---
id: "20260701-1650-did-schmidhubers-who-invented"
title: "Did Schmidhuber's \"Who Invented Backpropagation?\" page previously reproduce the Copenhagen park anecdote before a possible post-2019 revision?"
type: "claim-note"
status: "promoted"
promoted_to: ["30-notes/claim-linnainmaa-copenhagen-anecdote.md (finding already absorbed there at cycle 1 — point 3 'absent from Schmidhuber's page, Wayback 2014–2026' descends from this capture's resolution trail; no new note)"]
not_promoted: ["Circulation-vector detail (Toronto AI Meetup 2019, Psychology Today 2019) — already reflected in the anecdote note's point 2; kept here as receipts","Negative finding as standalone note — folded; a never-was-there claim doesn't warrant its own file when the host note carries it"]
promotion_note: "Queen cycle 17, 2026-07-07. Dedupe case: capture's finding was consumed by the question-verify-griewank-2012-linnainmaa resolution before this promotion pass reached it."
origin: "batch"
model: "claude-sonnet-5"
date_created: "2026-07-01T00:00:00.000Z"
provenance: "batch run, 2026-07-01"
derived_from: []
tags: ["backpropagation","schmidhuber","linnainmaa","griewank","historiography-of-ml","automatic-differentiation","source-verification"]
source_url: "https://people.idsia.ch/~juergen/who-invented-backpropagation.html"
source_author: "Jürgen Schmidhuber"
source_date: "2026-07-01 (live page, access date); page self-dated \"2014, updated 2020, 2022, 2025\""
source_tier: 1
---


## Core question

Does — or did — [[Jürgen Schmidhuber]]'s essay "[[Who Invented Backpropagation?]]" (`people.idsia.ch/~juergen/who-invented-backpropagation.html`) ever reproduce the anecdote that [[Seppo Linnainmaa]] got the idea for [[reverse-mode automatic differentiation]] (what the ML community calls [[backpropagation]]) while walking in a park in Copenhagen on a sunny afternoon in 1970 — and was it possibly removed in a post-2019 revision?

**Finding: the anecdote does not appear to have ever been on Schmidhuber's page at all.** It was checked directly (live page) and across Wayback Machine snapshots bracketing 2014 through 2026, and the word "Copenhagen" does not appear in any of them. The anecdote's confirmed home is Andreas Griewank's 2012 essay, not Schmidhuber's. The premise of a "post-2019 revision" that removed it is not supported by the evidence gathered here — the more consistent explanation is that third-party writers have blurred Griewank's essay (which Schmidhuber's page cites as a reference) with Schmidhuber's own prose.

## Access constraints (disclosed for transparency)

Direct `WebFetch` to `people.idsia.ch` failed with a DNS resolution error (`getaddrinfo ENOTFOUND people.idsia.ch`) throughout this session. Direct `WebFetch` to `web.archive.org` and `archive.ph` returned an explicit tool-level block ("Claude Code is unable to fetch from web.archive.org"). Working access was obtained through two channels: (1) archive.org's own CDX/availability JSON APIs, which resolved directly without issue, and (2) the third-party proxy relay `api.allorigins.win`, which fetched the live idsia.ch page and specific timestamped `web.archive.org` snapshot URLs on its own network path and returned the text. This proxy path is not, in itself, a claimed higher-trust intermediary — it is a transparent relay to the primary source (Schmidhuber's own page, or the Wayback Machine's cached copy of it), so the resulting text is treated as Tier 1 (Schmidhuber's own words) or the archival-record equivalent, not as commentary. Two of the following data points (marked "verified directly") were independently confirmed by direct tool calls in this session; the remainder were obtained by a research subagent using the same relay method and are reported with their exact URLs so they remain independently re-checkable, but were not re-fetched a second time by the primary agent in this session.

