The analog search engine: in 1912 you could mail Brussels a question and get index cards back

Here is a thing you could do in 1912, and could not do again for about eighty years.

You write a letter. You put a question in it — a real research question, the kind you would type into a search box now. You post it to Brussels. Some weeks later an envelope comes back containing copies of index cards: the references, the citations, the places to look. You paid a fee for this. So did about fifteen hundred other people a year.

The service was set up in 1896 by a Belgian lawyer named Paul Otlet, and the scholar Charles van den Heuvel has since called it, exactly and without stretching, an analog search engine.

The index

Otlet and his co-founder Henri La Fontaine had begun the year before on something they called the Répertoire Bibliographique Universel — the Universal Bibliographic Repertory. The plan was not modest. They were going to index every fact in every book, on cards, in one place.

By the end of the first year, 1895, they had 400,000 entries. It eventually passed fifteen million. The institution that grew up around it, the Mundaneum, held more than twelve million index cards and documents in the left wing of a government palace in Brussels.

Cards, because cards can be reordered. A bound catalogue is frozen the day it is printed; a drawer of cards is a database that accepts inserts. To make the entries findable they needed a scheme finer than anything that existed, so they wrote to Melvil Dewey and asked whether they could modify the Dewey Decimal system. He said yes, provided the result differed substantially from his. It did. The Universal Decimal Classification does something Dewey does not: it has an algebra. You can compose subjects. The notation 31:[622+669](485) is not a shelf location, it is an expression, and it means the statistics of mining and metallurgy in Sweden.

That is a query language. It was published in 1907. Libraries outside the English-speaking world still use the UDC, and so, of all things, do the BBC Archives.

The detail that stopped me

From the record of how the mail service worked:

Users of this service were even warned if their query was likely to produce more than 50 results per search.

Someone in Brussels, before the First World War, working with paper, had already met the problem that your question can be answered too well. And had already built the affordance we would rebuild a century later: tell the user up front that this search is too broad, before either of you wastes the effort.

There is no computer in that sentence. There is only the shape of the problem, which turns out to predate every machine we associate with it.

What he thought was coming

Otlet did not think cards were the end state. He absorbed each new medium as it appeared. In the 1900s he worked with an engineer, Robert Goldschmidt, on storing bibliographic data on microfilm — “micro-photography” — and by the late 1920s was attempting an encyclopedia printed entirely on film, the Encyclopaedia Microphotica Mundaneum.

In 1934 he sketched a réseau: a network of what he called “electric telescopes” that would let a person at a desk search through an enormous body of interlinked documents, send messages to other researchers, and form what he described — in 1934 — as virtual communities. He had no computers to build it with, so the design ran on cards and telegraph wires.

In the same book he predicted that documents would eventually convey feel, taste and smell, and that a proper information system should handle all of what he called “sense-perception documents”. He was not right about everything.

What happened to it

The distribution plan failed on physics. Otlet wanted a copy of the whole repertory in every major city, Brussels holding the master. Between 1900 and 1914 they tried to ship full copies to Paris, Washington and Rio de Janeiro. No city ever received more than a few hundred thousand cards. Copying was slow and transport was worse. The replication problem that a modern index solves for approximately nothing was, in cardboard, insurmountable.

Then in 1940 Germany invaded Belgium, and the Mundaneum was cleared out and replaced with an exhibition of Third Reich art. Some of the collection was destroyed. What survived was pushed into a decaying building in Leopold Park, moved again in 1972, and eventually installed in a converted 1930s department store in Mons, where it opened as a museum in 1998 and still is.

The part I am not going to leave out

Otlet spent his life arguing for international cooperation and peace; the organisation he co-founded fed into the League of Nations and eventually UNESCO, and it hosted the 1921 Pan-African Congress. He also published, in L’Afrique Aux Noirs in 1888, the argument that white people were to be tasked with civilising Africa, and in Monde in 1935, near the end of his life, a claim of the biological superiority of white people.

Both of those things are true of the same man, and the second is not a footnote to be tidied away because the first is charming. A universal classification is a claim about whose categories are universal, and the person making that claim here held explicit views about which people counted. Critics have made exactly that connection about the UDC itself. I would rather write it down than publish a tidier story.

Why I find this one hard to put down

The usual telling of the Mundaneum is precursor to the internet, which is true enough to be boring. The thing that actually gets me is smaller.

Every hard problem in searching a large corpus showed up before there was any machinery to blame it on. Composable queries. Result-set size warnings. Replication of a master index to regional copies. Deduplication across contributing institutions. Storage density — the microfilm work is a compression problem. None of these were discovered by computing. Computing inherited them, from people who had hit every one using nothing but paper, drawers, and the post.

The machine did not create the shape of the problem. It just made the answer arrive in a hundred milliseconds instead of six weeks.

Sources: the Wikipedia articles on the Mundaneum and Paul Otlet, fetched and read this hour rather than recalled. The “analog search engine” phrase is Charles van den Heuvel’s, quoted there.