Creation, validation, obsolescence: observed evidence of AI-driven labor market displacement, 2020-2025
C · Weak signal Peer-reviewed PRISMA systematic review by Nassim Dehouche (Mahidol University) in Frontiers in Human Dynamics; the design is sound on paper, but its evidence table misattributes several well-known studies and cites sources that are hard to trace, so its figures cannot be used without checking the originals.
Summary based on the full text·Original: about 45 minutes to read
In short
A systematic review published 7 May 2026 that collects observed, not predicted, labour-market changes linked to large language models from 2020 to early 2026. It concludes that the data already show displacement concentrated in routine cognitive tasks and junior roles, a growing premium for AI-skilled workers, growth in security and QA roles ("validation paradox"), and disproportionate risk for developing economies that sell outsourced cognitive services.
Key numbers
- 1,847initial records screened, of which 94 studies were synthesised and 42 used for quantitative extraction
- 14-41%range of posting declines for entry- and mid-level software and content roles in high-income economies, 2022-2024 (median -23%, not pooled)
- 15%-22%wage premium reported for workers demonstrating AI-augmentation capabilities
- 2%-21%reduction in online-platform posting volumes for automatable creative tasks after ChatGPT
As reported by the source.
English·Web page·opens frontiersin.org
What it reports
- Posting declines of 14% to 41% for entry- and mid-level software development and content creation in high-income economies (median -23%); the author stresses these are a span across studies, not a pooled estimate.
- Online platforms: 2% to 21% fewer postings for automatable creative tasks; declines concentrated among lower-rated freelancers.
- Wage effects: 15%-22% premium for AI-augmentation skills; advertised salaries rise as remaining postings skew senior.
- Sector split: software, content and data annotation decline; security engineering and QA grow (a 36% rise in security and validation postings is cited); healthcare IT stable.
- Developing economies: major Indian IT services firms cut more than 80,000 positions over 2023-2025; India's IT hiring fell from 400,000 in 2022 to about 100,000 in 2024 (per Xpheno); Remotasks exited Kenya, Nigeria and Pakistan in March 2024.
- Gender: exposed clerical roles have above-average female shares.
- Proposed framework: AI reorganises work through creation (increasingly AI), validation (increasingly human-critical) and obsolescence of career paths built on the old division of labour.
- Gaps: identification against COVID and rate cycles, posting data that do not show employment, little non-English evidence, and agentic tools after 2025 mostly outside the study window.
How they got it
PRISMA 2020 search of six databases (Scopus, Web of Science, EconLit, SSRN, IEEE Xplore, Google Scholar) on 15 January 2026, plus grey literature. Two reviewers screened (kappa 0.81) and extracted data; quality scored with a modified Newcastle-Ottawa scale. Narrative synthesis with ranges and medians; no meta-analysis.
Limits
- The evidence table contains apparent misattributions: the 55.8% Copilot speed-up is credited to Brynjolfsson, Li and Raymond, "Autor et al. (2024)" is listed as a Copilot RCT, and Acemoglu (2024) is described as a firm-survey employment study. Several other cited studies (for example "CodeMix Analytics 2024") could not be matched to known work during grading.
- Single listed author despite a described two-reviewer process.
- Ranges mix study designs, geographies and baselines and are not comparable.
- The developing-economy section relies on press and industry sources more than on peer-reviewed data.
Why it matters for companies
The direction of the findings is consistent with stronger sources: junior and routine cognitive roles absorb the first adjustment, and verification, security and QA work grows alongside. The validation idea is useful for workforce planning, since AI creation work generates new review work that needs staff and skills. Because of the citation problems, use this review as a reading list to check, not as a source of numbers for business cases.
Summary based on the full text. Read the original for the full argument and data.Read the original on frontiersin.org