Thread by @andy_matuschak

When I first read As We May Think* as a teenager, I was astonished by how much it predicted of the computer age in 1945—but recently I’ve been feeling wistful about some pieces it predicts which never came to pass. [thread]

* theatlantic.com/magazine/archi…

Bush recognized that we think associatively. He thought we’d amplify creative work by building and navigating associative trails across materials. Many see this as predicting the WWW, but on a web site, you can only consume the links the author put there; you can’t add your own.

Some people think this observation means Bush predicted the wiki. Yes, on a wiki, you can add a link to the text yourself… but that link would appear for _everyone._ There’s no notion of _personal_ associative markup.

Bush’s associations are bidirectional, while web links are one-way. I suspect bidirectional links make much more sense in a personal context than in contexts where all links must be shared by all; that may be part of why the web’s have stayed unidirectional.

And of course, even that’s just the web. What about associations between pages of an ebook and a paragraph of a PDF? Between an email from a colleague and some drawings it reminded me of?

Bush also suggested that this web of links wouldn’t just be between others’ published works—that subordinates your own work to others’. Your own materials (drawings, drafts, voice recordings, etc) are also stops on the associative “trail.”

If you want to arrange your own works relative to a disparate collection of others’ works, modern OSes offer folders and tags instead of links. In many cases, these are better: many associations form a clique or a tree, not a trail as Bush emphasizes.

OS-level tags solve a problem Bush observes: that indexing systems usually assume there’s a “best” categorization of an item, but they often have many associations. You want many-to-many. But tagging a 100-page PDF or a 100-comp PSD isn’t very helpful. Memex links are page-level!

But tags and folders are just sets. The structure is unordered, but your associations usually aren’t. It’s not just a theoretical problem: tags/folders are jumbled messes after a couple dozen items. Probably certain files are higher “degree,” but you’ll never know.

Bush also describes how you could acquire trails from others—another notion we’re missing. Others’ trails could be applied to materials you already have, so you could see a colleague’s associative structures alongside your own, on the same files.

Are there any good analyses of the structural reasons why we’re here and not there? Key problems with associative linking as attempted in past large-scale systems, etc? Other interesting attempts besides Xanadu? LiquidText is neat; fun to imagine OS-level adaptations… [fin]