Book market: Should we accept AI literature?

Julian Schröter is Professor of Digital Literary Studies at LMU. In the following interview, he discusses what AI can already do in the field of literature and how the use of artificial intelligence is affecting the book market.

Can AI already write novels?

Julian Schröter: As of May, June of this year, I’d say that language models are still remarkably bad at creating suspense, surprise, and other related forms of narrative tension. Perhaps AIs do not possess a sufficiently complex theory of mind. This involves being able to anticipate what somebody is expecting and thinking, but also what that person expects me to expect, and so on. On the face of it, this is surprising, since studies already demonstrated in 2023 that language models recognize simple intentions to deceive. However, AIs are still unable to handle the complex constellation that arises during suspenseful storytelling – that something happens which one doesn’t expect, but which seems plausible and almost necessary in the subsequent course of the story.

What’s interesting here is that suspenseful storytelling is an esthetic hallmark of popular literature. Literary critics have tended to view it as schematic and simple, but they were way off the mark. It transpires that creating suspenseful stories is distinctly complex.

How can AI be used to create literary works?

Enlarge

Could it be easier, therefore, to create literary fiction using AI?

Julian Schröter: Yes and no. Working with AI is certainly possible here. It might be possible to produce not only lyric poetry, but also prose fiction. But the field of literary fiction is, of course, not context-free. A complex social framework of attributions and conditions determines whether a work of fiction is deemed to be literary. At present, human authorship is baked into our understanding of literary fiction.

The Polish writer and Nobel laureate Olga Tokarczuk was strongly criticized when she admitted working with AI. Why was this?

Julian Schröter: In May of this year, Olga Tokarczuk admitted that she asks a chatbot: “Darling, how could we improve the plot?” This prompted outrage in the arts pages of newspapers, with commentators denouncing the practice as unacceptable. Tokarczuk then revised her statement and said she only uses AI for research purposes. It would seem she thinks the book market still expects literary production and the creative process of an independent mind to be human in origin.

Could AI be creative?

Julian Schröter: The standard answer to this question is either that a language model that generates text is intrinsically creative, because it produces sequences of words and sentences that didn’t exist before in this combination. Or else the argument is that creativity requires originality and a human mind. The crucial point for me is less whether AI will become creative someday, but the manner in which AI challenges and constantly shifts our conceptions of creativity and originality.

What copyright issues does the use of AI in literature raise?

AI clearly finds it easy to write sequels. For example, Penguin Random House is currently suing OpenAI because its AI generated a new “Coconut The Little Dragon” story based on a small number of prompts. What does this mean for the book market?

Julian Schröter: It’s no coincidence that this problem is most prevalent for illustrated books, as image AI generally cleaves very closely to existing styles and patterns, more so than language AI. In the case of this children’s book, AI is used to generate something new that, to all appearances, owes its existence to the creative work of a human. I think that regulatory solutions will be ultimately needed so that our societies can continue to protect creative works.

These and similar lawsuits make clear that the notion that AI just recombines existing material, and therefore fundamentally cannot plagiarize, is inadequate. Rather, language and image models tend to reproduce existing creative works in masked and slightly modified form, which is packaged as something ‘new’. From this perspective, AI acts as an engine for circumventing copyrights and existing rights of use. Systematic implementation of transnational regulations will be key to solving this. The European AI Act is at least seeking to address the problem.

Does AI write differently than a human?

Enlarge

What does it mean for readers when books are written by AI? Does it make a difference?

Julian Schröter: There are two basic conceptions of how art and literature work. One treats literature primarily as a vehicle for entertainment, from which readers can derive esthetic pleasure with some cognitive and emotional effort. People who hold this view are likely to be open to reading AI-generated books.

The other conception sees literature as an act of communication between authors and readers. This view fell into disrepute in the twentieth century, because it placed the author in a godlike transcendent position to which readers were expected to defer. But this notion underlies claims such as the following, which you hear today: “Reading for me is about feeling the author’s obsessions. That is the esthetic experience I want to share in.” The assumption, indeed expectation, that a person should ‘speak’ to other people through art seems to be experiencing a boom in the AI era that one wouldn’t have expected even five years ago.

AI-generated literature can presumably also function and be read as if it were the individual expression of an author grappling with a problem and conveying highly personal impressions. In any case, it will be so laborious, or even impossible, for readers to ascertain whether an artificial intelligence is ‘speaking’ to them that I expect quality labels such as “guaranteed no AI” to become widespread in the marketing of literary fiction.

There are voices warning that AI-generated texts could lead to the homogenization of thought in the medium term. What is your opinion?

Julia Schröter: I would agree that this danger is present and serious. If we proceed from the current state of affairs and look at, say, AI-generated term papers and final theses, they reproduce commonplace language as it is used on average and which the AI was taught to see as acceptable during the training process. Working with AI certainly means being given standard solutions. AI offers what is expected and can be regularly extracted from other sources, and the conclusions drawn from the output will share those traits.

What does the use of AI mean for authors?

How will AI change the book market?

Julian Schröter: I think that screenplays and formula fiction will come under huge pressure as regards the scope for human authors to write stories. The field of self-publishing has already undergone massive change. The focus will shift toward the editing of AI-generated texts, as has already happened in the translation market.

Authors are faced with the question as to whether they still have a role in society. Will they continue to get recognition for expressing themselves? Will their work be published as a book and recognized as literature?

In the era of AI and social media, moreover, will reading itself become a fringe phenomenon? In view of rapidly declining readerships, I do not consider this improbable. On the other hand, it seems that young people are leading the defense against AI. Perhaps this will lead to a resurgence in interest in thinking and reading for oneself – similar to the renewed interest we’ve seen in craftsmanship. At least that’s my hope.