Musk’s text generator wrote a play, the Czechs staged it. I watched…

100 years of work

The experiment had two main goals, to check how the algorithms would cope with such an ambitious task, and at the same time pay tribute to the Czech creators of the word robot, which has just celebrated its 100th birthday. It was used for the first time in a play titled “RUR” (from 1921) by Karel and Josef Čapek, which tells the story of a rebellion by humanoid robots that destroyed the human race.

The idea for the experiment was conceived by Tomáš Studeník, the Czech, as he called himself, “radical innovator”. It was he who found Robert Rose, a computational linguist at Charles University, who had experience working with the GPT-2 generator. After the team was assembled, the team started work, and the artistic side was handed over to the team of the Švanda Theater, headed by the director Daniel Hrbek. Interestingly, Hrbek, as he himself mentioned on the panel, was one of the skeptics of the entire project.

A man is needed, but …

All in all, it must be clearly said that the GPT-2 itself did not cope with the whole task, but at the same time the authors emphasize that it did surprisingly well. How to reconcile these two assessments? Well, the complicated emotional structure of the plays made the generator get lost when it let go, change the genders of the characters, forget that the main character is a robot, not a human.

To avoid this, the team split the play into eight separate scenes and limited the number of characters to two for each such episode. The computer gave the general assumptions of the scene, the characters and the first two lines of the text. In the end, the scientists received a series of dialogues, which were then just polished by the theater team, and a play based on the text was staged that you can watch if you have an hour of free time. According to Robert Rosa, 90% of the script is plain GPT-2 text, for the rest the theater people made slight adjustments.

Kinky AI

Once again, it has been confirmed that artificial intelligence, fed by human creativity, is quickly turning towards sex and violence, some threads are quite hardcore in this respect. However, it is worth remembering that this is more about the shape of our civilization than the quality of the algorithms themselves. At the same time, the layer of, let’s call it a philosophical one, despite a lot of chaos in the threads, is unexpectedly deep.

At the same time, one of the most surprising threads generated by GPT-2, concerning the topic of euthanasia, turned into a grotesque scene with … sticking fingers into a certain inelegant part of the body. Well, so far we still don’t understand much of what is happening in neural networks …

The “technological” structure of art

For anyone who has had contact with things generated by AI and / or has read more about it, it is fascinating to observe those moments where you can see algorithms trying to get out of a certain task in a human way, but their difference causes the dialogues to lead to the goal around with surprising digressions. There are also scenes where it seems that AI just loses the plot and then comes back to it via short cuts.

It would be nice if the creators put only the script on the website, preferably with marked fragments corrected by a human. It would then be possible to realistically assess to what extent the work of GPT-2 actually produced results in the form of a relatively coherent plot. What’s more, we could also assess to what extent the text itself is responsible for the surprisingly high level of the entire performance, and to what extent it was achieved by a really interesting arrangement and direction.

It is also worth adding here that the text came from the GPT-2 algorithms, i.e. a generator that is already a few years old. The newest GPT-3, which I wrote about recently in the context of the essay experiment, has multiple neural network connections and could do better. Another thing is that he also has more “mental” problems;), such as counseling a depressed person to commit suicide. Who knows, if his performance would not turn out even spookier.

When will the artificial director arrive?

Here it must also be clearly stated that the artists from the Prague theater really stood up to the task. The minimalist set design is very atmospheric, the acting performances are interesting, and you cannot say a bad word about music. All this is human added value that AI will not be able to replace for a long time.

Where to watch

Art, as befits an experiment of this type, can be watched online. After the performance, an over an hour-long discussion panel with the people behind the entire project is added. If you are interested, I am posting the entire broadcast below, you can also jump on the project website: https://www.theaitre.comwhere you will find more information about people and institutions taking part in this interesting adventure.

Source: AntyWeb by

*The article has been translated based on the content of AntyWeb by If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!