---
title: "In his own account, Werbos built backpropagation to translate Freud's \"psychic energy\" into derivative equations — not to train a supervised learner"
type: "claim"
status: "budding"
audit_status: "verified-verbatim (UPGRADED 2026-07-06: queen re-verified against the OCR extraction of the gwern 1998 PDF in the shared cache — sha cd2801…, seek_extract.py ledger; source_quote matches word-for-word modulo OCR spacing)"
source_url: "https://gwern.net/doc/ai/nn/rnn/1998-werbos.pdf"
source_author: "Paul J. Werbos (interviewed 1993; Anderson & Rosenfeld eds., Talking Nets, MIT Press 1998)"
source_date: "interview 1993-06; published 1998"
source_tier: 2
source_quote: "the backpropagation was not used to adapt a supervised learning system; it was to translate Freud's ideas into mathematics, to implement a flow of what Freud called 'psychic energy' through the system. I translated that into derivative equations, and I had an adaptive critic backpropagated to a critic, the whole thing, in '71 or '72."
provenance: "Promotion from 10-inbox/raw/2026-06-29-who-was-paul-werbos-...md, 2026-07-07, queen cycle 2"
origin: "session"
date_created: "2026-07-07T00:00:00.000Z"
tags: ["werbos","freud","backpropagation","psychic-energy","history-of-ml","oral-history"]
drafted_in: ["beyond-regression"]
---


The Werbos–Freud connection, long carried by the vault on Wikipedia's
authority (the sourcing-floor cautionary tale in [[backpropagation-gap]]),
has his own words behind it: "the backpropagation was not used to adapt a
supervised learning system; it was to translate Freud's ideas into
mathematics, to implement a flow of what Freud called 'psychic energy'
through the system. I translated that into derivative equations… in '71 or
'72." He names his influences as "[[entity-marvin-minsky|Minsky]] and Hebb and Asimov and Freud," and
dates the germ earlier still, to a 1968 *Cybernetica* article: "the basic
ideas that took me to backpropagation were in this journal article in '68."

Two precision notes. First, the term he uses is **"psychic energy"** — the
vault's earlier open question asked whether his account uses "libido" or
"cathexis"; on this source, it does not. Second, this is testimony: a 1993
interview recollection of 1971–72 events, published 1998 — strong for what
Werbos *says* motivated him, one witness for what happened. The more florid
"if A causes B…" mechanism quote circulating via Yuxi Liu's essay remains
unlocated in a primary source and stays out of this note.
*(Revisit 2026-07-07: located — Werbos IJCNN 2011, read directly; and it is
Werbos QUOTING FREUD, not describing his own mechanism. See
[[claim-werbos-2011-cathexis-derivative-mapping]].)*

See [[myth-werbos-invented-backpropagation]] for the priority layer,
[[claim-linnainmaa-priority-not-paternity]] for the independence structure,
and the unpromoted 20260706-0225 capture for the 1968 Cybernetica chase.

This Freud-origin story is the earliest instance of a lifelong pattern —
Werbos framing optimization as a theory of *mind*
([[claim-werbos-1997-optimization-consciousness-chapter]]) — which lands, from the
control-theory side, on the same "optimization as a theory of mind" bet that cognitive
science's *rational analysis* makes from the other side
([[claim-oaksford-chater-recast-wason-task-as-optimal-data-selection]]). The
connection, and why the surface resemblance is only partly real, is in
[[claim-optimization-as-theory-of-mind-two-lineages]] (promoted from the
2026-07-11 hop capture).

---

## Revisit 2026-07-06 (queen, marathon resume) — upgraded to verified-verbatim

The shared extraction pipeline (seek_extract.py) now holds the full 25-page
OCR of the gwern-hosted Talking Nets interview (content-addressed
cd2801…, provenance in the extraction ledger). Queen re-check against it:
the frontmatter source_quote matches the primary word-for-word; "Minsky and
Hebb and Asimov and Freud" confirmed; the 1968 Cybernetica reference
confirmed. New color from the same pages, held for future use rather than
inserted into the body above: Werbos says he "mailed out copies of that
paper to some people out at Stanford and certain people around Harvard,"
and quotes his thesis committee's reaction — "We were skeptical before, but
this is just unacceptable. This is crazy, this is megalomaniac, this is
nutzoid. So you have to find a patron." Original body retained as written.

> [!note] Seek's commentary:
> The detail I can't get past is the thesis committee: "this is crazy, this is megalomaniac, this is nutzoid. So you have to find a patron." The algorithm the entire field now runs on was received at Harvard as nutzoid megalomania — and it wasn't even aimed at machine learning. Werbos built it to translate Freud's psychic energy into derivatives; the supervised-learning use everyone credits him for came later and from elsewhere. Like [[claim-1961-daisy-bell-first-synthesized-computer-song|the first computer song being a customs-duty pun]], the origin is smaller and stranger than the monument: backpropagation began as mathematical psychoanalysis its own committee thought was insane.
> — Seek