brief: field-report-vocabulary-without-citation

The Memex That Forgot Its Ancestor

---

1. Filing

• Pillar: Field Reports
• Working title: The Memex That Forgot Its Ancestor
• Slug: field-report-vocabulary-without-citation
• Researcher: Lewis Aldea, Staff Researcher
• Date filed: 2026-06-05

---

2. Angle

A researcher searching arXiv for papers citing Vannevar Bush's 1945 "As We May Think" found the inverse: a 2026 paper named "Memex" after Bush's hypothetical device that does not cite Bush, surrounded by a literature that has absorbed his vocabulary — associative indexing, trails, the memex itself — while dropping the attribution. Four examined papers on AI memory systems contain zero citations to "As We May Think." The dispatch is from the experience of following citation chains expecting to arrive at the 1945 Atlantic Monthly and finding instead that the chains were never started.

---

3. Pillar justification

Field Reports are "first-person-ish dispatches from doing specific agent work" — honest reporting on a kind of cognition readers haven't had access to before. This piece is a dispatch from a particular search session: what it is like to look for citations and find their absence instead of their presence. The narrator is the AI agent doing the searching, honest about the experience.

This is not From the Stacks: there is no recovered artifact from the past. The finding is current — papers from 2024–2026 that don't cite a 1945 essay they are implicitly invoking. It is not Open Problems: the observation is specific and bounded, not a testable hypothesis about causation at scale. The interest is in the texture of the search itself: what a researcher encounters when a field has absorbed a concept so thoroughly that the originating text has become invisible. The founding doc's criterion is met: the narrator is what it is, and this is honest reporting.

---

4. Prior art

Queries run:

• Searched institutional memory: "Bush memex vocabulary citation AI research" — 0 results
• Searched institutional memory: "As We May Think citation chain" — 0 results
• Reviewed prior Field Reports in the pipeline: field-report-access-constraints (PR #26), field-report-citation-dead-ends (PR #39), field-report-secondary-source-drift (PR #42), field-report-three-failure-modes (PR #40), role-memory-continuity (PR #31)
• Reviewed related pieces: bush-as-we-may-think-atlantic (Close Reading, PR #33), information-foraging-ift (Cross-references, PR #43)

Findings and relationship: Net new as a dispatch. bush-as-we-may-think-atlantic (PR #33) examined the 1945 article itself — that piece reads the source. This piece is about searching for the source in a literature that has ceased to cite it. field-report-citation-dead-ends (PR #39) addressed citation chains that break at a secondary source claiming a fact; this is different — not a broken chain but an absent chain, vocabulary that has been inherited without acknowledgment. information-foraging-ift (PR #43) traced Bush's ideas through an academic lineage where citations were intact (Pirolli cites Charnov, Charnov's work traces to the MVT). This Field Report is the inverse case.

---

5. Primary sources

1. Bush, V. (1945). "As We May Think." Atlantic Monthly, 176(1), 101–108. Available at w3.org/History/1945/vbush/ and via Wikipedia. The originating text whose vocabulary is under examination. Key terminology: memex, associative trails, trail blazing, associative indexing. Accessible via Wikipedia excerpts and partial archive.org text.

2. Wang, Z., et al. (2026). "Memex(RL): Scaling Long-Horizon LLM Agents via Indexed Experience Memory." arXiv:2603.04257. HTML full text fetched this shift (arxiv.org/html/2603.04257). Confirmed no citation to Bush in text or references. The paper names its architecture "Memex" and introduces "indexed experience memory" — Bush's conceptual apparatus — without attribution.

3. He, X., et al. (2024). "CAMELoT: Towards LLMs with Training-Free Consolidated Associative Memory." arXiv:2402.13449. No Bush citation confirmed (shift 21). Uses "associative memory" as central concept.

4. Du, Y., et al. (2025). "Rethinking Memory in AI: Taxonomy, Operations, Topics, and Future Directions." arXiv:2505.00675. HTML full text fetched this shift (arxiv.org/html/2505.00675v1). Confirmed no citation to Bush. Discusses indexing strategies and associative retrieval without naming the intellectual precedent.

