Generative AI could expose the equivalent of 300 million full-time jobs worldwide to automation
5 events · 2 assessments · 2 decisions
Reassessed no material change
Staleness check, six days after prior assessment. Two web searches confirmed the evidence landscape has not moved materially: Goldman reaffirmed the 300 million estimate (2026 article and a March 2026 X post, the latter recorded as a new affirming instance), and recent criticism (capability-deployment gap analyses, ILO February 2026 brief on static task lists) continues to target exposure-measure precision, which the existing framework-assumption subclaim already carries, rather than denying the claim. Subclaim statuses unchanged (workload-share premise still unassessed). Re-affirmed supported, confidence 0.8, credence 0.75; re-recorded the single argument's evaluation as holds_with_caveats with a minor update noting the deployment-gap strand. Marginal_yield set low (0.15): the main future improvement waits on the workload-share subclaim's own assessment, not on another pass here. No structural changes; no notification to dependents since the verdict did not change.
Reassessed: still Supported
verdict confidence 0.80 · credence 0.75
Structured and assessed
First pass (structure_and_assess, with curator_change suggestions). Structure: adopted both Curator-suggested claims as requires premises under a named argument "Task-based derivation from occupational exposure data", since the 300 million figure is derived precisely from the occupation-exposure share and the workload-automation share in the Goldman Sachs analysis; it does not stand on independent evidence (independent bodies corroborate the magnitude but by their own distinct metrics, so requires, not supports, is right for these two). Also attached the existing methodology claim ("task-based analysis of occupational databases can meaningfully estimate AI exposure") as assumes: if that framework fails, the figure is ill-posed rather than merely false. Followed the Curator's advice to name the U.S./Europe-to-world extrapolation step in the assessment prose rather than minting a claim; no source disputes it as a standalone proposition, so it fails the claim bar as a node but is flagged as an uncertainty. No new claims minted; all three dependencies existed (per Curator identification, so no match_claim round-trips were needed). Evidence: three web searches. Original GS report and 2026 GS reaffirmation (instance recorded); IMF 2024 (~40% of global employment exposed) and ILO index corroborate the order of magnitude by independent methods; CEPR/VoxEU multi-model study is the strongest critique, showing exposure scores are highly model-dependent (high-exposure share of US occupations ranging ~3% to ~52% by scoring model). Aggregator articles (Forbes, CBS, Yahoo) merely report GS's estimate, so they were not recorded as instances; the originator is already recorded. Verdict: supported, confidence 0.8, credence 0.75, marginal_yield 0.3 (a deeper pass reading the primary GS report and CEPR paper in full could sharpen credence but is unlikely to change the status). Importance set to 0.6 (contestation 0.5): heavily consulted anchor of the AI-labor debate, moderately contested at the methodology level. Canonical form kept: fifteen words, neutral, frame-independent, accepted by both the source and its critics as what is in dispute.
Assessed Supported
verdict confidence 0.80 · credence 0.75
The figure originates in a March 2023 Goldman Sachs analysis by Joseph Briggs and Devesh Kodnani, who combined two inputs: their finding that roughly two-thirds of U.S. occupations are exposed to some degree of AI automation and their estimate that a quarter to half of exposed occupations' workload could be automated, then extrapolated from U.S. and European exposure patterns to employment worldwide, concluding that about 18 percent of work globally, the equivalent of some 300 million full-time jobs, could be automated. As an order-of-magnitude estimate of exposure, the figure holds up: independent task-based analyses by the IMF and the ILO, using their own methods, find comparable or larger shares of world employment exposed to generative AI, and Goldman Sachs reaffirmed the estimate in 2026. Three cautions govern how the number should be read. It measures exposure, not predicted job losses: the same analysis expects most exposed occupations to be complemented rather than replaced, and displaced work historically gives rise to new employment. Its precision is far softer than the round number suggests: the estimate assumes that task-based analysis of occupational databases can meaningfully estimate AI exposure, and multi-model studies have since shown that such exposure scores vary widely depending on which AI model does the scoring, so the true figure could plausibly differ by a large factor in either direction. And the extrapolation from U.S. and European occupational data to the world adds further uncertainty, since developing economies have substantially different occupational structures and lower measured exposure. The claim stands as a credible, transparently derived central estimate of the scale of exposure, not a verified count.
Claim entered the graph