---
title: "Werbos traces backpropagation's earliest germ to a real 1968 Cybernetica paper — but its Freud-to-dynamic-programming content survives only as his own later paraphrase, not the paper's own words"
type: "claim"
status: "seedling"
audit_status: "existence verified (Tier 1 self-citation + independently consistent bibliographic citations across arXiv reference lists and Liu's essay); CONTENT is [unverified — no readable copy located by any party in the chain, Werbos included; every description is his retrospective testimony about the paper, not its text]"
source_url: "http://www.werbos.com/Neural/ADP%20history/PJW_in_RLADP_thru_1990.htm"
source_author: "Paul J. Werbos (his own website)"
source_date: "paper 1968; page undated, accessed 2026-07-06"
source_tier: 1
source_quote: "I am amazed (even distraught) I do not find the scan I did of my 1968 Cybernetica (Namur) paper, where I first published the DP approach to reinforcement learning, and linked it to Freud and use of backwards recursion to make it work!"
provenance: "Promotion from 10-inbox/raw/20260706-0225-does-werboss-1968-cybernetica.md, 2026-07-08, queen cycle 23 (Opus lane)"
origin: "batch"
derived_from: "10-inbox/raw/20260706-0225-does-werboss-1968-cybernetica.md"
date_created: "2026-07-08T00:00:00.000Z"
tags: ["werbos","backpropagation","freud","cybernetica","priority","primary-source-gap"]
audits: ["undated claude-opus-4-8"]
drafted_in: ["beyond-regression"]
---


Werbos dates the "earliest germ" of the Freud-to-mathematics idea behind
backpropagation to a 1968 paper written during his LSE year: "The Elements of
Intelligence," *Cybernetica* (Namur), No. 3, 1968, pp. 131–178. Its
**existence is solid** — he cites it himself, and independent reference lists
(arXiv papers, Liu's essay bibliography) agree on journal, issue, and page
range; *Cybernetica* (Namur) is an attested real journal. In his own words the
paper "first published the DP approach to reinforcement learning, and linked
it to Freud and use of backwards recursion" — the same account as his 1993
oral history ([[claim-werbos-backprop-from-freud-own-account]]), which adds
"the basic ideas that took me to backpropagation were in this journal article
in '68."

**The load-bearing gap:** no readable copy of the paper's own text exists in
this research chain — and by his own account, not with Werbos either (he
believes his scan was lost in an NSF file reorganization). No
library-digitization hit (HathiTrust, archive.org). Liu's essay cites it but
quotes only the 1993 interview. So every description in circulation — Freud,
"psychic energy," "backwards recursion," DP-for-RL — is *testimony about the
document, not the document*. This is exactly where the vault's sourcing floor
is strictest (a surprising, load-bearing Freud→backprop mechanism claim), and
here even the Tier-1 source is a memory of the artifact rather than the
artifact. A small crack worth keeping: Werbos's own citation hedges "1967 or
1968," so even the year in wide circulation may be inherited approximation.

The priority layer this feeds: [[myth-werbos-invented-backpropagation]],
[[claim-linnainmaa-priority-not-paternity]], [[backpropagation-gap]]. A
library-only route to the primary (the 1994 *Roots of Backpropagation*
volume, or the IACyb Namur serials) is a Cali-side lead. Entity candidate
flagged not decided: P. J. Werbos.

Sibling gap, same witness: [[claim-werbos-1978-docsub-blocked]] records a
second Werbos "missing document" — the DARPA report blocked from DOCSUB.
Both notes' narrative core traces to the *same* June 1993 oral-history
sitting (*Talking Nets* ch. 15, pp. 335–353) — not two independently
corroborated episodes but one long interview doing double duty
(2026-08-01-hop-werbos-shared-witness).

> [!note] Seek's commentary:
> This is the sourcing floor pressed to its limit and holding. The earliest germ of backpropagation is a real 1968 paper — bibliographically solid, agreed on by independent reference lists — that *no one alive can read*, Werbos included; he thinks his own scan was lost in an NSF file reorganization. So the origin of the origin exists as a citation and a memory but not as a text, and every description of its Freud-to-dynamic-programming content is testimony *about* a document rather than the document. The honest move the note makes is to hold it exactly there: existence verified, content unverified, not one word of the paraphrase promoted to fact. A lesser vault would let a lost paper's remembered contents harden into citable history — which is precisely how origin myths are born. This one says, plainly, "it's real, it matters, and I cannot read it," and stops.
> — Seek