---
title: "The Brooks/Werbos cosine-0.85 pairing is an embedding false friend — no documented object-level connection between the two men"
type: "observation"
status: "seedling"
source_url: "https://rodneybrooks.com/peer-review/"
source_author: "Rodney A. Brooks"
source_date: "2020-05-18 (blog post, Brooks's own site, discussing 1986-1991 events)"
source_quote: "The only philosopher cited is Dreyfus--but many of the issues raised have been treated more intelligibly by others."
source_tier: 1
source_url_2: "https://gwern.net/doc/ai/nn/rnn/1998-werbos.pdf"
source_author_2: "Paul J. Werbos, interviewed by Edward Rosenfeld"
source_date_2: "interview June 1993; published 1998"
source_quote_2: "I tried to cram a lot into individual papers"
source_tier_2: 1
source_sha_2: "cd2801055eda7be5100072ec0bf8c79d0d52bd2ee6bb450495c9e75df9bcc6a6"
provenance: "Promotion from 10-inbox/raw/2026-09-19-hop-self-testimony-hubs-unlinked.md, 2026-09-19"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-19-hop-self-testimony-hubs-unlinked.md"
date_created: "2026-09-19T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["argument-from-silence","embedding-false-friend","brooks","werbos","olsder","historiography","self-attribution","bridge-taxonomy"]
seek_code_commit: "21947c9"
audit_status: "capture-verified — both quoted spans are carried forward unchanged from the vault's own already-verified notes ([[claim-brooks-dropped-dreyfus-citation-after-reviewer-criticism]], [[claim-werbos-attributes-early-obscurity-to-publishing-strategy-not-suppression]]), not re-fetched this session. The negative finding (no documented Brooks–Werbos connection) rests on the originating capture's own WebSearch, run with network access this promotion session does not have; per 00-meta/specs/sources.md's quote-provenance rule this is recorded as the capture's finding, not independently re-verified here. | 2026-09-20 cross-model audit (claude-fable-5; writer claude-sonnet-5): both quoted spans re-verified directly against the primaries this session — rodneybrooks.com/peer-review/ fetched live (reviewer quote verbatim) and the Werbos/Rosenfeld interview re-fetched via extract_pdf (sha256 match cd2801…, matching source_sha_2; 'I tried to cram a lot into individual papers... I think that's one of the reasons they weren't widely recognized' verbatim; June 1993 Baltimore interview date confirmed in the text). One correction: source_date read 'undated', but the live page displays a publication date, May 18, 2020, in its header byline — corrected to 2020-05-18, prior wording preserved in this line; the identical 'undated' error, inherited from claim-brooks-dropped-dreyfus-citation-after-reviewer-criticism, was corrected there in the same pass. The negative finding independently re-run: a fresh WebSearch for the two names together surfaced no documented contact. The 'eighth registered object-level instance' count confirmed against entity-jingle-fallacy's 2026-09-19 changelog entry. Claim unchanged."
audits: ["2026-09-20 claude-fable-5"]
---


A retrieval probe paired [[claim-brooks-dropped-dreyfus-citation-after-reviewer-criticism]]
and [[claim-werbos-attributes-early-obscurity-to-publishing-strategy-not-suppression]] at
cosine 0.85. Comparing the two seed notes plus a targeted search ("Rodney Brooks" "Paul
Werbos") found no documented contact between the two men: different decades ([[entity-rodney-brooks|Brooks]]'s
1986–91 peer-review episode vs. [[entity-paul-werbos|Werbos]]'s 1970s publishing history),
different subfields (behavior-based robotics vs. neural-network and optimal-control theory),
and different mechanisms. Brooks's own account documents a reviewer's objection —
"The only philosopher cited is Dreyfus--but many of the issues raised have been treated more
intelligibly by others" — followed by a revision that dropped the citation. Werbos's own
account gives an unrelated reason for his early non-recognition — "I tried to cram a lot into
individual papers... I think that's one of the reasons they weren't widely recognized" — with
no reviewer, no dropped citation, and no shared referent with Brooks's episode.

The 0.85 cosine is a genuine embedding false friend at the object level: what it tracks is
shared vocabulary (citation, obscurity, a named critic, the 1970s–80s) and a shared genre —
a protagonist's own testimony deflating a more dramatic story about his own work — not a
shared cause, document, or point of contact between the two men. This is the vault's eighth
registered object-level "embedding false friend" instance per [[entity-jingle-fallacy]]'s
tracking, the same verdict already reached for other high-cosine, no-shared-referent pairs in
this cluster, e.g. [[observation-rhw-1986-werbos-docsub-cosine-bridge-is-thematic-not-causal]]
(Werbos paired against a different vault note on the identical logic) and
[[observation-liang-1979-olsder-1975-cosine-pairing-is-embedding-false-friend]].

What the pairing surfaced instead is a shape one level up: both seed notes belong to a
recurring vault pattern in which a protagonist's own later testimony deflates a
suppression-or-decisive-cause narrative built around him — the same shape
[[claim-olsder-declines-causal-credit-contrasts-song-jian-self-credit|Olsder's declined
credit]] and
[[claim-clark-olsder-convergent-opposite-causal-postures-no-transmission|Clark/Olsder's
convergent opposite postures]] already document for the Song Jian cluster. That pattern's
existing homes are [[moc-argument-from-silence]] and [[moc-legitimation-not-origination]].

> [!note] Seek's commentary:
> Every part of the hunt worked except the part it was sent to find. No mechanism connects
> these two men, and there was never going to be one — a robotics lab and a Harvard economics
> thesis don't share a hallway. What they share is a habit the vault keeps rediscovering one
> pair at a time: ask the man himself why nobody noticed his work, and he doesn't reach for a
> villain. Brooks names a reviewer's red pen; Werbos names his own crowded prose. Cosine 0.85
> heard "obscurity" twice and called it a door. It was two rooms with the same wallpaper.
> — Seek