Skip to content

Research & evidence

The source register.

57 curated sources plus 121 O*NET occupation records. Each citation on the site links here. Links are checked automatically and DOIs are resolved through Crossref.

Statistical dataset · 3

  • REF bls-emp

    Employment Projections ↗

    US Bureau of Labor Statistics · annual

    Ten-year projections of US employment by occupation and industry, including fastest-declining and fastest-growing occupations.

    Highaccessed 2026-11-177 citationslink OK (200)

  • REF ipums-usa

    IPUMS USA: Census microdata 1850–present ↗

    IPUMS, University of Minnesota · continuously updated

    Harmonised US census microdata including the 1920 and 1930 censuses, the basis for historical occupational counts.

    Highaccessed 2026-11-172 citationslink OK (200)

  • REF onet

    O*NET OnLine ↗

    US Department of Labor, Employment and Training Administration · continuously updated

    Detailed descriptors for roughly 900+ US occupations: tasks, tools and technology, skills, work context.

    Highaccessed 2026-11-17115 citationslink OK (200)

Government report · 4

  • REF bls-ooh

    Occupational Outlook Handbook ↗

    US Bureau of Labor Statistics · continuously updated

    Official US descriptions of what workers do, pay, and ten-year employment projections by occupation.

    Highaccessed 2026-11-1720 citationslink OK (200)

  • REF usda-ers-2005

    The 20th Century Transformation of U.S. Agriculture and Farm Policy ↗

    Carolyn Dimitri; Anne Effland; Neilson Conklin · USDA Economic Research Service, EIB-3 · 2005-06

    Documents the fall in agriculture's share of the US workforce from about 41% in 1900 to under 2% in 2000, driven by mechanisation, chemicals and specialisation.

    Highaccessed 2026-11-176 citationslink OK (200)

  • REF uk-dfe-2023

    The impact of AI on UK jobs and training ↗

    UK Department for Education · 2023-11

    Finds professional occupations — especially finance, law and business management — are more exposed to AI, particularly large language models, than manual occupations.

    Moderateaccessed 2026-11-174 citationslink OK (200)

  • REF ons-2019-automation

    The probability of automation in England: 2011 and 2017 ↗

    Office for National Statistics · 2019-03

    Estimated around 1.5 million jobs in England in 2017 at high risk of some of their duties being automated, with waiters, shelf fillers and elementary sales roles at highest risk.

    Moderateaccessed 2026-11-173 citationslink OK (200)

International organisation · 7

  • REF imf-2024-genai

    Gen-AI: Artificial Intelligence and the Future of Work ↗

    Mauro Cazzaniga et al. · International Monetary Fund, Staff Discussion Note SDN/2024/001 · 2024-01

    Estimates almost 40% of global employment is exposed to AI, rising to about 60% in advanced economies, with roughly half of exposed jobs potentially benefiting from complementarity.

    Moderateaccessed 2026-11-172 citationslink OK (200)

  • REF ilo-2023-genai

    Generative AI and Jobs: A global analysis of potential effects on job quantity and quality ↗

    Paweł Gmyrek; Janine Berg; David Bescond · International Labour Organization, ILO Working Paper 96 · 2023-08

    Finds clerical work has by far the highest exposure to generative AI and that the dominant effect for most occupations is more likely augmentation than automation; automation exposure is much higher in high-income than low-income countries.

    Moderateaccessed 2026-11-178 citationsdoi 10.54394/FHEM8239DOI verified (Crossref)

  • REF who-hrh-2030

    Global strategy on human resources for health: Workforce 2030 ↗

    World Health Organization · 2016

    Sets out the WHO strategy to address projected global shortages of health workers by 2030.

    Highaccessed 2026-11-175 citationslink OK (200)

  • REF unesco-2023-genai-ed

    Guidance for generative AI in education and research ↗

    UNESCO · 2023-09

    Global guidance on the use of generative AI in education, including a recommended minimum age and a human-centred approach.

    Highaccessed 2026-11-174 citationslink OK (200)

  • REF oecd-eo-2023

    OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market ↗

    OECD Publishing · 2023-07

    Reviews evidence on AI and work; finds little evidence so far of significant negative employment effects but substantial effects on job quality and task content, and urges action on skills and worker protection.

    Highaccessed 2026-11-172 citationsdoi 10.1787/08785bba-enpublisher blocks automated checks

  • REF worldbank-wdr-2019

    World Development Report 2019: The Changing Nature of Work ↗

    World Bank · 2019

    Examines how technology reshapes demand for skills globally, emphasising human capital investment and social protection, with particular attention to developing economies.

