.g., BERT, LSTM) and LLMs (e.g., GPT-4) to analyse job descriptions and predict automation risks (Xu et al., 2025; Hampole et al., 2025).Scenario Planning and Policy Focus: Think tanks like TBI and IPPR, and organisations like IMF emphasise scenario-based modelling to inform policy, highlighting the need for reskilling and labour market reforms (TBI, 2024; IPPR, 2024, Korinek, 2023). They assume different ‘initial conditions’ in adoption to estimate different conclusions in employment, wages, productivity ect. Studies consistently find that AI’s impact varies by skill level, with low-skilled workers facing higher displacement risks and high-skilled workers benefiting from augmentation (Brynjolfsson et al., 2023; Chen et al., 2024). Many methodologies struggle with static assumptions, lack of causal evidence, and overreliance on theoretical models. For instance, Acemoglu & Restrepo (2022) assume fixed comparative advantage, while Webb (2020) ignores adaptation. As of year 2025, conclusions of some of the pre-2020 research is already redundant in terms of what professions will remain or not. The pre-2020 conclusion that creative and intellectual occupations will remain in high demand, has already been disproven.
Papagiannaki et al. (Fri,) studied this question.