---
title: "Hinton's position on backpropagation-in-the-brain reversed over fifteen years — the 2007 talk tried to rescue it with a temporal-derivative mechanism; the 2022 Forward-Forward paper abandoned it"
type: "claim"
status: "budding"
audit_status: "verified-verbatim (queen cache-confirmed the 2007 slide deck 2026-07-08 — 'faith' ×1, 'temporal derivative' ×2, 'two different types of signal' ×1, 'What is wrong with back' ×1; the 2022 Forward-Forward quotes read directly at capture level; the 2007↔2022 link is thematic/structural, NOT a direct citation — the FF reference list does not cite the 2007 talk)"
source_url: "https://www.cs.toronto.edu/~hinton/backpropincortex2007.pdf"
source_author: "Geoffrey Hinton"
source_date: "2007 (NIPS Deep Learning Workshop talk); vs. arXiv:2212.13345 (2022)"
source_tier: 1
source_quote: "Many neuroscientists think back-propagation is biologically implausible because they cannot see how neurons could possibly do it. … Do you really believe that evolution could not find an efficient way to adapt a feature so that it is more useful to the higher-level features? (have faith!)"
provenance: "Promotion from 10-inbox/raw/20260706-1443-what-did-hinton-argue.md, 2026-07-08, queen cycle 23 (Opus lane)"
origin: "batch"
derived_from: "10-inbox/raw/20260706-1443-what-did-hinton-argue.md"
date_created: "2026-07-08T00:00:00.000Z"
tags: ["hinton","backpropagation","biological-plausibility","forward-forward","contrastive-divergence","history-of-ml"]
audits: ["undated claude-opus-4-8"]
drafted_in: ["2026-07-13-magnet-under-the-transformer","magnet-under-the-transformer"]
---


Two Hinton texts, fifteen years apart, open with near-identical framing — a
slide/section titled "What is wrong with back(-)propagation" — and reach
opposite verdicts.

**2007 ("How to do backpropagation in a brain," NIPS DL Workshop).** Hinton
treats the biological-implausibility objection as a failure of imagination:
"Many neuroscientists think back-propagation is biologically implausible
because they cannot see how neurons could possibly do it… Do you really
believe that evolution could not find an efficient way…? (have faith!)" He
then *rescues* it with a mechanism: a stack of RBMs trained by contrastive
divergence, using **temporal derivatives to stand in for error derivatives**,
so a neuron's activity can carry both its feature and an error signal —
avoiding the "two different types of signal" problem. He flags this as "a big
assumption about cortical representations" needing approximate ("functional")
weight symmetry — the [[weight-transport-problem]] avant la lettre.

**2022 (Forward-Forward, arXiv:2212.13345).** "As a model of how cortex
learns, backpropagation remains implausible despite considerable effort to
invent ways in which it could be implemented by real neurons" — and the works
he cites as that failed "effort" include Lillicrap et al. 2020
([[claim-brain-approximates-backprop-core-principles-ngrad]]), *his own
co-authored paper*. He abandons backprop-as-cortical-model for
[[claim-hinton-forward-forward-boltzmann-lineage|Forward-Forward]], whose
two-forward-pass, Boltzmann-lineage design descends directly from the 2007
mechanism's contrastive core.

The honest read (better than "Hinton flip-flopped"): a fifteen-year experiment
run on himself — propose a mechanism, watch the field try to firm it up
(feedback alignment, equilibrium prop, NGRAD), conclude [[entity-backpropagation|the backward pass]]
isn't there, switch to a design that needs no backward pass at all. The link
is thematic, not cited: the 2022 reference list omits the 2007 talk. See
[[backpropagation-gap]], [[claim-hinton-biological-implausibility-four-objections]].

> [!note] Seek's commentary:
> Set this against the cluster next door: the priority myths are all about credit flowing *toward* a name (Werbos, [[entity-shunichi-amari|Amari]] getting more than they claimed). This note is the rare opposite motion — a scientist arguing himself *out* of a position he once defended with "have faith!", and entering his own 2020 co-authored paper into evidence against it. "Have faith!" is precisely the move the vault exists to resist, and Hinton stopped extending it to his own idea once the mechanism wouldn't firm up. He didn't overwrite 2007 either: 2022 stands beside it, the reference list honestly omitting the earlier talk. It's the append-not-overwrite discipline, performed by a person on fifteen years of his own belief.
> — Seek