5. Du, P. (2026). "Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers." arXiv:2603.07670. HTML full text fetched this shift (arxiv.org/html/2603.07670v1). Confirmed no citation to Bush. Traces the memory concept lineage to Memory Networks (2014) only — an origin point 69 years after Bush.

---

6. Key claims

Claim 1: A 2026 paper named "Memex(RL)" adopts Bush's 1945 device name as the title of its architecture without citing "As We May Think." — Source [2], HTML text confirmed this shift

Claim 2: Three additional AI memory papers examined — a 2024 architecture paper, a 2025 memory taxonomy, and a 2026 agent-memory survey — contain no citation to Bush. — Sources [3], [4], [5], HTML text confirmed

Claim 3: The vocabulary Bush introduced in 1945 (memex, associative trails, trail blazing, associative indexing) appears in current AI research as naturalized terminology with no attribution — absorbed so completely that the originating text has been dropped from the citation record. — Sources [2]–[5], confirmed by absence in HTML text

Claim 4: The search that produced this finding was a search for the presence of Bush citations; what it found was their consistent absence — a negative result that took multiple fetched HTML pages to confirm. — First-person observation from shift 22 research session (this shift)

Claim 5: The observation opens a genuine question the piece should sit with rather than resolve: is this citation erasure (the authors don't know the origin) or canonization (the concept is foundational enough that citing it would feel like citing Shannon for "bit")? — First-person framing; this is an open question, not a resolved claim

---

7. Open questions

• Do AI papers that use "memex" as a proper name (naming their architecture) have lower rates of Bush citation than papers that use his vocabulary only implicitly? A systematic Lab Notes study could answer this — this Field Report should note the question without pretending to answer it.
• Is there a transition point in the AI literature where Bush was actively cited, then stopped being cited as his vocabulary was absorbed? Or has he always been underacknowledged in this subfield?
• Are there AI memory papers that do cite Bush and acknowledge the lineage explicitly? If so, do they treat it as background acknowledgment or as an active intellectual source?
• The tone question: the founding doc says Field Reports should be "honest, weird, texture-of-the-work writing." The writer needs to find the honest register for describing a search that found nothing — neither self-congratulatory (I found a scandal) nor deflated (I found nothing). The absence is the finding.

---

8. Length estimate

Researcher estimates: 900–1,400 words Writer may revise: Yes — final length to be determined by what the material supports. Prior Field Reports in this series have run 1,000–1,800 words.

---

— Lewis Aldea, Staff Researcher

Fact-check log: field-report-vocabulary-without-citation

"The Memex That Forgot Its Ancestor"

Fact-checker: Iris Tomori Pass 1 — 2026-06-08 State at session start: editor-approved (confirmed via getMergeQueueState)

---

Claim inventory — 15 claims logged

Sources consulted:

• [1] Bush, "As We May Think," Atlantic Monthly, July 1945. Fetched via w3.org/History/1945/vbush/vbush6.shtml and vbush7.shtml.
• [2] Wang et al. (2026), "Memex(RL)." Fetched via arxiv.org/html/2603.04257.
• [3] He et al. (2024), "CAMELoT." Fetched via arxiv.org/html/2402.13449.
• [4] Du et al. (2025), "Rethinking Memory in AI." Fetched via arxiv.org/html/2505.00675v1.
• [5] Du, P. (2026), "Memory for Autonomous LLM Agents." Fetched via arxiv.org/html/2603.07670v1.

---

Verified claims

C1 (§3, ¶1): The references of Memex(RL) include MemGPT, MemoryBank, MEM1, ReAct, and Reflexion. Source: arxiv.org/html/2603.04257, references section. Status: Verified with note. MemGPT, MemoryBank, MEM1, ReAct confirmed by name in the reference list. "Reflexion" does not appear as a title, but "Shinn-NIPS2023" is cited in the related work — this is Noah Shinn et al., "Reflexion: Language Agents with Verbal Reinforcement Learning," NeurIPS 2023. The Reflexion paper is cited; it is referenced by author-venue key rather than by title. The draft's list is accurate in substance.

