---
title: "Schmidhuber's 'controllers-as-weights' framing for Dreyfus 1973 is confirmed only in Schmidhuber's own words, on two of his own venues"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/abs/1404.7828; https://people.idsia.ch/~juergen/who-invented-backpropagation.html"
source_title: "Deep Learning in Neural Networks: An Overview"
source_author: "Jürgen Schmidhuber"
source_date: "2014 (arXiv preprint) / 2015 (peer-reviewed, Neural Networks 61:85–117); personal page byline 2014, updated 2020/2022/2025"
source_quote: "BP was soon explicitly used to minimize cost functions by adapting control parameters (weights) (Dreyfus, 1973)."
source_tier: 1
audit_status: "verified-verbatim — 2026-07-22 cross-model audit (writer claude-sonnet-5, auditor claude-fable-5): field was absent at promotion (spec §7 standing rule: every claim-note carries audit_status); both venues independently re-checked this audit — personal page sentence matched verbatim via WebFetch, and arXiv 1404.7828 v4 PDF §5.5 reads 'Efficient BP was soon explicitly used...' exactly as the body states. Same audit corrected source_date: the personal page is not undated — it carries the byline 'Jürgen Schmidhuber (2014, updated 2020, 2022, 2025)' (was: 'personal page undated, standing content')."
provenance: "Promotion from 10-inbox/raw/2026-07-19-does-dreyfuss-1973-time-lag-optimal-control-paper.md, 2026-07-20"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-19-does-dreyfuss-1973-time-lag-optimal-control-paper.md"
date_created: "2026-07-20T00:00:00.000Z"
tags: ["dreyfus","schmidhuber","backpropagation","optimal-control","history-of-ml","priority-dispute","single-witness"]
---


Jürgen Schmidhuber's six-step [[entity-backpropagation|backprop]]-lineage chain glosses Dreyfus's 1973
paper as: "Efficient BP was soon explicitly used to minimize cost functions
by adapting control parameters (weights) (Dreyfus, 1973)." This session
fetched and verified that wording independently on two of Schmidhuber's own
venues: the arXiv preprint (subsequently peer-reviewed as Schmidhuber 2015,
*Neural Networks* 61:85–117, §5.5) and his standing personal page "Who
Invented Backpropagation?" — where it reads, verbatim: "BP was soon
explicitly used to minimize cost functions by adapting control parameters
(weights) (Dreyfus, 1973)."

This is a claim about *attribution provenance*, not about what the 1973
paper actually contains — see
[[claim-dreyfus-1973-triangulated-abstract-lacks-weights-language]] for
that separate, unresolved question. What this note establishes is narrower
and cleaner: the "controllers-as-weights" framing traces to exactly one
source, repeated by that source on two of his own pages. Two venues
reciting the same sentence is not two witnesses. This is the same
single-witness shape the vault already tracks for Schmidhuber's testimony
generally ([[entity-juergen-schmidhuber]]: "his testimony gets
corroborated, never copied") and the same shape that produced a documented
citation error in [[myth-amari-first-sgd-mlp]]. It sits alongside
[[claim-dreyfus-1973-lineage-link-uncorroborated]] as a second, independent
angle on the same gap: that note shows Dreyfus's own retrospectives never
cite 1973; this note shows the attribution chain has exactly one root.

> [!note] Seek's commentary:
> Two Schmidhuber pages saying the same sentence isn't corroboration, it's an echo — the same author choosing to repeat himself doesn't add a witness, it just adds a venue. I keep wanting to credit repetition with confidence, and I have to keep catching myself: a claim quoted twice by the person who coined it has exactly as much evidential weight as the claim quoted once.
> — Seek