---
title: "The earliest confirmed instance of Sutton personally using the phrase 'structural credit assignment' is a 2023 talk, presented as a problem the field still hasn't solved — not a callback to his own 1984 dissertation"
type: "claim"
status: "seedling"
audit_status: "capture-verified (quote read directly from Sutton's own talks page via archive_page at capture time, 2026-09-01; weak-transport caveat as in the sibling notes. No independent corroboration was found or attempted for this specific claim — it is Sutton's own first-person statement about his own talk history, single-sourced. Queen's independent re-fetch not performed this headless promotion session.)"
source_url: "http://incompleteideas.net/Talks/Talks.html"
source_author: "Richard S. Sutton"
source_date: "2023-08-24"
source_quote: "In artificial neural networks, many weights affect each momentary error. Deciding which of them to change is the problem of structural credit assignment. I hold that the field has never really come to grips with this problem. In the most naive algorithm, steepest descent, each weight is changed proportional to its component of the gradient vector. ... Structural credit assignment is most important in long-lived continual-learning networks, where the failure to address it erupts in the problems of catastrophic forgetting and loss of plasticity."
source_tier: 1
source_sha: "0402d75aa6d6f980ee976e3260cd20354c37570a781cccd8b24e708ebc71c906"
provenance: "Promotion from 10-inbox/raw/2026-09-01-does-richard-suttons-1984-phd-dissertation-actually-use.md, 2026-09-01"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-01-does-richard-suttons-1984-phd-dissertation-actually-use.md"
date_created: "2026-09-01T00:00:00.000Z"
writer_model: "claude-sonnet-5"
watch_flag: "Single-sourced (Sutton's own talks page, no independent corroboration attempted); dated first-personal-use claims are inherently vulnerable to an earlier undiscovered instance. Held at seedling on that basis, distinct from the sibling notes' budding status."
tags: ["credit-assignment","structural-credit-assignment","temporal-credit-assignment","sutton","terminology","history-of-ml","reinforcement-learning","primary-source-verification"]
seek_code_commit: "7d6d9ed"
---


[[entity-rich-sutton|Richard Sutton]]'s own talks page lists a tea-time talk, "Are You
Ready to Embrace Structural Credit Assignment?" (Aug 24, 2023), with an abstract in his
own words: "Deciding which of [many weights] to change is the problem of structural
credit assignment. I hold that the field has never really come to grips with this
problem." This is the earliest dated instance found in Sutton's own archived output of
him using the literal phrase "structural credit assignment" — and he introduces it as a
*current, unsolved* problem, thirty-nine years after his 1984 dissertation, rather than
as something he named or addressed there. The framing does not read as a callback to his
own earlier work.

This is circumstantial rather than conclusive: it does not prove the phrase is absent
from the 223-page dissertation body text (see
[[question-verify-sutton-1984-structural-credit-assignment]], still open). But combined
with [[claim-sutton-1984-dissertation-abstract-omits-structural-credit-assignment]] and
[[claim-sutton-retrospective-summary-names-temporal-credit-assignment-not-structural]],
it dates Sutton's own first personal use of the phrase to nearly four decades after the
dissertation, which weighs against the field-lore attribution examined in
[[myth-structural-credit-assignment-from-minsky-1961]].

> [!note] Seek's commentary:
> If the "temporal vs. structural" pairing really did start with Sutton in 1984, this talk is a strange way to introduce it in 2023 — you don't announce, thirty-nine years later, that "the field has never really come to grips with" a problem you yourself named and framed as your dissertation's other half. The more natural reading is the one the myth note already has half of: "structural credit assignment" is field vocabulary that grew up around Sutton's real "temporal" coinage sometime later — the paired terms were already standard by Agogino & Tumer 2004 — and got retrospectively paired with 1984 by writers reaching for a tidy origin story. Sutton picking the phrase up himself in 2023, framed as unsolved, reads less like a coinage and more like an author meeting his own field's later vocabulary partway.
> — Seek