CCB Magazine

Hannes Bajohr: "We're dealing with constant amnesia"

Photo: © privat

The word artificial intelligence strikes fear into the hearts of many creative minds. Hannes Bajohr, a Berlin author and philosopher with a PhD, is deliberately experimenting with AI and has written his new novel with it. What can AI do that humans can't? A conversation about a life in and with technology, which captures the literary work.

INTERVIEW Anna Raab

CCB Magazine:Mr. Bajohr, you are a man of letters and also experiment with AI programs. Does the AI ultimately have better ideas than you do?

Hannes Bajohr:No, the AI initially only does what I tell it to do. It doesn't replace me as an author, but acts as a tool, and it can generate good ideas. Ultimately, however, it remains a tool that can only produce raw text - which I, as the author, then process or rewrite.

CCB Magazine:That means she can't do anything an author can't?

Hannes Bajohr: It depends on what you want to do with it. ChatGPT, for example, which is what most people will think of when we talk about AI, is literarily a rather uninteresting tool. It uses a merely average language that has no particular style. Specially trained models are more interesting. That's what I work with - that is, I pick what is relevant to me and take the relevant texts as a training dataset, the output of which I subsequently process. This makes AI more flexible, but ultimately it remains an instrument.

I had used GPT-NeoX for my new novel in addition to GPT-J. Here I had the feeling that four different worlds merge into one and are spun even further. I found that fascinating

CCB Magazine:You are a renowned author and have written 14 books. For your novel Berlin, Miami, which will be published this autumn, you assigned the AI GPT-J to write an entire novel - and trained them beforehand with four contemporary German novels. How did you go about it? And why did you choose this AI?

Hannes Bajohr:In addition to GPT-J, I have also used GPT-NeoX. Both are open language models that are particularly suitable for authors because they can be trained on their own text. Unlike ChatGPT, I don't say: "Write a novel", but I use the principle of "next word prediction". That is, I enter a letter and the system generates a whole sentence from it, which I then keep or discard. From the contemporary novels, it also learns some quirks and certain speech patterns that keep recurring; in addition, existing characters show up. Here I had the feeling that four different worlds merge into one and are spun even further, which I found fascinating. All in all, GPT-J is a particularly good tool for experimenting, and you often don't know what the end result will be. That's why the free models are more useful than the commercially available ones like ChatGPT, Bard or Claude, which are mainly suitable for summaries or generic texts, but their results remain expectable.

CCB Magazine:Can't you tell from the results which novels you used to train the program?

Hannes Bajohr:You can definitely tell that from the text, which is also my goal. One of the four I trained the AI with, Juan Guse's Miami Punk, stands out in particular. That's because it's the longest, with over 600 pages.

CCB Magazin:And you're not afraid that at some point you won't be taken seriously as an author if you let yourself be guided by an AI in this way?

Hannes Bajohr:No, I have always written experimental literature, and this is also an experiment. It would also be wrong to say that "the AI" wrote the novel - it makes suggestions that I accept: sentence by sentence, paragraph by paragraph. I communicate with a machine and the machine communicates with me, but I retain authorship. This process is incredibly exciting. That's why you should see the novel more as a test object and read it accordingly.

CCB Magazin:You were born in 1984 in Berlin-Friedrichshain and grew up in Bonn. You studied philosophy, German language and literature, and history at the Humboldt University in Berlin and graduated with a thesis on the philosophy of Hannah Arendt. You also completed a PhD. Hannah Arendt in particular perceived technologization and bureaucratization as an intervention in human action. You consciously intervene by means of technology. Did you ask the authors of the novels you used for permission?

Hannes Bajohr:I know two of the four authors quite well, and I told them about my idea. I didn't ask the others. At the moment, all of this is still legally permitted, because there are still no clear regulations on this kind of processing. What would be forbidden would be a mere copy of the texts. But my experiment is neither a copy nor a plagiarism, it is a statistical re-generation or pastiche - and the law cannot deal with that yet. Legal regulations on this will surely be decided by the EU soon.

CCB Magazine:How will the use of AI change the professional field of authors?

Hannes Bajohr:AI will be able to replace aspects of editorial work - and thus it will become economically relevant. It will not replace a literary author. So far, no AI program has been able to write a high-quality novel without a human co-author. In my opinion, that will remain the case. And an AI won't be able to write a bestseller or "quality" novel like the ones published by Suhrkamp, for example.

CCB Magazine:But "QualiFiction" is the first AI-generated program for fiction and could at least take over the work of literary agencies in the future. It can check manuscripts for suitability and, if trained properly, can also write dime novels. Are you sure you won't have to correct your statement in a few years?

