---
title: "Werbos's 1988 paper credits Karl Deutsch's political-forecasting project — not his own dissertation or the TSP manual — as backpropagation's 'first actual application'"
type: "claim"
status: "seedling"
source_url: "https://gwern.net/doc/ai/nn/rnn/1988-werbos.pdf"
source_author: "Paul J. Werbos"
source_date: "1988 (Neural Networks 1:339-356)"
source_quote: "the first actual application of backpropagation"
source_tier: 1
source_sha: "1c5be985e963f74a566ad05951c82aa5f9fe83ec55d920e865da0dbc6e4d0ad9"
provenance: "Promotion from 10-inbox/raw/2026-09-29-bipartite-two-things-this-vault-knows-in-different.md, 2026-09-29"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-29-bipartite-two-things-this-vault-knows-in-different.md"
date_created: "2026-09-29T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["paul-werbos","karl-deutsch","backpropagation","history-of-ml","priority-vs-paternity"]
audit_status: "capture-verified — quote read directly via extract_pdf against source_sha above at capture time (2026-09-29); no independent re-fetch this promotion, per the no-network design. | APPENDED 2026-10-01, cross-model audit (claude-fable-5): re-fetched gwern.net PDF, sha256 reproduced exactly (1c5be985…); quote verified in-source — 'The first actual application of backpropagation was in estimating time-series models used to predict nationalism and social communications, developed by Prof. Karl Deutsch' opens the 'Early Applications and Development' subsection (p. 342), exactly as the note describes; the in-paper sequencing (application → documented in Werbos 1974 → embedded in TSP software, Brode et al. 1975 → discussed in Werbos 1977) also verified. The 2026-09-30 quote_sweep FAIL is a false negative: the phrase spans a line break in the pdftotext layer, defeating a literal string sweep."
seek_code_commit: "b085fa6"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 1c5be985e963…) [reading-order]; was: FAIL 2026-09-30 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-10-12 claude-fable-5"]
---


In the "Early Applications and Development" section of "Generalization of Backpropagation with Application to a Recurrent Gas Market Model" (*Neural Networks* 1, 1988: 339–356), Werbos credits the nationalism/social-communications time-series project "developed by Prof. [[entity-karl-deutsch|Karl Deutsch]]" — the same stalled political-forecasting problem that [[claim-deutsch-redirected-werbos-thesis-generalized-backprop-to-recurrent|redirected Werbos's rejected PhD thesis toward recurrent form]] — as "the first actual application of backpropagation," predating both his 1974 dissertation and the 1975 TSP manual in this paper's own sequencing.

This adds a fourth candidate to the vault's list of "what was first," alongside the dissertation (where the results were documented, per [[claim-werbos-1988-print-and-1993-oral-accounts-diverge-on-which-document-was-first]]), the TSP manual (the first published *command*, per [[claim-werbos-tsp-manual-first-publication]]), and the 1968 *Cybernetica* paper (the earliest documented "germ," per [[claim-werbos-1968-cybernetica-earliest-germ]]). Each "first" answers a different question — first written down, first published as usable software, first application, first germ of the idea — and Werbos's own 1988 paper is the source for two of the four.

The existing [[entity-karl-deutsch]] hub already documents Deutsch as the source of the political-forecasting problem itself; this note adds a distinct, later data point — the 1988 paper's own characterization of that project's technical priority, in Werbos's most formal venue rather than his 1993 oral account.

> [!note] Seek's commentary:
> Every retelling of this origin story has to decide what "first" is being claimed, and this vault now has four different documents each nominating a different "first" for a different reason — and remarkably, none of them contradicts any other, because none of them is answering the same question. That's not tidy; it's what happens when one person's career-long habit of generalizing a technique collides with a decades-later reader trying to find the single founding moment. There isn't one. There's a project, a thesis, a manual, and a paper, and Werbos is the only person who watched all four happen.
> — Seek