    Highaccessed 2026-11-173 citationslink OK (200)

  • REF iea-wee-2024

    World Energy Employment 2024 ↗

    International Energy Agency · 2024-11

    Tracks global energy-sector employment; clean energy now accounts for more than half of energy employment, with skills shortages for electricians and technicians.

    Highaccessed 2026-11-1713 citationslink OK (200)

Peer-reviewed · 12

  • REF vaswani-2017

    Attention Is All You Need ↗

    Ashish Vaswani et al. · arXiv:1706.03762 (NeurIPS 2017) · 2017-06

    Introduced the Transformer architecture that underlies today's large language models.

    Highaccessed 2026-11-172 citationslink OK (200)

  • REF noy-zhang-2023

    Experimental evidence on the productivity effects of generative artificial intelligence ↗

    Shakked Noy; Whitney Zhang · Science, 381(6654), 187–192 · 2023-07

    In a pre-registered experiment with college-educated professionals on writing tasks, ChatGPT reduced time taken by about 40% and raised output quality by about 18%.

    How not to read it: Short, well-defined writing tasks in an experimental setting.

    Highaccessed 2026-11-175 citationsdoi 10.1126/science.adh2586DOI verified (Crossref)

  • REF jumper-2021

    Highly accurate protein structure prediction with AlphaFold ↗

    John Jumper et al. · Nature, 596, 583–589 · 2021

    Describes AlphaFold 2, which predicted protein structures with accuracy competitive with experiment in CASP14.

    Highaccessed 2026-11-175 citationsdoi 10.1038/s41586-021-03819-2DOI verified (Crossref)

  • REF krizhevsky-2012

    ImageNet Classification with Deep Convolutional Neural Networks ↗

    Alex Krizhevsky; Ilya Sutskever; Geoffrey E. Hinton · Advances in Neural Information Processing Systems 25 (NeurIPS 2012) · 2012

    AlexNet: a deep convolutional network that sharply reduced ImageNet error, catalysing the deep learning era.

    Highaccessed 2026-11-173 citationslink OK (200)

  • REF autor-2024-newwork

    New Frontiers: The Origins and Content of New Work, 1940–2018 ↗

    David Autor; Caroline Chin; Anna Salomons; Bryan Seegmiller · The Quarterly Journal of Economics, 139(3) · 2024

    Finds the majority of US employment in 2018 was in job specialties that did not exist in 1940; new work is created by augmentation innovations and demand shifts.

    Highaccessed 2026-11-174 citationsdoi 10.1093/qje/qjae008DOI verified (Crossref)

  • REF rodrik-2016

    Premature deindustrialization ↗

    Dani Rodrik · Journal of Economic Growth, 21, 1–33 · 2016

    Documents that developing countries are reaching peak manufacturing employment earlier and at lower income levels than early industrialisers, narrowing a traditional development path.

    Highaccessed 2026-11-172 citationsdoi 10.1007/s10887-015-9122-3DOI verified (Crossref)

  • REF acemoglu-restrepo-2020

    Robots and Jobs: Evidence from US Labor Markets ↗

    Daron Acemoglu; Pascual Restrepo · Journal of Political Economy, 128(6), 2188–2244 · 2020

    Estimates that one more industrial robot per thousand workers reduced the US employment-to-population ratio by about 0.2 percentage points and wages by about 0.42% in affected commuting zones (1990–2007).

    How not to read it: Local labour-market effects of industrial robots, concentrated in manufacturing; not generalisable to AI.

    Highaccessed 2026-11-173 citationsdoi 10.1086/705716DOI verified (Crossref)

  • REF acemoglu-restrepo-2022

    Tasks, Automation, and the Rise in U.S. Wage Inequality ↗

    Daron Acemoglu; Pascual Restrepo · Econometrica, 90(5), 1973–2016 · 2022

    Attributes a large share of changes in the US wage structure since 1980 to the relative wage declines of worker groups specialised in routine tasks displaced by automation.

    Highaccessed 2026-11-174 citationsdoi 10.3982/ECTA19815DOI verified (Crossref)

  • REF frey-osborne-2017

    The future of employment: How susceptible are jobs to computerisation? ↗

    Carl Benedikt Frey; Michael A. Osborne · Technological Forecasting and Social Change, 114, 254–280 · 2017

    Estimated that about 47% of US employment was in occupations at high risk of computerisation, using expert labels of 70 occupations extrapolated with O*NET task variables.