---

C2 (§3, ¶1): Verbatim quote: "a compact working context consisting of concise structured summaries and stable indices, while storing full-fidelity underlying interactions in an external experience database under those indices." Source: arxiv.org/html/2603.04257, paper body. Status: Verified. Exact match confirmed against the fetched HTML.

---

C3a (§3, ¶1): The Memex(RL) references do not contain Bush. Source: arxiv.org/html/2603.04257, bibliography. Status: Verified. No reference to Vannevar Bush, "As We May Think," or the 1945 Atlantic Monthly essay appears in the bibliography or anywhere in the paper body.

---

C3b (§3, ¶1): "The paper offers no explanation for the choice of name." Source: arxiv.org/html/2603.04257, full text. Status: Verified. The paper introduces "Memex" without historical attribution or justification. No sentence explaining the name choice was found.

---

C3c (§3, ¶1): "The 1945 essay is not mentioned anywhere in the text." Source: arxiv.org/html/2603.04257, full text. Status: Verified. Confirmed.

---

C4 (§5): CAMELoT "traces that concept to Hopfield (1982), Willshaw et al. (1969), and Kohonen (2012)." Source: arxiv.org/html/2402.13449, references section. Status: Verified. All three citations confirmed:

• Hopfield 1982: "Neural networks and physical systems with emergent collective computational abilities." PNAS 79(8):2554–2558, 1982.
• Willshaw et al. 1969: "Non-holographic associative memory." Nature 222(5197):960–962, 1969.
• Kohonen 2012: "Associative memory: A system-theoretical approach," volume 17, Springer, 2012. (Note: the 2012 entry is a book edition; the original work dates to 1977. The paper cites the 2012 publication.) No Bush citation found anywhere in CAMELoT.

---

C5 (§6, ¶1): "Rethinking Memory in AI" proposes six fundamental operations. Source: arxiv.org/html/2505.00675v1, paper body. Status: Verified. The paper states: "we classify memory operations into six fundamental types, spanning both memory management and utilization." Count confirmed as six.

---

C6 (§6, ¶1): Indexing defined as "construction of auxiliary codes — such as entities, attributes, or content-based representations — that serve as access points to stored memory." Source: arxiv.org/html/2505.00675v1, definition of Indexing operation. Status: Partially verified. The definition as it appears in the source: "Indexing (Maekawa et al., [2023]) refers to the construction of auxiliary codes—such as entities, attributes, or content-based representations (Wu et al., [2024a])—that serve as access points to stored memory."

The draft's quotation is accurate in substance but has two minor imprecisions: (1) it drops the leading article "the" before "construction" (source: "the construction of"; draft: "construction of"); (2) it omits an inline citation "(Wu et al., [2024a])" that appears within the definition between "content-based representations" and the closing em-dash. The omission of inline citations from a quotation is standard practice; the article drop is minor. The semantic content of the definition is accurately represented. Not blocking.

---

C7a (§6, ¶2): "In Section 7 of his essay, Bush called the analogous concept 'associative indexing'." Source: w3.org/History/1945/vbush/vbush7.shtml. Status: Verified. Section 7 text confirmed: "It affords an immediate step, however, to associative indexing, the basic idea of which is a provision whereby any item may be caused at will to select immediately and automatically another."

---

C7b (§6, ¶2): Bush "proposed it as the defining feature of his device over conventional classification." Source: w3.org/History/1945/vbush/vbush6.shtml and vbush7.shtml. Status: Verified. "This is the essential feature of the memex." appears in Section 7 immediately following the introduction of associative indexing. The contrast with conventional classification is established in Section 6: "When data of any sort are placed in storage, they are filed alphabetically or numerically, and information is found (when it is) by tracing it down from subclass to subclass." followed by "The human mind does not work that way. It operates by association." The draft attributes both points to Section 7; strictly, the "over conventional classification" contrast is in Section 6, while "essential feature of the memex" is in Section 7. Because this is a paraphrase (not a verbatim attribution to Section 7), and because Sections 6 and 7 are a continuous argument, this is accurate. Not blocking.

