Four categories of claim
Each passage carries a coloured rule and a label:
A · Historical fact. Supported by reliable historical sources: primary texts in scholarly translation, or established historical scholarship. Example: Boethius named the mathematical arts the quadrivium [6].
B · Empirical evidence. Supported by documented research, measurement or observation, with the study's own caveats. Example: METR's estimated 3.61-month doubling time for AI time horizons, with its confidence interval [43].
C · Philosophical proposition. A conceptual or normative argument. It can be well or badly reasoned, but not proved by data. Example: that education should aim at flourishing rather than employability.
D · Future scenario. A conditional projection that states its assumptions. Example: the three planning horizons, and Shapiro's post-labor framework [58].
Primary research and commentary
The bibliography separates peer-reviewed research, primary research from institutes (such as METR, which publishes outside journals), classical and historical texts, scholarly books and reference works, intergovernmental reports, and commentary. Commentary, including David Shapiro's essays, is used for frameworks and scenarios, never as evidence for a measured fact.
The working assumption
For planning, the project adopts a 90% project-level confidence that advanced AI will substantially automate cognitive work and progressively challenge the necessity of human labour across much of the economy. This is an editorial planning assumption. It is not an empirical probability, not a forecast of a technological singularity, and it is not attributed to METR, Shapiro or any researcher. The framework is designed to remain valuable if the assumption proves wrong or the timing is slower.
How sources were checked
- Each bibliography URL is checked online with
npm run verify, which reports HTTP status and page title. Publishers that block automated clients were checked against Crossref metadata (for DOIs) or by catalogue record. - The METR update and all five Shapiro essays were read directly from their publishers on 22 November 2026.
- The Power–Wisdom Gap paper's full text was blocked to automated access; we used its Crossref-registered title, authors, date and abstract, and say so where we discuss it.
- Quotations are kept under fifteen words; summaries are our own wording.
- The build fails if any page cites a source that is not in the bibliography.
Known limitations of this project
- The framework's structure (ten domains, ten capabilities) is a reasoned proposal, not an empirically validated taxonomy.
- Coverage of non-Western traditions relies mainly on reference works; specialist review is needed.
- AI capability evidence changes monthly. The METR figures here are as of 20 November 2026.
- The brief proposed several sources we could not use as written: see open questions.
The full research document, taxonomy rationale and decision log are in the project repository.