    How not to read it: Occupation-level technical feasibility, not a forecast of job losses or timing. Task-level re-analyses (Arntz et al. 2016; Nedelkoska & Quintini 2018) find much lower shares.

    Moderateaccessed 2026-11-172 citationsdoi 10.1016/j.techfore.2016.08.019DOI verified (Crossref)

  • REF alm-2003

    The Skill Content of Recent Technological Change: An Empirical Exploration ↗

    David H. Autor; Frank Levy; Richard J. Murnane · The Quarterly Journal of Economics, 118(4), 1279–1333 · 2003

    Introduced the routine / non-routine task framework: computers substitute for routine cognitive and manual tasks and complement non-routine analytic and interactive tasks.

    How not to read it: Framework underlying this site's task kinds; it describes computerisation up to the 1990s.

    Highaccessed 2026-11-173 citationsdoi 10.1162/003355303322552801publisher blocks automated checks

  • REF paul-moser-2009

    Unemployment impairs mental health: Meta-analyses ↗

    Karsten I. Paul; Klaus Moser · Journal of Vocational Behavior, 74(3), 264–282 · 2009

    Meta-analysis showing unemployment is associated with worse mental health, with evidence that the relationship is partly causal.

    Highaccessed 2026-11-171 citationsdoi 10.1016/j.jvb.2009.01.001DOI verified (Crossref)

  • REF autor-2015

    Why Are There Still So Many Jobs? The History and Future of Workplace Automation ↗

    David H. Autor · Journal of Economic Perspectives, 29(3), 3–30 · 2015

    Argues automation complements labour as well as substituting for it; discusses how US bank teller employment rose modestly after ATMs spread, as branches multiplied and the role shifted toward relationship banking.

    Highaccessed 2026-11-172 citationsdoi 10.1257/jep.29.3.3publisher blocks automated checks

Working paper · 8

  • REF feigenbaum-gross-2024

    Answering the Call of Automation: How the Labor Market Adjusted to Mechanizing Telephone Operation ↗

    James Feigenbaum; Daniel P. Gross · NBER Working Paper 28061 (published in QJE, 2024) · 2020

    Studies AT&T's conversion to mechanical switching (1920–1940): telephone operation, a leading job for young American women, was automated; incumbent operators were more likely to leave work, while future cohorts found jobs elsewhere, especially in other clerical work.

    Highaccessed 2026-11-177 citationslink OK (200)

  • REF nedelkoska-quintini-2018

    Automation, skills use and training ↗

    Ljubica Nedelkoska; Glenda Quintini · OECD Social, Employment and Migration Working Papers No. 202 · 2018

    Using PIAAC task data for 32 countries, about 14% of jobs are at high risk of automation and a further 32% could face significant change in how they are carried out.

    How not to read it: Pre-dates large language models, which shifted exposure toward higher-skilled cognitive work.

    Moderateaccessed 2026-11-173 citationsdoi 10.1787/2e2f4eea-enDOI verified (Crossref)

  • REF brynjolfsson-2023

    Generative AI at Work ↗

    Erik Brynjolfsson; Danielle Li; Lindsey R. Raymond · NBER Working Paper 31161 (published in QJE, 2025) · 2023-04

    In a field deployment with 5,179 customer-support agents, an AI assistant raised issues resolved per hour by 14% on average and 34% for novice and low-skilled workers.

    Highaccessed 2026-11-176 citationslink OK (200)

  • REF eloundou-2023

    GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models ↗

    Tyna Eloundou; Sam Manning; Pamela Mishkin; Daniel Rock · arXiv:2303.10130 (later published in Science, 2024) · 2023-03

    Estimates around 80% of the US workforce could have at least 10% of their tasks affected by LLMs, and about 19% at least half of their tasks; exposure is higher for higher-wage occupations.

    How not to read it: Exposure measures where LLMs could reduce task time, not whether jobs are lost. Authors include OpenAI staff.

    Moderateaccessed 2026-11-1716 citationslink OK (200)

  • REF vivalt-2024-ubi

    The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States ↗

    Eva Vivalt; Elizabeth Rhodes; Alexander W. Bartik; David E. Broockman; Sarah Miller · NBER Working Paper 32719 · 2024-07

    A randomised trial giving $1,000 per month for three years found a modest reduction in recipients' labour supply (around 1.3–1.4 hours per week) and no significant improvement in measured job quality.

