Artificial intelligence and our lifes

When artificial intelligence tells stories, strange things happen. The German-Austrian writer Daniel Kelman tried it. What are machines good at in prose, and what aren’t?

In early 2020, German-Austrian writer Daniel Kelman typed the first sentence of a very promising short story into his computer:

“I was looking for an apartment. It didn’t go well.”

“The first thing he told me was, ‘Hey man, you got a nice ass and you’re not afraid of anything.’

So, the story should continue like this.

“You’re right,” Kelman tried to stay on topic, “but because of the apartment …”

However, the algorithm did not react to that anymore.

It neither succeeded nor failed

This experience with machine prose, Daniel Kelman described in February in the “Stuttgart Speech on the Future”, a new series of lectures in which personalities from the world of science, culture and politics reflect on the future.

His speech entitled “My Algorithm and Me” is now published as a book. Here Kelman describes his experiment: from February 2000, when he traveled to Silicon Valley to meet CTRL and its creator Brian Macken, at the invitation of “Open Austria”, the official representative of Austria there.

According to Kelman, the story writing experiment didn’t actually fail. Several beautiful absurd fragments were created. But the algorithm is incomprehensible and helpless when it comes to the plots and constellations of figures.

Daniel Kelman introduced the algorithm as a disoriented “secondary user” of data from human-written texts.

In a direct relationship with one algorithm, you immediately realize that “you are not dealing with a man in a suit,” Kelman said in his speech, “but with something much stranger – an entity for solving problems without an inside.” In his usual, concise way, Kelman sums it up as follows: “Notice that there is no one inside.”

Algorithm in theater

We are now going to the stage of the small Schwanda Theater in Prague. The piece “AI: When a Robot Writes a Game” (“Artificial Intelligence: When a Robot Writes a Piece”) had its virtual premiere. Behind the play, which was 90 percent written by artificial intelligence, is a group of linguists, computer and theater experts.

Here, too, it sometimes resulted in absurd changes of action, but then the algorithm pulls an ingenious move from the depths of its “unconscious”, as Kelman calls it.

“I’ll reveal the secret of acting to you. But I’m not sure I want to reveal it,” the robot said.

Learn from a writing experiment

These are just two of the many writing experiments around the world that teach us a lot about the artificial intelligence that creates text.

First, artificial intelligence gives humanity what it has just brought into the world. CTRL is partly “fed” by Redit data. Almost all topics are discussed on this favorite internet platform, and sometimes it can be offensive. If the developers hadn’t banned some expressions, the CTRL algorithm could very quickly start reproducing hate speech. Because he himself has no understanding of what people feel is inappropriate.

Second, artificial intelligence does not have reason, as humans do. Regardless of all the training and data, she lacks a point when telling a story, when telling a joke or trying to find a suitable metaphor. CTRL also does not know many unwritten rules, because it requires knowledge of the context – that is, the body, feelings and experience.

What can algorithms do then? They can calculate very well and make statistically normal texts on that basis. For example, the word “I” is more likely to be followed in one text by the word “mine” or “I believe”; and words like “poison” or “lamp” are less likely to follow, Kelman explained in his speech.

Artificial intelligence lacks “narrative consistency,” as Kelman puts it. Czech linguist Rudolf Rosa, from the team behind the Prague play, called it a lack of “aboutness”, ie an intention recognizable to the reader. Part of the problem is that the algorithm used by Kelman, just like the one used in Prague, suffers from amnesia. They only remember the last 500 to 1,000 words, and they forget everything that was said before. CTRL drops regularly after one page of text.

Written by michael

Vibration correction in quantum devices

The new Android virus is pretending to be a system update