The paper that nobody read

An idea published in 1945 described a machine that stored documents and let a reader follow links between them, building trails that others could then follow. It was read, admired, and largely ignored for decades.

The essay is now treated as a prediction of the web, which flatters it slightly. The author was describing a microfilm device with levers, and the mechanics would never have worked. What he got right was not the machine but the problem: that the quantity of recorded knowledge had already passed the point where any index could help, and that the useful unit was the connection between documents rather than the documents themselves.

The idea reappeared repeatedly over the following forty years in forms that did not spread. One system in the 1960s had linking, collaborative editing and a form of version history, demonstrated publicly in 1968 to an audience that found it remarkable and did not adopt it. The components existed. What did not exist was a reason for any particular person to start, since a network of links is worthless until other people have made links too.

What eventually worked was the version that was easy to join badly. A page could link outward without permission, to documents that knew nothing about it, and a broken link merely failed rather than preventing publication. Earlier systems were more capable and required agreement first.

The lesson appears throughout technology. The design that wins is frequently not the best one but the one with the lowest cost of being the first to try it.