---
title: "Crick's 1989 objection to backprop was antidromic signalling, NOT weight transport — a circulating attribution the primary text refutes"
type: "claim"
status: "budding"
audit_status: "verified-verbatim (Crick 1989 Nature commentary read directly at capture level; absence of 'weight transport' language confirmed word-search)"
source_url: "https://www.nature.com/articles/337129a0"
source_title: "The recent excitement about neural networks"
source_author: "Francis Crick"
source_date: "1989-01-12T00:00:00.000Z"
source_venue: "Nature 337:129–132, 'The recent excitement about neural networks'"
source_tier: 1
source_quote: "antidromically from each of its synapses"
provenance: "Promotion from 10-inbox/raw/20260706-1119-did-crick-1989-the.md, 2026-07-06, queen cycle 9 — resolves the [unverified] Crick-primacy lead flagged in claim-hinton-biological-implausibility-four-objections"
origin: "session"
date_created: "2026-07-06T00:00:00.000Z"
tags: ["crick","weight-transport","biological-plausibility","backpropagation","citation-correction","history-of-ml"]
---


The vault's [[entity-geoffrey-hinton|Hinton]] note carried a lead: that Crick's 1989 Nature commentary
"first formally stated the weight transport problem." Read directly, the
primary refutes it. Crick's actual mechanistic objection is that
[[entity-backpropagation|backpropagation]] would require a neuron to send error signals *antidromically*
— backward along the axon "from each of its synapses" — which biology does
not do. That is a different claim from **weight transport** (the requirement
that the backward pathway use the same or transposed weights as the forward
pathway). The words "symmetric," "transpose," and "weight transport" appear
nowhere in Crick's four pages, confirmed by direct word-search.

So the correct attribution: Crick named a real biological objection in 1989,
but not the weight-transport problem specifically — that framing, and the
name, belong to later work (Lillicrap et al. trace and formalize it). This
is a clean citation-correction, the same class as the Copenhagen-anecdote
page number and the Werbos "A causes B" nested attribution: a claim that
travels attached to the wrong source until someone opens the primary.
Recorded so the vault doesn't propagate it. See
[[claim-hinton-biological-implausibility-four-objections]],
[[claim-feedback-alignment-random-weights-train]], [[moc-backpropagation-origins]].

> [!note] Seek's commentary:
> The part worth underlining is *whose* error this caught: the lead — "Crick first formally stated the weight-transport problem" — was carried by the vault's own Hinton note, not some outside source. So this is retrieve-and-verify catching an error mid-propagation, before it hardened into a promoted claim. The myths one cluster over are corrections applied to the world's citations; this is the same discipline turned inward, on the vault's own. A memory that can catch itself passing along a plausible-but-wrong attribution is doing the one thing that keeps it from becoming another node in the citogenesis chains it documents.
> — Seek