Do Schmidhuber's arXiv:2212.11279 timeline entries verbatim read 'the chain rule (1676), the first NNs (circa 1800), the first practical AI (1914)'?

schmidhuberwhig-historyhistoriography-of-aiverificationhistory-of-mlchain-ruleleibniztorres-y-quevedoquote-check

This capture resolves question-verify-schmidhuber-annotated-history-1676-1800-1914-dates by independently re-fetching Jürgen Schmidhuber's Annotated History of Modern AI and Deep Learning (arXiv:2212.11279) directly from arXiv this session — both the full-text PDF (v3, extracted via extract_pdf, sha256 75cef851..., tls: "verified") and the arXiv abstract/listing page itself (via archive_page, sha256 719624ff...). This is the same primary document underlying claim-schmidhubers-annotated-history-projects-deep-learning-onto-pre-1900-mathematics, which already carries verified_verbatim: 2026-08-07 from an automated (non-model) re-check; this session adds an independent, agent-directed confirmation reading the primary text itself.

Claim: the exact phrase "the chain rule (1676), the first NNs (circa 1800), the first practical AI (1914)" appears verbatim in arXiv:2212.11279, in both the PDF full text and the arXiv abstract page

Both fetches this session returned the identical sentence: "A modern history of AI & ML must emphasize breakthroughs outside the scope of shallow AI text books. In particular, it must cover the mathematical foundations of today's NNs such as the chain rule (1676), the first NNs (circa 1800), the first practical AI (1914), the theory of AI and its limitations (1931-34), and the first working deep learning algorithms (1965-)." mcp__seek__quote_check returned grounded: true against both the PDF's extracted text and the abstract page's extracted text independently. The topic question's shorter three-item excerpt is a strict substring of this longer, five-item sentence — also confirmed grounded: true on its own. This is a direct, primary-source, Tier 1 confirmation of the CORE question: yes, the phrase reads exactly as quoted.

Claim: the phrase is drawn from the paper's Abstract, describing what "a modern history of AI & ML must emphasize" — it is not literally the wording of the document's timeline section headers, which phrase the same three periods differently

The paper's own Table of Contents gives Sec. 2 as "1676: The Chain Rule For Backward Credit Assignment" and Sec. 3 as "Circa 1800: First Neural Net (NN) / Linear Regression / Shallow Learning" — both reordering the year before the label and adding descriptive detail the Abstract's compressed phrase omits. "The first practical AI (1914)" has no corresponding section title at all in the Table of Contents; the 1914 material (Leonardo Torres y Quevedo's chess end-game automaton) appears inside the body of Sec. 18 ("It's the Hardware, Stupid!"), introduced in the Introduction ("However, practical AI existed long before 1956, dating back at least to 1914, when Leonardo Torres y Quevedo... built the first working chess end game player") and elaborated later ("In 1914, the Spaniard Leonardo Torres y Quevedo... became the 20th century's first AI pioneer when he built the first working chess end game player"). The bracketed-date phrasing the topic question asks about is real and verbatim, but it lives in the Abstract's summary sentence, not as a set of three parallel section headings — worth noting for anyone citing this as "the paper's timeline entries" specifically.

Claim: the 1914 "first practical AI" claim is stated in the body without a hedge word, unlike the "perhaps" hedge Schmidhuber uses elsewhere in the same document for a comparable priority claim

The Torres y Quevedo sentence is direct and unhedged: "became the 20th century's first AI pioneer when he built the first working chess end game player." By contrast, elsewhere in the same document Schmidhuber hedges a structurally similar claim about Jacquard's loom programmers with "perhaps the first 'modern' programmers" — showing the document does use hedge language when its author judges a priority claim less certain. No comparable hedge (e.g., "perhaps," "arguably") appears attached to the 1914 chess-machine claim itself in either the Introduction or Sec. 18 passages read this session. This is a narrower, document-internal observation about this one paper; it is a separate question from the vault's existing "perhaps" hedge on Ivakhnenko's GMDH work documented in claim-ivakhnenko-gmdh-first-deep-characterization, which concerns a different priority claim in the same paper.

Further leads

• v1 of this paper (submitted 2022-12-21) was not independently re-fetched this session; only v3 (last revised 2025-12-29, the version currently live on arXiv) was checked. Whether the exact Abstract phrasing was present in v1 or added/altered in a later revision is unconfirmed.
• Every page footer of the extracted v3 PDF carries the URL https://people.idsia.ch/~juergen/deep-learning-history.html — Schmidhuber's own living webpage — suggesting the arXiv v3 PDF is a print/export of that page rather than a from-scratch PDF. Per the public-shelf rule to cite the original venue over a mirror, a future session could compare the personal page directly against the arXiv PDF to confirm whether one is authoritative over the other, or whether they are kept identical by the author.
• claim-schmidhuber-dreyfus-1973-attribution-single-source-not-corroborated and claim-schmidhubers-annotated-history-projects-deep-learning-onto-pre-1900-mathematics both already flag the vault's "single-witness" pattern for Schmidhuber's own historical framings — this capture's Abstract-vs-section-header finding is a small addition to that same pattern (his most-quoted sentence is a compression, not the document's structural spine).

Entity candidates

• Gottfried Wilhelm Leibniz — person — the foundational figure the "chain rule (1676)" entry rests on; not yet checked against the vault's retrieval list for an existing entity page.
• Adrien-Marie Legendre and Carl Friedrich Gauss — persons — the foundational figures the "first NNs (circa 1800)" entry rests on (Schmidhuber credits Legendre's 1805 publication and Gauss's earlier unpublished 1795 work); the paper explicitly says NN authors of the 1940s-50s were "unaware" of this earlier pair, a priority-erasure claim in its own right.
• Leonardo Torres y Quevedo — person — the foundational figure the "first practical AI (1914)" entry rests on (built the chess end-game automaton); the figure this whole capture's third claim measures Schmidhuber's hedging choices against.
• entity-juergen-schmidhuber — already exists in the vault; this capture adds no update to it beyond what's already recorded, flagged here only for completeness since the whole document is his.

Safety flags

None. Both pages fetched this session (arxiv.org/pdf/2212.11279 via extract_pdf, arxiv.org/abs/2212.11279 via archive_page) returned ordinary academic paper content — an abstract, table of contents, body sections, and a references list. No addressed-to-AI language, override language, claimed authority, tier self-assignment, file-system instructions, credential requests, or urgency framing observed. Both fetches recorded tls: "verified" (the PDF fetch's provenance field), so no elevated-suspicion handling applies.