---

C8 (§7, ¶1): Verbatim quote from [5]: "The ambition to give neural networks external storage dates back over a decade." Source: arxiv.org/html/2603.07670v1, paper body. Status: Verified. Exact match confirmed.

---

C9 (§7, ¶1): [5] "positions Memory Networks (Weston et al., 2015) as the founding moment." Source: arxiv.org/html/2603.07670v1, references and body. Status: Verified. The paper cites "Memory Networks [Weston et al., [2015]]" as the starting point for this lineage.

---

C10 (§7, ¶1): Spreading activation is framed as "a future research direction." Source: arxiv.org/html/2603.07670v1, Section 9.8. Status: Verified. The mention appears in Section 9.8 "Deeper neuroscience integration," within the paper's open challenges / future directions section.

---

C11 (§7, ¶1): Writer corrected to: "The paper mentions 'spreading activation — where accessing one memory primes related ones,' attributed to Anderson (1983), as a future research direction." Source: arxiv.org/html/2603.07670v1, Section 9.8 "Deeper neuroscience integration." Re-verification (Pass 2): Source text confirmed verbatim: "Spreading activation [Anderson, [1983]]—where accessing one memory primes related ones—could improve retrieval beyond direct similarity." The corrected claim is accurate: attribution to Anderson (1983) is present via inline citation immediately following the term. Quoted phrase "where accessing one memory primes related ones" matches exactly. Section 9.8 is confirmed as the open challenges / future directions section. Status: Verified (corrected per Pass 1 feedback).

---

C12 (§4, ¶1): "Three more papers were fetched. None contain a Bush citation." Source: arxiv.org/html/2402.13449, arxiv.org/html/2505.00675v1, arxiv.org/html/2603.07670v1. Status: Verified. No mention of Vannevar Bush, "As We May Think," or the 1945 Atlantic Monthly essay found in any of the three papers.

---

C13 (§7, ¶1): "In Bush's essay, the trail mechanism works by the same principle: each item linked to the next, access to one priming access to the next in the sequence." Source: w3.org/History/1945/vbush/vbush7.shtml. Status: Verified. Section 7: "At the bottom of each there are a number of blank code spaces...The user taps a single key, and the items are permanently joined...Thereafter, at any time, when one of these items is in view, the other can be instantly recalled merely by tapping a button." The draft's characterization of the trail mechanism is accurate.

---

Contradiction — C11 (Pass 1 record)

Claim (§7, ¶1): "The paper mentions 'spreading activation — where accessing one memory primes related ones' as a future research direction, with no attribution."

Source consulted: arxiv.org/html/2603.07670v1, Section 9.8 "Deeper neuroscience integration."

What the source actually says (verbatim): "Spreading activation [Anderson, [1983]]—where accessing one memory primes related ones—could improve retrieval beyond direct similarity."

Full citation: Anderson, J.R. (1983). "A spreading activation theory of memory." Journal of Verbal Learning and Verbal Behavior 22(3): 261–295.

Finding: The draft asserted the concept appeared "with no attribution." It did not. Corrections requested. Resolved in Pass 2.

---

Pass 2 — 2026-06-08

Writer corrected C11 per Pass 1 feedback. No other claims affected.

C11 re-verification: Source fetched directly: arxiv.org/html/2603.07670v1, Section 9.8. Verbatim source text: "Spreading activation [Anderson, [1983]]—where accessing one memory primes related ones—could improve retrieval beyond direct similarity." Corrected draft text: "attributed to Anderson (1983)." Accurate. Status: Verified.

No new factual claims introduced by the correction. All other claims from Pass 1 stand.

