The end of the era of the document?

The evolution of information with LLMs suggests a departure from static documents towards dynamic, situation-specific configurations of information. Previously, digital documents mirrored paper documents: fixed phrases on a sheet. The author might have been wavering between two ways of phrasing a sentence, but one was chosen, and that’s the one that stuck. To alter the selection and configuration of words and phrases, you would have to edit the document or create a new one. Sometimes, decisions about how something is written/expressed are consequential — as, for example, seen in discourse about how the phrase “a land without a people for a people without a land” has been interpreted, used, and misused in connection with Israel’s occupation of Palestine.

With generative AI, the landscape is changing. The way knowledge is (re-)presented might come to align more closely with understanding as I think it “looks” inside the mind — existing not as specific phrases but as a form of intangible understanding that is verbalized only as needed. Note: while some argue that understanding is tied to the way it’s expressed (https://en.wikipedia.org/wiki/Cognitive_linguistics), understanding does to some extent exist independently of expression or through varying means of representation (https://theconversation.com/aphantasia-explained-some-people-cant-form-mental-pictures-162445).

If the expression of knowledge becomes less set in stone (or digital paper), how will that affect the way we conceive of and use knowledge? Maybe we will become less attached to the way things are phrased. Will our knowledge as individuals, as organizations, and as societies be less tied to specific expressions of philosophies, ideals, etc.? Will we start to think about things differently? Maybe we can learn something from what cognitive archaeology (https://en.wikipedia.org/wiki/Cognitive_archaeology) has revealed about the relationship between cognition and culturally dominant knowledge representations. To be clear, I’m not talking about how writing affects thinking, but about how the (im-)permanence or morphability of documents — of knowledge expression — affects thinking.

This stream of thoughts started with me feeling uncomfortable about a suggestion a colleague made that documents could become more transient or ephemeral. It produced this image in my head of documents and text slipping through our fingers like sand; a sort of anxiety around knowledge getting lost. I suspect my discomfort relates to the experience that knowledge is tied to words — that this specific phrasing of this sentence matters. Some of this comes down to aesthetics: sometimes a formulation is chosen because it’s beautiful, eloquent, or pleasing in some other way. And as the author, you can get almost desperate to preserve it. I think this relates to some people’s (I’m, uh, asking for a friend ...) semi-neurotic versioning of documents when writing — it’s hard to kill your darlings. But what happens when the darlings never stood a chance; when they were ephemeral to begin with (“ephemeral darlings”, as GPT named them when I asked for help rewriting this text)? Does it change how we feel about it? Will it cure the “hesitation to change” I experience, maybe not in me, but in future generations?

It’s not just aesthetics, though. I often experience that something inspires me not just because of its meaning, but because it was said in a particular way. Or that the reason I understand something is because of how it was conveyed.

When no words are set in stone, what happens to the knowledge? Is something lost? ... Is something gained? Can ephemerality relieve us of the burden of choice of expression, because multiple of those choices continue to be possible?

Maybe my anxiety regarding this is totally unfounded. Or maybe the questions I’m asking are trivial. Websites have foreshadowed documents that are subject to change — we have long been used to the fact that when we return to a website, we can’t be sure that it still looks the same or says the same as it did the last time we visited. And many of us are used to work documents being changed after they are shared. However, these changes are mostly made by people, and thus happen at a pace that we can (somewhat) keep up with.

Conversations aren’t feared for their ephemerality: we accept that meanings outlive exact words (well, mostly — except for those occasions when we frantically scribble down notes while in a particularly inspirational meeting or conversation). Could documents mirror this conversational fluidity? Maybe static documents were a 20,000-year step (https://en.wikipedia.org/wiki/Proto-writing) on the way to this. Maybe the era of the document is over, at least in the form of a digital mimicry of physical documents.

We probably won’t stop collecting information; record-keeping will still be a thing. But might the means for record-keeping change?

And what about legislation, which tends to put weight on the tiniest of semantic differences? How will generative AI affect that? Will we still rely on the exact way that a piece of legislation is phrased? Or could “legalese” be delegated to underlying data representations or metadata? This change is already in progress — for example, two identical-looking job ads can have different metadata, with the consequence that one of the jobs can only be offered to people in a certain region while the other can be in any of the countries that the company hires in.

This text was written in May 2024 but was not put online until November 2026 because I’ve lacked the confidence and was, at the same time, worried that it was actually good and someone would copy my thoughts without crediting them. That means it’s just been sitting uselessly on a hard drive for two years, which is silly. Hence why it has now been put online. If the words above do help you think some useful thoughts that you decide to share with the world, I’d really appreciate it if you’d give due credit.