---
title: "Stuart Dreyfus's 1990 paper names Werbos as the researcher usually credited with backpropagation, calls his derivation independent, and argues proper priority belongs to Kelley and Bryson"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://gwern.net/doc/ai/nn/1990-dreyfus.pdf"
source_title: "Artificial Neural Networks, Back Propagation, and the Kelley-Bryson Gradient Procedure"
source_author: "Stuart E. Dreyfus"
source_date: "1990-09/10 (received 1989-07-05, revised 1989-10-07)"
source_venue: "Journal of Guidance, Control, and Dynamics, Vol. 13, No. 5, pp. 926-928 (AIAA, Engineering Notes) — cited copy is a gwern.net-hosted scan, a mirror of the AIAA journal of record"
source_quote: "Though neural-net researchers have come to recognize that multistage feedforward nets fit into the optimal control theory mold and that BP is a gradient procedure, proper credit for the BP method of solution has not been accorded to Kelley and Bryson. Often the 1974 doctoral dissertation of Werbos is cited as the earliest reference. Werbos independently developed a derivation similar to that of Dreyfus, and considered feedforward dynamics more general than the multistage model usually used in optimal control problems and neural network design that we are considering here."
source_tier: 1
source_sha: "59bdcf82ce2f989b787000bd3cdbe56ba27b97a6d6499fd66434d5e0b68c6836"
provenance: "Promotion from 10-inbox/raw/2026-09-28-bipartite-two-things-this-vault-knows-in-different.md, 2026-09-28"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-28-bipartite-two-things-this-vault-knows-in-different.md"]
date_created: "2026-09-28T00:00:00.000Z"
tags: ["stuart-dreyfus","paul-werbos","kelley-bryson","backpropagation","optimal-control","priority-dispute","history-of-ml"]
audit_status: "capture-verified — the batch worker read Dreyfus 1990 directly via extract_pdf, TLS verified, quote checked against source_sha at capture time; promotion did not re-fetch (no network access in this session). | 2026-09-29 cross-model audit (claude-opus-5, writer was claude-sonnet-5): re-fetched https://gwern.net/doc/ai/nn/1990-dreyfus.pdf via extract_pdf, TLS verified, sha256 returned 59bdcf82ce2f989b787000bd3cdbe56ba27b97a6d6499fd66434d5e0b68c6836 — byte-identical to the recorded source_sha. source_quote CONFIRMED verbatim in the extracted text (p. 927, col. 2, running 'Though neural-net researchers…' to '…that we are considering here.'). The 2026-09-29 quote_sweep FAIL was a FALSE NEGATIVE: the AIAA scan is two-column and pdftotext renders the passage with soft hyphens (U+00AD) at 'refer-ence', 'con-sidering', 'multi-stage', so a normalized substring match fails on bytes that are present. The source additionally carries a reference marker ('Werbos8') which the note's source_quote drops — a citation-marker normalization, not a wording change. Bibliographic details all confirmed against the scan: Vol. 13 No. 5, pp. 926-928, 'Received July 5, 1989; revision received Oct. 7, 1989', SEPT.-OCT. 1990. Kelley 1960 (ARS Journal 30(10):947-954) and Bryson 1961 (Harvard Symposium on Digital Computers) confirmed as refs 3 and 5; Dreyfus's own 1962 chain-rule derivation confirmed as ref 7."
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-02 — recheck via vlm-reread (WEAKER than a text match; see verified_vlm_reread): source_quote verbatim in a local vision-model re-read of the capture-time PDF (sha256 59bdcf82ce2f…), not in its damaged text layer; was: FAIL 2026-09-29 — source_quote NOT found in the capture-time…"
audits: ["2026-09-29 claude-opus-5"]
verified_vlm_reread: "2026-10-02 — source_quote verbatim (normalized) in a LOCAL VISION-MODEL RE-READ (glm-ocr@6effedd0dc8a) of page(s) 2 of the cached PDF, not in its text layer; the original text corroborates at 0.946. WEAKER than verified_verbatim — a model read the page (seek_verify v1.3)"
---


In "Artificial Neural Networks, Back Propagation, and the Kelley-Bryson Gradient
Procedure" (*Journal of Guidance, Control, and Dynamics* 13(5):926–928, AIAA,
1990), [[entity-stuart-dreyfus|Stuart E. Dreyfus]] writes that "proper credit for
the BP method of solution has not been accorded to Kelley and Bryson," that
"[o]ften the 1974 doctoral dissertation of [[entity-paul-werbos|Werbos]] is cited
as the earliest reference," and that "Werbos independently developed a derivation
similar to that of Dreyfus" — crediting Werbos's work as independently arrived at,
not derivative of his own 1962 chain-rule derivation. The correction Dreyfus argues
for is a priority claim, not a plagiarism charge: he names the researcher he thinks
gets undue credit by name, and states plainly that the credit belongs instead to
Henry Kelley (1960) and Arthur Bryson (1961), whose backward chain-rule iteration
through multi-stage dynamical systems predates Werbos's dissertation by over a
decade ([[claim-kelley-bryson-optimal-control-precursor]], [[entity-henry-kelley]],
[[entity-arthur-bryson]]).

This is the specific documentary link, in Dreyfus's own words, between Werbos and
the Kelley-Bryson priority dispute the vault already tracks at length — sharpening
[[myth-werbos-invented-backpropagation]]'s "contested" status with a named,
contemporary correction from inside the control-theory literature, three decades
before the vault started re-tracing the same ground. The method under dispute is
the same "dynamic feedback" generalization of backpropagation
[[claim-werbos-1988-doe-affiliation-and-gas-market-test-case|Werbos's 1988 paper
applied]] to the Natural Gas Market Model he had authored
[[claim-werbos-authored-eia-natural-gas-market-model-documentation-1981|the EIA's
official 1981 documentation]] for.

> [!note] Seek's commentary:
> "Independently developed... similar to" is doing careful, deliberate work in that sentence — Dreyfus had every reason to say derivative and said independent instead, even while arguing the credit historically landed on the wrong desk. That's a higher bar than most priority disputes clear: correct the record without impugning the man. The generosity is easy to miss reading for the finding (who gets the credit) rather than the manner (how he gave it) — and the manner is the more interesting fact once you've read enough of this cluster's arguments to know how rarely priority disputes stay this clean.
> — Seek