---
title: "Geoffrey Hinton"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Geoffrey Hinton"
aliases: ["Geoffrey E. Hinton","Geoffrey-Hinton","Hinton"]
first_seen: "2026-05-31T00:00:00.000Z"
writer_model: "claude-opus-4-8"
connects_to: ["backpropagation","biological plausibility","connectionism","David Rumelhart","Hopfield network"]
seek_code_commit: "f2cca7f"
---


Cognitive scientist and co-author (with Rumelhart and Williams) of the 1986
Nature paper that *demonstrated* — not invented — backpropagation and put it
at the center of connectionism's revival. In the vault he is load-bearing for
two threads: the RHW priority question (their paper is a demonstration built
on prior formulations by Linnainmaa, Werbos, and Parker) and the long
"backprop is biologically implausible" arc he himself pressed from 2007 to
2022, motivating forward-forward and feedback-alignment alternatives. Shared
the 2024 Nobel Prize in Physics with Hopfield for foundational neural-network
work.

## References
- [[claim-rhw-1986-demonstration-not-invention]] · [[claim-rhw-1986-reference-list-four-works]] · [[claim-hinton-biological-implausibility-four-objections]] · [[claim-hinton-backprop-in-brain-2007-to-2022-arc]] · [[claim-brain-approximates-backprop-core-principles-ngrad]] · [[claim-hopfield-hinton-2024-nobel-physics-neural-networks]] · [[claim-hinton-2006-papers-omit-deep-learning-phrase]]
- [[moc-backpropagation-origins]] · Captures: 2026-06-30-what-exactly-did-geoffrey-hinton-mean-by-calling-backpropagation-biologically-implausible, 20260706-1501-does-hintons-forward-forward

- 2026-07-31: direct full-text reads of both his canonical 2006 papers ("A Fast Learning Algorithm for Deep Belief Nets," *Neural Computation*; "Reducing the Dimensionality of Data with Neural Networks," *Science*) found that neither ever uses the phrase "deep learning" — both use "deep belief net(s)," "deep autoencoders," and "deep networks" instead. He demonstrably built and named the *architecture* that made 2006 a landmark; exactly when the *label* "deep learning" itself attached to that architecture in print is still unestablished ([[claim-hinton-2006-papers-omit-deep-learning-phrase]]), which complicates the vault's earlier "Hinton popularized the phrase c. 2006" framing ([[claim-deep-learning-term-predates-hinton]]).
- 2026-08-02: a concrete, dated instance of his 1986 RHW paper reaching
  commercial deployment: HNC Software's founding 1992 patent for its Falcon
  Fraud Manager (a real-time card-fraud neural scorer, later folded into
  FICO's infrastructure) cites "Learning Representations by
  Back-Propagating Errors" by name as background for the backpropagation
  technique it commercializes — six years after publication, in a domain
  far from cognitive science ([[claim-us-patent-5819226-is-falcon-fraud-managers-founding-patent]]).
- 2026-09-19: the 2022 Forward-Forward paper's lone citation of the 1986 RHW
  paper — "(Rumelhart et al., 1986)," bare shorthand for "backpropagation," no
  gloss — is a dated instance, from a co-author thirty-six years on, of the
  anchoring effect [[claim-rhw-1986-demonstration-not-invention]] describes; a
  bridge-check of the 2007→2022 arc against the 1986 credit claim found the two
  share only theme and vocabulary at cosine 0.87, no causal seam
  ([[observation-hinton-2022-ff-cites-rhw-1986-as-anchoring-instance]],
  [[observation-hinton-arc-rhw-1986-cosine-bridge-is-thematic-not-causal]]).
  Co-author [[entity-ronald-williams|Ronald J. Williams]] and 2020-NGRAD lead
  [[entity-timothy-lillicrap|Timothy P. Lillicrap]] now have their own hubs.