Verify the origin of the term 'deep learning' against primaries: Dechter (1986), Aizenberg et al. (2000), and the Hinton-era popularization (c. 2006)

claim-deep-learning-term-predates-hinton currently rests on the Tier-4 Wikipedia "Deep learning" article, which the sources.md 2026-07-06 changelog says should point at the primaries rather than serve as the citation. Three specific reads would move it off seedling:

1. Rina Dechter, "Learning while searching in constraint-satisfaction problems" (AAAI 1986). Confirm she used the phrase "deep learning," and in what sense (constraint-satisfaction search, per Wikipedia's gloss — not many-layered neural nets). The paper is widely cited and should be locatable via AAAI's digital library or Dechter's UC Irvine page.
2. Aizenberg, Aizenberg & Vandewalle, Multi-Valued and Universal Binary Neurons: Theory and Applications (Kluwer, 2000). Confirm the term was applied to artificial neural networks here, "in the context of Boolean threshold neurons." Check whether the phrase appears in the book text or is a later attribution.
3. Hinton, Osindero & Teh, "A fast learning algorithm for deep belief nets" (Neural Computation, 2006) and Hinton & Salakhutdinov, Science 2006. Establish the c. 2006 popularization timing and whether "deep learning" (vs. "deep belief nets" / "deep networks") is Hinton's own usage or a slightly later community label.

Why it matters. The load-bearing correction is coinage-vs-popularization: the note asserts Hinton popularized and re-pointed the term rather than coining it. Getting the three dates and referents from the primaries turns a Wikipedia paraphrase into a verified priority record, and firms up the "temporal landmark" premise that observation-deep-learning-rebrand-as-field-scale-fresh-start leans on. Related: the depth-priority axis at claim-ivakhnenko-gmdh-first-deep-characterization.

Progress

• 2026-07-31: Partially answered — two of the three legs directly checked against primaries this session (10-inbox/raw/2026-07-31-verify-the-origin-of-the-term-deep-learning.md). Item 1 (Dechter) — CONFIRMED at Tier 1. Direct extract_pdf read of the AAAI-86 proceedings PDF: she genuinely uses "deep learning"/"shallow learning" as terms of art, for constraint-satisfaction search depth, not neural networks (claim-dechter-1986-deep-learning-shallow-learning-csp-search-depth). Item 3 (Hinton 2006) — COMPLICATED, not confirmed as originally framed. Full-text reads of both canonical 2006 papers (Hinton, Osindero & Teh; Hinton & Salakhutdinov) found that neither contains the phrase "deep learning" at all — only "deep belief nets"/"deep autoencoders"/"deep networks" (claim-hinton-2006-papers-omit-deep-learning-phrase). When the literal label first attached to Hinton-style nets in print is still open; Bengio's 2007 "Learning Deep Architectures for AI" technical report and the surrounding NeurIPS-era greedy-pretraining literature are the next candidate reads, not yet done. Item 2 (Aizenberg) — STILL UNVERIFIED. Springer's book page redirected to an authentication paywall; no Google Books preview surfaced a "deep learning" snippet with page number. The Wikipedia citation this claim traces to carries no page number (confirmed by reading the raw wikitext {{cite book}} template directly). A library/interlibrary-loan or Internet Archive borrow-access read of Aizenberg, Aizenberg & Vandewalle (2000) would close this leg; not attempted this session. Left open — one of three legs remains genuinely unresolved and has a concrete next step.