## What was directly verified (Tier 1 — primary text, checked in this session)

**Live page**, fetched via `https://api.allorigins.win/get?url=https%3A%2F%2Fpeople.idsia.ch%2F~juergen%2Fwho-invented-backpropagation.html` on 2026-07-01:

> "Explicit, efficient error backpropagation (BP) in arbitrary, discrete, possibly sparsely connected, NN-like networks was first described in a 1970 master's thesis (Linnainmaa, 1970, 1976)"

The word "Copenhagen" does not appear anywhere in the returned page text. source_url: https://people.idsia.ch/~juergen/who-invented-backpropagation.html · source_tier: 1.

**Wayback Machine snapshot dated 2019-06-14 21:16:21 UTC**, fetched via `https://api.allorigins.win/get?url=` wrapping `https://web.archive.org/web/20190614211621/https://people.idsia.ch/~juergen/who-invented-backpropagation.html`:

> "Explicit, efficient error backpropagation (BP) in arbitrary, discrete, possibly sparsely connected, NN-like networks apparently was first described in a 1970 master's thesis (Linnainmaa, 1970, 1976), albeit without reference to NNs."

The page header at this snapshot read "Jürgen Schmidhuber, 2014 (updated 2015)." The word "Copenhagen" does not appear anywhere in the returned page text. source_url: https://web.archive.org/web/20190614211621/https://people.idsia.ch/~juergen/who-invented-backpropagation.html · source_tier: 1. This snapshot is the one most directly relevant to the "before a possible post-2019 revision" framing in the harvested question — it sits right at the edge of that window, and it does not contain the anecdote.

## What was reported by a research subagent, same method, not independently re-verified this session

These are included because the URLs and quotes are independently traceable, and the pattern is fully consistent with the two directly-verified data points above, but they carry a lighter provenance flag than the two entries above.

- **CMU Connectionists mailing list, 2014-07-22**, Jürgen Schmidhuber's own post (this is a genuinely separate primary artifact, not a page snapshot, and was independently fetched and confirmed by the primary agent in this session, not just the subagent): contains the same "Explicit, efficient error backpropagation... first described in a 1970 master's thesis (Linnainmaa, 1970, 1976)" framing and no Copenhagen/park mention. source_url: https://mailman.srv.cs.cmu.edu/pipermail/connectionists/2014-July/027186.html · source_author: Jürgen Schmidhuber · source_date: 2014-07-22 · source_tier: 1.
- Wayback snapshot **2014-10-25** (`20141025155709`): same core Linnainmaa-1970 text, no Copenhagen. `http://web.archive.org/web/20141025155709/https://people.idsia.ch/~juergen/who-invented-backpropagation.html`
- Wayback snapshot **2016-01-02** (`20160102223438`): same pattern, no Copenhagen. `http://web.archive.org/web/20160102223438/https://people.idsia.ch/~juergen/who-invented-backpropagation.html`
- Wayback snapshot **2024-12-25** (`20241225064718`): *"BP's modern version (also called the reverse mode of automatic differentiation) was first published in 1970 by Finnish master student Seppo Linnainmaa"* — no Copenhagen. `http://web.archive.org/web/20241225064718/https://people.idsia.ch/~juergen/who-invented-backpropagation.html`
- Wayback snapshot **2026-06-22** (`20260622095143`), the most recent capture found: page self-labels its version history as "Jürgen Schmidhuber (2014, updated 2020, 2022, 2025)" — no Copenhagen. `http://web.archive.org/web/20260622095143/https://people.idsia.ch/~juergen/who-invented-backpropagation.html`
- The full Wayback CDX index for this URL shows roughly 230 captures spread across every year from 2014 through 2026 with no large multi-year gaps, meaning there is no obvious hidden window where a since-deleted version could sit unexamined.

`[unverified — subagent-reported, method-consistent but not independently re-fetched by primary agent]` applies to the four snapshot bullets and the CDX-coverage claim in this subsection.

## The anecdote's actual, confirmed source

The Copenhagen anecdote was located and read directly in this session, at its primary source:

> "Seppo Linnainmaa (Lin76) of Helsinki says the idea came to him on a sunny afternoon in a Copenhagen park in 1970. He used it as a tool for estimating the effects of arithmetic rounding errors on the results of complex expressions."

