One click to talk to Dostoyevsky, Kafka, and Co.? AI is supposed to make it possible. But does it really work?
Having a conversation with literary idols who have been dead longer than oneself has been a dream for many authors, Germanists, and bookworms. A dream that now seems to be coming true. On various AI platforms, users can converse with all sorts of figures from literary history. Engaging in a deep discussion about faith and salvation with Dostoyevsky seems to be within reach.
This is not even complicated. On the platform ChatGPT, it works seamlessly with so-called CustomGPTs. These are specialized chatbots that users can provide with specific instructions – such as acting like a particular author. This seemingly allows for a conversation with the dead titans of written words.
In practice, the conversation is far less impressive than one might hope. At first glance, it remains questionable whether Dostoyevsky would have actually answered questions like the AI does. It is hard to imagine that the author would have responded with an internet-typical list when asked about his favorite books, reminiscent of Reddit, which was invented over a century after his death. Instead, behind the author’s voice, ChatGPT’s typical voice emerges, such as when the supposed Russian author asks at the end if one would like a detailed overview of the authors.
Admittedly, the experience heavily depends on the specific bot being used. For every well-known author in world literature, there are dozens of different CustomGPTs inviting conversation. A chatbot inspired by the Russian author Nikolai Gogol, while vague in its statements, does not tend towards listing. The text generated by it argues coherently over several lines that he would have probably liked Dostoyevsky’s books – if he hadn’t already been dead when the important novels were published. Whether Gogol would have actually argued this way remains unclear, as the bot, even after several inquiries, does not provide a thorough analysis of why the author might have liked Dostoyevsky.
Either vague and unconvincing or specific and artificial: the dream of a deep conversation with the classics quickly bursts. This is not a great surprise when one considers what Large Language Models actually are: stochastic machines that, to simplify, calculate texts based on probabilities, lacking intelligence or personality in the true sense. Thus, a Dostoyevsky bot is essentially nothing more than a technically impressive remix of passages from the countless texts with which the AI model was fed. So, no ghost in the machine after all.
But do users really aim for a deep conversation? Does anyone truly believe they can chat with the great titans of literature in a chat window? Or is it more about the author’s work? This is hinted at by the fact that most bots, when asked about their favorite novel, refer to “own” novels. For example, the English-speaking Kafka bot explains that its favorite novel is “The Trial” – and only upon further inquiry does it mention that it also appreciates Flaubert’s “Sentimental Education.” Whether this is true, or whether someone, somewhere, has written that the novel reminds them of Kafka, remains unclear, as the source for the statement is not disclosed in typical AI fashion.
The same bot also hints that most users are not interested in talking to the author. With CustomGPTs, preset questions can also be asked, focusing on understanding the work: “How can I interpret The Trial?” “What are the themes of your short stories?” “What did you want to convey with The Metamorphosis?” Is the AI there to reverse the author’s death? Moving away from interpretation, toward the author explaining their own work?
This is not possible, of course. Because the AI cannot interpret either, it can only calculate probabilities, as mentioned earlier. A stochastic model that spits out text based on the data with which the program was fed. In Dostoyevsky’s case, the statement is still not from the author himself, but rather a code recycling fragments from interpreters like Joseph Frank or Mikhail Bakhtin based on structures contained in the training data.
But perhaps the goal is not a truly valid interpretation in the author’s sense. Maybe it is about making a convincing argument, scoring a good grade on homework – or avoiding having to read the book for an assignment. Many of the bots – especially those used more frequently – do not imitate the author’s personality but are explicitly labeled as aids for understanding and interpretation. A digital library with chatbots, some say: a plagiarism machine. Not a resurrected spirit. So, to get closer to your favorite author, only one thing remains: reading. And if modern technology is a must, then on the smartphone. Depending on the definition, that could also be a ghost in the machine.
