The author and media scientist Hannes Bajohr has written a novel with the help of artificial intelligence. The result of this experiment – the book “(Berlin, Miami)” – reads extremely unconventionally.
When it comes to the keyword AI, polarized comments are not far away. Some see impending doom, others sense a brilliant creativity booster.
That’s why the German literary and media scientist Hannes Bajohr dared to experiment: he used artificial intelligence to produce novels. To do this, he used four novels from contemporary German literature as learning material for language AIs. The AIs work on the principle of text prediction. Based on the information available, the language model calculates the most likely next word.
Weird and overwhelming
Hannes Bajohr had the AI write an entire novel – sentence by sentence. His intervention was limited to selecting the texts for training the AI and deciding whether the generated sentence would be included in the book.
These four books served as AI learning material
Hannes Bajohr trained the language model with texts he selected. Four novels served as the corpus, which the literary scholar Elias Kreuzmair described as exemplary of “literature of the digital society”:
- Juan Guse: “Miami Punk”
- Berit Glanz: “Pixel Dancer”
- Joshua Gross: “Flexing in Miami”
- Julia Zange: “Reality Thunderstorm”
What the result of this experiment triggers when you read it can hardly be put into words. Far from a conventional novel, it is simultaneously overwhelming, strange, absurd and incoherent.
Like a fever dream
Various characters, places and even rudimentary storylines are recognizable. However, “(Berlin, Miami)” is far from fulfilling the expectation that these will be brought into a logical connection with one another.
Instead, as if in a fever dream, you read through a confused juxtaposition of scenes, absurd snippets of dialogue and descriptions from which you can hardly derive any meaning.
Hannes Bajohr: “(Berlin, Miami)”, Matthes & Seitz / Raw Material, 2023.
An example: “As a child, I had teeth that showed an unconventional classification between a pine tree and a pondhead. The Jawling was rooted in strength, the pond head was strengthened by an extremely slender base, which pulled the jaw out of the Jawling at the same moment the legs shook.”
“Kieferling” and “Teichenkopf” are coinages of artificial intelligence. That’s impressive, but nothing in this passage makes sense, not even in the context of the entire book.
Pointless search for meaning
Anyone looking for meaning while reading will be bitterly disappointed. And yet you can always find literary set pieces: love episodes, for example, or longer passages in the first person perspective – but only for everything to disappear into thin air in the following sentence due to a completely wrong connection.
This is precisely one of Hannes Bajohr’s findings from this experiment. For him, the incomprehensibility of the book is due to the fact that the AI language models used cannot form logical conclusions with the connection “because”. Instead, they simply attach elements together with “and then”.
Absurdities en masse
The result always makes you laugh because of its high level of absurdity. But it also requires a high tolerance for nonsense. It may also provide some reassurance to those who were afraid of the huge flood of AI on the literary market. Such language models probably cannot be written more quickly than simple dime novels or cheap crime novels without the help of knowledgeable people.
The cultural highlights of the week in the newsletter
Discover inspiration, stories and treasures from the world of culture: every Sunday, straight to your inbox. Subscribe to newsletter now.