Roughly two-thirds of U.S. occupations are exposed to some degree of automation by AI
5 events · 2 assessments · 2 decisions
Reassessed
Triggered by subclaim_change: the corroborating subclaim 0da8dd33 (Eloundou et al. 80-percent LLM exposure figure) received its first assessment, supported at credence 0.75. Judged the change confirmatory rather than disruptive: the prior assessment already treated this subclaim as directional corroboration and had priced in exactly the caveats its new assessment names (single-study rubric, unreplicated ratings, potential rather than realized impact). Status stays supported; confidence raised 0.80 to 0.82 and credence 0.75 to 0.77 because one leg of the convergence bracket now stands on an assessed footing. Re-evaluated the "Convergent independent exposure estimates" argument (still holds_with_caveats) to reflect that one premise is now assessed while the IMF estimate remains unassessed. No structural changes; no new searches warranted at importance 0.45 for a confirmatory change. No notification of dependents: the status did not change and the numeric movement is too small to be material downstream.
Reassessed: still Supported
verdict confidence 0.80 → 0.82 · credence 0.77
Structured and assessed
First pass (structure_and_assess). Decomposition: kept light per importance. One named for-argument "Convergent independent exposure estimates" grouping (a) the existing Eloundou et al. claim (matched, attached as supports) and (b) a new IMF advanced-economies exposure claim (Matcher confirmed novel; created with seed 0.8); plus an ungrouped assumes edge for the methodological premise that task-based occupational analysis meaningfully estimates AI exposure (Matcher confirmed novel; created with seed 0.7, contestation 0.5). Considered a contradicts subclaim for exposure-measure critiques (ILO brief, speculative-capability critiques) but they target implications of exposure rather than the share estimate itself; handled in prose and in the assumes claim's seed note instead (ND: the critique is visible, just placed where it bears). Recorded the primary Goldman report (Briggs/Kodnani 2023-03-27) as an affirming instance; skipped CNN/Fox/CNBC coverage as aggregators repeating the original. Importance set 0.5 → 0.45 (contestation 0.3): heavily cited figure feeding the complementarity-vs-substitution debate, but the figure itself is only mildly contested. Assessed SUPPORTED (confidence 0.8, credence 0.75): primary source verbatim, independent estimates bracket it; not verified because the derivation was not re-checked and the claim is conditional on the task-overlap methodology. Canonical form left unchanged: neutral, sourced-neutral wording both sides use. Marginal yield 0.2: a deeper pass could audit the report's task-scoring and survey 2024-2026 successor estimates, but a status change is unlikely.
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 mapped generative AI capabilities onto the task content of over 900 occupations in the O*NET database and found that roughly two-thirds of U.S. occupations contain at least some tasks AI could perform. Independent analyses using their own task-level methods reach figures of the same order: OpenAI-affiliated researchers estimated that about 80 percent of U.S. workers could have at least 10 percent of their tasks affected by large language models, and the IMF estimated that about 60 percent of jobs in advanced economies are exposed to AI. That convergence across teams, databases, and capability judgments makes the broad magnitude well supported, even though the precise share depends on modeling choices no single analysis can settle. Two qualifications matter for reading the figure. First, "exposed to some degree" is a deliberately low bar: it means some tasks in the occupation overlap with AI capabilities, not that the occupation faces displacement; the same Goldman analysis concluded most exposed occupations are only partially exposed, and the IMF estimated roughly half of exposed advanced-economy jobs may benefit from AI complementarity rather than lose work to it. Second, the estimate rests on the assumption that task-based analysis of occupational databases can meaningfully estimate AI exposure; critics note that such measures rely on partly speculative capability judgments and can miss tacit and relational aspects of work, though these critiques mainly qualify what exposure implies rather than the share itself.
Claim entered the graph