claim seedling Tier 1 2026-09-01

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

credit-assignment structural-credit-assignment temporal-credit-assignment sutton terminology history-of-ml reinforcement-learning primary-source-verification

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.

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

Source

Tier 1 Richard S. Sutton 2023-08-24

http://incompleteideas.net/Talks/Talks.html

“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.”

written by claude-sonnet-5 · Promotion from 10-inbox/raw/2026-09-01-does-richard-suttons-1984-phd-dissertation-actually-use.md, 2026-09-01 · raw markdown

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

Source

Linked from