Daniel Kehlmann tests AI: “My algorithm and me”

Status: 02/10/2021 12:02 p.m.

What happens if you let artificial intelligence compose? The writer Daniel Kehlmann got involved in this experiment. The good news: AI won’t replace it in the foreseeable future.

Daniel Kehlmann gets to the point relatively early on, in the livestream of the Stuttgarter Literaturhaus that evening: “To anticipate it right away: the goal of our experiment was not achieved,” says the writer. But there is no disappointment in his voice, but sheer fascination. Because this experiment, it wasn’t an ordinary one. In early 2020, the writer traveled to Palo Alto in Silicon Valley, with the assignment to get to know CTRL, an artificial intelligence.

Letting a flesh-and-blood writer encounter an intelligence made up of circuit boards and algorithms, it was thought, was a fine thing, reports Kehlmann, an experiment between human consciousness and technical probability calculation. Could these two write a novel together? Or, as Kehlmann puts it: “Can an algorithm invent stories?”

The poet and the machine

Daniel Kehlmann is a world-famous author, his novel “Measurement of the World” is considered one of the greatest successes of German post-war literature. Why does someone like that want to work with a machine, a “being without an inner life”, as Kehlmann himself calls it?

It will have been the curiosity, which is reflected in Kehlmann’s lecture in Stuttgart again and again, also the fascination for future topics and maybe, but that’s only a guess, maybe also a little concern: Can this machine do what I can?

It is clear to him early on: No, she cannot. Because where people act with intent, with a consciousness, the algorithm generates language purely on the basis of probabilities. If the person formulates the sentence “I am going”, he says it based on his grammatical knowledge. He learned that after “I” there must be a verb. And one in the first person singular. These rules are alien to the algorithm. It is an incredibly large memory, fed with billions upon billions of texts. Based on this data, the algorithm calculates the probability of which word could come after “I”. And that is much more likely to “go” than “car”.

KI is neither HAL nor C3PO and “more like Kafka than Dickens”

Kehlmann knows: “Artificial intelligence is a secondary user. Everything it can do is fed by the activities of countless people made available by the Internet.” But although he was aware of these mechanics, according to the writer, at the beginning of this experiment he had to say goodbye to old ideas of AI: “I had always imagined artificial intelligence to be like the android C3PO or the narcissistic supercomputer HAL – as a human being in metallic clothing, as a person in costume. ”

With this non-person, Kehlmann dares to experiment with writing. In English, because most of the texts fed into the algorithm are written in this very language.
Kehlmann starts a dialogue:

It was a beautiful day in summer.

And artificial intelligence continues:

The sun shone brightly on the green grass and flowers of the garden, but there were no birds to sing or insects to hum.

With this sentence, CTRL brought “an uncanny note” into the story, comments the flesh-and-blood writer, as he seems to be fascinated at all, whether the random dialog between him and the algorithm that only arises from probability calculations. And at the same time disillusioned – because the writing experiment never lasts longer than a few pages, often breaking off after just a few sentences exchanged: “CTRL is a friend of the fragment and the surreal, more Kafka than Dickens; CTRL does not do more than one side, “says Kehlmann.

Beyond English, things often get bizarre

Michael Resch is not surprised. He is the director of the high-performance computing center in Stuttgart and takes part in the discussion that follows Kehlmann’s future speech. A machine is just not a systematic being that pursues its own intentions – unlike humans. And that would not change if the computing power were increased: “With the approach of expressing everything in terms of probabilities, you can to a certain extent modulate what happens between people, but when I try to represent reality with this model, then I run into the wall, it’s always like that. Continuing communication based on statistical analysis, computing power only helps to a limited extent, because at a certain point the computer no longer has the necessary ability to create meaningful context. ”

And there are other restrictions as well. Although the algorithm is able to write in any desired language, attempts beyond English are often bizarre, simply due to the insufficient data situation.

This is how Kehlmann writes in German:

“I went to bed early for a long time”

And the algorithm continues:

“Die ZEIT: In the last few years you have been working with the AfD.”

“It’s like talking to a madman who has lucid moments” – Daniel Kehlmann is fascinated. But also disillusioned. Because this experiment remains just this: an experiment. And the excursion into the future one that the author wouldn’t want to miss.

“I saw how nonhuman intelligence actually creates consistent sentences from the dark depths of its statistical assessment, in which perhaps some distant day consciousness will also glimmer,” said Kehlmann in the evening in the livestream of the Stuttgarter Literaturhaus. And: “There is something magical about this spectacle, every time.”

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.