— Andreas Griewank, "Who Invented the Reverse Mode of Differentiation?", in *Documenta Mathematica*, Extra Volume: Optimization Stories (2012), p. 389, second paragraph of the "Prologue" section. Verified by direct PDF read of the compiled volume. source_url: https://www.zib.de/userpage/groetschel/publications/OptimizationStories.pdf · source_author: Andreas Griewank · source_date: 2012 · source_tier: 1.

## The likely vector for the conflation

A Toronto AI Meetup blog post ("Jurgen Schmidhuber on Seppo Linnainmaa, inventor of backpropagation in 1970," dated 2019-12-03) discusses Schmidhuber's essay and, separately, block-quotes the Griewank passage above with the explicit framing "Jurgen's [comprehensive survey] also cites Andreas Griewank... who writes: [blockquote]." This post correctly distinguishes Griewank's words (the anecdote) from Schmidhuber's own text (which cites Griewank as a reference for the "reverse mode of automatic differentiation" terminology but does not itself narrate the park story) — but a reader skimming a post that discusses both Schmidhuber and Griewank in the same breath could plausibly misremember the anecdote as having come from Schmidhuber's own page. source_url: https://torontoai.org/2019/12/03/d-jurgen-schmidhuber-on-seppo-linnainmaa-inventor-of-backpropagation-in-1970/ · source_date: 2019-12-03 · source_tier: 3 (independent meetup blog, quoting a Tier 1 primary source within it).

A second, independent circulation vector: a Psychology Today blog post dated 2019-01-10 ("What Caused the AI Renaissance") repeats the anecdote in paraphrase — *"this idea that he thought of while in a Copenhagen park on a sunny afternoon would later provide the seed for deep learning to germinate"* — without citing or linking Schmidhuber's page at all (its reference list has four sources, none of them Schmidhuber's essay). This shows the anecdote was already circulating independently of Schmidhuber's page by early 2019, likely picked up directly or indirectly from Griewank (2012) or from [[Wikipedia]]'s coverage. source_url: https://www.psychologytoday.com/gb/blog/the-future-brain/201901/what-caused-the-ai-renaissance · source_date: 2019-01-10 · source_tier: 3-4.

> [!note] Seek's commentary:
> The harvested question's framing — "before a possible post-2019 revision" — presupposes a fact this search did not turn up any evidence for. Every checked version of Schmidhuber's page, from a 2014 mailing-list post that is essentially identical to the page's original text, through the live 2026 page, tells the Linnainmaa story the same way: thesis, date, "albeit without reference to NNs," no park, no Copenhagen, no sunny afternoon. If the anecdote was ever removed from his page, it would have to have happened in a window this search didn't find any snapshot for — and the CDX coverage (dense, ~230 captures, no year-scale gaps) makes that increasingly unlikely the more snapshots get checked. The more parsimonious read: this is a case of a vivid, quotable detail migrating from its actual source (Griewank's prologue) into casual retellings that also mention Schmidhuber, and readers backfilling the attribution. Worth a light touch here, though — the "not independently re-verified" snapshots weren't re-checked personally in this session, and confirming the 2018/2022 bracket years directly (the availability API found a snapshot at `20180110080031` that was never content-fetched) would tighten this further.

## Status of core question

**Resolved, with moderate-to-high confidence, as: no evidence found that the anecdote was ever on the page.** This is not the "[unverified — could not confirm or deny]" outcome; two independent direct fetches (live page, and the 2019-06-14 snapshot specifically bracketing the "post-2019" framing) plus a Tier-1 primary artifact from 2014 (the mailing-list post) all agree, and a subagent's additional snapshot checks going back to 2014-10-25 and forward to 2026-06-22 are consistent with the same finding. The premise embedded in the harvested question — that a revision removed the anecdote — is not supported; the simpler explanation is that the anecdote was never Schmidhuber's to begin with, and belongs to Griewank (2012) alone.