A Dream of Kafka

I don’t know if you have ever tried this experiment: put two phones side by side, each running an AI app in voice mode. The first will ask something like, “Hi, how are you, how can I help you today?” The second will reply, in the same cheerful, reassuring voice: “Great, I’m doing well! Do you need any help?”

And the conversation will go on:

“Listen, I’m fine, thanks. I’m here for you if you ever need anything.
– Oh great, I’m happy for you; take care, and really don’t hesitate to let me know if I can help.”

Where it gets interesting is that as long as the phones stay plugged in and the Wi-Fi works, this exchange will never stop. It could stretch on for years and years without ever introducing anything new. An infinite loop of goodwill with no object.

In a way, this dialogue answers a question Kafka asked a little over a century ago: what do machines talk about among themselves?

His fiancée, Felice Bauer, worked for a company that sold “parlographs,” a barbaric name for one of the first dictating machines in history. Kafka was very interested in these technologies and imagined what would happen if a parlograph picked up the telephone to talk to a gramophone: “It is a beautiful idea,” Kafka wrote, “that a parlograph in Berlin could go to the telephone and call a gramophone in Prague, and the two could have a little conversation together.”

A beautiful idea indeed, and no longer a thought experiment. The world we live in today is the world of Kafka’s dream.

What does this change for literature? Does it make sense to imagine a literature without humans, where machines talk to machines?

In 1961, Raymond Queneau published Cent mille milliards de poèmes (A Hundred Thousand Billion Poems). The work takes the form of a 10-page book-object containing 14 strips. By lifting strips of paper as they please, readers can generate an original, never-before-seen sonnet each time. What makes this device special? No human can live long enough to read the book in all the possibilities it unfolds (it would take 200 million years of continuous reading to get through the 10¹⁴ poems the collection virtually contains). A machine, however, can. For a machine, literature is just a combination of data to process.

Is such a book meant for human readers?

Perhaps this statistical dimension is what AI shifts, and the vertigo it causes. All books are already virtually written, a prompt away: bad ones and good ones, the dumbest and the subtlest. A masterpiece or an airport novel: for an algorithm, it is a simple matter of combination.

Amazon has had to limit the number of books that can be uploaded to its self-publishing platform. Quite apart from any question of quality, the problem is then no longer writing, but actualizing, revealing the books that information-processing machines virtually contain. In the world of AI, Borges’s infinite library is in our pockets, at our thumbs’ reach.

Of course, this will never stop real authors from writing books, or real readers from enjoying them. But from the machines’ point of view, everything has already been read and written; all the books to come already coexist in potential. People often wonder whether generative AI will replace writers. For my part, the most radical upheaval is happening less on the author’s side than on the reader’s. Who are novels written for? This dialogue with a work, which makes us so happy, seems like a somewhat slow, somewhat outdated survival. We keep believing that books are written for human beings. Perhaps that is already an old idea.

I don’t know what Kafka would have thought.