Cherrypicker | Literature
Does AI dream of electric books?

Artificial intelligence (AI) is here to stay, and we will have to learn to live with this omnipresent technology. Hannes Bajohr has conducted a bold experiment in narrative literature by getting an AI language model to write a novel.

By Hendrik Nolde

© Rohstoff Artificial intelligence is a topic that was impossible to avoid last year. The release in particular of the freely accessible ChatGPT language model by the US firm OpenAI in November 2022 sparked a huge global boom in AI. In the space of just a few months, hundreds of millions of users tried their hand at generating texts automatically with the chatbot. Hardly surprisingly, this sudden and massive incursion of AI into the everyday lives of ordinary people also fuelled the media discourse about its possible impacts on the way we live together as a society. How will teachers deal in future with the problem of students handing in AI-generated essays? What risks does generative AI entail for social media, which are already saturated with disinformation in any case? And will it in fact still be possible at all in the near future to distinguish between AI-generated texts and those written by humans?

Being a text medium, literature cannot ignore such questions. The literary scholar Hannes Bajohr has already spent a long time intensively studying the relationship between literature and digital technology. Leaving theory behind, he has now taken the bold step of putting an idea to the test: with the aid of several AI language models, he has created a novel (Berlin, Miami) in a bid to ascertain whether KI is capable of telling stories. The outcome is a challenging and awkward text that can also serve as an experiment by which to reflect on one’s own reading experience.

The collapse of the concept of language

Any attempt to summarize the plot of (Berlin, Miami) is doomed to failure. It is hard to identify any coherent narrative strands. Instead, the novel is characterized by associative connections, gaps and almost rhythmically recurring linguistic set pieces and neologisms. The narrator comes across as highly unreliable, contradicts themself constantly and at times even interrupts themself. Dialogues peter out or end up essentially as monologues. Thus the collapse of language that critics of AI sometimes warn against is as it were demonstrated on both a thematic and stylistic level.

The atmosphere of the depicted world is vaguely futuristic and latently dystopian. There is frequent mention of dubious institutions such as the Ãää Party and a looming breakdown of the world, without the narrator apparently ever requiring any explanation for the underlying reasons. Technology plays a central role; the development of a computer program named DARIA is described repeatedly, which can be read as an allusion to the ELIZA language model developed by Joseph Weizenbaum in 1966 – regarded to be the earliest form of AI. At its strongest moments, the novel reveals how the narrator AI is plagued by existential questions about its own raison d’être: “How is a system that is its own system supposed to constantly understand its own development if it cannot see its own development?”

Death of the author and birth of the reader

Bajohr’s AI novel fluctuates between profundity and nonsense, and as such is also reminiscent of earlier (human) literary traditions that played and experimented with language as a way of tapping into narrative forms – like the Dadaists of the early twentieth century or the theatre of the absurd. Samuel Beckett would probably have had fun with ChatGPT.

Whether or not it can be considered a joy to read (Berlin, Miami) is extremely subjective. The lack of coherence can definitely be frustrating if one is not capable of detaching oneself from one’s usual reading habits and expectations. If however one accepts that the confusion one feels while reading may actually be intentional, a fascinating literary experience can ensue. The moment one waives any entitlement to an all-encompassing understanding of the tale being told, one can certainly lose oneself in language that, though awkward and clumsy, is fascinating precisely for being so, and can deeply immerse oneself in an obscure world in which terms like co-yoga, pond-head and three-U-ten-foam effect sound entirely natural. The question of whether AI can tell a story must therefore likewise remain unanswered, or at least be put off for another day.