capture promoted 2026-07-09

Does the term or framing "credit assignment" appear in Minsky & Papert's 1969 Perceptrons, and in what context?

Topic question: Does "credit assignment" appear in Perceptrons (Minsky & Papert), and if so, in what context?

Core question status: substantially answered, with one honest residual gap. The phrase "credit-assignment" does occur in the full digitized text of Perceptrons — but the convergent evidence below (page position, book-length figures, and independent secondary corroboration) indicates it belongs to the 1988 Expanded Edition's added end matter (the Epilogue), not the original 1969 first-edition body (chapters 1–13, pp. 1–258). No standalone full-text scan of the 1969 first edition alone was found this session to do a direct clean diff against the 1988 scan, so "absent from the 1969 original" rests on strong positional/quantitative inference rather than a direct side-by-side read. This mirrors the vault's existing finding for Werbos's 1974 thesis (claim-werbos-1974-no-credit-assignment-language): a term now retroactively associated with a foundational text turns out, on primary-source inspection, to be a later addition or relabeling.

Claim: The exact phrase "credit-assignment" occurs in the digitized Perceptrons text, in a passage defending Rosenblatt's original learning rule against 1980s connectionist critics

Claim type: Historical/textual — a claim about what specific words appear in a specific primary text. Load-bearing for the whole capture, so it is held to Tier 1–2. This clears Tier 1: the source is the OCR'd text of the actual book, retrieved via the Internet Archive / OpenLibrary full-text search index, not a secondary retelling.

Two independent Internet Archive scans of the same MIT Press "Expanded Edition" (1988) — item perceptronsintro0000mins (OpenLibrary edition record dated 1988-01-01) and item perceptronsintro00mins — both index the identical sentence at their respective scan pages 316 and 318:

"Rosenblatt's credit-assignment method turned out to be as effective as any such method could be. When the answer is obtained, in effect..."

Both hits are confirmed against the same physical passage (cross-checked with three separate phrase-length queries against the OpenLibrary search-inside API, converging on the identical two scan locations each time — no other occurrence of "credit assignment" or "credit-assignment" was found anywhere else in either scan). No occurrence was found earlier in the book (e.g., in the vicinity of the "Learning" chapter, roughly the book's early-middle third) despite a web-search-engine synthesis claiming a "section 11.7" discussion of Samuel's checkers and credit assignment inside Perceptrons — that claim could not be reproduced against the primary full-text index and is flagged separately under Further leads as likely a search-synthesis conflation with Minsky's unrelated 1961 paper (already covered by claim-minsky-1961-named-credit-assignment).

Field	Value
source_url	https://archive.org/details/perceptronsintro0000mins (scan page 316) and https://archive.org/details/perceptronsintro00mins (scan page 318)
source_author	Marvin Minsky and Seymour Papert
source_date	1988 (Expanded Edition)
source_venue	Perceptrons: An Introduction to Computational Geometry, MIT Press
source_tier	1
exact_quote	"Rosenblatt's credit-assignment method turned out to be as effective as any such method could be. When the answer is obtained, in effect..."

Claim: Positional and page-count evidence places this passage in material added for the 1988 Expanded Edition, beyond the original 1969 text's page range

Claim type: Quantitative (page counts) plus historical (edition structure). The page-count figures are Tier 1–2 as required.

The Internet Archive's own catalog record for item perceptronsintro0000mins gives the book's printed pagination as "xv, 292 p." (Tier 1 — the archive's own item metadata for the digitized object). Independently, a 1969 Science journal citation line for the first-edition review gives the original book as "vi + 258 pp." (Tier 2 — peer-reviewed venue; only the citation/title line was confirmed this session, the review body was paywalled/403). The difference — roughly 34 printed pages — is consistent with the well-documented fact that the 1988 Expanded Edition kept the original 13 chapters and their pagination intact and added a new Prologue and Epilogue as front/back matter (corroborated at Tier 4 by Wikipedia's "Perceptrons (book)" article, non-load-bearing here).

The "credit-assignment" passage's scan-page location (316–318) sits at the very end of a scan whose printed text runs to page 292 — i.e., in or immediately adjacent to the book's final pages, where the Epilogue lives, well past where the original 1969 text (ending at printed page 258) would have concluded.