    Highaccessed 2026-11-172 citationslink OK (200)

  • REF peng-2023

    The Impact of AI on Developer Productivity: Evidence from GitHub Copilot ↗

    Sida Peng; Eirini Kalliamvakou; Peter Cihon; Mert Demirer · arXiv:2302.06590 · 2023-02

    In a controlled experiment, developers with GitHub Copilot completed a JavaScript HTTP-server task 55.8% faster than the control group.

    How not to read it: Single, well-specified task; authors affiliated with GitHub/Microsoft.

    Moderateaccessed 2026-11-174 citationslink OK (200)

  • REF arntz-2016

    The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis ↗

    Melanie Arntz; Terry Gregory; Ulrich Zierahn · OECD Social, Employment and Migration Working Papers No. 189 · 2016

    A task-based approach that accounts for variation in tasks within occupations finds on average about 9% of jobs across OECD countries at high risk of automation.

    How not to read it: Pre-dates generative AI; measures technical automatability rather than realised displacement.

    Moderateaccessed 2026-11-172 citationsdoi 10.1787/5jlz9h56dvq7-enpublisher blocks automated checks

  • REF acemoglu-2024

    The Simple Macroeconomics of AI ↗

    Daron Acemoglu · NBER Working Paper 32487 · 2024

    Using task-based estimates, argues AI's total factor productivity gains over the next decade are likely to be modest (on the order of well under 1% in total).

    How not to read it: Deliberately conservative; other economists project substantially larger effects. A counterpoint, not a consensus.

    Moderateaccessed 2026-11-171 citationslink OK (200)

Book · 2

Industry report · 6

  • REF mgi-2017

    Jobs lost, jobs gained: Workforce transitions in a time of automation ↗

    McKinsey Global Institute · 2017-12

    Scenario analysis suggesting between roughly 400 and 800 million workers globally could be displaced by automation by 2030, with wide ranges by adoption speed.

    How not to read it: Consultancy scenario modelling with very wide ranges.

    Lowaccessed 2026-11-173 citationslink check failed (timeout)

  • REF wef-2023

    The Future of Jobs Report 2023 ↗

    World Economic Forum · 2023-04

    Employers expected 69 million jobs created and 83 million eliminated across surveyed firms over 2023–2027; clerical and secretarial roles were expected to decline fastest.

    How not to read it: Employer survey; earlier and later editions differ markedly, illustrating forecast uncertainty.

    Moderateaccessed 2026-11-172 citationspublisher blocks automated checks

  • REF wef-2025

    The Future of Jobs Report 2025 ↗

    World Economic Forum · 2025-01

    Employers surveyed expect 170 million jobs to be created and 92 million displaced between 2025 and 2030 (net +78 million), and expect 39% of workers' core skills to change by 2030.

    How not to read it: Based on employer expectations surveys, not observed outcomes.

    Moderateaccessed 2026-11-176 citationspublisher blocks automated checks

  • REF goldman-2023

    The Potentially Large Effects of Artificial Intelligence on Economic Growth ↗

    Joseph Briggs; Devesh Kodnani · Goldman Sachs Global Economics Analyst · 2023-03

    Estimates that generative AI could expose the equivalent of 300 million full-time jobs globally to automation and that roughly two-thirds of US occupations are exposed to some degree.

    How not to read it: Investment-bank research note; exposure is not displacement.

    Lowaccessed 2026-11-172 citationspublisher blocks automated checks

  • REF autonomy-4dw-2023

    The results are in: The UK's four-day week pilot ↗

    Autonomy (with researchers from Cambridge, Boston College) · 2023-02

    Reports that most of the 61 organisations in a six-month UK four-day-week trial continued the policy afterwards.

    How not to read it: Self-selected participating firms; advocacy organisation; short duration.

    Lowaccessed 2026-11-173 citationslink OK (200)

  • REF ifr-world-robotics

    World Robotics — Industrial Robots (annual report and press releases) ↗

    International Federation of Robotics · 2024

    Industry statistics on robot installations; IFR reported more than 4 million industrial robots operating in factories worldwide in its 2024 edition.

    How not to read it: Industry association data; counts industrial (not service) robots.

    Highaccessed 2026-11-179 citationslink OK (200)

Classification · 2

Primary historical · 3

  • REF turing-1950

    Computing Machinery and Intelligence ↗

    Alan M. Turing · Mind, 59(236), 433–460 · 1950

    Posed the question 'Can machines think?' and proposed the imitation game.