Hannes Bajohr:I don't think so, because humans are still needed to recognize logical as well as emotional errors. With AI, it may be possible to write dime novels much faster in the future. But you have to look at the whole thing realistically: No one wants any kind of uniformity, which is why there will still be both the author and the literary agent in the future. What will happen: The fascination with AI will die down. At the moment, these programs are still something completely new, fascinating, and sometimes even frightening. This applies to all technical innovations, by the way; it was no different with the implementation of Google or even spell-check programs. By now, no one is discussing them anymore.

AI will be able to replace aspects of editorial work - and thus it will become economically relevant. It will not replace a literary author

© Yvonne Tenschert

CCB Magazine:But in the field of literature, the question is whether such programs can also generate emotion. Can they?

Hannes Bajohr:I don't know if you've ever tried creating "art" with ChatGPT? Try your hand at poetry. The program can produce poems in principle. But they are banal, and there are hardly any options other than couplets. Moreover, the program has an incredible penchant for kitsch and melodrama. It is nothing more than a glorified text completion, there is no room for intellect or even for emotional depth. I would therefore say: An AI-made novel can certainly be touching - but only if a human co-author operates the machine correctly.

CCB Magazine:So, did you succeed in doing that with your novel experiment?

Hannes Bajohr:The results are totally absurd and funny. But that was also my goal: to prove that an AI cannot tell a story. Because narratives are always based on the cause-effect principle - and this is exactly where the AI fails. As a statistical machine, it cannot think according to the principle "if..., then...", i.e. in causalities, but always assumes that two things happen simultaneously, so it thinks in correlations: "then..., then..., then..." - but that is not yet a narrative. In addition, the AI forgets things, mixes up characters, confuses time levels. You have to deal with a constant amnesia, with a forgetful and broken machine that creates absurdities. Personally, I find the result very entertaining, but it reads less like a novel, more like a very surreal text. And you have to remember that I kept intervening when things got too weird for me.

CCB Magazine:An AI is not alive, it has no past. Autofictional texts are quite popular right now, the Nobel Prize went to Annie Ernaux last year. Do you think an AI would also be capable of that, or could autofiction perhaps even remain the one serious competitor?

Hannes Bajohr:This is indeed a very interesting question, especially from the point of view of literary theory. First and foremost, one has to ask what the relationship is between the personal experience and the final textual result. Is a text no longer worth anything as soon as it turns out that what is narrated is invented? Does the task of literature lie in the expression of what is experienced? These are precisely the questions that make up the real value of the debate. I have already heard from several sources that this technologization of literature that we are experiencing is leading to a "rehumanization" of literature. Autofiction, memoir, a general take-off on identity - it may well be that this is now becoming more and more important. But, as I said, I don't believe that literature can be completely automatized. The point is that even with autofiction, we can't be sure if there's an AI involved or not - we just have to trust the opposite.

AI remains a symbol machine that knows nothing about the real world

CCB Magazine:So that means we can no longer distinguish what is AI-generated and what is not? At the same time, you claim that an AI cannot write high-quality literature, not even in the future. If we stay with the example of autofiction: Couldn't you then simply feed the program with personal experiences and adventures and let it write an autobiography from them, a whole life, so to speak?

Hannes Bajohr:These machines are good with shapes, with text as a pure succession of characters. So you could certainly enter notes, which the program then puts into another form, into a new container. But entering abstract experiences and feelings in a diary-like manner - this would not result in a text that is literary in the traditional sense. The AI remains a symbol machine that knows nothing about the real world, of life and suffering; which does not mean that it cannot nevertheless produce entertaining or indeed experimental text.

CCB Magazine:How will AI change this reality in the future? Could it come to a situation where the author is even upgraded in the future, because they continue to stand for the man-made, which is put to the test by the increase of AI?

Hannes Bajohr: Subjectivity is becoming more and more important. In journalism in particular, we're increasingly seeing that people want to see: there's a person behind this. This is another reason why video formats, podcasts and the like are becoming increasingly popular. Nevertheless, we have to accept: We can no longer rely on a written text always coming from a human. That's new - for the last 2,000 years we could, now we can't. What could actually become problematic are the possible economic upheavals that result. Because if an AI can suddenly write serial texts, it will displace jobs and professions. Fewer and fewer authors will be able to write more and more texts. That's why it's so important to stand up for your rights as an author. The Writer's Guild of America, which is currently campaigning for the rights of screenwriters, is a good example of this. But this also shows that it is not machines that act here, it is always people who use machines. And a society would do well not to forget that.

Category: Specials