Metaphorical Intelligence: On Generative AI And The Library Of Babel

Paul Campos, the author at the Lawyers, Guns & Money blog who I’ve mentioned the most times here, has a post up today linking to some recent incidents with LLMs and listing some ideas about LLMs that he’d like to see explored in more detail.

One of those things is the connection with Borges’ various metaphorical libraries.

It’s not the world’s least obvious connection: LLMs are the current hotness in infinite text generators, and Borges’ interest was in humanity facing the concept of infinite texts.

So they go together pretty well.

They’re the peanut butter and chocolate of humanity’s interface with text.

I discussed this metaphorical similarity last month in “Generative AI: A Drunken Walk Through The Library Of Babel“.

There, I preferred the idea that the LLM is best viewed not as the Library itself, but as an unreliable guide for navigating the Library: the LLM can find you a text that resembles the text you describe to it, but whether that text has any actual value outside of similarity in words is not part of its process.

But Campos’ post did get me to wonder who else has written on this.

So I did a small search.

Now, this is DuckDuckGo so the search database isn’t as big as Google’s but their non-AI search tools are better (and Google just announced that they’re shutting down their non-AI search anyway) so it’s what we have to work with these days.

A few people have connected LLMs to Borges before:

there’s an LLM called “Babel”, in reference to Borges’ Library
there’s a generated video with the text created by an LLM “conditioned to produce text in the style of Jorge Luis Borges”
there’s a blog post noting the similarity between LLM training and a Borges story about a writer re-creating “Don Quixote” by immersing himself in Cervantes’ life
there are several blog posts that do little more than name-check Borges in the context of something about LLMs

But there doesn’t seem to be anything that discusses the linkage with any real depth.

This isn’t surprising.

First, while there will be some overlap between “people who understand Borges’ well enough to write an essay” and “people who understand LLMs well enough to write an essay”, that overlap is probably smaller than you think.

But second, and probably more importantly, it’s not really a connection that will reward a technical discussion; from both the computer science side and the literary criticism side, there’s not much more to say than “huh; that’s amusing”.

And then you move on.

To really get into it, you have to move on from observation to metaphor; I didn’t even get through a whole post on it before going off into metaphor.

So you’d probably need to deal with it in a Borges (or Lem) style of fictional technical discussion to be interesting about it for more than a few paragraphs.

And the overlap between people who can do that and people who understand both Borges and LLMs well enough is very small.

Which is a shame, ’cause that’s be a fun read.