The Memex Lineage

Begin in July 1945, in *The Atlantic Monthly*. Vannevar Bush, the wartime science administrator who has just finished running the U.S. effort that produced radar, the proximity fuze, and the Manhattan Project, writes a 9,000-word essay called 'As We May Think.' He names a problem: the rate at which humans are recording knowledge has vastly outstripped the rate at which any human can consult it. He names a cause: the hierarchical filing systems we use fight against the way the mind actually retrieves information, which is associative. He proposes a device — the Memex — that stores all your books and lets you build *associative trails* between items, sharable to colleagues. He predicts a profession of trail blazers. He does not solve the maintenance problem. The conceptual primitive — the associative trail — is established, and the rest of the lineage runs from this essay.

↓ is built on by

Seventeen years later, at Stanford Research Institute, Doug Engelbart publishes *Augmenting Human Intellect: A Conceptual Framework*. The framework extends Bush from passive retrieval to *interactive augmentation* — a tight reciprocal loop between human and machine, where the machine extends what the human can hold in mind, and the human steers what the machine pays attention to. Augmentation, not automation. In 1968 his team gives the *Mother of All Demos* — the first public demonstration of hypertext, the mouse, real-time collaborative editing, video conferencing, and direct-manipulation graphics — all in 90 minutes. Most of what he showed took the industry twenty years to ship. The bootstrap principle — using the augmentation tools to improve the augmentation tools, recursively — is the part the industry never picked up.

↓ is paralleled by

While Engelbart is building working systems, Ted Nelson is designing the document substrate they should run on. He coins *hypertext* in 1965, then *hypermedia* and *transclusion*. His Project Xanadu specifies an unusually demanding set of seventeen requirements: bidirectional links that never break, transclusion of source material with its context preserved, version preservation across all edits, automatic micropayment to authors at the level of every quoted passage. *Computer Lib / Dream Machines* (1974) and *Literary Machines* (1981) elaborate the vision. Most of Xanadu has never shipped publicly. Nelson is the figure in the lineage who refuses to let the lossy form of hypertext be confused with the faithful one. Whether the world has been right to choose the lossy form is a question he has spent sixty years arguing was answered wrong.

↓ is shipped (lossily) by

At CERN, Tim Berners-Lee designs and implements HTTP, HTML, and URL together in eighteen months. The web ships in 1991. The compromises are deliberate: one-directional links, no transclusion, no persistent identifiers, no version preservation, no built-in author payment. The trades make the system shippable on existing infrastructure with no central authority — and the trades make the web possible at internet scale. Bush's vision deploys, finally, at the cost of most of what Nelson and Engelbart had asked for. The deferred costs of those trades — link rot, vanished sources, the attention economy patching the missing payment layer — are the price the future is now paying. Berners-Lee has spent the last twenty years working on Solid and the Semantic Web, attempting to repair the gaps the original deployment left.

↓ is maintained at scale by

Twelve years after the web ships, Jimmy Wales and Larry Sanger launch Wikipedia. Within five years it is the largest encyclopedia in human history. Within a decade it is the most-visited reference site in the world. Wikipedia answers Bush's open question: who maintains the trails. The answer is *thousands of volunteers governed by editorial policies* — neutral point of view, verifiability through reliable sources, no original research. The governance produces uniform usefulness at the cost of flattening traditions that disagree. But the maintenance is real, and it works, and it has continued to work for twenty-five years. Wikipedia is the institutional layer of the lineage — the proof that population-scale collaborative knowledge maintenance is possible, with the specific tradeoffs its institutional design requires.

↓ is now extended by

Eighty years after Bush's essay, the maintenance problem is solved at the personal scale. The LLM is cheap enough to read 15 pages, update cross-references, flag contradictions, and integrate a new source — at near-zero marginal cost per maintenance pass. The trail blazers Bush predicted are now a tool: an LLM doing the maintenance work, in partnership with humans who supply attention, judgment, and curated corpora. This platform — 0mn1.one — is one place where this work is being done. The wiki you are reading is the substrate; the trails (like this one) are the associative paths Bush imagined; the LLM that maintains the wiki is the trail blazer made functional. The lineage closes a long arc here, with substantial work still ahead — the empirical literature on LLM-augmented cognition is young. But the structural pattern Bush named in 1945 has, eighty years later, an instantiation that works for one person at a time, on commodity hardware, with maintenance at a cost the user can afford. The Memex is finished — for now.