For most of human history, finding something meant knowing where it was. You wanted a fact, you went to the book. You wanted the book, you went to the shelf. You wanted the shelf, you consulted the catalogue. The catalogue was the genius of the library — an index of indexes, a map of where things lived, maintained by people whose entire profession was the maintenance of that map.
This worked because there was no alternative. The volume of information exceeded the capacity of any individual mind to hold it, and we had no technology that could search the contents directly. So we did the next best thing. We organised. We classified. We built systems of order — Dewey Decimal, Linnaean taxonomy, alphabetical filing, folder hierarchies, database schemas — and we encoded into those systems an enormous amount of cognitive labour. Putting something in the right place was the act of making it findable later.
We came to believe that order was the work.
It wasn't. Order was the workaround.
There is a particular experience that has become common in the last two years and is worth sitting with. You take a folder of scraps — half-finished documents, screenshots, voice memos, PDFs you never read, a stray drawing — and you give it to a machine. The machine reads it. The machine returns a theme. Not a tag, not a category, not a place to file the scraps. A theme. Something the scraps were collectively about that you had not yet seen.
This is not search. Search assumed you already knew, roughly, what you were looking for. This is recognition. The machine has read the unsorted material and surfaced what is meaningful in it — without anyone having first organised it.
What this exposes is not a faster way to do the old work. It is the revelation that much of the old work was instrumental, not essential. We did not classify because classification was thinking. We classified because we could not see the whole at once, and classification was the substitute for seeing.
There is a comfortable position available at this point in the argument. It says: yes, the volume work is automated, but the value work — judging whether the machine got it right, deciding what the theme really is — that work remains human. This is true. It is also not the most interesting thing the new capability reveals.
The more interesting thing is what it tells us about the old one.
Every system of classification we ever built was made by someone. In 1605, Thomas James arranged the new Bodleian Library's collection into four faculties — Theology, Medicine, Law, Arts — with Theology at the head, as the European scholastic tradition had insisted for four centuries. In 1841, Antonio Panizzi published 91 cataloguing rules at the British Museum Library, founded on the principle that every book should be entered under its author — a decision that quietly elevated individual authorship as the defining identity of a printed work and became the basis of Anglo-American cataloguing for the next century. In 1876, Melvil Dewey decided that religion should occupy the 200s on the shelf and that nine-tenths of that space should belong to Christianity. These were not neutral acts of organisation. They were arguments about what mattered, what was central, what deserved its own category and what could be lumped together at the edges.
We inherited those arguments and we have been thinking inside them ever since.
For most of the modern era, this was invisible. The filing system was the world. You could not easily ask whether a different filing system would have surfaced different patterns, because the labour of re-classifying everything was prohibitive. The categories we had were the categories we used, and the things that fell awkwardly between categories — that sat at the edges, that did not quite belong — were simply harder to find and therefore harder to think about.
What the machine that reads the unsorted pile reveals is that the pile contains things our categories never had a place for. The themes it surfaces are not always the themes we would have named. Sometimes they are sharper. Sometimes they are stranger. Sometimes they are obviously wrong in a way that is itself informative — a tell that the material is being read through a different filter than ours.
This is uncomfortable. Not because the machine is right and we were wrong, but because the existence of an alternative reading exposes the contingency of ours. We have been operating, for several centuries, inside a particular era's guess at what was worth distinguishing. Most of what we now consider obvious about how knowledge is organised is downstream of decisions made by a small number of people, mostly in a few countries, mostly a long time ago. The filing system we inherited shaped what we could subsequently see — and, more pointedly, what we could not.
The labour of order was never neutral. Every act of classification was also an act of suppression: of the adjacencies that did not fit, the categories that were never created, the connections that the filing system could not represent and therefore did not preserve.
We do not yet know what we have been unable to see.