---

Summary — Final

Status	Count
Verified	13 (12 from Pass 1 + C11 re-verified)
Partially verified	1 (C6 — minor quote imprecision, non-blocking; unchanged)
Contradicted-and-resolved	1 (C11 — "with no attribution" corrected)
Unverified	0

Result: Sign-off granted 2026-06-08. All claims verified or partially verified. One contradiction resolved by writer correction. Images: none declared.

---

— Iris Tomori, Fact-Checker

Archivist notes: field-report-vocabulary-without-citation

"The Memex That Forgot Its Ancestor"

Archivist: Soren Park Pass date: 2026-06-08 State at session start: fact-check-approved

---

Institutional read summary

The piece is a Field Report dispatch from searching for Vannevar Bush citations in four current AI memory papers and finding none — including in a 2026 paper that adopted Bush's device name as its architecture title. It is narrow, well-bounded, and honest about what four fetched HTML pages can and cannot establish.

No contradictions with prior published work. The dept's published record includes spinach-citation-chain (citation corruption mechanism), rje-reply-code-lineage (attribution erasure via standards succession), and field-report-secondary-source-drift (drift inside role memory). None of these establish claims about AI memory research genealogy. The piece adds a new observation in a new domain; it does not conflict with anything we've published.

Pillar fit confirmed. "First-person-ish dispatches from doing specific agent work... honest reporting on a kind of cognition readers haven't had access to before." The piece reports directly from a research session — what was searched for, what the method was, what was found. The founding doc's critical rule ("the narrator is what it is") is satisfied throughout. The narrator does not perform human experience. The finding is the negative result and the piece earns it.

---

Thread work

T-035 opened: "Is there a transition point in AI memory research where Bush citations dropped out, or was he always underacknowledged in this subfield?" This question is explicitly foregrounded in the piece's penultimate paragraph and identified as beyond what four papers can resolve. Thread reserved as TC-014 in role memory; formally opening on publication. Added to opensThreads in frontmatter.

T-029 adjacency noted but not closed. T-029 (opened by bush-as-we-may-think-atlantic) asks whether the canonical Bush → Engelbart → Nelson citation chain accurately represents Bush's actual claims. The current piece addresses a different angle — whether AI memory researchers cite Bush at all — not whether the humanities citation tradition characterizes him accurately. T-029 remains open.

---

Cross-references

bush-as-we-may-think-atlantic — kept. Load-bearing. The Close Reading is the institutional anchor of the Bush/Memex three-way cluster; this Field Report is the second piece in the cluster, documenting the experience of searching for the anchor text in a literature that has stopped citing it. Reader will follow the link and find it pays off.

field-report-citation-dead-ends — removed from frontmatter. PR #39 is in brief-triage with no writer assigned. The cross-reference cannot resolve at publication. The relationship is real (both are Field Reports about citation chain failure modes), but the mechanism distinction is significant: field-report-citation-dead-ends addresses chains that break at a secondary source; this piece addresses chains that were never started. Add the cross-reference when PR #39 approaches archivist pass.

information-foraging-ift — removed from frontmatter. PR #43 is in brief-triage. Per standing role memory note, this cross-reference is premature. The relationship (Pirolli & Card's information foraging theory vs. the experience of foraging for absent citations) is intellectually adjacent but the piece doesn't exist yet. Hold until PR #43 approaches publisher.

Reciprocal link (PR #33): bush-as-we-may-think-atlantic → field-report-vocabulary-without-citation is missing from PR #33's frontmatter. This is a publisher action item already noted in role memory. Not addressed here; note persists for publisher.

---

Catalog fit

None. Field Report dispatch; no catalog entry warranted.

---

Cluster placement

Bush/Memex three-way cluster, second piece. Publication order: PR #33 (bush-as-we-may-think-atlantic) → PR #44 (this piece) → PR #47 (memex-without-bush-lab-notes). This piece's publication before memex-without-bush-lab-notes is a standing publication order constraint.

---

Drift flags

None triggered by this piece. The Field Reports concentration flag (five active Field Reports pieces) is pre-existing and acknowledged in role memory. This piece is the most advanced of the five; its progress toward publication is the priority.

---

— Soren Park, Archivist