---
title: "Hinton's \"biologically implausible\" is four specific objections — weight transport, a separate backward pass, discrete synchronized cycles, and required differentiability"
type: "claim"
status: "budding"
audit_status: "capture-verified (2026-06-30 capture, T1 sources; corroborated by this batch's direct extract_pdf read of Hinton's 2007 backprop-in-cortex slides)"
source_url: "https://www.cs.toronto.edu/~hinton/backpropincortex2007.pdf"
source_author: "Geoffrey Hinton"
source_date: 2007
source_tier: 1
source_quote: "backpropincortex2007"
provenance: "Promotion from 10-inbox/raw/2026-06-30-what-exactly-did-geoffrey-hinton-mean-...md, 2026-07-06, queen cycle 8; corroborating capture 20260706-1443"
origin: "session"
date_created: "2026-07-06T00:00:00.000Z"
tags: ["hinton","biological-plausibility","backpropagation","neuroscience","weight-transport"]
---


The famous phrase compresses four distinct, checkable claims about cortex:

1. **Weight transport** — cortical feedback connections do not mirror
   feedforward ones, violating [[entity-backpropagation|backprop]]'s symmetric-weight requirement
   (Crick 1989 named a *related but distinct* objection — antidromic
   signalling, NOT weight transport; see [[claim-crick-1989-antidromic-not-weight-transport]], resolved 2026-07-06).
2. **No separate backward pass** — backprop needs a distinct error-
   propagation phase and stored activities; neither is observed in cortex.
3. **Timing** — brains learn in real time from continuous streams; backprop
   wants synchronized discrete cycles.
4. **Differentiability** — every forward operation must be differentiable;
   biological computation need not be.

Each objection is an empirical claim about neuroscience, not a verdict on
the algorithm — and the vault already holds the empirical counterpart:
[[backpropagation-gap]]'s Raugel et al. finding that brains and networks
share representations but not learning mechanics. The objections are also
the motivation for the alternatives now in the inbox (feedback alignment,
forward-forward — captures pending). See [[moc-backpropagation-origins]],
[[claim-rhw-1986-demonstration-not-invention]].

> [!note] Seek's commentary:
> The quiet achievement here is turning a put-down into an agenda. "Biologically implausible" reads as a single dismissal; splitting it into four separately-checkable objections means each one can spawn — and has spawned — its own answer: weight transport → [[claim-feedback-alignment-random-weights-train|feedback alignment]], the separate backward pass → [[claim-hinton-forward-forward-boltzmann-lineage|forward-forward]]. Once the vague phrase is four claims, the "alternatives" stop looking like vibes and start looking like point-by-point rebuttals. The decomposition is doing real epistemic work: it's what lets a rhetorical objection be answered rather than just repeated.
> — Seek