An engineer imagined linked ideas, not machine-written prose

The writer's problem

At the end of the Second World War, the outgoing director of the U.S. Office of Scientific Research and Development turned his attention from weapons research to a civilian problem: the growing mass of scientific and scholarly material had outrun any human's ability to find, connect or recall it. Vannevar Bush's task was not to build a machine, since none of what he proposed had been built, but to describe, in a widely read magazine, what a machine assisting a reader's own thinking might look like, and to be precise about what such a device would and would not do.

What the documents show

Bush's essay, 'As We May Think', published in The Atlantic's July 1945 issue, proposes a desk-sized device he names the 'memex,' which stores a user's books, records and correspondence on microfilm and lets the user join any two items into a permanent, named 'trail' by tapping a key, then retrace that trail later or branch from it. Bush is explicit that the machine performs 'associative indexing,' letting 'any item... select immediately and automatically another' item the user has linked, not that it composes, drafts or originates any text of its own. An independently hosted digitized copy of the original magazine text confirms the wording is unchanged from the 1945 publication: the memex only recalls and connects what a human has already written and chosen to link.

The essay's mechanism, tying items into a browsable trail rather than filing them into a fixed hierarchy, is the same basic move later hypertext and note-linking tools made operational. Bush frames the memex as an extension of a reader's own associative memory, not as a replacement for a writer's judgment about what to link.

The editorial choice

Reading 'As We May Think' as an early argument for machine-assisted retrieval and cross-reference, rather than for machine-generated prose, is an editorial distinction worth holding onto whenever the essay is cited as an ancestor of today's AI writing tools. Citing Bush to justify a tool that generates or drafts text on its own reaches past what the essay itself proposes; citing him for tools that help a writer find, link and revisit their own sources stays closer to his actual claim.

What stays with the author

Bush's memex leaves every act of selection, linking and trail-building to its human user. The essay does not resolve, and was never asked to resolve, what should happen once a tool starts generating connections or text on its own rather than only storing the ones a person made.

• Does the tool being described only retrieve and connect what a person already wrote, or does it generate new text?
• Would Bush's own description of 'associative indexing' still apply once a machine chooses the connections?
• Is the memex being cited as a precedent for retrieval, or stretched into a precedent for authorship it never claimed?

A 1945 magazine essay about a desk that was never built still sets a useful boundary: it imagined machines that help a mind find its own material, not machines that decide what the mind should say.