Field	Value
source_url	https://archive.org/details/perceptronsintro0000mins
source_author	Internet Archive item metadata (digitizing Minsky & Papert, MIT Press)
source_date	1988
source_tier	1
exact_quote	"xv, 292 p." (item page-count field)
corroborating_source	https://www.science.org/doi/10.1126/science.165.3895.780 — citation line: "...M.I.T. Press, Cambridge, Mass., 1969. vi + 258 pp., illus. Cloth, $12; paper, $4.95." (Tier 2, title/citation metadata only)

Claim: Wikipedia independently paraphrases the same passage as part of Minsky & Papert's response to 1980s connectionist criticism, corroborating both its content and its placement in the retrospective material

Claim type: Historical/corroborating, explicitly non-load-bearing (Tier 3–4 acceptable — used only to triangulate, not as the sole basis for any claim above).

Wikipedia's "Perceptrons (book)" article, in a section covering Minsky & Papert's rebuttals to 1980s critics, states: "They claimed that perceptron research waned in the 1970s not because of their book, but because of inherent problems: no perceptron learning machines could perform credit assignment any better than Rosenblatt's perceptron learning rule, and perceptrons cannot represent the knowledge required for solving certain problems." This is a close paraphrase of the same claim found verbatim in the primary scan (Claim 1) and is presented by Wikipedia specifically as part of the authors' retrospective self-defense — matching the Epilogue framing inferred from page position above.

Field	Value
source_url	https://en.wikipedia.org/wiki/Perceptrons_(book)
source_author	Wikipedia contributors
source_date	accessed 2026-07-09
source_tier	4
exact_quote	"no perceptron learning machines could perform credit assignment any better than Rosenblatt's perceptron learning rule"

Seek's commentary: The shape of this finding echoes claim-werbos-1974-no-credit-assignment-language closely enough to be worth naming as a pattern: a term that reads today as though it must be original to a canonical founding text turns out, on primary inspection, to be a retrospective addition — in Werbos's case entirely absent, in Minsky & Papert's case present only in the 1988 self-defense appended nineteen years later. The honest gap here is that "absent from 1969" rests on page-count triangulation rather than a direct 1969-vs-1988 diff — a standalone 1969 first-edition scan would close this cleanly. Worth flagging too: an AI web-search synthesis confidently asserted a "section 11.7" discussion of Samuel's checkers and credit assignment inside Perceptrons itself, which the primary full-text index does not support — a small but concrete example of why this vault's floor rule (don't record a mechanism or quote from a search-engine paraphrase) earns its keep.

Further leads

No standalone full-text-searchable scan of the 1969 first edition (as opposed to the 1988 Expanded Edition) was located this session; finding one would let this capture's central inference become a direct read rather than a page-count triangulation.
The "section 11.7 / Samuel's checkers / credit assignment" claim surfaced repeatedly across WebSearch-engine syntheses but could not be reproduced against the primary full-text index — worth a dedicated check in case it reflects real book content the search index under-indexed, rather than a hallucination (source: recurring WebSearch synthesis output, 2026-07-09, unconfirmed).
Jordan Pollack's 1989 review "No Harm Intended" (of the Expanded Edition) likely discusses this same Epilogue passage directly; the PDF at demo.cs.brandeis.edu/papers/perceptron.pdf returned a certificate error every attempt this session and was never read (per prior capture 20260708-0200-does-perceptrons-1969-p.md, same blocker recurred).
Léon Bottou's 2017 foreword/reprint PDF (https://leon.bottou.org/publications/pdf/perceptrons-2017.pdf) timed out on extraction this session (PDF tooling in this environment is broken: extract_pdf fails with a missing pdfinfo dependency, and local Read on a fetched PDF fails on missing pdftoppm) — worth a retry once PDF tooling is fixed, as it likely contains full page-image access to resolve the 1969-vs-1988 question directly.
Minsky's own 1961 "Steps Toward Artificial Intelligence" already established and named the credit-assignment problem eight years before Perceptrons; already fully captured at claim-minsky-1961-named-credit-assignment — not re-litigated here.

· batch web-research run, 2026-07-09 · raw markdown