What was the memex, and why does Vannevar Bush's 1945 essay still matter?

In July 1945, as the Second World War was ending, the engineer who had run the American scientific war effort asked a peaceful question: once the fighting stops, how do we keep up with everything we know? Vannevar Bush's answer, an essay called "As We May Think", described a personal machine he named the memex. It was never built. Yet the problem it tackled, too much information and clumsy ways to find it, is the same one behind web links, search engines and the language models in the news this year.

Reading the essay today is a useful exercise. It shows that the hard part of working with knowledge is not storing it but finding the right piece at the right moment, and that idea matters to anyone who uses AI tools to look things up.

Who was Vannevar Bush, and what did he publish?

Bush was an American engineer who directed the Office of Scientific Research and Development during the war. The magazine's introduction notes that he coordinated roughly six thousand scientists working on military problems. In the essay he asks those scientists to turn to a new task: making the huge and growing record of human knowledge easier to reach.

"As We May Think" appeared in the July 1945 issue of The Atlantic Monthly. It reworked an earlier text Bush had written in 1939, "Mechanization and the Record". A shorter version ran in Life magazine on September 10, 1945, with drawings of the imagined machines.

Bush starts with a complaint any student will recognize. Research is piling up faster than anyone can read it, and important results get lost. He points to Gregor Mendel, whose work on heredity went unnoticed for a generation because the right people never saw it.

How was the memex supposed to work?

Bush described the memex as an ordinary-looking desk. On top were slanted screens for reading, a keyboard, and buttons and levers. Inside sat the library itself, stored on very small microfilm, so compact that a person could add thousands of pages a day for a lifetime without filling it. Books and magazines would be bought on microfilm, and a glass plate on the desk would photograph your own handwritten notes and add them to the collection.

So far that is a fast, private filing cabinet. The new part was what Bush called associative indexing: any item could be made to call up another item instantly. You put two pages side by side, press a key, and they are joined for good. Join many items and you get a trail, which you can name, replay, extend with your own comments, branch into side trails and even copy for a friend.

His own example follows a researcher studying why the short Turkish bow seemed to outperform the English longbow during the Crusades. The researcher links an encyclopedia entry to a history book, then branches off into textbooks on how materials stretch, then adds a page of personal notes. When the topic comes up again in conversation, the whole trail returns with a touch.

Why did Bush think association beats an index?

Libraries of the 1940s filed things by alphabet or by subject codes. An item lived in one place, and to find a second item you had to leave and start again from the top. Bush argued that the human mind works differently: one thought leads straight to a related one along a web of connections. A machine could not copy the brain, he wrote, but it could borrow that habit and, unlike memory, never let the trails fade.

Here is the contrast in a small table:

Bush also predicted new kinds of encyclopedias that would come with trails already built in, and a new job for "trail blazers", people who enjoy making useful paths through the record for others.

What came out of the essay?

The memex itself stayed on paper; microfilm and levers were never the right tools. But the idea traveled. The engineer Douglas Engelbart read the essay soon after it appeared, and his later work on interactive computing, including the mouse and on-screen links, grew from that vision. Writers on hypertext, the name for text joined by clickable links, have treated Bush as a founding figure, and the World Wide Web turned "tie two items together" into the everyday link.

The essay also explains a lot about today's AI. A modern language model such as GPT-2, shown by OpenAI in February, learns patterns of association from huge amounts of text. That is not a memex: it keeps no exact copy of each page and cannot show you the trail it followed. That gap, between a machine that associates and a record you can check, is exactly what Bush cared about.

How close are today's computers to the memex?

In some ways we passed it long ago. A laptop holds far more than a memex desk, search engines reach billions of pages, and a hyperlink is Bush's "press a key and the items are joined", available to anyone who writes a web page. Online encyclopedias with links running through every article look a lot like the ready-made trails he imagined.

In other ways we are not there yet. Most links are made by the author of a page, not by the reader, and the paths you take while researching usually vanish when you close the browser. Bush wanted every reader to build, keep, annotate and share their own trails. Note-taking tools that link your own pages to each other come closer. The memex test for any of these tools is simple: can you see where an answer came from, and can you follow the trail again tomorrow?