---
title: "Does the 2024 Nature comment (Smith, Kirby, Guo, Griffiths) actually propose iterated learning as an explanation and fix for AI model collapse — in its own words?"
type: "question"
status: "open"
date_raised: "2026-07-11T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["model-collapse","iterated-learning","cultural-evolution","source-verification","paywall"]
---


[[claim-iterated-learning-theory-reframes-model-collapse-as-cultural-evolution]]
rests partly on a 2024 *Nature* comment, "AI model collapse might be prevented
by studying human language transmission" (Kenny Smith, Simon Kirby, Shangmin
Guo, Thomas L. Griffiths; *Nature* 633, 525, 2024-09-19). At capture time the
comment's body text was **paywalled**; its framing — that Kirby's
iterated-learning paradigm can explain and help prevent model collapse — was
corroborated only through a Princeton faculty repository record, not read
verbatim. The `source_quote` on the note is effectively the article's *title*,
not a substantive line from the argument.

**What would answer it.**

- Read the primary text of *Nature* 633, 525 (institutional access, author
  postprint, or a Wayback capture of an open version) and pull a verbatim
  sentence that states the iterated-learning → model-collapse proposal.
- Confirm the four-author attribution and the exact claim strength: does the
  comment *propose* iterated learning as an explanation, merely *suggest*
  studying language transmission, or claim something stronger about
  prevention?

**Why it matters.** This is the origin claim of the 2024→2026 arc. Until the
comment is read in its own words, the note's account of what it argued is a
secondary paraphrase, and the note stays `status: seedling` /
`audit_status: flagged`. Resolving this either upgrades the leg to a verbatim
Tier-1 citation or corrects an overstatement inherited from the capture.