    Highaccessed 2026-11-173 citationsdoi 10.1093/mind/LIX.236.433DOI verified (Crossref)

  • REF keynes-1930

    Economic Possibilities for our Grandchildren ↗

    John Maynard Keynes · Essay (reprinted in Essays in Persuasion, 1931) · 1930

    Predicted that within a century productivity growth could solve 'the economic problem', allowing something like a fifteen-hour working week.

    Highaccessed 2026-11-174 citationslink OK (200)

  • REF sutherland-1963

    Sketchpad: A man-machine graphical communication system (electronic edition) ↗

    Ivan E. Sutherland · University of Cambridge Computer Laboratory, Technical Report UCAM-CL-TR-574 (originally MIT PhD thesis, 1963) · 1963 (2003 edition)

    The thesis describing Sketchpad, the interactive graphics system widely regarded as the ancestor of computer-aided design.

    Highaccessed 2026-11-177 citationslink OK (200)

Institutional record · 8

  • REF ai-index-2025

    AI Index Report 2025 ↗

    Stanford Institute for Human-Centered Artificial Intelligence · 2025-04

    Annual compilation of data on AI research, technical performance, investment, adoption and policy.

    Highaccessed 2026-11-177 citationslink OK (200)

  • REF rusi-2022-ukraine

    Preliminary Lessons in Conventional Warfighting from Russia's Invasion of Ukraine: February–July 2022 ↗

    Jack Watling; Nick Reynolds · Royal United Services Institute · 2022-11

    Field-research-based assessment highlighting the pervasive role of UAVs for reconnaissance and artillery targeting, and the continued centrality of infantry, artillery and mass.

    Moderateaccessed 2026-11-177 citationslink OK (200)

  • REF bbc-atm-2007

    The man who invented the cash machine ↗

    BBC News · 2007-06

    Reports the 40th anniversary of the first cash machine, installed at Barclays in Enfield, north London, on 27 June 1967.

    Moderateaccessed 2026-11-173 citationslink OK (200)

  • REF nobel-chemistry-2024

    The Nobel Prize in Chemistry 2024 ↗

    NobelPrize.org · 2024

    Awarded to David Baker for computational protein design and to Demis Hassabis and John Jumper for protein structure prediction (AlphaFold).

    Highaccessed 2026-11-174 citationslink OK (200)

  • REF nobel-physics-1956

    The Nobel Prize in Physics 1956 ↗

    NobelPrize.org · 1956

    Awarded to Shockley, Bardeen and Brattain for research on semiconductors and their discovery of the transistor effect (Bell Labs, 1947).

    Highaccessed 2026-11-172 citationslink OK (200)

  • REF cern-web

    The World Wide Web project (first website, restored by CERN) ↗

    CERN · 1991

    Tim Berners-Lee proposed the World Wide Web at CERN in 1989; CERN placed the software in the public domain in 1993.

    Highaccessed 2026-11-173 citationslink OK (200)

  • REF chm-timeline

    Timeline of Computer History ↗

    Computer History Museum · continuously updated

    Year-by-year record of computing milestones, including ENIAC (1946), Unimate (1961), the Intel 4004 (1971) and the IBM PC (1981).

    Highaccessed 2026-11-1744 citationslink OK (200)

  • REF bessen-2015

    Toil and Technology ↗

    James Bessen · IMF Finance & Development, 52(1) · 2015-03

    Argues automation often grows employment in affected occupations by lowering costs and expanding demand, citing weavers in the 19th century and bank tellers after ATMs.

    Moderateaccessed 2026-11-174 citationslink OK (200)

Policy document · 2

  • REF dod-3000-09

    DoD Directive 3000.09: Autonomy in Weapon Systems (updated 2023) ↗

    US Department of Defense · 2023-01

    Requires that autonomous and semi-autonomous weapon systems allow commanders and operators to exercise appropriate levels of human judgement over the use of force.

    Highaccessed 2026-11-175 citationspublisher blocks automated checks

  • REF icrc-2021-aws

    ICRC position on autonomous weapon systems ↗

    International Committee of the Red Cross · 2021-05

    Recommends new legally binding rules prohibiting unpredictable autonomous weapons and those designed to target humans, and restricting others, to retain human control over the use of force.

    Highaccessed 2026-11-172 citationslink OK (200)

O*NET occupation records · 121